Contents 1. Introduction
Ship detection is of great importance for maritime surveillance which includes ship traffic monitoring, fisheries control, border surveillance, counter piracy, and so on. With the rapid development of satellite sensor systems, ship detection from remote sensing imagery is a crucial application for maritime surveillance. Remote sensing using Earth Observation can potentially detect most ships, including cooperative and non-cooperative targets in large areas where traditional surveillance methods, for example, Automatic Identification Systems (AIS) and coastal radar, are not satisfied. Also, the combined use of a traditional surveillance methods and satellite imaging could be more effective. Synthetic Aperture Radar (SAR) with capacity to image day and night under most meteorological conditions, has become the state of the art technique for ship detection [1]. At the same time, the electro-optical (EO) satellite images can be used to detect ships with more detailed information [2]. Conventional optical image processing usually utilizes higher-resolution panchromatic image with spatial techniques [2], [3] to detect ships and fails to take advantage of the spectral and thermal information, which makes detection performance vulnerable to clutter. Multispectral image can provide better discrimination between ship and clutter, and spectral signatures that are possibly unique to the ship itself [4]. There are several papers in the open literature treating methods for ship targets detection on multi-spectral data. [5] proposed a novel method using visual attention based on biquaternion to detect ships, but the false alarm rate rose when clutter contained. [4] proposed an automatic ship detection method in high resolution commercial images. However, ship detection is much more difficult in low resolution multi-spectral images.
