Classification of hyperspectral remote sensing imagery by k-nearest-neighbor simplex based on adaptive C-mutual proportion standard deviation metric

Abstract

The k-nearest-neighbor simplex (kNNS) based on an adaptive C-mutual proportion standard deviation metric for classification of hyperspectral remote sensing imagery was proposed. By analyzing spectral characteristics on the samples of the same and different classes, a C-mutual proportion standard deviation metric is put forward which innovates a novel metric on distance and similarity measures for pattern recognition. Combined with the adaptive adjusting algorithm, this metric is used for the classification of hyperspectral remote sensing imagery. The traditional kNNS classification algorithm is improved by this metric and the adaptive adjusting algorithm, and its classification accuracy is enhanced. Three experiments with different types of hyperspectral imagery are conducted to evaluate the performance of the proposed algorithm in comparison to the other five classification algorithms. The experimental results demonstrate that the proposed algorithm is superior to other algorithms on overall accuracy and kappa coefficient.