Design of a Deep Learning-Based Ship Navigation Status Monitoring Model and Validation of Its Anomaly Detection Performance

Abstract

Addressing the limitations of traditional ship navigation status monitoring, including manual inspection dependence, delayed anomaly identification, high false alarm rates, and inadequate adaptability to complex maritime environments, this paper systematically investigates a deep learning-based framework for navigation status monitoring and abnormal behavior detection. As modern maritime surveillance increasingly relies on electromagnetic sensing networks and wireless communication technologies for reliable information acquisition and transmission, intelligent monitoring has become essential for navigation safety. First, based on ship dynamics, deep learning, and computer vision theories, a comprehensive technical framework integrating data acquisition, preprocessing, feature extraction, state modeling, and anomaly detection is established to define core navigation state indicators and abnormal behavior categories. Second, by fusing Automatic Identification System (AIS) data, navigation sensor data, and video surveillance information, a multisource dataset (MSS-2024) covering 15 navigation states and 8 representative abnormal behaviors is constructed. Third, a dual-branch Transformer-CNN fusion model incorporating attention mechanisms and multi-scale feature fusion is developed to enhance state representation and anomaly recognition under complex conditions. Finally, extensive experiments based on real-world vessel navigation data validate the proposed approach, demonstrating its effectiveness in improving recognition accuracy, response efficiency, and robustness, while providing technical support for intelligent maritime electromagnetic monitoring and navigation safety assurance.