Intelligent maritime vessel detection and classification toward offshore wind energy zone surveillance using YOLOv11 and explainable hybrid machine learning for autonomous maritime monitoring

The rapid global expansion of offshore wind energy facilities creates new surveillance challenges within coastal transition regions. Motivated by this, the Offshore Zone Risk Weighted Detection (OZRWD) framework is proposed for offshore surveillance and evaluated on the SeaShips coastal dataset. To overcome sea clutter and structural occlusions, OZRWD YOLOv11 integrates a squeeze and excitation channel attention mechanism. On SeaShips, it achieves 0.971 mAP@0.5, outperforming baseline YOLO variants and significantly enhancing the detection of morphologically challenging categories like fishing boats. Following detection, an XGBoost classifier analyses bounding box attributes to attain 93.7% accuracy. This is paired with SHapley Additive exPlanations (SHAP) to provide auditable, feature level decision transparency, bridging the gap between deep learning perception and maritime regulatory compliance. A novel Zone Risk Scoring (ZRS) module then fuses predicted vessel class, spatial proximity, and detection confidence into a continuous operational risk indicator. With processing times of 12.3 ms per frame, this real time, explainable pipeline represents a viable approach for potential deployment in offshore wind energy areas. Future work may extend this architecture via Multi Agent Reinforcement Learning (MARL) for autonomous patrol coordination.