Multi-Frequency and Multi-Mission SAR Imagery for Maritime Surveillance by Deep Learning

In the ambit of Maritime Situational Awareness, (MSA), this paper presents a dataset for vessel detection that encompasses COSMO-SkyMed (X-band), SAOCOM (L-band), and Sentinel-1 (C-band) Synthetic Aperture Radar (SAR) imagery compiled over the same areas with a minimal temporal gap, shorter than 15 minutes, between successive acquisitions. The dataset is further enriched by incorporating auxiliary AIS information and it is organized and annotated to fit the requirements of an Object Detection (OD) architecture. Indeed, thanks to its ability to effectively learn complex and high-level representations from raw data Deep Learning (DL) has emerged as a valuable tool for combining multiple sources or types of features to create a more informative and comprehensive representation. Furthermore, a novel data augmentation technique tailored for SAR-based vessel detection is presented. The main goal of the paper is thus to establish an unprecedented single-look complex (SLC) multi-frequency dataset for SAR ship detection. The dataset is the starting point to construct a universal framework for ship detection that transcends the limitations of frequency dependency, fostering a comprehensive and versatile approach to ship detection across various SAR frequencies. Serving as the inaugural dataset of SLC SAR multi-frequency data, this study also aims at stimulating further exploration in this promising research domain.