Настройки

Укажите год
-

Небесная энциклопедия

Космические корабли и станции, автоматические КА и методы их проектирования, бортовые комплексы управления, системы и средства жизнеобеспечения, особенности технологии производства ракетно-космических систем

Подробнее
-

Мониторинг СМИ

Мониторинг СМИ и социальных сетей. Сканирование интернета, новостных сайтов, специализированных контентных площадок на базе мессенджеров. Гибкие настройки фильтров и первоначальных источников.

Подробнее

Форма поиска

Поддерживает ввод нескольких поисковых фраз (по одной на строку). При поиске обеспечивает поддержку морфологии русского и английского языка
Ведите корректный номера.
Ведите корректный номера.
Ведите корректный номера.
Ведите корректный номера.
Укажите год
Укажите год
Применить Всего найдено 17189. Отображено 100.
30-10-2024 дата публикации

Система идентификации личности по изображению лица, сформированного сверточной искусственной нейронной сетью

Номер: RU2829289C1
Автор: Арлазаров Владимир Викторович (RU), Николаев Дмитрий Петрович (RU), Шешкус Александр Викторович (RU), Славин Олег Анатольевич (RU)
Принадлежит: Общество с ограниченной ответственностью "СМАРТ ЭНДЖИНС СЕРВИС" (RU)

Изобретение относится к системе идентификации личности по изображению лица, сформированного сверточной искусственной нейронной сетью. Техническим результатом является обеспечение идентификации личности по изображению лица. Система содержит модуль приема цифрового описания изображения лица личности с информационного выхода сверточной искусственной нейронной сети, модуль селекции адресов записей идентификационных данных личностей в базе данных сервера системы, модуль памяти, модуль приема адресов считывания данных, модуль сравнения данных, модуль приема идентификационных данных из сервера базы данных, модуль выдачи документальных данных личности и сервер базы данных. 4 ил.

Подробнее
Номер записи: 1
11-12-2024 дата публикации

Способ диагностики пронации пяточной кости

Номер: RU2831665C1
Автор: Виндерлих Марина Евгеньевна (RU)
Принадлежит: Федеральное государственное бюджетное образовательное учреждение высшего образования "Марийский государственный университет" (RU)

Изобретение относится к медицинской технике, а именно к способам ортопедической диагностики пронации пяточных костей стоп. Способ диагностики пронации пяточных костей у детей, в котором проводят клинический осмотр нижних конечностей. Далее делают фотографию дорсальной поверхности нижних конечностей в положении стоя на горизонтальной поверхности. Затем фотоизображение обрабатывают с помощью входящей в библиотеку PyTorch Keypoint RCNN модели машинного обучения путем выделения ключевых точек и линий дорсальной поверхности голеностопного сустава и стопы. Далее находят координаты точек, используя Keypoint RCNN модель, а обучение данной модели произведено с помощью машинного обучения. Данная модель натренирована на поиск 17 ключевых точек дорсальной поверхности нижних конечностей, а обучение произведено с помощью PyTorch и Scikitleam. Затем анализируют и рассчитывают углы пронации пяточных костей. Техническим результатом является доступность и повышение диагностической точности определения угла ...

Подробнее
Номер записи: 2
04-02-2025 дата публикации

СПОСОБ ПОСТОБРАБОТКИ СЕЙСМИЧЕСКИХ ДАННЫХ

Номер: RU2834175C1
Автор: Воцалевский Збигнев Владиславович (RU), Мифтахов Руслан Фанисович (RU), Шевырёв Никита Алексеевич (RU), Безухов Никита Сергеевич (RU), Ефремов Игорь Иванович (RU), Богоедов Даниил Александрович (RU), Базанов Андрей Константинович (RU)
Принадлежит: Общество с ограниченной ответственностью "ГРИДПОИНТ ДАЙНАМИКС" (RU)

Изобретение относится к области геофизических исследований и, в частности, к сейсмической интерпретации, является технологией постобработки сейсмических данных с использованием методов машинного обучения для улучшения их качества и повышения точности структурной интерпретации геологических структур. Изобретение может быть использовано для целей разведочной геофизики. Согласно изобретению осуществляют генерацию исходных синтетических сейсмических данных с широким диапазоном частот вейвлета, используемого для генерации. Затем добавляют в исходные синтетические сейсмические данные различные типы шумов, артефактов обработки структурных и тектонических объектов. Далее осуществляют обучение конволюционной нейронной сети на датасетах с шумами и без. После применяют обученную таким образом нейронную сеть для дальнейшей очистки реальных данных и повышения разрешающей способности сейсмической записи путём увеличения частотного диапазона реальных данных. Затем проводят анализ обработанных данных для ...

Подробнее
Номер записи: 3
23-06-2023 дата публикации

Convolutional neural network compression method and device combining dynamic pruning and conditional convolution

Номер: CN116306808A
Автор: GAO ZHIPENG, MO ZIJIA, RUI LANLAN, YANG YANG, ZHAO CHEN, YU XINLEI, FENG FUTENG
Принадлежит:

The invention provides a convolutional neural network compression method and device combining dynamic pruning and conditional convolution, and the method comprises the steps: inputting a training sample set into a convolutional neural network, and obtaining the input features and output features of each convolution layer in the convolutional neural network, calculating the channel significance similarity between the input features of each convolutional layer and the sample similarity between the output features of each convolutional layer, constructing a first total loss function, and carrying out the iteration of the convolutional neural network through employing the first total loss function, so as to obtain a first lightweight model and a plurality of second lightweight models, and when the predicted value of the first lightweight model does not reach a set threshold value, combining the combination coefficients of the convolutional layers of the second lightweight models and the convolutional ...

Подробнее
Номер записи: 4
23-06-2023 дата публикации

Polypeptide function prediction method and device

Номер: CN116312750A
Автор: LAI REN, RONG MINGQIANG, GU ZHIXIN, LI ZONGXU
Принадлежит:

The invention discloses a polypeptide function prediction method and device. The method comprises the steps of data set construction and data preprocessing; establishing a multi-label prediction model, wherein the multi-label prediction model is composed of a feature embedding module, a multi-scale convolutional neural network CNN module, a BiGRU module for processing a context-related sequence and a classification module; training the constructed network model by using the obtained data set, and improving model feature extraction and parameter prediction performance by using an optimization algorithm; and the performance of the multi-label prediction model is evaluated through five indexes of calculation precision, coverage rate, accuracy, absolute true value and absolute false value. According to the method, association between the tags is effectively utilized, the accuracy of polypeptide function prediction is improved, the method is high in operability and practicability, can be applied ...

Подробнее
Номер записи: 5
23-06-2023 дата публикации

Trusted graph neural network node classification method

Номер: CN116304906A
Автор: WANG BO, WANG WEI, GUO YINGLONG, XU ZHENYU, LIU YANBEI, LI ZHISHENG, JIA ZHIYANG, YANG MINGKAI
Принадлежит:

The invention discloses a credible graph neural network node classification method, and belongs to the technical field of graph semi-supervised classification, and the method comprises the following specific steps: inputting graph data, carrying out the feature learning and feature extraction of the graph data through a graph convolutional neural network, and obtaining the embedded representation of final graph node data; the embedded representation is input into a centroid module to be processed, and different types of centroids are obtained through the embedded representation with label nodes; inputting the embedded representation of the label-free node and the centroid of each category into an RBF module to obtain the distance between the label-free node and the centroid, and predicting the category of the label-free node; uncertainty loss and gradient penalty loss are added through an RBF module, cross entropy loss is combined to jointly serve as a target function, a loss function is ...

Подробнее
Номер записи: 6
23-06-2023 дата публикации

Auxiliary diagnosis system and method based on medical big data

Номер: CN116313061A
Автор: ZHANG XIULING
Принадлежит:

The invention provides an auxiliary diagnosis system and method based on medical big data, and relates to the technical field of medical data auxiliary diagnosis, and the method comprises the steps: obtaining medical data, carrying out the preprocessing of non-structural data in the medical data through a data preprocessing module, and generating a diagnosis recognition model through a model generation module, and training the generated diagnosis and recognition model through a model training module according to a pre-acquired training data set, and inputting the preprocessed medical data into the trained diagnosis and recognition model through a diagnosis and recognition module to realize auxiliary diagnosis and recognition. According to the auxiliary diagnosis system and method based on the medical big data, auxiliary diagnosis based on the medical big data can be achieved, use by doctors is facilitated, non-structural data in the medical data can be converted, and use by auxiliary diagnosis ...

Подробнее
Номер записи: 7
12-05-2023 дата публикации

Data enhancement and segmentation method and device for metal corrosion image

Номер: CN116109518A
Автор: SONG WEI, ZHU TIANCHENG, ZHENG TAO, ZHU SHIQIANG, SHI HAOLEI
Принадлежит:

The invention discloses a data enhancement and segmentation method and device for a metal corrosion image, and the method comprises the steps: obtaining a metal corrosion image, carrying out the segmentation of the metal corrosion image, obtaining a plurality of segmented sub-images, carrying out the enhancement operation of each sub-image, and recombining the segmented sub-images into a new image with the size consistent with the size of an original metal corrosion image, according to the metal corrosion segmentation method and system, the metal corrosion data set is constructed through the metal corrosion segmentation model, data expansion is carried out on the metal corrosion data set, the number of the images is increased, the diversity of the metal corrosion images is improved, and the metal corrosion segmentation method and system can be used for metal corrosion segmentation of the to-be-segmented image and the metal corrosion segmentation of the to-be-segmented image. Therefore, ...

Подробнее
Номер записи: 8
30-06-2023 дата публикации

Coal flow monitoring system and method based on laser scanner

Номер: CN116363193A
Автор: WU KUNSONG, GUO KAI, ZHENG LIN, HONG YUEBIN, LIU XIANCHUN, YANG YANG, SUN XINJIA
Принадлежит:

The invention discloses a coal flow monitoring system based on a laser scanner and a method thereof, which adopt an image processing technology and an artificial intelligence technology to determine the volume of transmitted coal through a three-dimensional model based on the transmitted coal in unit time. And carrying out convolution kernel-based spatial domain feature filtering on the monitoring image representing the apparent state of the coal through a convolutional neural network to capture explicit and implicit mode features in a local neighborhood in the monitoring image, so as to accurately estimate the density value of the coal. Therefore, the accurate mapping model of the apparent state characteristics of the coal and the coal density value is constructed, so that a more accurate coal flow value is obtained based on the density estimation value and the volume of the transmitted coal.

Подробнее
Номер записи: 9
23-06-2023 дата публикации

Image aesthetic description and positioning method and system based on multi-modal mutual attention

Номер: CN116310690A
Автор: GONG YONGSHUN, HUANG JIN, ZHANG XINXIN, SHI YANG, YIN YILONG
Принадлежит:

The invention provides an image aesthetics description and positioning method and system based on multi-modal mutual attention, and relates to the technical field of aesthetics quality evaluation.The method comprises the steps that an image, image comment data, image description data and various image data describing position information are obtained, and the image is preprocessed; obtaining text information and image information in the image data, inputting the text information and the image information into the neural network model, and extracting high-level semantic vector features of a text for the text information; for image information, extracting deep semantic features of an image through a convolutional neural network; and introducing an attention fusion module, fusing the high-level semantic vector features and the deep semantic features, and generating image description and image region positioning. Combination of text features and image features is fully considered, and subtle ...

Подробнее
Номер записи: 10
16-05-2023 дата публикации

Identity recognition method for position independence

Номер: CN116127431A
Автор: ZHANG XIANGGANG, ZENG JING
Принадлежит:

The invention discloses an identity recognition method for position independence, and belongs to the field of target recognition. According to the invention, the target identity is identified through the mobile phone motion data, and another path for target identity identification is opened up. The method comprises the following steps: firstly, acquiring inertial sensor data when a mobile phone is located at hands, waist, trouser pockets and upper arms, then training an identity recognition network according to the data and corresponding targets, and finally recognizing the target identity of the newly acquired data by adopting the identity recognition network. According to the invention, a new method for identity recognition is developed, and the recognition preparation rate reaches 87.8%.

