An Improved YOLOv5 Algorithm for Ultra-Thin Silicon Wafer Defect Detection

The photovoltaic industry has long been a focal point of national industrial policy and has received substantial support from government authorities. As a fundamental component of solar panels, photovoltaic silicon wafers are prone to various defects during manufacturing, which can significantly impair the photoelectric conversion efficiency of solar cells. In this paper, we propose an improved YOLOv5-based defect detection algorithm tailored for ultra-thin silicon wafers. Our dataset consists of 340 defect-labeled wafer images collected from production lines, encompassing three common defect types. To enhance detection performance, we integrate an SNConv module for efficient feature extraction and an SE (Squeeze-and-Excitation) attention mechanism to improve defect feature representation. The experimental results show that our improved algorithm can effectively detect defects in ultra-thin photovoltaic silicon wafers while ensuring detection speed, and is expected to be applied in practical production.