FS-YOLO: a lightweight algorithm for efficient defect detection of photovoltaic panels

Engineering notes

Engineering notes will be added by the Power for Solar editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为光伏效率、钙钛矿太阳能电池、储能技术、太阳能热利用、BIPV、并网技术等高价值论文补充中文说明。

Original abstract

Efficient and accurate defect detection in solar cells is essential for maintaining the operational reliability of photovoltaic systems, yet existing deep learning models often encounter a bottleneck in balancing computational efficiency with the precise capture of fine-grained geometric features. To address these challenges, this paper proposes an improved lightweight defect detection model based on the YOLO11 architecture. The proposed method incorporates a C3k2_Faster module into the backbone to minimize computational redundancy through Partial Convolution (PConv), while a Dynamic Snake Convolution (DySnakeConv) block is integrated into the detection head to enhance the model's sensitivity to the complex geometric morphologies of micro-cracks. Experimental results on a specialized solar cell dataset demonstrate that the proposed model achieves a parameter count of 2.28M and a computational complexity of 5.4 GFLOPs, both of which represent a reduction compared to the baseline YOLO11n's 2.58M parameters and 6.3 GFLOPs. Despite the lightweight architecture, the model maintains a high inference speed of 194.59 FPS and demonstrates competitive performance in core categories, particularly achieving an mAP50 of 0.7461 for the challenging "cracked" defect class. This research provides an efficient and robust solution for deploying real-time defect inspection systems on resource-constrained edge devices in industrial environments.

Links and sources

Comments