Ship Detection in SAR Images: a Segmentation-Based Approach Massimo Sciotti, Pierfrancesco Lombardo Dept. INFOCOM- University of Rome “La Sapienza”, Via Eudossiana 18, 00184 Rome ITALY Ph. +39-06-44585472, Fax +39-06-4873300, e-mail: pier@infocom.uniroma1.it Abstract—This paper deals with the problem of detecting ship targets in medium and high resolution SAR images. Achieving a controlled false alarm rate is a major problem for the presence of a highly non-homogeneous sea clutter environment due to the highly variable environmental and weather conditions in closely spaced areas. After highlighting the problems of conventional techniques, a new approach is proposed based on cascading a segmentation stage and a local CFAR detection stage. The former estimates the homogeneous back-scattering regions, while the latter detects the ship targets inside the fairly homogeneous identified regions. I. INTRODUCTION It is well known that Synthetic Aperture Radar (SAR) are well suited to the monitoring of sea surface conditions and activities, as surface wind, swell structure, pollution conditions, and presence of man-made objects. Specifically, the capability of detecting ships and estimating their velocity was reported in many studies [1,2]. Unfortunately, the repetition time of tenths of days of the present systems (ERS-2/ENVISAT) is a major limit for the application of SAR in operational sea traffic control. This problem will be strongly reduced by next- generation SAR designed to have both wide coverage and high repetition time, as the Cosmo/Skymed satellite constellation aiming at a repetition time of few hours, [3]. From a signal processing point of view, the major technical problem is the large non-homogeneity of typical SAR images of sea, originated from the presence of several regions with definite differences in back-scattering behaviour. This is likely to give rise to a large number of false alarms preventing from a fully automatic reliable operation, required to process the large amount of collected data. As an example, Fig. 1 shows a quick look ERS-1 image of the Mediterranean Sea around Elba Island (Italy) with 100 m resolution and more than 10 looks. Various features typical of non-homogeneous sea images are apparent: a) transitions between regions with different wind conditions, b) low wind spiral marks, c) effects of land, d) reflectivity alteration due to bathimetry, e) presence of ship wakes. Aim of this paper is to develop an automatic technique for the detection of ships against a generic non-homogeneous sea clutter, assuring a constant false alarm rate (CFAR). This work was supported by the Italian Space Agency (ASI) under the scientific research program Fig. 1. ERS-1 quick look image of Elba island (Italy), July 12 th , 1997. Typical features: a) transitions between regions with different wind conditions, b)spiral marks due to different wind conditions, c) effects due to land areas, d) back-scattering alterations due to bathimetry, e) presence of ship wakes. In particular, the problems of standard CFAR techniques for ship detection are analysed and a new segmentation-based automatic technique is proposed. Two types of SAR images are considered: (i) low resolution images with a large number of looks, as the quick-look images available soon after the satellite pass, [2], usable for real-time sea traffic control; and (ii) high resolution images that allow more accurate and reliable discrimination of ships and their classification. The detection technique for the latter case must deal with a larger level of non-homogeneity and take into account the intrinsic non-gaussianity of the back-scattering. Considering K- distributed sea clutter echoes a novel analytic expression for the false alarm probability (P fa ) of CFAR techniques is derived, which allows to set the threshold adaptively inside the (b) (e) (a) (c) (c) (d) wake false wake ship (e) (b) 0-7803-6707-3/01/$10.00 © 2001 IEEE 2001 IEEE RADAR CONFERENCE 81