Study on spectral similarity measure in hyperspectral remote sensing data

Based on the principles and methods of different algorithms, a new framework of spectral similarity measure in hyperspectral RS image is put forward. It includes five kinds of similarity measure approaches: geometric measure, encoding measure, probability measure, measure based on feature and measure based on transformation. Based on the analysis to current algorithms, some new approaches including spectral similarity measure based on spectral polygon, quaternary encoding, decimal encoding, tree-based transformation and wavelet transformation are proposed and experimented. It proves that those new approaches can be used to classification, retrieval and other processes effectively.