Neural network-based data fusion methods for automatic identification system and radar data in a shipboard intelligent monitoring system

Abstract

The study addresses the problem of improving the reliability of maritime situational monitoring by reconciling data from the Automatic Identification System (AIS) and radar. The relevance of the study stems from the need to ensure a reliable navigational picture when data from heterogeneous sensor sources are incomplete or contradictory. Real navigational logs of a target vessel were analysed; independent vessel tracks were constructed from each system, and discrepancies in position, speed, and timestamps were calculated. The results show that the two systems have different update rates and exhibit a systematic spatial offset, which substantially complicates their direct joint use without preprocessing. A neural network model for adaptive navigation data fusion is proposed. The model performs intelligent filtering of incoming data and detects anomalies without fixed a priori averaging of the input signals. Its architecture comprises two independent processing branches, one for radar data and the other for AIS data, and an adaptive fusion block that dynamically redistributes confidence weights according to the current uncertainty of each source. A mechanism for detecting instability in navigation data was implemented by analysing reconstruction-error statistics using a variance-based uncertainty criterion. This enables real-time assessment of incoming measurement quality and adaptive adjustment of source weights. Experimental validation on real navigational data demonstrated the operability of the proposed adaptive fusion model and its ability to generate a consistent target-vessel trajectory by dynamically redistributing confidence weights between the sources. The results confirm the functionality of the proposed architecture and its applicability to navigation tasks.