Détection de bateaux dans les images de radar à ouverture synthétique

Abstract

The main purpose of this thesis is to develop efficient algorithms and design a system for ship detection from Synthetic Aperture Radar (SAR) imagery. Ship detection usually involves through detection of point targets on a radar clutter background.The detection of a ship depends on the physical properties of the ship itself, such as size, shape, and structure; its orientation relative to the radar look-direction; and the general condition of the sea state. Our strategy is to detect all possible ship targets in SAR images, and then search around each candidate for the wake as further evidence.The objectives of our research are (1) to improve estimation of the parameters in the K-distribution model and to determine the conditions in which an alternative model (Gamma, for example) should be used instead; (2) to explore a PNN (Probabilistic Neural Networks) model as an alternative to the commonly used parameteric models; (3) to design a FC (Fuzzy Clustering) model capable of detecting both small and large ship targets from single-channel images or multi-channel images; (4) to combine wake detection with ship target detection; (5) to design a detection model that can also be used to detect ship targets in coastal areas. We have developed algorithms for each of these objectives and integrated them into a system comprising six models.The system has been tested on a number of SAR images (SEASAT, ERS and RADARSAT-1, for example) and its performance has been assessed.