Ocean-aware deep learning for civilian maritime object detection and tracking in complex ocean environments: a comprehensive review

Abstract

Reliable maritime object detection and tracking are essential for civilian ocean engineering applications, including vessel traffic monitoring, offshore infrastructure inspection, port management, marine debris surveillance, sea ice observation, hydrographic surveying, and environmental risk assessment. However, complex ocean environments present substantial challenges for automated perception systems because of dynamic sea surfaces, sea clutter, low SNRs, atmospheric interference, scale variation, occlusion, weak target signatures, and limited annotated datasets. This review examines recent advances in ocean-aware deep learning for maritime detection, segmentation, localization, and tracking using optical, infrared, Synthetic Aperture Radar (SAR), multibeam echosounder (MBES), satellite, airborne, unmanned aerial vehicle (UAV), unmanned surface vehicle (USV), and autonomous sensing data, together with Automatic Identification System (AIS) information. It synthesizes methodological developments in robust feature extraction, multi-scale representation, small-object detection, multimodal data fusion, weakly supervised learning, domain adaptation, and tracking under intermittent visibility. Existing limitations remain in sea-state annotation, cross-sensor benchmarking, long-term tracking datasets, operational validation, and real-time deployment. Future research should prioritize sea-state-aware architectures, physics-informed learning, SAR–optical–AIS–MBES–USV data fusion, synthetic data generation, lightweight implementation, and standardized evaluation protocols. These directions are critical for developing reliable civilian maritime monitoring systems under diverse sensor, weather, and ocean conditions.