Proposal of neural approach to maritime radar and automatic identification system tracks association

This study presents results of a study on association of targets tracked with an automatic identification system and marine tracking radar. This paper shows various approaches for solving association problem. Description of selected numerical methods is given, followed by an experimental evaluation, showing advantages and disadvantages for each of them. On the other hand, novel approach is proposed: namely, the use of neural networks for this task. Association is defined as a classification problem and suitable neural networks are selected and trained. Each pair of targets is being classified with neural filter, stating if association takes place. The main advantage of this approach is its flexibility in adjusting to current data. Neural networks can be trained during data processing to adjust to current situation, which is not the case in a priori defined numerical models. Experimental verification of the proposed solution and comparison of neural networks with numerical methods is included. The results obtained indicate that neural networks are an interesting alternative to numerical methods.