Adaptive Maritime Vessel Detection System Under Reduced Visibility Conditions

This paper presents a hybrid architecture for maritime vessel detection combining image quality enhancement via MPRNet with parallel YOLO-family detectors. The YOLO_COCO model handles standard visibility, while specialized YOLO_Fog addresses fog, precipitation, and spray under extreme low-contrast regimes. Analytical synthesis using average confidence metrics selects the most reliable detection. Multi-stage filtering ensures real-time processing. Experimental validation demonstrates robustness across three scenarios: heavy rainfall, dense fog, and intense sea spray. The specialized model successfully identifies vessels where the general-purpose detector fails, achieving automatic weather-based adaptation. The system integrates preprocessing, parallel detection with dynamic model selection, and reporting subsystem via Flask web service, enabling applications in navigation support, port surveillance, and environmental monitoring.