Solar Cell Defect Detection Algorithm Based On Improved YOLOv5s

Among many renewable energy sources,solar energy,as the only TW-level energy source,has received widespread attention and application.Solar cells are an important carrier of solar power generation.The defects,such as broken cells,cracks,finger failures,and silicon material defects might be found in the solar cells due to production process defects or human operation errors during the production and preparation process.These defects affect the life of solar cells and photoelectric conversion efficiency,so it is important to detect them before they leave the factory.Research on defect detection methods for solar cells has important scientific and engineering practical significance.Defect detection of solar cell surface by traditional methods is prone to problems such as inspection,leakage and improper positioning.However,with the development of artificial intelligence and machine learning technology,deep learning algorithms have been widely used in the defect detection of solar cells,and good detection results have been achieved.YOLO algorithm is one of the excellent target detection algorithms.It combines target location and target category prediction in its structure,achieves high accuracy and performance of target detection,and is more suitable for practical application.To make the defect detection of solar cells more accurate,based on the YOLOv5 algorithm,this paper improves the network structure,optimizes the network parameters,and becomes more accurate in detecting the defect of solar cells.The specific changes are as follows: First,the structure of the original model is optimized by adding an attention mechanism and improving its loss function to improve the detection accuracy of defects;Secondly,Bottleneck Transformer is used to replace the original C3 module.On the basis of adding attention mechanism to the original model and improving the loss function,the number of layers,network parameters,calculation amount and training time of the improved model are basically unchanged,but the detection accuracy is significantly improved,and the defect detection effect of solar battery image is also improved.The results show that the m AP value of the optimized algorithm is 76.7%,the accuracy of detecting defect-free solar batteries is93.1%,the m AP is 4.2% higher than the original YOLOv5 s network,and the mean accuracy(m AP)of the whole class is 13.7% higher than the main target detection algorithm SSD.Experiments show that this algorithm can detect solar cell defects more accurately,and provide a better detection algorithm for actual production.