基于面向对象的光学和Sentinel—1 SAR数据油菜种植空间分布信息提取方法研究

The Extraction Method of Rape Oil Planting Distribution Using Per-Field Classification Based Optical and Sentinel-1 SAR Images Data

为提高Sentinel—1 SAR数据作物种植分布提取精度,以湖北省江陵县为研究区域,运用资源三号卫星CCD融合数据(空间分辨率为2 m)提取田间边界对象,将对象内Sentinel—1 SAR后向散射系数取平均,以此消除相干斑点噪声的影响,再通过对各种地物纯像元SAR后向散射特征分析,发现3—4月油菜后向散射系数明显高于其他作物,运用得出的油菜分类阈值(4月22日VH极化的SAR后向散射系数大于2.1且小于3.5,且2月27日NDVI指数大于0.3),对满足条件的对象进行筛选,最后得出江陵县油菜种植空间分布信息。运用GPS定点对分类结果进行验证,得到其KAPPA系数为0.88,并运用其他两种传统分类方法(直接用SAR数据进行阈值分类、运用其他时段SAR数据进行对象提取)进行了比较,发现光学遥感数据提取对象、SAR数据确定对象属性的油菜空间分布提取方法的精度有一个质的提高。由于SAR数据不受云层影响,能定时获取,因此此方法很适合在多云雨地区的作物种植空间分布信息的提取。

The aim of this research was to improve the accuracy of crops distribution classification using Sentinel-1 SAR data.The method involves extracting the crop field border using ZY-3 satellite multi-band optical data based on image segment and merge methods,then averaging the RADAR backscatter coefficient of SAR data within each object which eliminates the influence of coherent speckle noise.Through the analysis of SAR backscattering characteristics of various ground objects which created pure pixels(identified by ground investigating using GPS),we found that the backscattering coefficient values were higher than other crops in March and April.We determined that the rape oil plant areas were characterized with SAR backscattering coefficient values being more than 2.1 and less than 3.5,and the NDVI index being greater than 0.3 on 27th February(filtered for the nonplanting areas).Using these rules and SAR data that eliminated the influence of coherent speckle noise,we calculated rape oil planting spatial distribution in Jiangling County.The classification results were verified using GPS within a 5.35 km2(length 2.57 km,width 2.08 km)area which had wheat and rape oil planted,and the KAPPA coefficient was 0.88.The accuracy of the spatial distribution of rape oil plants was improved over the use of traditional classification methods.The SAR data from Sentinel-1 is not affected by clouds,therefore data needed for this method can be obtained regularly.This method is suitable for crop cultivation spatial distribution information needed for business operations.