A Superpixel-Based Neighborhood Polarimetric Covariance Matrix for Polsar Ship Detection

Abstract

In a recent work, a neighborhood polarimetric covariance matrix [N] was proposed to detect ships from polarimetric SAR (PolSAR) imagery. However, its computational complexity is extremely high. Besides, the heterogeneity surrounding ship edges is also not well considered in [N]. To cure these draw-backs, we construct a novel superpixels-based neighborhood polarimetric covariance matrix [SN] in this paper. Specifically, the simple linear iterative clustering (SLIC) is first used to obtain superpixels. Then, the vector vmean corresponding to the mean value of superpixel is further computed so as to characterize the neighborhood information of each pixel in superpixel. Finally, by combining the original scattering vector v and vmean together, the vector t12 is built to calculate [SN]. The experiment tested on one L-Band ALOS PolSAR imagery shows that i) the polarimetric whitening filter derived from [SN] (i.e., PWFSN) has a better detection performance than that derived from [N] (i.e., PWFN); ii) the calculation process of [SN] takes much less time than that of [N].