Spectral-spatial classification of hyperspectral images with k-means++ partitional clustering

摘要

We propose and investigate a complex hyperspectral image classification method with regard to the spatial proximity of pixels. Key feature of the method is that it uses common and relatively simple algorithms to attain high accuracy. The method combines the results of pixel-wise support vector machine classification and a set of contours derived from k-means++ image clustering. To prevent redundant processing of similar data a principal component analysis is used. The method proposed enables the accuracy and speed of hyperspectral image classification to be enhanced.