Superstructure scattering distribution based ship recognition in TerraSAR-X imagery

Abstract

Benefiting from the improved resolution and polarization information of SAR data, ship recognition has attracted much attention during the last decade. This paper considers the ship recognition in TerraSAR-X imagery. We propose a novel feature extraction algorithm, named Superstructure Scattering Distribution (SSD), by investigating the ship's superstructure and corresponding electromagnetic scattering mechanism. In SSD, we first segment the image areas that correspond to the scattering of ship's superstructure using segmentation and edge detection methods. Once the superstructure areas of ship in SAR imagery have been obtained, we then calculate a novel feature parameter to describe the distribution and shape complexity of the ship based on the fractal dimension and other shape complexity descriptors. Extensive experiments have been carried out on ship datasets collected from TerraSAR-X SAR imagery. The results validate that the SSD feature can achieve better performance together with other physical features.