Defect detection of solar cells based on improved YOLOv5s

This study proposes an improved lightweight YOLOv5s neural network model for efficiently identifying various defects on the surface of solar cells. Firstly, ShuffleNetv2 is used as the backbone feature extraction network in the YOLOv5s network. Secondly, the Triplet Attention attention mechanism is introduced into the backbone network of YOLOv5s.Lastly,the two-dimensional activation function FReLU is used to replace Leaky ReLu. The experimental results show that the improved YOLOv5s model has an average accuracy value of 94.1%, ensuring the detection accuracy of defect targets; At the same time, the floating-point operation amount was reduced by 78.7%, and the model size was reduced by 64.8%, effectively improving the model's lightweight performance.