Подробнее
Номер записи: 11
16-05-2023 дата публикации

Game role recommendation method and system based on computer vision

Номер: CN116116012A
Автор: LU QIN, BIAN HUICONG
Принадлежит:

The invention discloses a game role recommendation method and system based on computer vision. The method comprises the following steps: in response to a game role recommendation request instruction, performing screenshot storage on a game role to-be-selected interface, and performing head portrait clipping on a selected role in a screenshot to obtain a clipped role head portrait; performing role identification on the cut role head portrait to obtain a serial number of the role; determining a selected role list according to the serial numbers of the selected roles; and according to the selected role list and the round information of the selected roles, realizing role recommendation. When a player uses the recommendation system, the player does not need to additionally open a browser webpage to fill the role by himself/herself, and role recommendation can be completed only by clicking a button of the floating window of the recommendation system.

Подробнее
Номер записи: 12
16-05-2023 дата публикации

Plant disease identification method, equipment, storage medium and device

Номер: CN116129189A
Автор: TIE JUN, ZHENG LU, LONG JUANJUAN, WANG JIANGQING, LI ZIMAO, SHE WEI, XU SHENGZHOU, MA JIATING
Принадлежит:

The invention discloses a plant disease recognition method, equipment, a storage medium and a device, and the method comprises the steps: carrying out the disease feature recognition of a plant image obtained after data enhancement preprocessing through a plant disease recognition model of improved knowledge distillation, and determining a plant disease image according to a feature recognition result. The plant disease identification model for improving knowledge distillation is a model obtained by combining a soft label of a corresponding teacher model with a hard label of a target plant actual data set through a knowledge distillation method and then training a student model; and determining a disease type based on the plant disease image. According to the invention, disease feature recognition is carried out on the plant image through the plant disease recognition model based on the improved knowledge distillation, and the plant disease image is determined according to the feature recognition ...

Подробнее
Номер записи: 13
30-06-2023 дата публикации

Machine learning device, machine learning method, and machine learning program

Номер: CN116368498A
Автор: YANG YINCHENG, TAKEHARA HIDEKI, KIDA SHINGO
Принадлежит:

A weight storage unit (50) stores weights of a plurality of filters for feature detection of a job. A continuous learning unit (20) continuously learns weights of a plurality of filters for an input job. The filter processing unit (30) locks weights for a predetermined ratio of filters among the plurality of filters after learning one task so that the filters are not used in learning of another task, and initializes weights for the other filters so that the filters are used in learning of another task. A comparison unit (40) compares the weights of a plurality of filters after learning two or more tasks, extracts, as a shared filter between tasks, a duplicate filter having a weight similarity equal to or greater than a predetermined threshold value, and retains one of the duplicate filters as the shared filter. And initializing weights of filters other than the shared filter.

Подробнее
Номер записи: 14
02-05-2023 дата публикации

Image depth denoising method based on residual self-coding

Номер: CN116051408A
Автор: ZHANG JIE, LU MIAOXIN, HUANG WENXIAO, ZHANG HUANLONG, ZHANG JIANWEI, WANG FENGXIAN, LI LINWEI, QU GUANG
Принадлежит:

The invention provides an image depth denoising method based on residual error self-coding. The method comprises the following steps of: grading noise by taking a standard deviation of the noise as a grading basis; expanding the number of pictures in the data set through a data enhancement strategy to obtain a training set; constructing an image denoising model, wherein the image denoising model comprises a coding module, a dense residual module and a decoding module which are connected in sequence; respectively adding the Gaussian noise in each noise level range in the step 1 into the pictures of the training set to obtain noise-containing images, and inputting the noise-containing images into an image denoising model to respectively carry out learning training to obtain trained image denoising models of different noise levels; and during image denoising application, inputting a noisy picture into the trained image denoising model, and directly reconstructing and outputting the denoised ...

Подробнее
Номер записи: 15
30-05-2023 дата публикации

Faster R-cnn context mechanism optimization method based on multilayer feature fusion

Номер: CN116189027A
Автор: WAN JIUDI, PAN CHUNJIE, ZHANG QIANJIN
Принадлежит:

The invention relates to a Faster R-cnn context mechanism optimization method based on multilayer feature fusion, and belongs to the technical field of target recognition. According to the method, on the basis of extracting feature vectors of a pre-selection box by using HyperNet traditionally, a multi-layer feature fusion module improved on the basis of being different from a traditional mode is added into a model, and semantic information of different levels of an image is extracted; a traditional sampling mode is improved into anti-pooling, so that the chessboard problem is avoided; and then, fusing the obtained feature vector with a context attention mechanism to obtain a multi-information feature vector, and finally carrying out identification and classification. According to the method, the accuracy of the model for identifying static non-animal bodies can be improved.

Подробнее
Номер записи: 16
02-05-2023 дата публикации

Method and system for acquiring medical image segmentation model, electronic equipment and medium

Номер: CN116051579A
Автор: YANG XIONG, MAO YUFEI, SU SAISAI, LI JINGDONG
Принадлежит:

The invention provides a medical image segmentation model acquisition method and system, a vascular medical image segmentation method, electronic equipment and a medium. The medical image segmentation model acquisition method comprises the following steps: acquiring a first training sample of a medical image segmentation model, wherein the first training sample comprises a first medical training image and a first label image corresponding to the first medical training image; shape constraint information of a target area in the first medical training image is acquired; and according to the first medical training image and the shape constraint information of the target area, training the initialized first neural network model until a first preset training ending condition is satisfied, and obtaining the medical image segmentation model. According to the invention, the segmentation efficiency and the segmentation precision of the medical image and the stability of the segmentation result can ...

Подробнее
Номер записи: 17
23-06-2023 дата публикации

Financial contract term extraction method and system based on incremental learning

Номер: CN116306606A
Автор: LEI QI, WU YEWEI, LI YI, LUO XING, KONG GUANQING
Принадлежит:

The invention discloses a financial contract clause extraction method and system based on incremental learning, and the method comprises the steps: obtaining clause data, and carrying out the preprocessing and standardization of the clause data; constructing a batch data generator, and generating a neural network input tensor; building a clause extraction model based on a one-dimensional CNN and a bidirectional LSTM network; inputting the clause training data into the clause extraction model for training, and performing incremental training on new data; and sending the clause test data into the trained clause extraction model for prediction, obtaining a sequence label of the paragraph, and extracting to obtain a predicted clause. The method has the advantages that class incremental learning is carried out in a knowledge distillation mode, the prediction effect of an old model can be reserved, and the recognition capability of a new class can be obtained. Meanwhile, the speed and lightweight ...

Подробнее
Номер записи: 18
02-06-2023 дата публикации

Super-division network, method, device and equipment based on novel re-parameterization structure

Номер: CN116205284A
Автор: GUO JIANJUN
Принадлежит:

The invention provides a super-division network, method, device and equipment based on a novel re-parameterization structure. The network comprises an input image module, a re-parameterization convolution module, a recombination pixel module and an output super-division image module which are connected in sequence. The input image module is used for inputting pictures; the re-parameter convolution module is used for super-division network reasoning of the picture and comprises a training time structure and a reasoning time structure; the structure during reasoning comprises a 3 * 3 convolution unit, and the structure during training comprises a re-parameter convolution unit 1 to a re-parameter convolution unit 3 which are connected in sequence; wherein the input channel of the re-parametric convolution 1 unit is 1, and the output channel is c; the input channel of the re-parameter convolution 3 is c, and the output channel of the re-parameter convolution 3 is scae2; the pixel recombination ...

Подробнее
Номер записи: 19
23-06-2023 дата публикации

Pet motion state identification method and system, medium and electronic equipment

Номер: CN116304826A
Автор: DONG YULONG, CHEN HONGSHANG, ZHANG WEI, ZHANG XIAOLIN
Принадлежит:

The invention provides a pet motion state identification method and system, a medium and electronic equipment, and the method comprises the steps: obtaining the motion data of a sample pet, and the motion data comprises acceleration information; calibrating the motion data according to the motion type to obtain calibration data; performing time-frequency domain conjoint analysis on the calibration data to obtain a time-frequency domain spectrogram corresponding to each motion type as a sample, and performing sample training to obtain a trained recognition model; and based on the trained identification model, according to the time domain and/or frequency domain spectrogram corresponding to the acceleration information of the pet to be identified, identifying to obtain a motion state corresponding to the pet. According to the method, the convolutional neural network model obtained by training the time-frequency domain spectrogram corresponding to the combination of various acceleration information ...

Подробнее
Номер записи: 20
23-06-2023 дата публикации

Fusion feature-based skin acne detection method and system

Номер: CN116309357A
Автор: SHENG YU, LIU CAN, ZHU YUNYUN
Принадлежит:

The invention discloses a fusion feature-based skin acne detection method and system, and the method comprises the steps: collecting a face visible light image containing acnes, marking the acnes, collecting a face infrared image containing acnes, marking the temperature distribution of the acnes, and constructing a training set; using the training set to train a pre-constructed acne classification model based on an improved Faster RCNN network; collecting a visible light image and an infrared image of the face of a patient, inputting the images into the acne classification model obtained through training, and obtaining face acne position distribution, area and temperature distribution results. According to the face acne classification method, the morphological features are collected, the skin temperature features are also collected, the deep learning model is trained through the fused features, and the face acne classification result is obtained based on the trained deep learning model ...

Подробнее
Номер записи: 21
30-06-2023 дата публикации

Polarized SAR image registration method based on attention mechanism

Номер: CN116363175A
Автор: XIANG DELIANG, DING HUAIYUE, CHENG JIANDA, HU CANBIN, SUN XIAOKUN
Принадлежит:

The embodiment of the invention discloses a polarized SAR image registration method based on an attention mechanism, and the method comprises the steps: carrying out the polarization whitening filtering of two different polarized SAR images, and obtaining the filtered images; adaptive threshold constraint, morphological corrosion and non-maximum suppression operations are performed on the two filtered images, and key points are determined; inputting data blocks with set sizes around each key point in the two polarimetric SAR images into two branches of a trained twinborn attention network for depth feature extraction to obtain a depth feature descriptor of each key point; matching the depth feature descriptors of the key points in the two polarimetric SAR images by using a violent matching method and a random sampling consensus algorithm to obtain a perspective transformation matrix; and the polarimetric SAR image is converted into a registered image by using the perspective transformation ...

Подробнее
Номер записи: 22
02-06-2023 дата публикации

Method and device for acquiring AO chartlet

Номер: CN116206040A
Автор: JI DAOMING, ZHUANG XINRUI, LIAO JINGTANG, XU ZIYA, LIU QIAN
Принадлежит:

The invention discloses a method and device for obtaining an AO map, relates to the field of image processing, and aims to improve the efficiency of obtaining a high-quality AO map. According to the scheme, the method comprises the steps that according to two-dimensional images of the 3D object at N visual angles and camera parameters of the N visual angles, a geometric information layer of the 3D object at each visual angle in the N visual angles is obtained, and the geometric information layer comprises a normal graph; n is greater than or equal to 2; inputting the geometric information layer of each visual angle into a trained neural network to obtain a single-visual-angle AO layer of each visual angle; performing UV parameterization according to the two-dimensional images of the 3D object at the N visual angles to obtain UV parameters of the grid model of the 3D object; and according to the obtained UV parameters, fusing the single-view-angle AO layer of each view angle to obtain an ...

Подробнее
Номер записи: 23
02-06-2023 дата публикации

Method and system for judging potential customer intention level based on enterprise WeChat

Номер: CN116204635A
Автор: CAO HUAIJIE
Принадлежит:

The invention discloses a potential customer intention level discrimination method and system based on enterprise WeChat, and the method comprises the steps: obtaining an enterprise WeChat session record, carrying out the decryption and data analysis processing, and generating target customer text data in combination with user basic information data; the original target client text is subjected to word segmentation and converted into a word sequence, the word sequence is converted into a word number sequence, words in each word list have unique numbers, and each element in the word number sequence is converted into a word vector form to obtain a word vector text sequence; inputting the obtained word vector text sequence into a constructed intelligent intention recognition model, outputting an intention classification probability of a target customer text, and obtaining a customer intention level discrimination result; and generating target data and pushing the target data according to the ...

