多源遥感精确提取光伏电站研究  

Research on Accurate Extraction of Photovoltaic Power Station from Multi-source Remote Sensing

作　　者：王卫[1,2,3,4] 蔡俊兴 田广增 成量 孔德庸 WANG Wei;CAI Junxing;TIAN Guangzeng;CHENG Liang;KONG Deyong(Yingdong College of Biology and Agriculture,Shaoguan University, Shaoguan Guangdong 512005, China;North Guangdong Soil and Land Research Center, Shaoguan University, Shaoguan Guangdong 512005, China;College of Natural Resources and Environment, South China Agricultural University, Guangzhou Guangdong 510640, China;College of Tourism and Geography, Shaoguan University, Shaoguan Guangdong 512005, China;Tourism and Urban Management College, Jiangxi University of Finance and Economics, Nanchang Jiangxi 330013, China)

机构地区：[1]韶关学院英东生物与农业学院,广东韶关512005 [2]韶关学院粤北土壤土地研究中心,广东韶关512005 [3]华南农业大学资源环境学院,广东广州510642 [4]韶关学院旅游与地理学院,广东韶关512005 [5]江西财经大学旅游与城市管理学院,江西南昌330013

出　　处：《北京测绘》2021年第12期1534-1540,共7页Beijing Surveying and Mapping

基　　金：广东省自然科学基金(2018A030307075);广东省科技创新战略专项(粤科函规财字〔2018〕1523号);韶关市科技计划(201644)。

摘　　要：光伏是主要的清洁能源之一,对光伏电站的监测是获取光伏电站运行、分布和变化的重要途径。以哨兵2号(Sentinel 2)和高分2号(GF-2)卫星影像为数据源,建立光伏电站识别指数,构建融合指数特征和纹理特征的多特征多尺度的图像,利用深度学习方法识别不同本底环境的光伏电站,结合空间关联方法精确提取光伏电站的面积。结果表明,利用深度学习方法对多特征多尺度图像的识别精度达到了95%,结合空间关联方法提取光伏电站的面积精度达到了97%,分别比多光谱图像和多特征图像提高了11.8%和6.5%。基于多特征多尺度的图像,利用深度学习方法的光伏电站监测,可为管理部门提供准确的分布和变化数据,并丰富了我国高分系列遥感的应用。Photovoltaic(PV)is one of the main clean energy.The monitoring of PV power station is an important way to obtain its operation,distribution and change.In this paper,based on Sentinel 2 and GF-2 satellite images,a photovoltaic power station identification index was established,and a multi-scale image with texture features was constructed.PV power stations in different background environments were identified by deep learning method,and the area of PV power station was accurately extracted by spatial correlation method.The results showed that the recognition accuracy of multi feature and multi-scale image was 95%by using deep learning method,and 97%by combining spatial correlation method,which was 11.8%and 6.5%higher than that of multi spectral image and multi feature image,respectively.Based on the multi feature and multi-scale image,the PV power station monitoring using deep learning method could provide accurate distribution and change data for the management department,and enrich the application of high resolution series remote sensing in China.

关 键 词：光伏电站 多特征提取 多尺度分割 多源遥感 深度学习 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置]