基于J-M距离的多时相Sentinel-1农作物分类

为准确获取云雨天气较多地区的农作物分布信息,以敦化市西北部的黑市乡和额穆镇为研究区,以高分一号(GF-1)卫星宽视场(wide field view, WFV)传感器和哨兵一号(Sentinel-1)为数据源,通过WFV数据提取耕地范围,利用J-M(Jeffries-Matusita)距离可分性进行Sentinel-1影像最优分类时相组合的判断,根据选取Sentinel-1数据通过最近邻分类器多次循环对研究区主要农作物的分布信息进行提取。研究结果表明,进行多次循环产生的分类结果优于单次循环结果,循环次数达到5次结果趋于稳定,总体分类精度为84.23%。可见利用Sentinel-1数据进行敦化市农作物分类是可行的,多次循环的最近邻算法有利于获取更精准的农作物分布信息。

Time-series classification of Sentinel-1 data based on J-M distance

In order to accurately obtain the distribution information of crops in areas with more clouds and rains, this paper takes Heishi Township and Emu Town in the northwest of Dunhua City as the research area, and uses GF-1 WFV and Sentinel-1 as the data source, extracts the range of cultivated land through WFV data,uses J-M distance separability to judge the optimal classification time-phase combination of Sentinel-1 image. According to the selection of Sentinel-1 data, the distribution information of the main crops in the study area is extracted by the nearest neighbor classifier multiple cycles. The results show that the classification results produced by multiple cycles are better than the single cycle results. The number of cycles reaches 5 times and the results tend to be stable, and the overall classification accuracy is 84.23%. It can be seen that it is feasible to use Sentinel-1 data to classify crops in Dunhua City.The multi-cycle nearest neighbor algorithm is beneficial to obtain more accurate crop distribution information.