Подробнее
Номер записи: 24
16-05-2023 дата публикации

Three-dimensional point cloud generation method and device of power transmission line and storage medium

Номер: CN116129039A
Автор: DU WEI, WANG JIAYING, YANG GUOZHU, ZHANG JIALIN, WU JIANXIONG, LI YURONG
Принадлежит:

The invention provides a three-dimensional point cloud generation method and device for a power transmission line, and a storage medium. The method comprises the steps of collecting a to-be-processed video stream of the power transmission line through shooting equipment on a preset aircraft; performing coordinate system conversion processing on the to-be-processed video stream to obtain three-dimensional coordinate data; performing three-dimensional live-action modeling processing on the three-dimensional coordinate data to obtain a power transmission line three-dimensional point cloud model corresponding to the power transmission line; and performing optimization processing on the three-dimensional point cloud model of the power transmission line to obtain an optimized three-dimensional point cloud model of the power transmission line. The to-be-processed video stream of the power transmission line is acquired, coordinate system conversion and three-dimensional live-action modeling are ...

Подробнее
Номер записи: 25
30-06-2023 дата публикации

Parallel deep convolutional neural network optimization method based on Spark and AMPSO

Номер: CN116362298A
Автор: MAO YIMIN, DAI JINGGUO, LUO QUANCHENG
Принадлежит:

The invention proposes a parallel deep convolutional neural network optimization method based on Spark and AMPSO, and the method comprises the following steps: S1, carrying out the pre-training of a DCNN model on a main node through employing an original data set, carrying out the pruning of the pre-trained DCNN model through employing an FP-FIS strategy, obtaining a compressed DCNN model, and distributing the compressed DCNN model to each calculation node; s2, dividing the original data set into data sets with the same size, and distributing the data sets to each computing node; s3, calling a map function and an MPT-AMPSO strategy on each mapper node, and performing parallel training on the compressed DCNN model by using the data set distributed on each node; and S4, calling a reduce ByKey function and a DLBNP strategy on each reduce node to carry out parallel updating on the compressed DCNN model parameters, and finally obtaining a parameter updating result. According to the invention ...

Подробнее
Номер записи: 26
30-06-2023 дата публикации

Expert data amplification method and device based on random network prediction error

Номер: CN116352700A
Автор: ZHANG ZONGCHANG, YU YANG, ZHOU ZHIHUA, HAN FUGUANG
Принадлежит:

The invention discloses an expert data amplification method and device based on a random network prediction error, and the method comprises the steps: training a prediction network through employing a random network distillation method on the expert data of a mechanical arm, and discriminating the data similar to the expert data of the mechanical arm through employing the prediction error of the prediction network; the mechanical arm carries out a series of operation sampling under the scene of the target task to obtain data, the part, similar to the expert data of the mechanical arm, in the sampled data is screened according to the prediction error, and then the data obtained through screening is used for amplifying the original expert data set of the mechanical arm. According to the method, under the condition that interaction with a mechanical arm expert is not carried out, data are sampled through interaction between the mechanical arm and the environment in the scene of the target ...

Подробнее
Номер записи: 27
16-05-2023 дата публикации

Three-dimensional medical image segmentation method and system based on long and short term memory self-attention model

Номер: CN116129107A
Автор: ZHANG XUMING, WEN MINGWEI, ZHOU QUAN
Принадлежит:

The invention belongs to the field of image segmentation in medical image processing, and discloses a three-dimensional medical image segmentation method and system based on a long and short term memory self-attention model.The three-dimensional medical image segmentation method can be used for segmenting a target area image in a medical image and comprises the following steps that (1) a training set sample is established; (2) processing a to-be-segmented original three-dimensional medical image to obtain a to-be-segmented sample; (3) constructing and training a three-dimensional medical image segmentation network based on long and short term memory self-attention; and (4) inputting a to-be-segmented sample into the network, and outputting a segmentation result of a target region in the to-be-segmented sample. According to the method, the CNN and the Transform are integrated to obtain a new model for real-time accurate segmentation of the target area (such as a tumor) in the three-dimensional ...

Подробнее
Номер записи: 28
23-06-2023 дата публикации

Transformer state prediction method based on multivariate perception and feature extraction

Номер: CN116304822A
Автор: ZHANG FAN, CHEN MEIQIN, JI SHENGCHANG, ZHANG YUKUN, LI JISHENG
Принадлежит:

The invention discloses a transformer state prediction method based on multivariate perception and feature extraction, and the method comprises the steps: collecting vibration signals of the surface of an oil tank of a transformer in a normal state and a fault state; the method comprises the following steps: preprocessing a vibration signal and visualizing the vibration signal into a gray feature image, storing a part of the gray feature image as a fault feature image library, using the remaining gray feature image as a model data set, and extracting features based on the model data set in combination with algorithms of a convolutional neural network and a long-short term memory network; a 3D convolutional neural network is used to encode and decode an original gray feature image, a feature image of a vibration signal in a fault state is used as a reference, a matched feature image decoded and output by a decoder is selected from all time steps, and the spatial similarity between the feature ...

Подробнее
Номер записи: 29
14-07-2023 дата публикации

Semantic communication method based on deep neural network

Номер: CN116436567A
Автор: ZHANG HUA, CAO ZIJIAN, WANG HAOTIAN
Принадлежит:

The invention discloses a semantic communication method based on a deep neural network. The semantic communication method comprises the following steps: inputting statement information into a semantic encoder to output semantic information; further encoding the semantic information into channel information by using a channel encoder; the channel information is input into a channel feature extractor to extract channel features; the channel features and the channel information are input into a channel decoder at the same time, and the channel features and the channel information are combined through bilinear transformation and then decoded to recover estimated semantic information; and recovering the estimated statement information by using a semantic decoder. Simulation results show that the semantic communication network provided by the invention is superior to a traditional reference and an existing semantic communication system in the aspect of semantic transmission precision under the ...

Подробнее
Номер записи: 30
14-07-2023 дата публикации

Image processing method and device for decoupling, denoising and detail enhancement

Номер: CN116433488A
Автор: HUANG JIE
Принадлежит:

The invention relates to the field of image processing, and discloses an image processing method and device for decoupling, denoising and detail enhancement. The method comprises the steps of inputting a to-be-processed image into a trained image processing model, so that a backbone network of the image processing model performs response value increasing and pixel value reducing processing on noise in the to-be-processed image to obtain a denoised image, and increasing pixel values and gradients of textures and deformation edges in the denoised image to obtain a processed image; according to the image processing method, the denoising capability and the detail enhancement capability can be decoupled, the denoising capability and the detail enhancement capability are respectively carried out on the low-resolution image, the denoising capability and the detail enhancement capability of the to-be-processed image are improved, and the denoising capability and the detail enhancement capability ...

Подробнее
Номер записи: 31
14-07-2023 дата публикации

Driver and conductor risk identification method and device

Номер: CN116434200A
Автор: SONG JIANHUA, ZHANG YAN, WANG SHIHUI, WANG YESHUAI, HE PENG, MA CHUANXIANG, HUANG CHEN, LI YAMIN, LYU SHUNYING, YANG CHAO
Принадлежит:

The invention relates to a driver and conductor risk identification method and device, and the method comprises the steps: inputting collected real-time pressure data into an emotion identification model, and determining a current emotion state; if the current emotional state is an emotional state of a preset type, continuously obtaining multiple groups of pressure data in a preset time period, inputting the pressure data into an emotion recognition model, and obtaining a plurality of corresponding target emotional states; according to the multiple target emotional states, determining an early warning counting result in a preset time period, and if the early warning counting result reaches a preset counting threshold, generating a risk early warning result; wherein the emotion recognition model is trained in the following mode: inputting sample pressure data with emotion label information into an initial recognition model for training until the recognition success rate of the trained recognition ...

Подробнее
Номер записи: 32
22-08-2023 дата публикации

Traffic jam prediction method based on tense graph convolutional neural network

Номер: CN116631200A
Автор: ZHANG HAO, ZHANG GE, HUA QIFAN, DONG KAILONG, GAO SHANGBING, LIANG KUN, KONG DECAI, ZHOU GUILIANG, ZHU HONGLAN
Принадлежит:

The invention discloses a traffic jam prediction method based on a tense graph convolutional neural network. The method comprises the steps of obtaining a data set image and performing preprocessing; a T-GCN-based prediction model is constructed and trained, the GCN maps the spatial features and relationships of the traffic flow between the observation stations into a graph, and the output of the GCN is input into a GRU model; the weight of each hidden state is calculated by using multilayer perception through a Softmax function, and congestion prediction is performed on a data set by using a trained T-GCN-based prediction module; and outputting a prediction result by using a full connection layer, and comparing the congestion prediction result. According to the invention, images of the road condition of the day are shot, the congestion condition of the current road is directly analyzed and the congestion condition of the subsequent road section is predicted in a graph convolution mode, ...

Подробнее
Номер записи: 33
22-08-2023 дата публикации

SAR vehicle target identification method based on improved PVT

Номер: CN116626637A
Автор: WANG ZHAOCHENG, ZHANG GUANG, WANG RUONAN, ZHAO SHUO, LIU LU, SHAO XUEYAN
Принадлежит:

The invention relates to an improved PVT-based SAR vehicle target identification method, which comprises the steps of firstly obtaining a data set and performing data expansion, secondly constructing a PVT target identification model based on an existing transform network structure, then improving the PVT target identification model to obtain a final target identification model, and finally training the target identification model to obtain a target identification result. And optimizing the feature extraction network according to the loss to obtain an optimized target recognition model. The SAR vehicle target recognition method has the advantages of high calculation efficiency and high accuracy, is suitable for SAR vehicle target recognition, solves the problem of small receptive field of a deep neural network during gesture image processing, and can obtain global information in a large picture.

Подробнее
Номер записи: 34
22-08-2023 дата публикации

Power terminal identification method based on CEEMDAN parameter increment cooperating with attention moving convolution

Номер: CN116628460A
Автор: YIN LINFEI, YAO XIAORUI
Принадлежит:

The invention provides a power terminal identification method based on CEEMDAN parameter increment and attention moving convolution. According to the method, a fully adaptive noise set empirical mode decomposition method, parameter knowledge guidance, incremental learning, a collaborative attention mechanism and a moving convolutional neural network are combined and applied to power terminal identification. The method comprises the basic steps of power consumption data acquisition, construction of a power terminal graph matrix under a completely adaptive noise set empirical mode decomposition method, training of a collaborative attention mechanism and a mobile convolutional neural network, identification under the guidance of parameter knowledge and incremental processing of unlearned power terminals. Incremental processing includes self-selection incremental characterization learning and learning rate control. According to the method, the modal aliasing problem can be effectively solved ...

Подробнее
Номер записи: 35
08-08-2023 дата публикации

Radio frequency fingerprint identification method based on semi-supervised prototype network

Номер: CN116563897A
Автор: LIN DI, YUAN MAOSEN, HU SU, WU WEIWEI, YANG GANG, MA SHANG, JIN CHUANXUE, YANG DIAN
Принадлежит:

The invention provides a radio frequency fingerprint identification method based on a semi-supervised prototype network, and belongs to the technical field of Internet of Things security information. Radiation source equipment signals of all classes cannot be contained in a limited training set, and samples of unknown classes may appear in a test set. However, when the situation occurs in a real scene, a trained closed set model does not have the capability of identifying unknown classes when identifying unknown signals. Especially in some special environments, for example, an unmanned aerial vehicle pretends to be authenticated equipment, identification errors may occur when radio frequency fingerprint identification is performed without obtaining a signal of the equipment. According to a radio frequency fingerprint identification method based on a semi-supervised prototype network, a convolutional layer is combined with an attention mechanism to carry out effective feature extraction, ...

