A new method for classification of POLSAR images

Abstract

The present paper proposes an unsupervised feature learning method for POLSAR image classification. The proposed method includes two steps. In these two steps, features are created and learned from scratch. The first is to learn dictionaries and encode features using scatter matrices. The dictionary is learned using a set of vectors that are known as hierarchical matching pursuit (HMP). The dictionary is learned with K-singular vector decomposition (K-SVD). Afterward, the sparse codes can be computed with orthogonal matching pursuit (OMP). The second step extracts features from the previous step. The results demonstrate that the features extracted and learned from this method led to more efficient POLSAR classification results than other existing similar methods.