高光谱遥感图像最大似然分类问题及解决方法

The MLC's Problem in Classification of Hyper-spectral RS Image and Its Solving Method

利用最大似然分类器对高光谱遥感图像分类时,由于波段数目多、波段间的相关度大,使协方差矩阵的行列式近似于奇异,因而导致了不合理的分类结果。本文调整了波段协方差矩阵对分类的影响,改进了最大似然分类判决函数,利用改进后的判决函数进行分类,试验证明这种方法是有效的。另外,根据图像特征和经验对图像进行波段选择,也是改善高光谱遥感分类效果的有效途径。

When classifying the hyper-spectral remote sensing images using the max-llkelihood classifier, the determinant of spectra＇s coy-matrix is singular approximately and an unreasonable result comes out because of the big number of the bands and the close correlation between them. We modify the deciding function of the max-likelihood classifier by adjusting the influence of spectra＇s coy-matrix, then classify the images. It is proved to be successful by the experiment. In addition, the spectral selection based on the image＇s feature and experience is also an effective way to improve the classification of hyper-spectral remote sensing images.