摘要
对无人机采集的光伏面板图像进行准确的分割提取,是提升光伏组件故障检测精度的前提。针对光伏面板红外图像的分割问题,首先对语义分割网络DeepLabV3+的空洞卷积率进行优化并引入深度可分离膨胀卷积,使模型进一步捕获全局和上下文信息;然后,设计了基于坎尼边缘检测算法和线段检测算法的边缘特征提取模块,获得细化的光伏面板边缘作为分割网络的补充特征,并通过四通道融合网络和并行融合网络实现了光伏面板的精确分割。实验结果表明,2种融合网络对光伏面板红外图像的分割精度高于DeepLabV3+,并且对不同场景下的光伏面板红外图像均能实现准确分割。
Accurate segmentation and extraction of photovoltaic panel images collected by the unmanned aerial vehicle(UAV)is a prerequisite for improving the fault detection accuracy of photovoltaic modules.To solve the segmentation problem of infrared images of photovoltaic panels,the cavity convolution rate of semantic segmentation network DeepLabV3+is optimized and depthwise separable dilated convolution is introduced to make the model further capture global and contextual information.Then,an edge feature extraction module based on Canny edge detection algorithm and line segment detector(LSD)is designed to obtain the refined edge of the photovoltaic panel as the supplementary feature of the segmentation network,and the accurate segmentation of the photovoltaic panel is achieved through the four-channel fusion network and the parallel fusion network.The experimental results show that the segmentation accuracy of the two fusion networks for infrared images of photovoltaic panels is higher than that of DeepLabV3+,and they can achieve accurate segmentation of infrared images of photovoltaic panels in different scenes.
作者
沈灵鑫
王银
李杰
李茂环
李小松
SHEN Lingxin;WANG Yin;LI Jie;LI Maohuan;LI Xiaosong(School of Electronic Information Engineering,Taiyuan University of Science and Technology,Taiyuan 030024,China;School of Software,Beihang University,Beijing 100191,China;Nanjing Dirui Technology Co.,Ltd.,Nanjing 211500,China)
出处
《控制工程》
北大核心
2025年第4期707-719,共13页
Control Engineering of China
基金
国家自然科学基金资助项目(61905172)
山西省科技成果转化引导专项项目(201904D131023)
山西省重点研发计划项目(201903D121130)
山西省研究生教育创新项目(2020SY422)。
