基于超像素统计量的随机森林遥感图像分类

Research on random forest remote sensing image classification based on superpixel statistics

针对遥感图像地物覆盖分类方法对图像空间分布信息利用不足的问题,提出一种基于超像素统计量的随机森林遥感图像分类方法。以北京市海淀区为研究区,选用Landsat-8卫星为主要数据源,通过改进SLIC超像素分割方法,使之适用于多光谱遥感图像中超像素的分割,提取超像素常见的六个统计量(最小值、最大值、均值、标准差、上四分位数、下四分位数)用于随机森林在遥感图像中的分类。实验结果表明,其对研究区遥感图像的总体分类精度为89. 01%,明显改善了对地物的错分和漏分现象,能够推广到Landsat-8遥感图像的地物覆盖分类工作中。

In order to solve the problem of insufficient utilization of the spatial distribution information in remote sensing image classification,this paper proposed a random forest remote sensing image classification method based on superpixel statistics.Taking Haidian district,Beijing as study area and Landsat-8 OLI（ operational land imager） image as the main data source,it utilized six statistics（ maximum,minimum,mean,standard deviation,upper quartile,lower quartile） calculated by superpixel segmentation algorithm as features to build remote sensing image classification model based on random forest. The experiments show that superpixel statistical features can significantly improve the overall classification accuracy of remote sensing image to 89. 01%,and effectively reduce omission error and commission error. This method can be extended to land cover classification from Landsat-8 remote sensing image.