Hyperspectral image principle component extraction method based on RFS and ART

Algorithms used to extract principle components of hyperspectral image are sensitive to noise and data distribution. A principle components extracting algorithm based on the region feature spectrum (RFS) and ART is presented. The algorithm firstly extracts region feature spectrum through spatial neighborhood clustering as input pattern vectors of the network, and then acquires the classificatory character adaptively. Finally, extraction is successfully achieved by using clustering spectral vectors. The experiments on hyperspectral images indicate that the size of data processed by network is reduced about 97%, and the extraction effect is obviously better than that by K-means algorithm.