Objective In the production process of electroluminescence EL materials for photovoltaic cells there are problems such as complex background large difference in defect scale and imbalance of defect categories which lead to missed and false detections in defect detection. To address these problems a photovoltaic cell defect detection algorithm based on an improved YOLOv8n is proposed. Methods Firstly the backbone of YOLOv8n was improved by replacing the C2f module with the C2f-Faster module. This change significantly reduced the number of model parameters while enhancing the model?? s detection speed and feature fusion capability. Secondly deformable convolution was introduced in the neck to increase the local receptive field thereby precisely locating defects and reducing recognition errors. Finally the wise intersection over union WIoU loss function was introduced to replace the complete intersection over union CIoU loss function of the original model. This change addressed the issue of label imbalance in the dataset and improved the model ?? s performance in detecting small-target defects. Results The improved YOLOv8n algorithm achieved an mAP@ 0. 5 of 93. 2% on the experimental dataset with a detection speed of 110 FPS and a computational load of only 6. 7 GFLOPs. Compared with the YOLOv8n benchmark algorithm the proposed method improved mAP@ 0. 5 by 2. 2%，reduced the number of parameters by 13. 3% and decreased computational load by 17. 2%. Conclusion Ablation experiments and comparisons with mainstream object detection models demonstrate the effectiveness of the improved model. The proposed model achieves higher accuracy while maintaining a lightweight structure and real-time performance. It exhibits clear advantages over other methods and meets the requirements for industrial deployment.