Research On Hyperspectral Unmixing Algorithm Based On Spatial Correlation Constrained Sparse Representation

Hyperspectral imaging technology obtains the spatial and spectral information of dozens or even hundreds different bands in the same time. Hyperspectral image has been applied in the identification and detection of ground target, the prediction of disasters and many other fields. Due to the resolution limitation of the sensors, the observation of one pixel in a hyperspectral remote sensing image may contain several disparate substances, which is called as a "mixed pixel". The mixed pixel problem has became the main obstacle to the development for quantification analysis of hyperspectral image.It is necessery to find a efficient method to solve the mixed pixel problem. In recent years, with the rapid development of compressed sensing and sparse representation, a sparse unmixing method has been proposed, which includes the sparsity constraint to the classical linear unmixing method. Sparse unmixing method has become a hotspot in the research area of hyperspectral unmixing. Focus on the application of sparse unmixing in hyperspectral unixing, this thesis has made a lot of reasearch, and the main works ande innovations are as follows:1) An adaptive total variation spatial regularization algorithm for sparse hyperspectral unmixing is proposed. Based on the analyses of the interpixel correlation in the hyperspectral imagery, an adaptive total variation spatial regularization is adopt to the sparse unmixing model, and it is solved by the alternating direction method of multipliers. In the alternating iteration, weighting factor of Total variation regularization term is no longer a global variable, which is determined by image local information and weighted by image edge information. Experimental results on both simulated and real data demonstrate that the proposed algorithm outperforms SUnSAL_TV algorithms, with a better spectral unmixing accuracy.2) As the endmembers selection criterion of SOMP and SMP is not optimal in the sense of minimizing the residual of the new approximation, a simultaneous adaptive backtracking-based orthotrogonal matching pursuit(SBAOMP) algorithm is proposed.The algorithm uses a block-processing strategy to divide the whole hyperspectral image into several blocks. Some potential endmembers are selected and added to the estimated endmember set in each block, then SBAOMP incorporates a backtracking process to detect the previous chosen endmembers’ reliability and deletes the unreliable endmembers from the estimated endmember set in each iteration. The endmembers picked in each block is associated as the endmember sets of the whole hyperspectral data. Finally, the abundances are estimated using the whole hyperspectral data with the obtained endmember sets.