Answers are grounded in this paper and cite their sources.
Algorithm improves defect detection accuracy to 94.2% in photovoltaic panels, highlighting advancements in dynamic convolution and loss function integration.
Aiming at the problems of insufficient accuracy and large parameter amount of the existing YOLO model in defect detection, an improved YOLOv8s algorithm was proposed for photovoltaic panel defect detection.On the basis of retaining the backbone network structure, the full-dimensional dynamic convolutional ODConv is introduced, ODConv is combined with C2f, and the C2f_ODConv module is constructed to replace C2f of the neck small target layer, which effectively improves the feature extraction ability of small targets. At the same time, the TripletAttention mechanism is embedded to enhance the model's attention to subtle defects.Finally, we introduce the Focaler-IoU loss function. It integrates Focal Loss and dynamically weights difficult samples to improve bounding box regression accuracy.The results show that the mAP of the improved model is increased from 90.6% to 94.2%, and the calculation amount is reduced to 8.8GFLOPs, especially in the "cover" defect detection, the mAP is increased by 9.1%. Compared with other mainstream algorithms, the proposed method has advantages in accuracy and speed, and is suitable for automatic inspection and defect identification tasks of photovoltaic power stations.
No takes yet. Share an insight, caveat, or question.
Song et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: