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.
