基于高分卫星遥感图像的土地利用分类技术研究  

Research of land use classification technology based on high-resolution satellite remote sensing images

作　　者：赵永娇 滑申冰 丁禹 刘晓琳 姚德贵 李哲 李鑫慧[1] ZHAO Yongjiao;HUA Shenbing;DING Yu;LIU Xiaolin;YAO Degui;LI Zhe;LI Xinhui(Nanjing University of Information Science and Technology,Nanjing 210044,Jiangsu,China;China Electric Power Research Institute,Beijing 100192,China;State Grid Henan Electric Power Company Electric Power Science Research Institute,Zhengzhou 450000,Henan,China)

机构地区：[1]南京信息工程大学,江苏南京210044 [2]中国电力科学研究院有限公司,北京100192 [3]国网河南省电力公司电力科学研究院,河南郑州450000

出　　处：《水利水电技术(中英文)》2024年第S2期279-284,共6页Water Resources and Hydropower Engineering

基　　金：国家电网有限公司总部科技项目资助(5500-202324180A-1-1-ZN)

摘　　要：土地覆盖分类能够反映城市发展中潜在的自然和社会过程。Sentinel-2A/B卫星有着较高的时间、空间和光谱分辨率,选择2018年11月—2019年11月吉林省舍力镇该卫星数据研究基于多时相的土地覆盖分类方法。首先根据植被和非植被的特性,分析归一化差值植被指数(Normalized Difference Vegetation Index,NDVI)时间序列数据的植被物候信息,缨帽变化湿度分量(Tasseled Cap Wetness,TCW)时序数据的非植被地物特征;然后根据NDVI和TCW时序序列曲线在典型地物下的时间特征信息,得到最佳时间段,构建NDVI和TCW时间序列数据。在此基础上,利用监督分类方法(最大似然法和支持向量机法)对土地覆盖类型进行分类。最大似然法结合时间序列数据在吉林省舍力镇区域土地覆盖分类总体精度达到了82.39%,kappa系数为0.78,支持向量机法结合时序数据的分类方法在吉林省舍力镇区域的土地覆盖分类总体精度达到了88.24%,kappa系数为0.83。研究结果表明时间序列数据对土地覆盖类型分类结果有着良好的效果。Land cover classification can reflect the underlying natural and social processes in urban development.Sentinel-2A/B satellite has high temporal,spatial and spectral resolution.This paper selects the satellite data of Sheri Town,Jilin Province from November 2018 to November 2019 to study the multi-temporal land cover classification method.In this paper,according to the characteristics of Vegetation and non-vegetation,vegetation phenological information of Normalized Difference Vegetation Index(NDVI)time series data was analyzed.The characteristics of non-vegetation feature in Tasseled Cap Wetness(TCW)time series data;Then,according to the time characteristic information of NDVI and TCW time series curves under typical ground objects,the best time period is obtained,and the NDVI and TCW time series data is constructed.On this basis,supervised classification method(maximum likelihood method and support vector machine method)were used to classify land cover types.The overall accuracy of land cover classification by maximum likelihood method combined with time series data reached 82.39%,and the Kappa coefficient was 0.78.The overall accuracy of land cover classification by support vector machine method combined with time series data reached 88.24%,and the Kappa coefficient was 0.83.The result of this paper show that time series data has a good effect on land cover type classification.

关 键 词：sentinel-2A/B 时间序列 土地覆盖类型分类 最佳时序数据组合 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]