In the field of new energy photovoltaics, efficient defect detection of Tunnel Oxide Passivated Contact solar cells is crucial for ensuring power generation efficiency and stability. Addressing the limitations of traditional manual vision and image processing techniques in terms of efficiency and accuracy under high-throughput and complex conditions, this study proposes an improved lightweight convolutional neural network model based on the YOLOv5 algorithm. The model integrates three enhancement approaches: Channel Attention plus Generative Adversarial Networks, Squeeze-and-Excitation, and Convolutional Block Attention Module to bolster the model’s capability to recognize key features. By automating feature extraction and incorporating attention mechanisms, the model significantly improves the efficiency and accuracy of defect detection in photovoltaic wafers. Experimental results demonstrate that the enhanced model excels in detecting small targets and complex backgrounds on the PVEL-AD dataset while maintaining low computational costs, indicating promising application prospects and practical value. This research not only enhances detection performance but also provides effective technical support for quality control and maintenance management in the photovoltaic industry.