Solar Cell Surface Defect Detection Based on Improved YOLO v5

A solar cell defect detection method with an improved YOLO v5 algorithm is proposed for the characteristics of complex image background, variable morphology, and large-scale differences. First, deformable convolution incorporated into CSP module to achieve adaptive learning scale perceptual field size; then, feature extraction capability model enhanced by introducing ECA-Net attention mechanism; finally, network structure one tiny prediction head added improve accuracy target at different scales. To further optimize algorithm, this paper uses Mosaic MixUp fusion data enhancement, K-means++ clustering anchor box CIOU loss function enhance performance. The experimental results show that achieves 89.64% mAP trained on EL dataset, which 7.85% higher than original speed reaches 36.24 FPS, can complete task more accurately while meeting real-time requirements.