Contextual sparse representation and its application in polarimetric SAR image classification

Recently, sparse representation (SR) has shown great potential in the realm of pattern recognition, which substitutes over-complete redundant function system for traditional orthogonal basis functions and provides great flexibility for adaptive sparse extension of signals. Polarimetric SAR classification is a high-dimensional nonlinear mapping problem. On the basis of the sparse characteristics of the features for polarimetric SAR image classification, in this paper a contextual sparse representation (CSR) model for polarimetric SAR image classification is proposed, which incorporates the intrinsic polarimetric information and the spatial contextual information in the sparse representation procedure. Results on Danish EMISAR datasets demonstrate the effectiveness of the proposed approach and it is proved that CSR-based polarimetric SAR image classification method can acquire accurate results with the great improvement of efficiency. Moreover, the influence of sparsity degree on classification results is further discussed and the comparison with SR-based polarimetric SAR image classification method also verifies the effectiveness of the proposed approach.