Three-dimensional geometrical feature estimation for ship classification through SAR images

Abstract

For the purpose of ship classification with SAR images, we put forward a way to extract the three-dimensional geometrical features of the ship in the case when the extracting and matching of the ship scattering centers are handicapped by the ocean movement and the scintillation of SAR images. An ellipsoid-ellipse model is used to approximate the ship and its SAR silhouette. With this model, the radar azimuths and the ship length are estimated in an almost unbiased way. Moreover, by introducing the length-width joint probability density function based on real samples, we got the width and height estimation by solving a constrained nonlinear Least Square problem. The estimations show satisfactory accuracy and robustness for the classification. The electromagnetic simulated images are used to validate the feasibility of the ellipsoid-ellipse model, as well as the efficiency of the algorithm.