Zhang, R; Liu, J; Li, S; Shi, X; Wu, P; (2025) Fine-grained maritime traffic perception via low-rank adaptive transformer and regional matching fusion. Journal of Marine Engineering and Technology 10.1080/20464177.2025.2578910. (In press).

Text JMET Manuscript.pdf - Accepted Version Access restricted to UCL open access staff until 30 October 2026. Download (10MB)

Abstract

Traditional waterway traffic information services and management primarily depend on equipment such as Automatic Identification Systems (AIS), shore-based radar, and video surveillance. However, these traditional methods suffer from temporal and spatial latency, as well as insufficient precision in heterogeneous data fusion, making it challenging to satisfy refined situational awareness demands in waterways. This study introduces a novel end-to-end multi-source data fusion framework for fine-grained maritime traffic perception, addressing the limitations of traditional methods in integrating heterogeneous data. We enhance an Edge-SAM instance segmentation model with low-rank matrix optimisation and a ship-sensitive attention mechanism, achieving pixel-level segmentation of waterway targets. A regional matching method is proposed to deeply fuse discrete AIS information with detected objects from surveillance videos, effectively tackling challenges such as temporal alignment and coordinate mapping. Experiments on a real-world inland waterway dataset demonstrate significant improvements over baseline models, with 21% higher IoU, 14% higher Dice, and 23% higher Precision. The method achieves robust data fusion accuracy and real-time processing capability, offering a reliable solution for intelligent waterway regulation and collision risk assessment. This approach not only enhances vessel segmentation accuracy but also provides a viable technical pathway for the efficient operation of intelligent maritime supervision systems.

Download activity - last month

Export as CSVXMLJSON

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as JSONXMLCSV

Archive Staff Only

View Item