Solar Panel Defect Detection Based on Improved YOLOv4

An improved YOLOv4 detection model is proposed to address the issues of low accuracy, slow detection speed and large model volume in defect detection of solar panels. Firstly, GhostNet is used to replace CSPLocknet-53 in YOLOv4 to achieve model lightweight. Secondly, introducing Depthwise Separable Convolution convolution into the model structure further reduces model parameters and improves model speed. Thirdly, an improved Efficient Channel Attention(ECA) mechanism is introduced into the model to improve detection accuracy. Finally, the S-T-ReLU activation function is used to replace the ReLU activation function in the original YOOv4 to further improve the detection accuracy. The results showed that the improved model had better detection performance. Compared with the original model, the mAP increased by 1.06%, the Floating Point Operation Per Second (FLOPs) decreased by 89.11%, the model volume decreased by 82.61%, the model parameter quantity decreased by 82.77%, and the Frames Per Second (FPS) increased by 35.34%, proving the effectiveness of the proposed algorithm.