Work

Deep Learning Object Detection for Maritime Security: A Critical Review of Task Evidence, Benchmark Comparability and Deployment Constraints

Year: 2026
Type: preprint
Abstract: Deep learning has been used successfully for object detection in optical, infrared,SAR, and multispectral images of maritime environments. Nevertheless, itsdeployment in maritime security systems remains challenging. This review anal-yses major object detection architectures, sensing methods, application domains,and deployment constraints in this domain. The most frequently representeddetection architectures in the reviewed papers include You Only Look Once(YOLO) detectors (199 papers) and attention-based and transformer models(123 papers). Tracking, multimodal fusion, threat reasoning, and detector secu-rity are much less represented. There are also major deficiencies in evaluationpractices. For example, 121 out of 299 papers reporting detection metrics donot provide an intersection-over-union threshold value. Inference latency, energyconsumption, and target hardware are also insufficiently reported in many stud-ies. Besides detection accuracy, we discuss how security-task evidence affectsbenchmark selection and detector deployment in communication-aware settings.The review provides a reporting checklist and a roadmap for future research.Future work should concentrate on multimodal benchmarks, small-object detec-tion, cross-sensor generalisation, trustworthy sensor fusion, edge intelligence, andrealistic evaluation of detectors.
Language: English
Open Access status: green
OpenAlex ID: W7233834979
Cites: 170
Cited by:
Related to: 10