Ship detection methods for maritime domain awareness using Sar satellite data

Abstract

Maritime Domain Awareness is the understanding of all aspects relating to the maritime domain that may have an effect on the security, economy or environment of a country bordering the sea. Large ships in South Africa have historically been monitored using ship-based transponder systems such as Automatic Identification System. These systems transmit geographical ship coordinates which allow operators to track ships. The greatest disadvantage of monitoring ships in this manner is that the transponders need to be installed and switched on in order to track ships. The monitoring of ships can be done in another manner by taking advantage of Synthetic Aperture Radar imagery which allows for the monitoring of large portions of the Earth. This imagery is generated using an active sensor which makes use of radar pulses to observe areas under any weather condition, day or night. In this dissertation, various ship detection systems configurations were tested using Synthetic Aperture Radar imagery and experimental conclusions were drawn for each configuration. The system was tested against simulated Synthetic Aperture Radar imagery and actual Synthetic Aperture Radar imagery located within the South African Exclusive Economic Zone. Tests were performed in order to evaluate which noise distribution model best models the Synthetic Aperture Radar imagery used in this study over South African coastal waters. The parameters and the effects of varying them within each system configuration were evaluated. Experimental results found that the K-distribution is the noise distribution model that best describes the noise of sea-water found within the Synthetic Aperture Radar imagery used in this study. Furthermore, it was determined that ship detection system configurations using the Constant False Alarm Rate prescreening method had the highest average detection accuracies and lowest false alarm rates amongst all of the configurations tested. The various detection system configurations had similar detection accuracies at low thresholds but varied significantly in terms of the number of false alarms.