Year: 2026
Type: article
Abstract: To tackle the trade-off between lightweight deployment and high precision for photovoltaic panel defect detection, this paper proposes a lightweight photovoltaic panel defect detection model, LPD‑YOLO. This model performs deep compression and optimization based on YOLOv8. It embed Ghost HGBlock into the backbone and adopt GSConv and VoVGSCSP in the detection head to cut model parameters. Additionally, an adaptive scale fusion attention module is integrated to achieve weighted fusion of multi-scale features by combining Zoom-Cat scale alignment, ScalSeq serialization and reorganization, and multi-branch attention, thereby enhancing the model’s feature extraction capability for small targets and complex defects. Finally, the model is compressed and fine-tuned using network slimming and hybrid knowledge distillation, significantly reducing computational overhead while preserving detection accuracy. Experimental results on the GB_HSP_ modified dataset show that the LPD‑YOLO model has only 0.44 M parameters and 0.9 GFLOPs computational complexity, while maintaining a competitive detection accuracy (mAP50) of 76.8%. Compared to mainstream lightweight models, its parameters and computational cost are reduced by more than 80%. It outperforms baselines including Faster R-CNN and YOLOv6–v12 with a well-balanced trade-off among accuracy, speed, and model lightweighting, demonstrating strong potential for efficient photovoltaic panel defect detection on resource-limited devices.
Source: Scientific Reports
Authors: Yuwen Liu, Shunxin Cao, Mengting Liu
Language: English
Subfield: Industrial and Manufacturing Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7222440739