Подробнее
Номер записи: 36
15-08-2023 дата публикации

Human track and trunk activity identification method and system

Номер: CN116597521A
Автор: LIANG QUANQUAN, ZHANG KUN, WU QINGZHE, WANG WENHUA, YU ZHIJIE
Принадлежит:

The invention belongs to the technical field of wireless sensing and action recognition, and discloses a human track and trunk activity recognition method and system. The method comprises the following steps: acquiring channel state information (CSI) data by using a transceiver carrying a Ubuntu system and a CSI-TOOL, enabling a receiving end to have three omnidirectional antennas, and performing quotient on the two antennas to obtain a basic signal of a CSI quotient model; for the denoising of a CSI quotient value, a Savitzky-Golay filter is selected, and the width and height of a peak value in a signal waveform can be reserved, so that the characteristic of a CSI quotient signal is ensured not to deviate; removing high-frequency noise by using discrete wavelet transform; a Doppler spectrogram and a CSI (Channel State Information) quotient pseudo-color graph are respectively generated aiming at the motion trail and trunk activity of a human, and action recognition and classification are ...

Подробнее
Номер записи: 37
08-08-2023 дата публикации

Automobile parking space recognition system and method based on convolutional neural network

Номер: CN116564126A
Автор: LI ZHUOTONG, LIU CHAOYANG, ZHANG JIA, WU MINGZHE
Принадлежит:

The invention relates to an automobile parking space recognition system and method based on a convolutional neural network, in particular to the technical field of automobiles, and the system comprises an acquisition module which is used for carrying out the real-time acquisition of images around an automobile through a camera; the acquisition module is used for acquiring images, collected in real time, around the vehicle; the identification module is used for identifying the parking space state around the vehicle; the pushing module is used for pushing the parking state of the target parking space to an automobile central control screen and interacting with a user through the automobile central control screen; the control module is used for controlling the vehicle to be automatically parked in the target parking space; and the feedback module is used for feeding back the specified parking space to the control module and the pushing module until the parking state of the parking space specified ...

Подробнее
Номер записи: 38
04-08-2023 дата публикации

Information extraction model acquisition method and system

Номер: CN116542250A
Автор: CHAI YIXUAN, WU HAIPANG, CHEN MING
Принадлежит:

The invention relates to the technical field of information, in particular to an information extraction model obtaining method and system, and the method comprises the steps: obtaining a confidence model and an integrated model comprising a plurality of auxiliary information extraction models; obtaining a candidate text set, and determining a candidate label of each candidate text in the candidate text set through the integration model; the candidate labels are labels, meeting preset conditions, of the candidate texts obtained through the integrated model; determining the confidence degree of the candidate labels through a confidence degree model, determining target labels meeting the confidence degree requirement in the candidate labels, and further obtaining a target annotation text set; the target label text set comprises candidate texts with target labels; and training a target information extraction model based on the text data set to obtain a trained target information extraction ...

Подробнее
Номер записи: 39
04-08-2023 дата публикации

Visual question and answer method and system based on region representation and visual representation

Номер: CN116542995A
Автор: XU HAO, GAO LING, SHENG NAN, SHI LIDA, ZHANG HONGDA
Принадлежит:

The invention discloses a visual question and answer method and system based on region representation and visual representation, and the method comprises the steps: constructing an image region feature map for each image to be subjected to question and answer based on image region features learned by a Faster R-CNN model; generating neighbor enhanced region representation based on the image region feature map; guiding neighbor enhanced region representation based on problem features learned by a GRU model to obtain final visual features; and obtaining a predicted answer based on the region representation, the visual features and the question features, and completing visual question answering. According to the method, a region feature map is constructed for each image to represent association and features among all regions in the image; meanwhile, on the basis of a random walking method with restart, neighborhood enhancement region representation is learned in the region feature map. In ...

Подробнее
Номер записи: 40
08-08-2023 дата публикации

Motion capture missing data recovery method combining graph neural network and bone length constraint

Номер: CN116563952A
Автор: ZHENG CHUANQIN, HONG YANKUN
Принадлежит:

The invention discloses a motion capture missing data recovery method in combination with a graph neural network and bone length constraint. The method comprises the following steps: S1, constructing a skeleton graph taking joints as graph nodes and taking natural connection of a human body structure as edges; s2, based on the graph neural network of the skeleton graph, an encoder takes the joint coordinate vector of the graph node as input, performs multilayer space graph convolution operation on the input data to generate a higher-level feature graph, and reconstructs the output of the higher-level feature graph by using a full-connection neural network; s3, dividing the human skeleton structure into a plurality of bone block chains, and forming a bone length constraint by excavating a bone length constraint relationship between nodes in the bone block chains; s4, based on a mark selection algorithm of skeleton length constraint, recovering missing point coordinates more conforming to ...

Подробнее
Номер записи: 41
25-07-2023 дата публикации

New energy automobile charging circuit board alarm system

Номер: CN116476690A
Автор: LIU LING, HE XIAOHU
Принадлежит:

The invention discloses a new energy automobile charging circuit board alarm system. According to the method, the CNN model and the BiLSTM model are fused to construct the model for predicting the charging data of the electric vehicle, the defect that a traditional LSTM model only considers a single historical factor is overcome, and the prediction precision of the model is improved. On the basis, the prediction data passes through a sliding window, the influence of wrong charging data on residual error change in the data transmission process is eliminated, a reasonable pre-alarm threshold value is set, and the charging state of the electric vehicle is monitored in real time. Experiments prove that the method can effectively send out a pre-alarm signal, take corresponding protection measures and reduce the car burning risk in the charging process of the electric car, so that the alarm accuracy is improved, meanwhile, an alarm module in the multi-stage alarm module is matched, an alarm can ...

Подробнее
Номер записи: 42
01-08-2023 дата публикации

Light field image super-resolution model based on generative adversarial network and training method thereof

Номер: CN116523757A
Автор: LIANG ZHIQING, XIE RUYUN, LI FENGXU, LONG YUXIN, ZHANG HONGYU, ZHANG HUAIYUAN, YI TENGDA, ZHANG HONGBO, LIU ZIJI, ZHENG XING
Принадлежит:

The light field image super-resolution model based on the generative adversarial network comprises a generative network and a discrimination network, and the generative network comprises an initial feature extraction module, a feature aggregation module and an up-sampling module; the initial feature extraction module comprises a three-dimensional convolutional layer and a plurality of deepened convolutional layers which are connected in sequence; the feature aggregation module is composed of a spatial feature aggregation module and an angle feature aggregation module, and the up-sampling module comprises a first convolution layer, a sub-pixel convolution layer, an activation layer and a second convolution layer which are connected in sequence; and the discrimination network is used for discriminating the difference between the generated image of the generative network and the real image and assisting the generative network model in training. According to the method, the image generation ...

Подробнее
Номер записи: 43
01-08-2023 дата публикации

Early diagnosis method for abnormity of power module of DCS (Distributed Control System) device

Номер: CN116520182A
Автор: CAO WUZHONG, XU WENCAI
Принадлежит:

A DCS device power supply module abnormity early diagnosis method comprises the following steps: obtaining an output ripple signal generated by a monitored power supply module in a preset time period, and load values of the monitored power supply module at a plurality of preset time points in the preset time period; an artificial intelligence technology based on deep learning is adopted, a deep learning technology is utilized, features are extracted from an output ripple signal and a load value of a switching power supply, and fault diagnosis of a monitored power supply module is realized by utilizing classification processing. Therefore, the switching power supply can be effectively monitored, the abnormal state of the power supply can be found in time, and the reliability and the safety of the power supply are improved.

Подробнее
Номер записи: 44
01-08-2023 дата публикации

Shielding target re-identification method based on contrast learning attention and correlation learning

Номер: CN116524536A
Автор: WANG PENG, GAO LIYING, JIAO BINGLIANG, CHEN HONGYU
Принадлежит:

The invention discloses an occluded target re-identification method based on comparative learning attention and correlation learning, and belongs to the technical field of mode identification. According to the method, a fine-grained correlation learning module is provided, the correlation between pixel-level features of to-be-compared images is calculated, then the pixel-level correlation is compressed into image-level similarity step by step through 4D convolution, the fine-grained pedestrian correlation is learned under the guidance of image matching loss, and the pedestrian correlation is obtained. The problems of information asymmetry and spatial dislocation between images are solved, and the accuracy of shielded pedestrian re-identification is further improved.

Подробнее
Номер записи: 45
22-08-2023 дата публикации

Microseismic event identification method and system based on multi-modal neural network

Номер: CN116626753A
Автор: ZHANG YAN, ZHANG YONGXUE, LIU XIAOQIU, DONG CHUNFENG, ZHANG CHI, ZHANG YIMING
Принадлежит:

The invention discloses a microseism event identification method and system based on a multi-modal neural network, and the method comprises the following steps: obtaining a time domain sample and an S domain sample, and constructing a training data set based on the time domain sample and the S domain sample; constructing a deep learning network model G, and setting a model loss function; training the deep learning network model G based on the training data set to obtain an identification model; and carrying out microseism event identification by using the identification model. According to the method, the deep learning convolutional neural network structure of the time domain mode and the S domain mode of the microseismic data is combined, the multi-mode information of the microseismic signals is fully excavated, the accuracy of the model is improved, the method can still have good performance under strong noise interference, events can still be well recognized, and the method is suitable ...

Подробнее
Номер записи: 46
15-08-2023 дата публикации

Permanent magnet synchronous motor rotor and forming detection method thereof

Номер: CN116593890A
Автор: SHAO MINGYUAN, ZHU GUOLIANG, WANG TAO, TANG ZHANGJUN, MATSUO SHIGERU, JIN BO, LI YINGJIE, BIAN SHUJUN, FEI LIMING, CHEN YU
Принадлежит:

The invention discloses a permanent magnet synchronous motor rotor and a forming detection method thereof. The method comprises the following steps: firstly, acquiring a detection image of a detected permanent magnet synchronous motor rotor, carrying out fan-shaped equal division on the detection image to obtain a plurality of fan-shaped region images, then, enabling the plurality of fan-shaped region images to respectively pass through a convolutional neural network model serving as a filter to obtain a plurality of fan-shaped region feature vectors, and then, obtaining a plurality of fan-shaped region feature vectors; the cosine similarity between any two of the plurality of fan-shaped region feature vectors is calculated to obtain a classification feature vector composed of a plurality of cosine similarities, then manifold surface optimization is carried out on the classification feature vector to obtain an optimized classification feature vector, and finally, manifold surface optimization ...

Подробнее
Номер записи: 47
22-08-2023 дата публикации

Fingerprint classification screening system based on convolutional neural network

Номер: CN116631013A
Автор: YANG CHEN, MA KUN, MENG ZHI
Принадлежит:

The invention discloses a fingerprint classification screening system based on a convolutional neural network, and belongs to the technical field of image recognition. The fingerprint classification and screening system comprises the steps of building a convolutional neural network, training a model with fingerprint classification labels, building a fingerprint classification database and building a fingerprint screening platform. According to the method, tensorflow is used for building an AlexNet network model, a large number of fingerprint images with classification labels are input to carry out network training, and the trained AlexNet network model can carry out reliable classification on fingerprints; the fingerprint screening platform established based on the trained AlexNet network has the main functions of inputting a fingerprint image, performing enhancement processing, refining processing and classification processing, extracting feature points, judging whether the fingerprint ...

Подробнее
Номер записи: 48
22-08-2023 дата публикации

3D human body posture estimation method based on complementary enhancement of key points and grid vertexes

Номер: CN116631064A
Автор: QIAN LIFEI, LUO LING, JIANG ZONGBO, ZHONG YUANHONG, XU QIANFENG, HUANG ZHIYONG, HAN SHU, ZHONG DAIDI, ZHOU QING
Принадлежит:

The invention particularly relates to a 3D human body posture estimation method based on complementary enhancement of key points and grid vertexes. The method comprises the steps that SMPL model parameters of all video frames are output through a regression prediction model; the processing step of the regression prediction model comprises the following steps: extracting a key point heat map, shallow layer features and final layer features through a backbone network; sequentially converting the key point heat map into 2D key point coordinates and 2D key point features, and extracting spatial-temporal features to generate 3D key point coordinates; generating a partial 3D vertex heat map through multi-scale feature processing, and converting the partial 3D vertex heat map into 3D vertex coordinates; key point-vertex embedded features are generated, and SMPL model parameters are output through an SMPL parameter regression module; and constructing an SMPL model based on the SMPL model parameter ...

