Photovoltaic Hotspot Detection Based on Improved YOLOv8l Algorithm
LIU Zewei
HUANG Yong
ZENG Xiaolong
GUAN Zhouyang
Abstract：This study addressed the challenge of detecting small and irregularly distributed photovoltaic hot spots by proposing an efficient detection algorithm based on the improved you only look once version 8 large(YOLOv8l)model.The proposed algorithm improved the neck network of the YOLOv8l model by introducing a detection layer for small-sized targets,which effectively enhanced the feature extraction capability and detection accuracy for micro-targets.Meanwhile,the traditional convolutional module in the backbone network was replaced by the receptive-field attention convolution(RFAConv),which,with the aid of the refined spatial attention mechanism,enabled the algorithm to focus more on capturing the global information of the detection targets.Furthermore,the original loss function of the model was replaced by the more effective inner-complete intersection over union(Inner-CIoU)loss function,which allowed the model to flexibly adjust the size of the auxiliary bounding boxes during the detection process,thereby improving the targeted detection accuracy.The results showed that the improved YOLOv8l algorithm achieved precision,recall,and mean average precision of 90.4%,93.2%,and 94.8%,respectively,in the detection of infrared photovoltaic hot spot images,representing an improvement of 3.4%,2.2%,and 2.3%over the original YOLOv8l algorithm.Therefore,this improved algorithm has demonstrated significant advantages in the field of photovoltaic hot spot detection.
Keywords：photovoltaic modulesYOLOv8lRFAConvsmall object detectionInner-CIoUfault detection
Publication Date：2024-09-20
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：6( 389-394 )