基于多时相Sentinel-1水稻种植范围提取

Extraction of paddy fields using multi-temporal Sentinel-1 images

通过遥感手段监测与提取水稻种植范围的是农业现代化管理的重要手段,南方地区春夏季节多云雨天气,难以获得有效的光学数据。为了准确提取多云雨地区的水稻种植范围,以江西省南昌县蒋巷镇为研究区,以Sentinel-1多时相数据为数据源开展研究。通过早稻生长关键物候期Sentinel-1 SAR数据各地类后向散射系数,计算不同物候期条件下水稻田与其他地类的J-M距离,分析不同物候期SAR数据组合情况下水稻田与其他地类的J-M距离变化,获得提取水稻种植范围的最佳物候期影像。分别采用随机森林法、最大似然法、支持向量机和神经网络法进行分类,并对比验证分类精度。结果表明,孕穗期(6月14日)、三叶期(4月21日)、移栽期(5月3日)、二季晚稻移栽期(7月26日)组合SAR数据为水稻田提取最优时相组合。采用随机森林方法进行分类能够获得较高精度,研究区地物分类总体精度达到0.943,Kappa系数0.932。研究对于多云雨地区采用SAR数据开展水稻田制图,在时相选择和分类方法上有一定的借鉴意义和参考价值。

The monitoring and information extraction of paddy fields using remote sensing techniques is an important means for modern agricultural management.However,it is difficult to obtain effective optical monitoring data of south China due to the frequent cloudy and rainy weather in spring and summer in this area.To accurately extract information on paddy fields in areas subject to frequent cloudy and rainy weather,this study investigated the paddy fields in Jiangxiang Town in Nanchang County,Jiangxi Province,using multi-temporal Sentinel-1 SAR data as the data source.Specifically,this study calculated the J-M distance between paddy fields and other land types in different phenological periods,analyzed the changes in the distance based on the backscattering coefficients of various land types in key phenological periods,and then obtained the best phenological images for the information extraction of paddy fields.Afterward,this study conducted ground object classification using methods such as random forest,maximum likelihood,support vector machine,and neural network and then compared and verified the classification accuracy.The results are as follows.The combined SAR data of the different stages including booting stage(June 14),trefeil stage(April 21),transplantion period(May 3),and transplanting peried of second season late rice(July 26)is the optimal temporal combination for the information extraction of paddy fields.Higher classification accuracy of ground objects in the study area can be obtained using the random forest method,with overall classification accuracy of up to 0.943 and a Kappa coefficient of 0.932.This study conducted the mapping of paddy fields in areas with frequent cloudy and rainy weather using SAR data and will provide important references for the temporal selection and classification.