SARFish: Space-Based Maritime Surveillance Using Complex Synthetic Aperture Radar Imagery

Synthetic Aperture Radar (SAR) imagery is an essential part of maritime surveillance since it can operate in most weather conditions, day and night. Space-borne SAR especially can provide a large surveillance coverage in which both ground-range detected (GRD) imagery at a reduced resolution and single-look complex (SLC) imagery at a finer resolution of the same observation scene are available. Recently, a large, fully labeled, Sentinel-1 GRD based dataset, named xView3-SAR, was released to promote machine learning (ML) research in detection and classification of ships in order to assist with monitoring illegal, unreported, and unregulated (IUU) fishing activities. In this paper, the xView3-SAR's companion SLC dataset, called the SARFish, is introduced. The aim is to promote the use of SLC data for improving IUU fishing ship detection and classification. An ML-based algorithm using the combined GRD-SLC SAR data is presented as a base-line model for demonstration. The algorithm consists of a constant false-alarm rate (CFAR) detector whose false-alarm rate is reduced by a cascade of classifiers trained using the GRD data, followed by a ship classifier trained using the SLC data. Preliminary results suggest that despite not using multi-looks to reduce sea-clutter spikes, the use of SLC images does not degrade ship detection. Furthermore, a very modest improvement in ship classification is observed.
^1
1
Under contract from Rheinmetall Defence Australia Pty Ltd, Adelaide, SA, Australia.