基于多时相Sentinel-1影像的江南地区水稻信息提取  

Rice Information Extraction in the Southern Yangtze River Region Based on Multi-temporal Sentinel-1 Images

作　　者：罗福正 杨柳 卢彬 黄永生 LUO Fuzheng;YANG Liu;LU Bin;HUANG Yongsheng(Qinghai Geospatial and Natural Resources Big Data Center,Xining 810000,China;Qinghai Provincial Key Laboratory of Geospatial Information Technology and Application,Xining 810000,China)

机构地区：[1]青海省地理空间和自然资源大数据中心,青海西宁810000 [2]青海省地理空间信息技术与应用重点实验室,青海西宁810000

出　　处：《地理空间信息》2024年第11期73-77,86,共6页Geospatial Information

基　　金：国家自然科学基金面上项目(2020-ZJ-927)。

摘　　要：为准确提取多云雨江南地区水稻种植信息,以江苏省丹阳市访仙镇为研究区,以多时相Sentinel-1数据为主、Sentinel-2数据为辅开展研究。通过计算水稻关键生育期内不同地物与水稻间的J-M距离,引入水稻移栽期内的归一化水体指数(NDWI)和收割完成后的归一化裸土指数(NDBI),获取水稻种植范围的最优特征组合;再以K近邻、随机森林和支持向量机(SVM)为基础建立分类模型,并对比验证了最优特征组合下不同模型的分类精度。结果表明,引入NDWI和NDBI的分蘖拔节期、孕穗抽穗期的VV极化后向散射系数以及孕穗抽穗期、乳熟成熟期VH极化后向散射系数为水稻提取的最优组合,采用SVM算法建立的分类模型的水稻识别精度最高,总体精度为0.916,Kappa系数达到0.828。To accurately extract rice cultivation information in the cloudy and rainy southern Yangtze River region,taking Fangxian Town,Danyang City,Jiangsu Province as the research area,we primarily utilized multi-temporal Sentinel-1 data to carry out research while supplementing with Sentinel-2 data.We computed the J-M distance between different terrain objects and rice during critical growth stages,and introduced the normalized difference water index(NDWI)during the rice transplanting period and the normalized difference bare soil index(NDBI)after harvesting to obtain the optimal feature combination for identifying rice cultivation areas at first.And then,we built classification models based on the K nearest neighbors,random forest and support vector machine(SVM)algorithms.Finally,we compared and validated the classification accuracy of different models under the optimal feature combination.The results reveal that the feature combination incorporating NDWI and NDBI during the tillering and heading stages in VV-polarized backscatter coefficients,as well as during the heading and maturing stages in VH-polarized backscatter coefficients,provide the optimal configuration for rice extraction.SVM algorithm achieved the highest accuracy in rice identification,with an overall accuracy of 0.916 and Kappa coefficient of 0.828.

关 键 词：水稻提取 水稻生育期 SVM Sentinel-1 Sentinel-2 

分 类 号：P237[天文地球—摄影测量与遥感]