Study of the classification efficiency of difficult-to-distinguish vegetation types using hyperspectral data

Abstract

The article is devoted to the effectiveness research of methods of controlled spectral and spectral-spatial classification of hyperspectral data. In particular, minimum distance, support vector machine, mahalanobis distance and maximum likelihood methods are considered on the example of vegetative cover types differentiation. Significant attention is paid to studying the dependence of the accuracy of data classification with listed methods on the spectral features number and their selection method. The perspectivity of complex processing of spectral and spatial features, considering the correlation of close pixels, is demonstrated. The experimental results obtained with various methods of forming training sets are presented.