Modified Otsu's algorithm: A new computationally efficient ship detection algorithm for SAR images

Abstract

In the context of high spatial resolution Synthetic Aperture Radar (SAR) images, this paper deals with a new computationally efficient algorithm which allows us to accurately detect ships of certain dimensions, which can span in a range supposed known. The topic of ship detection in SAR imagery has been extensively analyzed in literature. However, little is known about such application on SAR data acquired by COSMO-SkyMed (CSK) satellite constellation. Commonly, in SAR images, ship detection methods rely on sea (background) statistical characterization. Nevertheless, such approaches typically require a high computational load. Thus, the time required by classic detection algorithms (e.g. Constant False Alarm Rate - CFAR) may become critical in the context of ship detection for maritime surveillance. In this paper we present a new computationally efficient algorithm for ship detection in SAR images, which is not a CFAR approach. Our algorithm is obtained modifying the classical two-classes segmentation provided by Otsu's histogram thresholding algorithm. Such modifications are necessary to apply Otsu's algorithm in SAR images. They are based on the geometrical proprieties of the target to be detected, i.e. ships whose dimensions can span in a range a priori known. Proofs of the goodness of our algorithm are given on a CSK image acquired over New York.