北京测绘2025，Vol.39Issue(11)：1567-1573,7.DOI:10.19580/j.cnki.1007-3000.2025.11.002
基于改进DeepLabV3+的光伏板识别检测应用
Application on photovoltaic panel recognition and detection based on improved DeepLabV3+
摘要
Abstract
To improve the efficiency of photovoltaic panel resource surveys and defect detection,this study focused on a typi-cal region in Guangdong Province.By combining deep learning technology,a multi-scale photovoltaic panel recognition and detection system was developed,integrating satellite-based macro surveys and unmanned aerial vehicle(UAV)-based micro inspections.This study optimized and improved the DeepLabV3+model for semantic segmentation by incorporating deep residual network(ResNet-50)and convolutional block attention module(CBAM).It interpreted and analyzed multi-source remote sensing images to identify and extract photovoltaic panels.The results are compared with those of the original Deep-LabV3+,computer vision semantic segmentation models(U-Net),and fast semantic segmentation convolutional neural net-works(Fast-SCNN).The analysis demonstrates that the improved DeepLabV3+model has high photovoltaic panel recogni-tion accuracy,strong scene adaptability,and relatively high reliability.Additionally,the improved DeepLabV3+model is applied for intelligent detection of UAV images to analyze defects such as panel damage,scratches,hot spots,black edges,and no electricity.The results indicate that while the improved DeepLabV3+model has certain limitations in detecting minute scratch defects,its overall accuracy in photovoltaic panel defect detection is high.This provides accurate data guidance for photovoltaic power station operation and maintenance management,enhancing the informatization of the photovoltaic indus-try.关键词
深度学习/光伏板/遥感影像/无人机/缺陷检测Key words
deep learning/photovoltaic panel/remote sensing image/unmanned aerial vehicle(UAV)/defect detection分类
测绘与仪器引用本文复制引用
余德坤,易雅琴..基于改进DeepLabV3+的光伏板识别检测应用[J].北京测绘,2025,39(11):1567-1573,7.基金项目
广东省省级科技计划(2021B1212100003) （2021B1212100003）
广东省自然资源厅科技项目(GDZRZYKJ2023001). （GDZRZYKJ2023001）