Подробнее
Номер записи: 49
28-07-2023 дата публикации

Power transmission channel management and control device and method based on binocular vision and laser point cloud prior

Номер: CN116503345A
Автор: LI QINGWU, CHEN JUNFENG, DING CHENGLONG, ZHOU YAQIN, MA YUNPENG, LI XIAO, ZOU YANG, YANG XIAOWEN
Принадлежит:

The invention relates to the technical field of power transmission, and discloses a power transmission channel management and control device and method based on binocular vision and laser point cloud prior. Comprising a left eye camera, a right eye camera, a first fixing frame, a second fixing frame, a third fixing frame, a horizontal rotating shaft, a vertical rotating shaft, a rotary controllable holder and a microprocessor, the left eye camera and the right eye camera are symmetrically arranged and are respectively arranged on the second fixing frame through the first fixing frame; two ends of the second fixing frame are respectively connected with one end of a third fixing frame, and the other end of the third fixing frame is fixed on the rotary controllable holder; the horizontal rotating shaft is arranged on the second fixing frame, and the second fixing frame is connected with the two third fixing frames through the horizontal rotating shaft; the microprocessor is arranged on the ...

Подробнее
Номер записи: 50
09-06-2023 дата публикации

Digital face driving method

Номер: CN116246321A
Автор: SONG ZHEN, BAO YIHUA, CHEN JUNYU, LIU CHAO
Принадлежит:

The invention relates to the technical field of data processing, and discloses a digital face driving method, which comprises the following steps of preparing information of a person A and a person B in advance, collecting information data of the person A and the person B, and driving the digital person B by using BlendShape data. According to the digital face driving method, the face changing neural network is trained on the big data set, so that the original expression can still be kept after face changing. In this way, the BlendShape parameter extracted from the face of the B is used for driving the digital human of the B, and compared with an existing processing mode, the method is more natural and real. DeepFace Lab and FaceSwap can be used for the deep face changing of the DeepFake artificial intelligence, and other neural networks can also be used for face changing. Facial expression capture can but not limited to use of a facial expression capture helmet, and can use modes of a ...

Подробнее
Номер записи: 51
13-06-2023 дата публикации

Project recommendation method and device, equipment and storage medium

Номер: CN116257641A
Автор: YUAN DINGQIANG
Принадлежит:

The invention relates to artificial intelligence and digital medical treatment, and provides a project recommendation method and device, equipment and a storage medium. The method comprises the following steps: generating an interaction matrix based on interaction information of an initial user to an initial project, generating a user walk sequence and a project walk sequence based on the interaction matrix, inputting the user walk sequence into a user representation model to obtain a user representation vector, processing the project walk sequence based on the project representation model to obtain a project representation vector, and displaying the project representation vector. Generating a user context vector based on the multi-dimensional information of the target user, generating an item context vector based on the multi-dimensional feature of each initial item, and generating an interested item vector based on the item representation vector and the item context vector, the target ...

Подробнее
Номер записи: 52
15-08-2023 дата публикации

Unsupervised semantic segmentation method based on fine-grained feature grouping

Номер: CN116597151A
Автор: YU XIAODAN
Принадлежит:

The invention discloses an unsupervised semantic segmentation method based on fine-grained feature grouping. On one hand, features extracted by a convolutional neural network are extracted by using a capsule network, important information is reserved, interference of irrelevant information is reduced, and the overall segmentation effect is improved; and on the other hand, pre-segmentation is carried out on the picture by using super-pixel segmentation, and subsequent pixel features are guided to carry out fine-grained grouping by using the pre-segmented region boundaries, so that the problem of poor segmentation when the boundaries are fuzzy is solved. According to the method, the subsequent advanced semantic information is subjected to fine-grained feature grouping according to the segmentation boundary to serve as the input of the Capsule layer, so that the number of capsules can be reduced; central difference is carried out in each super-pixel block, so that detail information of segmentation ...

Подробнее
Номер записи: 53
26-05-2023 дата публикации

Anti-collision early warning system and method

Номер: CN116168565A
Автор: LAI JINGWEN, MA TIANLI, XIAO SHUNLI, PAN TAO, CHEN GUANXUN
Принадлежит:

The invention discloses an anti-collision early warning system and method.The system comprises laser radars, Internet of Vehicles equipment and a control terminal, the laser radars arranged on the roadside are about 6-9 meters away from the road surface and used for collecting point cloud data and transmitting the point cloud data to the control terminal, and the control terminal sends the point cloud data to the Internet of Vehicles equipment; the control terminal uses a convolutional neural network to perform object identification and classification on the point cloud data, uses a Kalman filter to track the identified object, and predicts the position of the object after a certain period of time in the future; judging whether a collision event occurs at the position of the object after a certain period of time in the future or not; and if yes, sending an alarm to the object through the vehicle networking equipment on the roadside. As the laser radar arranged at the high position has a ...

Подробнее
Номер записи: 54
09-06-2023 дата публикации

Electronic component defect identification method and device based on convolutional neural network

Номер: CN116245802A
Автор: ZENG NIANYIN, LI XINYU, WU PEISHU, HU LIWEI, LIU HAONAN, LI HAN
Принадлежит:

The invention relates to an electronic component defect identification method and system based on a convolutional neural network, and the method comprises the steps: collecting a multi-modal fusion image of a plurality of electronic components including defects, and taking the multi-modal fusion image as an original image set; determining the defect type and the defect position of each electronic component in the original image set, marking the defect type and the defect position, and dividing marked data into a training set and a test set; training a three-mode fusion convolutional neural network model by using the training set, testing the trained model by using the test set, and obtaining a trained model when an index meets a preset requirement; and inputting the multi-modal fusion image of the to-be-identified electronic component into the trained three-modal fusion convolutional neural network model to obtain the defect type and defect position of the to-be-identified electronic component ...

Подробнее
Номер записи: 55
06-06-2023 дата публикации

Lesion Chinese chestnut classification method and system based on hyperspectrum and convolutional neural network, and medium

Номер: CN116229260A
Автор: ZHONG NAN, ZHONG QIONGDA, PENG ZHITONG
Принадлежит:

The invention discloses a lesion Chinese chestnut classification method and system based on a hyperspectrum and a convolutional neural network, and a medium. The method comprises the steps that hyperspectral information of Chinese chestnuts with various degrees of lesions is acquired and preprocessed, and a hyperspectral data set of the Chinese chestnuts with various degrees of lesions is obtained; constructing a lesion Chinese chestnut recognition model based on an improved VGG16 convolutional neural network, and training the lesion Chinese chestnut recognition model according to the hyperspectral data sets of the lesion Chinese chestnuts of various degrees; and obtaining hyperspectral information of a to-be-detected Chinese chestnut sample, inputting the hyperspectral information into the trained lesion Chinese chestnut recognition model for detection, completing Chinese chestnut lesion recognition, and controlling a pneumatic classifier to classify lesion Chinese chestnuts according ...

Подробнее
Номер записи: 56
06-06-2023 дата публикации

Self-adaptive energy conversion device and method based on vibration spectrum of mining air duct

Номер: CN116232123A
Автор: QIAO TIEZHU, ZHANG SONGWEI, DONG HUIJIE
Принадлежит:

The invention provides an energy self-adaptive conversion device and method based on a mining air duct vibration frequency spectrum, and belongs to the technical field of mining air duct vibration frequency spectrum energy self-adaptive conversion. The technical problem to be solved is to provide an energy adaptive conversion device hardware structure based on the vibration spectrum of the mining air duct and an improvement of a conversion method. According to the technical scheme adopted for solving the technical problem, the system comprises an energy conversion device and an energy processing device, wherein the energy conversion device is used for collecting natural vibration frequency data of the mining air duct and converting vibration energy into electric energy; the energy processing device is used for carrying out self-adaptive matching on the electric energy output by the energy conversion device; the electric energy output end of the energy conversion device is electrically connected ...

Подробнее
Номер записи: 57
06-06-2023 дата публикации

Shale identification method based on improved Resnet

Номер: CN116229141A
Автор: LIU HONGQI, MA TONGLE, LIU WEI
Принадлежит:

The invention discloses a shale identification method based on improved Resnet, and the method employs an AdaGrad optimization algorithm for parameter updating, the algorithm can properly adjust the learning rate for each element of a parameter, learning is carried out at the same time, W represents a to-be-updated weight parameter, h is the sum of squares of the loss of the previous parameter with respect to the gradient, is the gradient of a loss function with respect to the weight, and is the sum of squares of the loss of the previous parameter with respect to the weight. The method is very ideal in application in a Lonmaxi shale reservoir section, the improved residual network is very suitable for being applied to image recognition, and the problem that a deep neural network cannot process image data is solved; according to the residual network identification method provided by the invention, the features after image processing are transmitted into the multi-layer perceptron to be output ...

Подробнее
Номер записи: 58
27-06-2023 дата публикации

Multi-fan wind speed and wind direction prediction method based on dynamic graph convolution and transformer

Номер: CN116341720A
Автор: HAN ZHIYING, YANG DUAN, SUN MAN, YUWEN SHAOJIA, TAN JINXIN
Принадлежит:

The invention discloses a multi-fan wind speed and wind direction prediction method based on dynamic graph convolution and transformer, and the method comprises the steps: carrying out the sampling and storage of a wind speed and wind direction X at a historical moment based on a sensor and a storage device at each fan of a wind power plant; modeling long-term spatial dependence and short-term spatial dependence of each fan of the wind power plant based on graph convolution and a graph self-attention network, constructing a dynamic graph convolution network, and obtaining spatial dependence relationship information of each moment of the wind power plant; on the basis of a self-attention mechanism of a transformer, completing modeling of a time dependency relationship of a wind speed and wind direction sequence; the graph convolutional network is embedded into a transform network, a space-time sequence prediction neural network is constructed, space-time dependency relationship modeling ...

Подробнее
Номер записи: 59
06-06-2023 дата публикации

Credit card user behavior prediction method based on attribute reconstruction and attention mechanism

Номер: CN116228400A
Автор: YANG LIANG, LIU JIAYAO, ZHANG QIULIANG, NIU BINGXIN, JIA YONGNA, ZHANG YAJUAN, GU JUNHUA
Принадлежит:

The invention discloses a credit card user behavior prediction method based on attribute reconstruction and an attention mechanism, and the method comprises the steps: firstly obtaining the historical data of a credit card user in a certain period of a bank system, and converting the historical data into graph data; nodes of the graph data represent credit card users, and edges represent relations among the users; obtaining graph topology information and a node attribute matrix; then, a node attribute reconstruction model is constructed based on the graph convolutional neural network, the node attribute reconstruction model comprises an encoder and a decoder, the encoder encodes graph topology information and node attribute information into a potential distribution matrix, and the decoder decodes the potential distribution matrix into a reconstructed node attribute matrix; finally, constructing a credit card user behavior prediction model, and inputting graph topology information and the ...

Подробнее
Номер записи: 60
05-05-2023 дата публикации

Method and device for predicting time sequence index

Номер: CN116070773A
Автор: ZHOU FAN, PAN CHEN, MA LINTAO, LIU YU, ZHANG JAMES, LIN WEITAO, HU XUANWEI, ZHU XINXIN, HU YUNHUA
Принадлежит:

The embodiment of the invention provides a method and device for predicting a time sequence index. According to the method, first, corresponding n first representations are obtained based on respective service features of n service objects in a target time period t; wherein the n business objects form nodes in the tree structure based on the range affiliation relation of the n business objects. Thirdly, performing convolution operation on the n first representations to obtain n second representations; wherein the second representation corresponding to any target business object in the n business objects is obtained by performing convolution operation on the first representation in a target convolution window, and the target convolution window comprises a plurality of business objects covered by a path from the root node to the target business object in the tree structure. Based on this, n object representations of the n business objects can be determined according to the n second representations ...

