01-08-2023 дата публикации
Номер: CN116523820A
Принадлежит:
The invention requests to protect a solar cell defect detection method based on an attention convolutional neural network. The method comprises the following steps: S1, collecting solar cell panel defect images; S2, preprocessing the solar cell panel defect images; S3, randomly dividing an original solar cell defect image library obtained in S2; s4, inputting the image into a YOLOv5 network based on attention; s5, filling or scaling the original cell defect image, fixing the image size of the input network, and normalizing the image data; s6, inputting the input image into the backbone network; s7, extracting three feature maps of a large scale, a medium scale and a small scale; s8, a feature fusion step; s9, respectively predicting bounding box information and confidence information by using a convolutional layer; and S10, the trained model is used for solar cell crack and fragment defect detection under real working conditions. The defect feature expression capability is improved, so ...
Подробнее