A New Bayesian Likelihood Ratio Test for Supervised Classification of Fully Polarimetric SAR Data: An Application for Sea Ice Type Mapping

One of the potential applications of polarimetric Synthetic Aperture Radar (SAR) data is the classification of land cover, such as forest canopies, vegetation, sea ice types, and urban areas. In contrast to single or dual polarized SAR systems, fully polarimetric SAR systems provide more information about the physical and geometrical properties of the imaged area. In this study, a new likelihood ratio test, called the Bayesian Likelihood Ratio Test (BLRT), is proposed. The BLRT test is mathematically derived and used for the supervised classification of polarimetric SAR data. The proposed BLRT test is based on the Chernoff error bound for the case of complex Wishart distribution. Furthermore, a new criterion that utilizes the spatial context information in the classification process is incorporated in to ensure the homogeneity of the output classes. In the proposed iterative approach, supervised classification can be performed using the BLRT test alone or by combining it with the derived spatial criterion. The combination produces more homogenous classification results, avoiding the noise produced by using BLRT only. The homogeneity of the resulting classes depends on the size of the defined window for the neighborhood pixels. The higher the size the smoother the resulting classes. The potential of the proposed BLRT test alone to discriminate between four sea ice types is examined for the sea ice types mapping using fully polarimetric C-band RADARSAT-2 data collected over Franklin Bay, Canadian Arctic. Moreover, the ability of the proposed BLRT jointly with the derived spatial criterion is investigated. Application for Arctic sea ice mapping shows that incorporation of the spatial criterion provides more promising classification results. Comparisons with classification results based on the Wishart classifier and the Expectation Maximization with Probabilistic Label Relaxation (EMPLR) algorithm are presented. Experimental results have shown the advantage of the proposed BLRT test alone or in combination with the derived spatial criterion against the two standard polarimetric SAR classification techniques. Higher overall classification accuracy was achieved using the BLRT test on two selected study areas (97.8% for the two study areas) in comparison to the overall classification accuracy using the Wishart classifier (97.2% and 90.4% for the first and second study areas, respectively). By including the proposed spatial criterion, the overall classification accuracy using the BLRT test is improved to 99.9% and 98.6% for the first and second study areas, respectively, while 99.0% and 91.2% overall classification accuracy is obtained using the EM-PLR algorithm.