Подробнее
Номер записи: 61
05-05-2023 дата публикации

YOLOX-based steel plate surface defect detection method, device and equipment

Номер: CN116071338A
Автор: HU JING, CHEN WEI
Принадлежит:

The invention relates to a YOLOX-based steel plate surface defect detection method, which comprises the following steps: constructing a defect detection model based on a YOLOX network, and initializing epoch, a learning rate and a model weight; inputting the training set into a defect detection model, preprocessing the training set, inputting the preprocessed training set into a backbone network, and outputting three effective feature maps with different sizes through a Focus module and four darknet modules which are sequentially connected in series; the three effective feature maps are respectively input into three CBAM modules with corresponding sizes, after adaptive feature correction is carried out, the three effective feature maps are respectively input into three branches of the bottleneck network, each branch comprises an RFB module and an ASPP module which are connected in series, and three enhanced feature maps are output and are respectively input into corresponding decoupling ...

Подробнее
Номер записи: 62
09-06-2023 дата публикации

Chicken farm supervision method and system based on 5G monitoring

Номер: CN116249036A
Автор: ZENG ZHIXIONG, LYU ENLI, CHEN ZHURONG, EUDEAN, YE ZICONG, LIU YUQI, PENG WANSEN, LU CHANG, CHEN WEIHAN
Принадлежит:

The invention aims to provide a chicken farm supervision method and system based on 5G monitoring, and belongs to the field of chicken farm supervision, and the method comprises the steps: collecting the breeding data of a chicken farm, and uploading the collected breeding data to a server through a 5G network; the server inputs the collected breeding data into a trained neural network model, and the neural network model outputs the health condition of the chicken; the database collects health conditions of chickens and reports the health conditions to a chicken farm user side, and a user supervises the chicken farm through the user side. The whole chicken farm is monitored through 5G monitoring, unmanned supervision can be achieved, and the working energy efficiency is improved.

Подробнее
Номер записи: 63
09-06-2023 дата публикации

Image tampering area positioning method based on DAUnet network and storage medium

Номер: CN116246151A
Автор: LIU ZHICHANG, GUO CHEN
Принадлежит:

The invention relates to the technical field of image processing, in particular to an image tampering area positioning method based on a DAUnet network and a storage medium, and the image tampering area positioning method comprises the following steps: S1, carrying out the feature extraction and channel lifting of a target image through a First Block layer, then carrying out the feature extraction through a Dense Block module, and carrying out the channel lifting of the target image; the extracted image features are sampled and compressed through a Transformation layer; s2, performing detail extraction on the compressed feature image through a cavity space pooling pyramid ASPP module; and S3, carrying out image up-sampling on the feature image subjected to detail extraction through a deconvolution module, enabling image semantic information of a lower layer in down-sampling to participate in processing of the deconvolution module, and finally carrying out dimension reduction processing ...

Подробнее
Номер записи: 64
02-05-2023 дата публикации

Transform-based traffic scene small sample target detection method and device

Номер: CN116052108A
Автор: TIAN YAN, ZOU NINGJIE, HU JIAYU, ZHU YUNFANG, JIANG ZHAOYI, WANG XUN
Принадлежит:

The invention discloses a traffic scene small sample target detection method and device based on Transform, and the method comprises the steps: retraining an existing target detection model, designing a multi-head self-attention mechanism in a Transform encoder, achieving pseudo-class embedding to associate samples of the same class and distinguish samples of different classes, and being capable of detecting a new class of a support set, and improving the detection accuracy. The problem of inaccurate generation of the candidate region in the small sample target detection task is improved; the Transform decoder makes the detector easy to train and strong in generalization ability by using pseudo-class information embedding, and solves the problems that a small sample target detection task is difficult to learn and easy to over-fit. Compared with other advanced methods, the method has considerable competitiveness.

Подробнее
Номер записи: 65
09-06-2023 дата публикации

Thyroid nodule fine-grained automatic identification and classification system based on weak supervised learning

Номер: CN116246099A
Автор: YANG JIE, CHEN HONG, LI JINFENG, WEI ANRUO, LI YUHAN, SU JIAHAO
Принадлежит:

The invention relates to the technical field of data processing, in particular to a thyroid nodule fine-grained automatic identification and classification system based on weak supervised learning, which comprises a server, a thyroid nodule fine-grained identification model is preset in the server, and the server is used for acquiring an identification image; outputting a fine-grained recognition result output by the thyroid fine-grained recognition model according to the recognition image; the thyroid fine-grained recognition model comprises a feature pyramid attention module used for processing a recognition image according to a feature pyramid attention algorithm; the space attention module is used for acquiring a binary mask and generating an attention map according to the binary mask to position a discrimination area; and the multi-network fusion module is used for acquiring output results of the plurality of backbone networks and fusing the output results to generate a fine-grained ...

Подробнее
Номер записи: 66
02-05-2023 дата публикации

Method and device for identifying safety state of seat safety device

Номер: CN116039559A
Автор: ZHANG LEI, CAI DEXUAN, LI PENG, TANG SHANZHENG
Принадлежит:

The invention discloses a method and device for recognizing the safety state of a seat safety device. The method comprises the steps that vehicle state information is obtained; when the vehicle state information meets a preset first-level reminding condition, a seat image is obtained; judging whether a passenger sits on the seat according to the seat image; when the passenger sits on the seat, whether the vehicle state information meets a preset secondary reminding condition or not is judged; when the vehicle state information meets a preset secondary reminding condition, safety belt lock catch state information is obtained; judging whether the passenger fastens a safety belt or not according to the seat image and the safety belt lock catch state information; and when the passenger does not fasten the safety belt, secondary reminding is carried out. According to the method and the device, the driving safety is improved.

Подробнее
Номер записи: 67
26-05-2023 дата публикации

Vector light field polarization transient resolving method based on convolutional neural network

Номер: CN116164842A
Автор: SHEN HUA, CHEN YUFAN, LI CHEN, ZHU RIHONG
Принадлежит:

The invention discloses a vector light field polarization state transient solution method based on a convolutional neural network. The method comprises the following specific steps: constructing a CNN for vector light field polarization state transient solution; building a rotating wave plate experiment system, obtaining a data set required by CNN training by using the rotating wave plate experiment system, and training the CNN; a vector light field polarization transient resolving system is obtained after a rotary wave plate experiment system is improved, a transient vector light field light intensity graph and a modulated transient vector light field light intensity graph are obtained through the vector light field polarization transient resolving system, the two images are input into a trained CNN, and the polarization direction and the elliptic polarization of the corresponding position of the transient vector light field are obtained through output. Therefore, the polarization state ...

Подробнее
Номер записи: 68
23-06-2023 дата публикации

Diving battery feature identification method based on interpretable neural network

Номер: CN116298915A
Автор: CHEN YUNXIA, ZHOU ZIYI, WANG CONG, GAO HAOTIAN
Принадлежит:

The invention relates to a diving battery feature recognition method based on an interpretable neural network, and the method comprises the following steps: 1, dividing an original charging and discharging data set of a lithium battery into a training set and a test set, and generating a label matrix of a diving battery and a normal battery; 2, constructing a feature matrix of the battery, and constructing a classification model based on a neural network; and 3, identifying the characteristics of the diving battery by applying gradient weighted class activation mapping. According to the method, the original charging and discharging data of the lithium battery are directly utilized for feature recognition, only data extraction is needed, other preprocessing methods are not needed for data processing, and feature recognition is more convenient. According to the method, the optimal classification features of the diving battery and the normal battery are quickly and accurately identified by ...

Подробнее
Номер записи: 69
23-06-2023 дата публикации

Graph neural network-based high-resolution remote sensing image forest land secondary classification method

Номер: CN116310826A
Автор: PENG SONG, LI JING
Принадлежит:

The invention provides a high-resolution remote sensing image forest land secondary classification method based on a graph neural network. The method comprises the following steps: reading a high-resolution remote sensing image needing forest land secondary classification; cutting the high-resolution remote sensing image into a plurality of image small images BlkImage according to a set overlapping ratio; in each small image BlkImage, i final forest land detection frames Box ''and i corresponding forest land categories Cls'' are identified; and combining the final forest land detection frames Box ''and the corresponding forest land categories Cls'' in all the small image pictures BlkImage, thereby obtaining all the forest land detection frames and the corresponding forest land categories in the high-resolution remote sensing image, and completing the second-level classification of the forest land. The high-resolution remote sensing image forest land secondary classification method based ...

Подробнее
Номер записи: 70
06-06-2023 дата публикации

Federal domain adaptive fault diagnosis method based on statistical feature fusion

Номер: CN116226784A
Автор: SU CHANGWEI, MIAO XUHONG, GAO SHENGYAO, WANG XUEREN, ZHOU TAO, LI YIBIN, ZHANG YANTAO, TANG YUHANG
Принадлежит:

The invention discloses a federated domain adaptive fault diagnosis method based on statistical feature fusion. A source domain client receives statistical features of a target domain client and statistical features of self-active source domain data and performs data standardization on a source domain data set; the central server sends the fault diagnosis model to all source domain clients, and then the source domain clients use two kinds of standardized source domain data as input to train a received global model in the source domain clients; the trained model is returned to the central server, and the central server carries out global model aggregation on the received models of all the source domain clients; when the training round number reaches a set value, training is finished, a final fault diagnosis model is obtained and sent to the target domain client side, the target domain client side obtains mechanical fault data to be diagnosed, and a fault diagnosis result is obtained through ...

Подробнее
Номер записи: 71
05-05-2023 дата публикации

Non-transitory computer readable medium and apparatus for blade wear monitoring

Номер: CN116071291A
Автор: WINTER SARAH, JEREMY J. FAUST
Принадлежит:

A non-transitory computer readable medium is provided that stores instructions that, when executed by at least one processor, cause the at least one processor to: obtain a first signal based on an input image using a trained machine learning model, the input image being an image of a plant cut by a blade, the first signal indicates the wear level of the blade; determining whether the level of the first signal is greater than or equal to a threshold; generating a second signal in response to determining that the level of the first signal is greater than or equal to the threshold; and outputs a second signal.

Подробнее
Номер записи: 72
27-06-2023 дата публикации

Preparation method and device of traditional Chinese medicine for inhibiting bacteria of underwear and underpants

Номер: CN116342551A
Автор: WU HAO
Принадлежит:

The invention discloses a preparation method and device of traditional Chinese medicine for inhibiting bacteria of underwear and underpants. The method comprises the following steps: firstly, enabling a detection image and a reference image of a traditional Chinese medicine raw material to be detected to pass through a twin network model comprising a first image encoder and a second image encoder to obtain a detection feature map and a reference feature map; calculating a difference feature map between the detection feature map and the reference feature map, and enabling the difference feature map to pass through a space attention module to obtain a difference space enhanced feature map; and enabling the difference space enhanced feature map to pass through a classifier to obtain a classification result used for representing whether the traditional Chinese medicine raw materials to be detected conform to regulations or not. Therefore, whether the traditional Chinese medicine raw materials ...

Подробнее
Номер записи: 73
27-06-2023 дата публикации

Image-text cross-modal retrieval method and device, equipment and storage medium

Номер: CN116340558A
Автор: SHU CHANG, CHEN YOUXIN
Принадлежит:

The invention relates to an artificial intelligence technology, and discloses an image-text cross-modal retrieval method, which comprises the following steps of: obtaining an image set and a text set, and partitioning each image in the image set to obtain a partitioned image set; extracting a feature vector of each block picture, calculating a position vector of each block picture, and obtaining a block vector set of the block picture set according to the feature vectors and the position vectors; performing text word segmentation on each text in the text set one by one, and performing vector conversion on the text word segmentation to obtain a word segmentation vector set; performing global pooling on the block vector set and the word segmentation vector set to obtain a picture global vector and a text global vector; and constructing an image-text cross-modal retrieval model according to the image global vector and the text global vector, and performing image-text cross-modal retrieval ...

