Typical sequence classification method in hyperspectral images with reduced bands

This work presents a new method for hyperspectral spectra classification based on the Typical Sequence (TS) derived from the Asymptotic Equipartition Theorem and Information Theory. Each Endmember (EM) of a scene is represented by a Hidden Markov Model (HMM) and a spectrum is classified in a given class if it can be considered a TS generated by the HMM associated with the EM related to the class. The Discrete Wavelet Transform (DWT) is used in the orthogonal decomposition of the original spectrum and the HMM parameters are estimated using this orthogonal decomposition. The proposed method is tested with AVIRIS spectra of a scene with 13 EM and the classification results show that 32 spectral bands can be used instead of the original 209 bands, without significant loss in the classification process.