Defect Detection of Photovoltaic Modules Based on Electroluminescence Images: A RCLM-YOLO Method

• Published in: International Symposium on Autonomous Systems (Online) pp. 1 - 6
• Main Authors: Guo, Shili, Song, Mingchao, Liu, Ye
• Format: Conference Proceeding
• Language: English
• Published: IEEE 23.05.2025
• Subjects: Accuracy; attention mechanism; Attention mechanisms; auxiliary regression box; Computational efficiency; Computational modeling; Defect detection; EL image; Feature extraction; Photovoltaic cells; Solar panels; Training; YOLOv8
• ISSN: 2996-3850
• DOI: 10.1109/ICAISISAS64483.2025.11052184

In this article, a solar cell defect detection method named RCLM-YOLO, an improved YOLOv8n algorithm is proposed for the characteristics of complex solar cell image backgrounds, diverse defect morphologies, and stringent real-time detection requirements. Firstly, the backbone network of YOLOv8 is redesigned using a decoupling approach, employing reversible connections to progressively separate low-level texture details from high-level feature information. Secondly, a Local Window Attention mechanism is introduced at the low-level layers to suppress background noise, while the SimAM module is integrated at the high-level layers to enhance the discrimination ability of high-level features. Finally, an Inner-CIoU auxiliary regression box is adopted to improve the learning capability of the model for medium-quality samples. Experimental results show that RCLM-YOLO achieves an improvement of 3.0% and 4.4% in mAP@0.5 and mAP@0.5:0.95, respectively, on the PVEL-AD dataset compared to YOLOv8n, reaching 92.7%. Additionally, the parameter count and computational cost are reduced by 25.6% and 15.9%, respectively.