The complex background interference in electroluminescence(EL) images of photovoltaic modules poses challenges for accurate defect identification. To address this
an improved deep learning model YOLOV5-PMD based on YOLOv5 is proposed. This model utilizes a lightweight convolution GSConv module to replace the Conv part of the Neck
enhancing accuracy while reducing the number of parameters. In addition
the convolutional CBAM attention mechanism is combined to improve the feature perception ability
and the EIoU loss function is replaced by the CIoU loss function to achieve more accurate defect location. On the public SolarMonocrystal dataset
the improved model reduces the training parameters from 7.025×10 to 6.695×10
and the average accuracy of mAP50 reaches 89.8%
which is 2.6% higher than that of the original model.
LIANG L, WANG X B, LI X F, et al.Research on fault object detection method for photovoltaic panel UAV inspection based on YOLOv5[C]//2023 9th International Conference on Computer and Communications(ICCC). Chengdu, China, 2023: 1766-1770.
CONG Z H, SUN H K, WANG S J, et al.YOLOv5-CPP: improved YOLOv5-based defect detection for photovoltaic panels[C]//2023 42nd Chinese Control Conference (CCC), Tianjin, China, 2023: 8294-8299.
GUAN Q J, YU K, WANG H R, et al.Lightweight protective clothing detection algorithm based on ghost convolution and GSConv convolution[C]//2023 International Conference on the Cognitive Computing and Complex Data (ICCD). Huaian, China, 2023: 73-77.