Infrared fault detection of photovoltaic module based on improved YOLOv8n

The current photovoltaic module defect identification means encounter difficulties in feature extraction, and the model calculation amount is too large, these two points directly affect the accuracy of photovoltaic module defect detection. In order to solve these problems, the improved algorithm based on YOLOv8n is proposed for the detection of infrared thermal imaging defects of photovoltaic modules. First, the CBAM attention mechanism was added to the original YOLOv8n nalgorithm, and the lightweight convolutional attention module was used to improve the proportion of small target features and improve the detection performance. Secondly, the SPPFCSPC module is introduced, which effectively integrates the pooling results through convolution, thereby improving the accuracy and efficiency of feature extraction. Finally, the GhostConv convolutional layer is selected to reduce the computational load to reduce the amount of model computation. The improved algorithm’s average detection accuracy on the detection of infrared images of photovoltaic modules mAP can achieve $91.4 \%$, which is $2.7 \%$ higher than the original YOLOv8n algorithm $88.7 \%$, and the convergence effect of the accuracy and recall rate is also improved compared with the original YOLOv8n algorithm model.