Hyperspectral image classification using spectral histograms and semi-supervised learning

Different classification methods have been applied to hyperspectral images during the last decade. Many of these methods have so far used pixel spectral signatures. Methods that include spatial information in the analysis achieve a better classification accuracy than those that only account for spectral signature of pixels. In this research, an algorithm that extracts regional texture information by computing spectral difference histograms over window extents in hyperspectral images was developed. The spectral angle distance was used as the spectral metric and different window sizes were explored for compute the histogram. The histograms were used in a semi-supervised learning framework that uses both labeled and unlabeled samples for training the Support Vector Machine classifier. Algorithm validation and comparisons are done with real and synthetic hyperspectral images. The method performs well with high spatial resolution images. The algorithm performs well under different Gaussian noise levels.