Research on optimization method of strip surface defect detection based on YOLOv7

Aiming at the problem of low target detection accuracy in strip surface defects caused by multi-type, obscure features in the targets of interest, and large range of scale variations, this paper proposes a strip surface defect detection algorithm with improved YOLOv7. The optimization method adopts lightweight CNeB convolutional network, two-dimensional fast wavelet transform decomposition, and loss function with modulation factor focusing on the loss of difficult samples to improve the feature extraction ability of YOLOv7, realize the robustness of the detection accuracy of defects at different scales under the premise of reducing the amount of network calculation, and further enhance the detection ability of the model for difficult samples. The experimental verification of the optimization method is carried out by using the NEU-DET strip surface defect dataset, and the results show that the optimization method can effectively improve the average accuracy of strip surface defect detection while ensuring the detection speed.