基于高光谱遥感图像的树种(树种组)分类

Classification of tree species or tree groups based on hyperspectral remote sensing images

高光谱遥感图像各波段间存在着高相关性和高冗余度,在处理方面上具有一定的困难。使用光谱一阶、二阶微分法对高光谱遥感图像进行处理,增强像元亮度值的差异,选择树种特征差异性较大的特征波段进行组合降维,再利用野外实地调查的样地作为分类训练样本进行分类,然后用位置精度评价对原始影像及光谱一阶、二阶分类结果进行精度评价及分析比较。结果表明,光谱一阶、二阶微分法所选取的波段总体分类精度与Kappa系数都比原始图像分类结果高,其总体分类精度分别为90.20%与91.30%,Kappa系数分别为0.85与0.86;而原始图像分类的总体分类精度与Kappa系数分别为80.20%与0.78。由此可知,光谱一阶、二阶微分法所选取的波段对毛竹、雷竹、杂竹、杉木、马尾松、常绿阔叶树和落叶阔叶树等7个树种(树种组)区分较为有效。

There are high correlation and high redundancy between each band of hyperspectral remote sensing images, which is difficult to deal with. This article uses the spectra of the first order and second order differential method to the hyperspectral remote sensing image pixels, brightness values, enhance the pixels, differences in brightness values, choose bigger differences between the characteristics of the band for dimension reduction, then using the field investigation of sample area classified as classification the training sample, location accuracy evaluation is used after the original classification figure and spectral classification figure for the first order and second order accuracy evaluation and comparative analysis. The results showed that the spectra of the first order and second order differential method selected band overall classification accuracy and Kappa coefficient is higher than the original image classification results, the total classification accuracy are 90.20% and 91.30% respectively, Kappa coefficient of 0.85 and 0.86 respectively;The overall classification accuracy and Kappa coefficient of original image classification were 80.20% and 0.78 respectively. Accordingly, the spectra of the first order and second order differential method selected wavelengths can effectively put the bamboo, LeiZhu, miscellaneous bamboo, fir, pinus massoniana, evergreen broadleaf and deciduous hardwood such as 7 distinguish forest tree species or tree species groups.