09-06-2023 дата публикации
Номер: CN116245860A
Принадлежит:
The invention provides a small target detection method based on a super-score-yolk network, which takes a yolk network as a basic network architecture, is combined with a super-score module, adds a position attention mechanism and a self-attention mechanism, and introduces a recursive cavity convolution module, and comprises the following steps: 1) taking a super-score module as the basic network architecture; the method comprises the following steps: 1, acquiring an image and preprocessing the image to generate an original image; step 2, a super-resolution fitting network is connected with an improved yolk target recognition network to form a super-resolution-yolk network model, an original image is processed, a super-definition image with a higher resolution is output, operation such as feature extraction, target frame regression classification and object detection recognition is performed on the super-definition image, a recognized image is obtained, and a data set for training is formed ...
Подробнее