In the photovoltaic cell production process, due to the positive and negative circuit screen printing defects with small target, uneven distribution and other characteristics, the current detection is still time-consuming and labor-intensive. The work aims to propose a screen printing defect detection method based on the YOLOv8 optimization algorithm to solve the problem of difficult detection of photovoltaic screen printing electrode defects. Based on the theory of machine vision, a battery cell image acquisition platform was built to collect images, label and divide image data sets, and conduct batch transformation processing to enhance the data set model based on the YOLOv series of algorithm comparison experiments, indicating that the YOLOv8 algorithm was more suitable for the detection of defects in small localized targets. Then, the shuffle attention mechanism (SA) of deep learning technology was introduced into the attention module of the YOLOv8 algorithm to effectively extract feature information, replace the original feature fusion module, and finally conduct ablation experiments with the original algorithm model. The results showed that the defect recognition accuracy was improved by 4.6 percentage points. The optimized algorithm can improve the defect recognition accuracy and effectively inhibit the chances of defective cells generated by screen printing entering the subsequent industrial process.