Foreign object detection on photovoltaic panels based on DHLNet

The accumulation of foreign objects on photovoltaic panels can significantly reduce their photoelectric conversion efficiency and may cause hotspot effects, leading to module aging and permanent damage, which threatens the safe and economic operation of PV power plants. To address challenges in foreign object detection, such as low image contrast, complex morphology, and sample imbalance, this study proposes the DHLNet model. The model is based on an innovative Dual-Flow Feature Pyramid Network, which improves the accuracy of small object detection in complex backgrounds through cross-stage information interaction and multi-level feature reconstruction, while maintaining a lightweight structure. To further enhance performance, a high-frequency enhancement module is integrated into the backbone network to increase the model's sensitivity to targets with blurred edges. In addition, a large-separable-kernel attention mechanism is introduced to enhance the extraction capability of key features. Experimental results show that DHLNet achieved [email protected] and [email protected]:0.95 scores of 79.9% and 58.8%, respectively, representing improvements of 3.1% and 2.2% compared with the original YOLO11n algorithm. The F1 score reached 0.761, an improvement of 1.3%. The model also supports real-time inference, making it suitable for on-site detection and online monitoring. These results verify the effectiveness and practical application potential of DHLNet in automated foreign object detection on photovoltaic panels.