Подробнее
Номер записи: 74
23-06-2023 дата публикации

Three-dimensional mass spectrum imaging method and device based on sparse sampling and depth generation model

Номер: CN116297786A
Автор: MA XIAOXIAO, LI DAN, QIAN YAO
Принадлежит:

The invention relates to the technical field of mass spectrum imaging, in particular to a three-dimensional mass spectrum imaging method and device based on sparse sampling and a depth generation model.The method comprises the steps that full sampling is carried out on part of tissue slices to obtain mass spectrum data, the sparse sampling process is simulated on the mass spectrum data to obtain simulated sparse sampling data, and the simulated sparse sampling data are obtained; and inputting and training the mass spectrum data and the simulated sparse sampling data to obtain a depth generation model, optimizing to obtain an optimal model structure and hyper-parameter setting, generating a trained depth generation model, and reconstructing the sparse sampling mass spectrum imaging to obtain complete three-dimensional mass spectrum imaging data. According to the embodiment of the invention, the three-dimensional mass spectrum imaging data can be obtained by combining the depth generation ...

Подробнее
Номер записи: 75
30-05-2023 дата публикации

Running posture recognition and correction method and system based on high-dimensional time sequence diagram network

Номер: CN116186561A
Автор: TAN GUANGXIN, ZHAO KAI, DU CHUNJIE, ZHAO DANXIA
Принадлежит:

The invention discloses a running posture recognition and correction method and system based on a high-dimensional time sequence diagram network, and the method comprises the steps: collecting the sensor information of each body part of a runner through a wireless transmission module, and carrying out the filtering processing and feature extraction of the collected sensor signals of a human skeleton diagram network; carrying out parameter learning and model training of a graph convolutional neural network on the high-dimensional time sequence graph network feature information to obtain a motion posture recognition model, carrying out real-time monitoring, carrying out comparative analysis according to feature information of professional athletes, judging wrong body force exerting parts of the runner in the motion process, sending correction signals to the body parts, and carrying out motion posture recognition on the wrong body force exerting parts. And reminding and correction are carried ...

Подробнее
Номер записи: 76
02-05-2023 дата публикации

Bird-eye view visual angle feature map processing method and device, electronic equipment and storage medium

Номер: CN116051489A
Автор: CHEN WEI, GUO XIANDA, ZHU ZHENG
Принадлежит:

The embodiment of the invention provides a bird's-eye view visual angle feature map processing method and device, electronic equipment and a computer readable storage medium, and relates to the field of automatic driving, and the method comprises the steps: obtaining an original image and an original depth map of a camera visual angle, and extracting the image features of the original image; performing depth prediction on the original depth map to obtain N target depth maps, and projecting the image features to a three-dimensional space according to the N target depth maps to obtain a three-dimensional projection point of each pixel in the original image; n is a positive integer; performing aerial view visual angle pooling on the three-dimensional projection point to obtain a first aerial view visual angle feature map; obtaining a second aerial view visual angle feature map corresponding to the laser radar; and carrying out feature fusion on the first aerial view visual angle feature map ...

Подробнее
Номер записи: 77
02-05-2023 дата публикации

Model training method, 3D garment rendering method, electronic equipment and storage medium

Номер: CN116051705A
Автор: YAN HAONAN, QIN WEINING, ZHAO XIN
Принадлежит:

The embodiment of the invention provides a model training method, a 3D garment rendering method, electronic equipment and a storage medium, and relates to the field of model rendering, and the training method of a texture map generation model comprises the steps: mapping a random parameter into hidden space data through a mapping network, carrying out affine transformation according to the hidden space data to obtain a texture map style vector, inputting the texture mapping style vectors into a plurality of migration layers; superposing random parameters on the output vectors of the convolution layers, inputting the output vectors superposed with the random parameters into a migration layer, taking data output by the migration layer as the input of the next convolution layer, and generating a predicted clothing texture map according to the output of the last migration layer; inputting the predicted clothing texture map and the real clothing texture map into a discriminator, and outputting ...

Подробнее
Номер записи: 78
23-05-2023 дата публикации

Pixel-level infrared image splicing method based on deep learning

Номер: CN116152055A
Автор: HAO ZIQIANG, LIU SHENGZUO, TANG YANFENG, XU XIAOYU, HONG YANG
Принадлежит:

The invention discloses a pixel-level infrared image splicing method based on deep learning, belongs to the technical field of image splicing, and aims to solve the problem that images obtained by an existing infrared image splicing method have ghosts and artifacts, and the method comprises the following steps: step 1, constructing a network model; step 2, preparing a data set: selecting a KAIST data set, adjusting the size of each image in the data set, and fixing the size of the input image; 3, training a network model: inputting the data set prepared in the step 2 into the network model constructed in the step 1 for training; step 4, selecting a minimum loss function and an optimal evaluation index; step 5, fine-tuning the model: training and fine-tuning the model by using an LTIR data set to obtain stable and available model parameters, and finally enabling the model to have a better infrared image splicing effect; and 6, saving the model: solidifying the finally determined model parameters ...

Подробнее
Номер записи: 79
06-06-2023 дата публикации

Eye socket lymphoma and inflammatory false tumor classification method, system, device and terminal

Номер: CN116228733A
Автор: ZHAO FENGJUN, KANG HAOBEI, ZHANG TAO, WANG BIN, CAO XIN, YI HUANGJIAN, PENG JINYE, HE XIAOWEI
Принадлежит:

The invention belongs to the technical field of medical image processing and computer-aided diagnosis, and discloses an orbital lymphoma and inflammatory false tumor classification method, system, device and terminal, in an obtained orbital DCE-MRI image, a tumor area and an eye cone triangular area in an original image are manually segmented and preprocessed; extracting features from the tumor area by using a neural network, and performing clustering statistical analysis on the features as tumor area features; performing feature extraction on the eye cone triangular region as a similarity judgment standard, and extracting eye cone region features by using a self-classification and self-recovery network; taking the original image as an input network, and combining the features of the tumor area and the eye cone triangular area to train a network model; and inputting the multi-modal orbit data into a classification model for processing, thereby realizing classification of orbit lymphoma ...

Подробнее
Номер записи: 80
14-04-2023 дата публикации

Knowledge guidance-based small sample target detection and identification method

Номер: CN115965968A
Автор: WU JINJIAN, ZHANG YANHE, CHEN LU
Принадлежит:

The invention provides a small sample target detection and identification method based on knowledge guidance. The method comprises the following steps: acquiring an image training set of a to-be-classified target; performing instance segmentation on a semantic level on each target image to serve as a live-action primitive, performing feature extraction through a preset residual neural network to obtain a feature vector of the live-action primitive, and constructing a feature matrix of the target image according to the feature vector and the number of basic primitives in a primitive space; according to the priori knowledge of the basic primitives and the semantic information and the priori knowledge between the real scene primitives, reasoning the priori logic knowledge and the semantic information in the form of a knowledge graph; according to the method, the knowledge-driven knowledge graph is constructed by associating the image features with the semantic primitives, so that the dependence ...

Подробнее
Номер записи: 81
14-04-2023 дата публикации

Modeling method of video feature extraction model based on space-time separation convolution and asymmetric convolution

Номер: CN115965892A
Автор: XING SUXIA, GUO ZHENG, YU ZHONGZHONG, TONG XIN
Принадлежит:

The invention discloses a modeling method of a video feature extraction model based on space-time separation convolution and asymmetric convolution, and belongs to the technical field of mode recognition, machine vision and video processing, and the method comprises the steps: building a C3D network model; the first four convolution layers of the C3D network are converted into space-time separable convolution, and 3D convolution is definitely decomposed into continuous operation of 2D space convolution and 1D time convolution; improving the last four convolutional layers of the C3D network to reduce model parameters, and improving 3D convolution into asymmetric convolution; and constructing a video feature extraction model based on space-time separation convolution and asymmetric convolution. Experiments show that the method has the advantages that the improved model has a better effect in the aspect of video feature extraction, model parameters can be effectively compressed, and the method ...

Подробнее
Номер записи: 82
04-08-2023 дата публикации

Transform-fused medium and large-sized complex component welding seam identification method

Номер: CN116543207A
Автор: LIU ZONGMIN, LI JIE, LI ZUFENG, AO WENGANG, WANG JIRUI, GU ZHENJIE, WANG SONG, HUANG GUOYANG
Принадлежит:

The invention discloses a Transform-fused method for identifying a welding seam of a medium-large complex component. The method comprises the following steps: S1, collecting image information of a welding seam sample; s2, constructing a component welding seam recognition model, inputting the welding seam sample image information into the component welding seam recognition model for model training, and obtaining a trained component welding seam recognition model; and S3, inputting the welding seam image information of the to-be-detected component into the trained component welding seam recognition model, and outputting a welding seam recognition result of the to-be-detected component. On the premise of ensuring the detection speed, the welding seam detection precision is greatly improved, the anti-interference capability is high, and the robustness is high.

Подробнее
Номер записи: 83
04-08-2023 дата публикации

Time enhancement information sequence recommendation method and system based on graph convolutional network

Номер: CN116542720A
Автор: CHEN JIANXIA, YU TIANCI, MAO LEI, LUO SHIJIE
Принадлежит:

The invention belongs to the technical field of sequence recommendation, and discloses a time enhancement information sequence recommendation method and system based on a graph convolutional network, and the method comprises the steps: combining an article embedding matrix and a position embedding matrix, and constructing a hidden layer representation of a sequence; constructing a time-enhanced user-article graph by using an adaptive window function based on the time-enhanced graph convolutional network, and distributing a time weight for each user-article interaction item by using a double window function; a self-attention layer based on filtering enhancement is constructed, the filtering enhancement layer is used in front of a self-attention module, and fast Fourier transform and a learnable filter are utilized to suppress noisy signals in sequence embedding; the processed user embedding and item embedding are aggregated, and then a score is output for each user-item pair. According to ...

Подробнее
Номер записи: 84
14-07-2023 дата публикации

Behavior recognition method based on space-time characteristic difference and correlation of skeleton data

Номер: CN116434339A
Автор: JIANG MIN, QING RUXIN, KONG JUN
Принадлежит:

The invention discloses a behavior recognition method based on space-time characteristic difference and correlation of skeleton data, and relates to the field of computer vision. A behavior recognition module built by the method comprises a space graph convolution module and a time graph convolution module; the spatial graph convolution module comprises a time feature learning unit, a channel feature learning unit and a time channel context topology unit, and the time feature learning unit and the channel feature learning unit are used for independently learning feature difference and feature correlation between joints in a time frame dimension and a channel dimension respectively; according to the method, a time channel context topology unit is used for learning the correlation of features between joints which cooperatively complete actions and learning the feature difference of part of joints which change due to movement, the time channel context topology unit learns global context feature ...

Подробнее
Номер записи: 85
14-07-2023 дата публикации

Cut tobacco production workshop safety management method and cut tobacco production workshop safety management system

Номер: CN116434455A
Автор: LIU YICHAO, TANG ZHE, FANG TING, ZHAO XIAOJUN, TANG ZHICHAO, ZHANG ZHIWU, WANG XUEJIAO, CHANG WENHUI, MA MINGLEI, SUN SHUO
Принадлежит:

The invention relates to a cut tobacco production workshop safety management method and a cut tobacco production workshop safety management system. The safety management method for the tobacco shred making workshop comprises the following steps: acquiring a photoelectric signal generated by an early warning mechanism, judging whether a person enters a bottom area of a turned tobacco basket according to a preset rule, and if so, generating a first alarm and a control instruction for controlling a basket turning machine to act; temperature information collected by an infrared collection device arranged near the flue is obtained, whether the flue temperature exceeds a threshold value or not is judged according to the temperature information, and if yes, a second alarm is generated; and acquiring image data acquired by an image acquisition device arranged in the high-risk working area, judging whether a person who does not wear a safety helmet exists in the high-risk working area or not according ...

Подробнее
Номер записи: 86
09-05-2023 дата публикации

Multi-view image classification method based on global filtering module

Номер: CN116091830A
Автор: LEI FANGYUAN, YANG YUHAN, JIANG JIANJIAN, KANG YOON-JI
Принадлежит:

The invention discloses a multi-view image classification method based on a global filtering module, belongs to the technical field of deep learning image classification, and solves the technical problem that traditional machine learning cannot meet classification and retrieval of current appearance patent images. The method comprises the following steps: acquiring a multi-view image data sample set, and dividing the multi-view image data sample set into a training set and a test set; obtaining multi-view image data in the training set, and performing block embedding processing to obtain a multi-view image block set; the multi-view image block set is used for generating image blocks, the image blocks are embedded and input into an MGViT network, and a forward propagation process of the MGViT network is defined; a global filtering module and a Transform module are defined, and a loss function used for iterative optimization is defined; inputting the training set into an MGViT network for ...

