基于HJ影像的面向对象土地覆被分类方法

Object oriented land cover extraction classification techniques on HJ Images

提出了基于HJ影像的面向对象技术土地覆被分类的一整套方法,即采用面向对象的影像分类技术,充分利用影像的光谱特征、空间特征、纹理、上下文关系,综合运用多层分割、整体分割、类内分割和局部分割模式,融入二叉树流程法,以多时相HJ影像为实验数据,对土地覆被类型进行自动提取.并以地形复杂的麻阳苗族自治县为例进行实验研究,结果表明:分类总体精度为82.88%,能够满足利用遥感影像进行土地覆被信息提取的精度要求,说明是可行的;由于中分辨率的HJ影像提供的光谱细节并不是很丰富,HJ影像面向对象土地覆被分类技术的光谱特征优势不明显,利用二叉树流程法和多种分割方法是降低误差提高分类精度的有效途径.

Put forward a set methods of object-oriented land cover classification based on HJ images. Taking full use of the features of HJ image including spectral characteristic, spatial feature, texture, and context, applying several segmentation modes comprehen- sively, such as multi-layers segmentation modes, whole segmentation mode, part seg- mentation mode and the mode of segmentation inside a class, and with binary tree process method introduced, it realizes to extract land cover information automatically from HJ-CCD images taking MaYang county with complex terrain as a case area. Result shows that the overall precision is 82.88~ and Kappa coefficient is 0. 8163, which can meet accuracy requirement of land cover information extraction using remote sensing im- ages. The experiment shows that land cover information extraction using HJ remote sensing images and object-oriented classification technology is feasible. Although HJ images cannot provide rich spectral details, it is effective to improve the classification accuracy to introduce of Binary tree process method and to combine a variety of segmen tation methods into the land cover information extraction process.