Learning recovered pattern from region-dependent model for hyperspectral imagery

The Compressive-Projection Principle Component Analysis (CPPCA) technique which recovers hyperspectral image(HSI) data from random projection efficiently, has been proved to be significant in decreasing signal-sensing costs at the sender. Inspired by the fact that the spectral signature of the same ground cover is similar, and two pixels of the neighborhood are likely to belonging to the same ground cover, this paper proposed a novel region-dependent approach CPPCA to recover HSI data. Due to the fact that the region map is critical to our proposed algorithm, herewith we employ a robust supervised Bayesian approach (LORSAL-MLL segmentation) which explores both the spectral and spatial information in an intuitive interpretation with small size samples to segment hyperspectral image into different regions. The CPPCA reconstruction procedure is then employed to each region independently other than each partition individually. The effectiveness and practicability of proposed region-dependent CPPCA (RDCPPCA) reconstructed algorithm is illustrated by real hyperspectral image data set with several criteria measurement.