Wishart distance-based joint collaborative representation for polarimetric SAR image classification

• Published in: IET radar, sonar & navigation Vol. 11; no. 11; pp. 1620 - 1628
• Main Authors: Geng, Jie, Wang, Hongyu, Fan, Jianchao, Ma, Xiaorui, Wang, Bing
• Format: Journal Article
• Language: English
• Published: The Institution of Engineering and Technology 01.11.2017
• Subjects: collaborative representation classifier; image classification; polarimetric SAR image classification; polarimetric synthetic aperture radar; PolSAR; radar imaging; radar polarimetry; Research Article; statistical analysis; synthetic aperture radar; Wishart distance‐based joint collaborative representation; synthetic aperture radar; polarimetric synthetic aperture radar; collaborative representation classifier; image classification; PolSAR; Wishart distance-based joint collaborative representation; radar imaging; statistical analysis; radar polarimetry; polarimetric SAR image classification
• ISSN: 1751-8784; 1751-8792
• DOI: 10.1049/iet-rsn.2017.0056

Inspired by collaborative representation classifier (CRC), a Wishart distance-based joint CRC (W-JCRC) is proposed for polarimetric synthetic aperture radar (PolSAR) image classification. Since that neighbouring pixels usually belong to the same category with high probability, they can be simultaneously represented via a joint representation model of linear combinations of labelled samples. The joint collaborative representation of neighbouring pixels can overcome the influence of speckle noise at the same time. Considering the statistical property of PolSAR data, a weighted regularisation term with revised Wishart distance is designed to contain the correlations between unlabelled and labelled samples. The coefficients of representation are estimated by an $l_2$l2-norm minimisation derived closed-form solution. In the experiments, three real PolSAR images are applied to evaluate the performance, and the experimental results demonstrate that the proposed method is able to improve classification accuracies compared with other state-of-the-art methods.