FEATURE EXTRACTION FOR POLSAR IMAGE CLASSIFICATION USING MULTILINEAR SUBSPACE LEARNING

Multiple informative polarimetric descriptors can be computed from direct measurements of polarimetric covariance matrix and target decomposition theorems. Under the tensor algebra framework, each pixel is modeled as a third-order tensor object by combining multi-features and incorporating neighborhood spatial information together. Typically, the tensor object is of high correlation and redundancy in both the spatial and feature dimensions. In this paper, we propose a feature extraction method using the multilinear principal component analysis to facilitate the classification process. Experimental results in comparison with principal component analysis, independent component analysis and linear discriminate analysis demonstrate that the classification accuracy is significantly improved since the extracted features by the proposed method are more discriminative.