Towards a behavior analysis of remote-sensed vessels

This paper analyzes the potentialities to classify vessels detected through optical and synthetic-aperture radar (SAR) satellite-borne platforms and estimate their motion. For classification, the discriminative power of a set of geometric features extracted from segmented remote-sensed images is valuated by clustering data derived from a set of accurate footprints belonging to either tanker or cargo ships. The same procedure is repeated on a few dozens of real, remote-sensed optical images. Concerning velocity estimation, which in this context is based on the detection and analysis of the wake pattern generated by the ship motion, a discussion concerning the accuracy of the wake detection task is presented. In particular, since wake patterns are usually hard to detect, a method is proposed to enhance the wake signal-to-noise ratio, based on a dedicated pre-filtering stage. Results returned by the proposed method are compared with those obtained adopting a standard literature approach, eventually observing that the introduction of the prefiltering stage improves the wake detection accuracy. A maritime surveillance system based on a pipeline of the modules described here represents a useful tool to support the authorities in charge of monitoring maritime traffic with safety, security and law enforcement purposes.