PolSAR Image Classification via D-KSVD and NSCT-Domain Features Extraction

Polarimetric synthetic aperture radar (PolSAR) image classification is a powerful and important application in remote sensing. In this letter, we propose a PolSAR image classification method based on discriminative dictionary learning (D-KSVD) model and nonsubsampled contourlet transform (NSCT)-domain features. The D-KSVD model is used in our experiment to classify PolSAR images and is more time efficient and accurate when compared with the sparse representation classifier (SRC) used by other researchers to the PolSAR image classification. This is due to the fact that the D-KSVD model employs linear classifier rather than compute the reconstruction error employed by SRC. On constructing an effective dictionary of D-KSVD, we use NSCT to extract polarization features from the PolSAR coherency matrix. However, the low-frequency coefficients of NSCT domain have better discrimination ability than others, and we consider them as the classification features; hence, we adopt them into the D-KSVD model to obtain a dictionary which accommodates a linear classifier. Our experimental results of two real PolSAR images indicate that the proposed PolSAR image classification method based on D-KSVD model and NSCT-domain features extraction approach achieves better results with time efficiency and accuracy.