Defect Detection of Photovoltaic Cells Based on SBC-YOLOv8n

The surface defects of photovoltaic cells are characterized by fine scales and significant overlap with background features, leading to frequent missed detections and false alarms in traditional detection algorithms. Moreover, the robustness of defect recognition in existing methods urgently requires improvement. This paper proposes an improved model based on YOLOv8n, termed SBC-YOLOv8n (SE-BiFPN-CA-YOLOv8n), to enhance recognition accuracy and stability in complex industrial scenarios, thereby strengthening the detection capability for minute defects on photovoltaic cell surfaces. First, a dual-attention mechanism combining coordinate attention and squeeze-and-excitation attention is integrated before and after the SPPF module in the backbone network. This enables the feature extraction ability for minute defects from two dimensions: spatial location perception and channelwise weight adjustment. Second, the original neck network structure is replaced with a weighted bidirectional feature pyramid network (BiFPN), which simplifies nodes and adopts a bidirectional weighted fusion mechanism to reinforce multi-scale feature interaction while suppressing background interference. Finally, to address class imbalance, the focal loss function is optimized by increasing the focusing parameter γ from 2 to 3 to enhance the attention on hard-to-classify samples, and adjusting the class balance factor α from 0.25 to 0.5 to substantially raise the loss weight of defect samples while reducing that of the background. The improved SBC-YOLOv8n model achieves an mAP0.5 of 80.2% on the test set, representing a 4.2 percentage point improvement over the original YOLOv8n model. Meanwhile, precision, recall, and F1 score are significantly enhanced, effectively improving the detection capability of minute defects and overall robustness while maintaining the model’s real-time performance.