Подробнее
Номер записи: 87
09-05-2023 дата публикации

Model training method based on motion capture

Номер: CN116091972A
Автор: LI SENHE
Принадлежит:

The invention discloses a model training method based on motion capture, and relates to the technical field of behavior analysis. Comprising the following steps: pre-selecting an smplx model; marking motion nodes in the plane data set of each frame based on the material video; using a re-projection loss function to restore shielded motion nodes, and establishing a planar model and a projection three-dimensional model; setting a plurality of motion data sets; three-dimensional projection ambiguity is eliminated based on limit actions and postures of a human body; a discriminator is arranged, adversarial learning is carried out according to the whole-body actions and the whole-body postures of the space model, the priori of human motion is obtained, and a trained model is obtained. The joint points of the shielded part are predicted and restored by using a reprojection loss function, so that all the joint points are clearly connected; meanwhile, through the confrontation loss of the model ...

Подробнее
Номер записи: 88
06-06-2023 дата публикации

Community discovery method of graph convolutional neural network based on node structure features

Номер: CN116226467A
Автор: SHI ZHENQUAN, SUN FAN, SHI QUAN, LUO QICAI, ZHANG WENGJIAN, HUANG ZILING, FENG JI, WANG BOWEN
Принадлежит:

The invention discloses a community discovery method of a graph convolutional neural network based on node structure features. Firstly, different types of nodes are mapped to the same space, and node features are obtained by using a GCN model. Then, the structural features of the nodes are considered from the two aspects of isomerism and overlapping performance; and then, stacking the target node representations under the three angles, and generating a final target node representation by using an MLP (Markup Language Protocol). And finally, performing community division on the nodes by using a k-means clustering algorithm. According to the method, node structure features are considered from the two aspects of isomerism and overlapping, node characterization is considered from the two aspects of node features and the node structure features, a new angle is provided for node characterization learning, the node characterization quality is improved, the community discovery accuracy is guaranteed ...

Подробнее
Номер записи: 89
26-05-2023 дата публикации

Information identification method, identification device, computer equipment and storage medium

Номер: CN116166172A
Автор: XING YONGXIN, YAN JUNCHAO, ZHANG ZHEYAO, TONG HAOPENG
Принадлежит:

The invention provides an information recognition method and device, computer equipment and a storage medium, and belongs to the technical field of artificial intelligence and target recognized.The information recognition method comprises the steps that according to touch information fed back by a touch screen, touch pixel position information is determined; according to a preset corresponding relation between pixel points of the touch screen and elements in a preset array, adjusting the elements corresponding to the pixel position information in the preset array to obtain a target array; and inputting the target array into a pre-trained target neural network for processing to obtain a gesture recognition result.

Подробнее
Номер записи: 90
30-05-2023 дата публикации

Hip joint segmentation method and device, electronic equipment and computer readable storage medium

Номер: CN116188492A
Автор: ZHANG YILING, LIU XINGYU
Принадлежит:

The invention provides a hip joint segmentation method and device, electronic equipment and a computer readable storage medium. The hip joint segmentation method comprises the following steps: acquiring a hip joint image to be segmented; inputting the hip joint image into a preset hip joint segmentation model, and outputting a hip joint segmentation result; wherein the hip joint segmentation model is obtained through model training based on an improved LadderNet network, and the improved LadderNet network is formed by adding an attention mechanism network in jump connection between each pair of adjacent decoder branches and encoder branches in each layer of the LadderNet network and is used for enhancing segmentation of hip joint image boundary features; moreover, a deep aggregation pyramid network is added to the last layer of the LadderNet network, and is used for extracting convolution features of different levels and different scales, and mining the feature information of the last layer ...

Подробнее
Номер записи: 91
06-06-2023 дата публикации

Aided driving AI middle platform system based on confidence learning and multi-task learning

Номер: CN116229398A
Автор: LUO XIN, WANG WENXIAO, CAI DENG
Принадлежит:

The invention discloses an AI (auxiliary driving) system based on confidence learning and multi-task learning, and the system comprises a data processing module which is used for receiving original data, carrying out the data marking and screening through the confidence learning, and constructing a data set; the model construction module is used for constructing an encoder and a decoder of a multi-task model and providing a context information transmission path and a fusion pooling enhancement module for the model with relatively high task correlation, so that contexts of the decoder are interacted; the model training module is used for defining and training parameters, and converting and outputting a model after training to specified times or precision; the model deployment monitoring module is used for carrying out equipment end deployment on the optimal model and carrying out real-time monitoring at a cloud end; and storing and returning scene data with a poor model effect to form a ...

Подробнее
Номер записи: 92
30-05-2023 дата публикации

Text similarity analysis method and system based on residual fusion and stack integration

Номер: CN116186556A
Автор: YUAN LING, LIU HAOWEN, XIANG FENG
Принадлежит:

The invention discloses a text similarity analysis method and system based on residual fusion and stack integration. A text similarity analysis method based on residual fusion and stack integration comprises the steps that a to-be-analyzed text is input into a target text similarity judgment model for similarity judgment, and a judgment result is obtained; the target text similarity judgment model is obtained by fusing a representation type SBERT model and an interactive type ABCNN model; if the to-be-analyzed text is judged to be the similar text, inputting the to-be-analyzed text into a target text similarity calculation model for similarity calculation to obtain a target text similarity value; the target text similarity calculation model is obtained by performing parallel heterogeneous integration on a plurality of different interactive pre-training models. The technical problem that an existing text similarity analysis model is difficult to have high retrieval efficiency and accuracy ...

Подробнее
Номер записи: 93
11-07-2023 дата публикации

Pancreatic cancer diagnosis method and system, medium and electronic equipment

Номер: CN116416225A
Автор: QIAN XIAOHUA, LI XINYUE, GUO RUI
Принадлежит:

The invention provides a pancreatic cancer diagnosis method and system, a medium and electronic equipment. The pancreatic cancer diagnosis method comprises the following steps: acquiring an abdomen plain-scan CT image; dividing a pancreas area in the abdomen plain-scan CT image into image blocks, and extracting image block features and prediction probability values of the image blocks; based on the image block features and the prediction probability values of different case areas, respectively constructing corresponding case level graphs, wherein the different case areas comprise a non-pancreatic cancer case pancreatic area, a pancreatic cancer case tumor area and a pancreatic cancer case non-tumor area; and training a graph neural network for pancreatic cancer prediction based on the graph, so as to obtain a pancreatic cancer prediction result based on the trained graph neural network. According to the pancreatic cancer diagnosis method and system, the medium and the electronic equipment ...

Подробнее
Номер записи: 94
25-08-2023 дата публикации

Energy consumption prediction method based on deep learning and spatio-temporal joint knowledge distillation

Номер: CN116644660A
Автор: ZHANG YANYU, JIN CHANGXIN, ZHAN TENG
Принадлежит:

The invention discloses an energy consumption prediction method based on deep learning and spatio-temporal joint knowledge distillation, and relates to the technical field of energy consumption data prediction. An energy consumption prediction system is constructed, characteristics of related energy consumption data are collected, preprocessed and extracted through a data processing module, the related energy consumption data comprise building structures, meteorological data and people flow data, and characteristics capable of representing spatio-temporal characteristics of predicted objects are extracted through the data processing module according to the related energy consumption data. A deep learning prediction model is trained by using features representing spatio-temporal features of a prediction object, parameters of the prediction model are continuously optimized according to a defined loss function in the training process, and the prediction model is compressed and optimized through ...

Подробнее
Номер записи: 95
18-07-2023 дата публикации

Rotating target detection method fusing optical and SAR image multi-modal information

Номер: CN116452936A
Автор: XU KAI, LIU SIYUAN, WANG ANLING, WANG ZIYU, LIANG DONG
Принадлежит:

The invention relates to a rotating target detection method fusing optical and SAR (Synthetic Aperture Radar) image multi-modal information, and compared with the prior art, the method can effectively solve the problem that a target cannot be accurately identified due to various interferences in the remote sensing image acquisition and transmission process. Meanwhile, the method can also solve the problems that different types of targets have different shapes, sizes, colors and other characteristics or have overlapping, shielding and the like. The method comprises the following steps: extracting features of an SAR image data set and an optical image data set; performing cross-modal multi-scale feature fusion; the two-stage rotation prediction head positions and classifies targets at different angles. The method can effectively solve the problem that a single-mode target cannot be accurately detected due to weather, illumination, object color and the like, and the problem that the target ...

Подробнее
Номер записи: 96
25-08-2023 дата публикации

Super-resolution reconstruction method and device based on deep learning

Номер: CN116645270A
Автор: YANG PENG, CHEN ZHANGQUAN, SUN YUANKANG, LI BING
Принадлежит:

The invention provides a super-resolution reconstruction method and device based on deep learning, and the method comprises the steps: firstly putting a current frame of a short video and corresponding front and rear frames of the short video into a bidirectional optical flow prediction network to generate corresponding front and rear optical flow diagrams, and carrying out the image registration of the front and rear frames according to the optical flow diagrams, so as to effectively promote the propagation of inter-frame information; secondly, aggregating the registration frame and the current frame of the short video by using a learnable gating structure so as to fully distinguish the importance of inter-frame information; and finally, by constructing bidirectional optical flow prediction loss and super-resolution reconstruction loss, it is ensured that the generated super-resolution reconstruction result has higher image quality and time coherence.

Подробнее
Номер записи: 97
25-08-2023 дата публикации

Transform-based oil extraction violation identification method under long tail data

Номер: CN116645626A
Автор: LIANG HONG, CAO GUOQING, ZHANG QIAN, ZHONG MIN, WANG FENGHUA
Принадлежит:

The invention discloses an oil extraction violation identification method based on Transform under long tail data, and belongs to the field of computer vision, and the method comprises the following steps: collecting a video stream of an oil field construction operation site, and making an oil extraction operation site violation data set; a Grabcut-based enhancement algorithm is used to enhance the middle and tail class data in the training set; an oil extraction violation identification model based on a Transform network is constructed; designing an overall sample sampling strategy of the oil extraction violation identification model; designing a training weight distribution strategy of the oil extraction violation identification model; training and optimizing the oil extraction violation identification model; and monitoring to obtain a field work video in real time, identifying the field work video based on the trained oil extraction violation identification model to obtain a violation ...

Подробнее
Номер записи: 98
28-07-2023 дата публикации

Multi-page report structuring method and device, electronic equipment and storage medium

Номер: CN116502609A
Автор: WU KAIYUAN, XU XIAO, LI XIANG
Принадлежит:

The invention relates to the field of data processing, and discloses a multi-page report structuring method and device, electronic equipment and a storage medium, and the method comprises the following steps: carrying out report sorting on a multi-page report, identifying a textbox position in the sorted report, calculating a translation position of the textbox position, and carrying out translation operation on the textbox position by utilizing the translation position; performing report updating on the sorted report by utilizing the translation textbox, configuring an identification field of an updated report, extracting a prefix field in the updated report, and identifying a position relationship corresponding to the prefix field; performing field matching on the prefix field and the identification field; when the matching fails, updating the prefix field by utilizing the position relationship; and when the matching is successful, extracting a suffix value corresponding to the prefix ...

Подробнее
Номер записи: 99
08-08-2023 дата публикации

Synchronous localization and mapping (SLAM) algorithm, terminal and storage medium

Номер: CN116563478A
Автор: XIE NINGWEI
Принадлежит:

The embodiment of the invention discloses a synchronous localization and mapping (SLAM) algorithm, a terminal and a storage medium. The method comprises the following steps: acquiring a training data set by the terminal; performing self-supervised training processing according to the training data set and the initial network model to obtain a target network model; wherein the initial network model comprises a global attention module; the global attention module is used for learning importance degrees corresponding to different image blocks and importance degrees corresponding to different channels; performing feature extraction processing on the target image information based on the target network model to obtain a target transformation matrix; and constructing point cloud data based on the target transformation matrix.

Подробнее
Номер записи: 100