Key points are not available for this paper at this time.
Ensuring the safe and efficient operation of solar photovoltaic (PV) cells is critical for solar energy collection systems. Surface damage to PV cells can cause significant economic losses and must be detected and resolved promptly. To overcome the limitations of existing algorithms, including loss of critical boundary information, difficulties in detecting small targets, and inefficient multi-scale feature fusion, we propose the MFL-YOLO model. This novel approach integrates the C3k2MEIS module to enhance the extraction of key features and significantly differentiate between foreground and background regions. To address challenges in multi-scale feature fusion for detecting multiple defect types, we introduce the FDPN framework, which substantially improves multi-scale fusion capability. Additionally, our lightweight LSCBD head reduces parameter count and computational requirements while enhancing small target detection performance. Experimental results demonstrate that MFL-YOLO achieves 37. 1% AP₅₀ on our proprietary dataset, outperforming the original YOLO11n baseline. Specifically, precision, AP₅₀, and AP₅₀–₉₅ are improved by 16. 0%, 7. 8%, and 10. 1%, respectively. Furthermore, our model demonstrates exceptional performance on both the public Kaggle dataset and the PVEL-AD dataset. Comparative analysis shows that MFL-YOLO outperforms other mainstream detection models, highlighting its substantial potential for practical PV cell defect detection applications.
Luo et al. (Tue,) studied this question.