基于FMF-YOLOv5的光伏组件红外图像故障诊断  

Infrared Image Fault Diagnosis of Photovoltaic Modules Based on FMF-YOLOv5

作　　者：张莉莉 王修晖 ZHANG Lili;WANG Xiuhui(Key Laboratory of Electromagnetic Wave Information Technology and Metrology of Zhejiang Province,College of Information Engineering,China Jiliang University,Hangzhou 310018,China)

机构地区：[1]中国计量大学信息工程学院浙江省电磁波信息技术与计量检测重点实验室,杭州310018

出　　处：《计算机工程与应用》2025年第2期327-334,共8页Computer Engineering and Applications

基　　金：浙江省自然科学基金(LY20F020018);浙江省重点研发计划(2021C03151)。

摘　　要：针对红外图像对比度较低、故障特征不明显的问题,提出全新的融合注意力机制(fusion attention mechanism,FAM),增强有效故障特征信息。创建新的融合金字塔池化(fusion spatial pyramid pooling,FSPP),增强特征提取能力。引入一种改进多层次融合卷积(multi-level fusion convolution,MFConv),利用MFConv构建的多层次跨阶段局部网络(multi-level cross stage partial network,MCSP)模块代替CSP模块,在提高少量模型参数量情况下,增加模型检测准确性。实验结果表明,在IoU阈值为0.5的情况下,该方法的平均精度(mAP)达到了93.1%。为光伏系统提供了可靠、高效的故障检测解决方案,从而使其成为提高系统性能和降低维护费用的实用解决方案。Aiming at the problems of low contrast of infrared image and not obvious fault characteristics,the paper firstly proposes a new fusion attention mechanism(FAM)to focus on important fault characteristics.Secondly,a new fusion spa-tial pyramid pooling(FSPP)is created to enhance feature extraction capabilities.Finally,a new multi-level fusion convolu-tion(MFConv)is introduced,and a multi-level cross stage partial network is built by using MFConv.The MCSP module replaces the CSP module to maintain accuracy while increasing the number of model parameters with a small amount.The experimental results show that the average accuracy(mAP)of the proposed method reaches 93.1%when the IoU thresh-old is 0.5.This method provides a reliable and efficient fault detection solution for photovoltaic systems,making it a prac-tical solution to improve system performance and reduce maintenance costs.

关 键 词：目标检测 光伏故障 特征融合 融合注意力 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]