基于无人机巡检的光伏面板缺陷识别方法  

Photovoltaic panel defect identification method based on UAV inspection

作　　者：韩虎虎 祁鑫 王鹤飞 李惠翔 齐吉祥 HAN Huhu;QI Xin;WANG Hefei;LI Huixiang;QI Jixiang(Safety Production Department,Guodian Siziwangqi Photovoltaic Power Generation Co.,Ltd.,Ulanqab 011800,China)

机构地区：[1]国电四子王旗光伏发电有限公司安全生产部,内蒙古乌兰察布011800

出　　处：《电子设计工程》2023年第6期176-179,184,共5页Electronic Design Engineering

摘　　要：在光伏发电过程中,为解决因迭代步数快速增加而引发的交叉熵过量损失问题,提出基于无人机巡检的光伏面板缺陷识别方法。根据适宜机型选择结果,规划无人机巡检路径,通过计算维度灾难系数值的方式,实现基于无人机巡检的光伏面板缺陷量提取。在此基础上,分别定义光伏面板的纹理特征与颜色特征,根据核函数构造原理,完成光伏面板的缺陷表现行为识别。实例分析结果表明,与YOLOv2网络型诊断模型相比,无人机巡检识别方法可在抑制迭代步数快速上升行为的同时,控制交叉熵损失量数值,有效避免了过量损失行为的出现。In the process of photovoltaic power generation,in order to solve the problem of excessive cross entropy loss caused by the rapid increase in the number of iteration steps,a photovoltaic panel defect identification method based on drone inspection is proposed. According to the selection results of suitable models,the UAV inspection path is planned,and the photovoltaic panel defect extraction based on the UAV inspection is realized by calculating the value of the dimensional disaster coefficient. On this basis,the texture characteristics and color characteristics of photovoltaic panels are defined respectively,and the defect performance behavior identification of photovoltaic panels is completed according to the principle of kernel function construction. The example analysis results show that compared with the YOLOv2 network-based diagnosis model,the UAV inspection and identification method can suppress the rapid increase in the number of iteration steps while controlling the value of the cross-entropy loss,effectively avoiding the occurrence of excessive loss behavior.

关 键 词：无人机巡检 光伏面板 缺陷识别 维度灾难系数 纹理特征 颜色特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]