基于NSCT和LSSVM的Landsat ETM+图像土地覆盖分类

摘要

A new method of remote sensing image land cover classification based on NonSubsampled Contourlet Transform (NSCT) and Least Squares Support Vector Machines (LSSVM) is proposed.This method selects best band features dynamically to composite a multispectral image, and then fuses it with panchromatic image based on NSCT and stationary wavelet transform for better spatial resolution of multispectral image, performs land cover classification on the fused image based on LSSVM.The experimental results show that the proposed method can effectively preserve spectral information and improve spatial information of the remote sensing image,provide more reliable classification features of ground objects,improve classification precision, and outperform the traditional classifier based on minimum distance, maximum likelihood, Bayesian, back propagation neural network.%提出一种新的基于非下采样Contourlet变换(NSCT)和最小二乘支持向量机(LSSVM)的遥感图像土地覆盖分类方法.该方法动态选择最优的多光谱图像的波段特征进行组合,基于NSCT和IHS对多光谱图像和全色图像进行融合,增强多光谱图像的空间分辨率,基于LSSVM对融合图像进行分类.实验结果表明,提出的方法在保留多光谱图像光谱信息的同时,增强了图像的空间细节表现能力,提供更加可靠的地物分类特征,提高了分类精度,并且优于传统的基于最小距离法、最大似然法、贝叶斯分类法和BPNN分类法的遥感图像分类方法,该方法是有效可行的.