A Method for Improving the YOLOv5 Algorithm for Photovoltaic Hot Spot Detection
JIANG Chengchen
HE Jianqiang
LU Qun
WANG Jiangfeng
YIN Yuxiang
LUO Yang
Abstract：A method for improving the detection of hot spots on photovoltaic(PV)strings using an enhanced YOLOv5 algo-rithm is proposed.The presence of hot spots can lead to damage in PV string arrays.To enhance the recognition capability of un-manned aerial vehicle(UAV)inspection systems for hot spots on PV strings,the YOLOv5 algorithm is refined to improve the accu-racy and efficiency of hot spot detection.The improvement is achieved through the use of Puzzle Mix for data augmentation,which fo-cuses on small targets in the dataset image enhancement model.Additionally,a 3D non-local SimAM module is introduced into the Backbone to enhance the weight of hot spots in feature extraction,suppressing background interference weight.The CIoU(Complete Intersection over Union)loss function is employed to obtain a more precise training model and achieve high-precision localization.The enhanced algorithm is compared with other algorithms through experiments conducted on a self-made hot spot dataset.The re-sults indicate that the proposed method enhances the detection capability of hot spots on PV strings.This approach can serve as a technical reference for the inspection of PV power stations.
Keywords：YOLOv5hot spotconvolution neural networkobject detection
Publication Date：2023-10-20
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：5( 2277-2281 )