A Novel Ship Detector Based on the Generalized-Likelihood Ratio Test for SAR Imagery

Abstract

Ship detection with synthetic aperture radar (SAR) images, acquired at different working frequencies, is presented in this paper where a novel technique is proposed based on the generalized-likelihood ratio test (GLRT). Suitable electromagnetic models for both the sea clutter and the signal backscattered from the ship are considered in the new technique in order to improve the detector performance. The GLRT is compared to the traditional constant false alarm rate (CFAR) algorithm through Monte-Carlo simulations in terms of receiver operating characteristic (ROC) curves and computational load at different bands (S-, C-, and X-). Performances are also compared through simulations with different orbital and scene parameters at fixed values of band and polarization. The GLRT is then applied to real datasets acquired from different sensors (TerraSAR-X, Sentinel-1, and Airbus airborne demonstrator) operating at different bands (S-, C-, and X-). An analysis of the target-to-clutter ratio (TCR) is then performed and detection outcomes are compared with an automatic identification system data when available. Simulations show that the GLRT presents better ROCs than those obtained through the CFAR algorithm. On the other side, results on real SAR images demonstrate that the proposed approach greatly improves the TCR (between 22 and 32 dB on average), but its computational time is 1.5 times slower when compared to the CFAR algorithm.

Keywords

• Maximum likelihood estimation
• radar cross-sections
• radar detection
• ship detection
• synthetic aperture radar