A novel marine radar targets extraction approach based on sequential images and Bayesian Network

This research proposes a Bayesian Network-based methodology to extract moving vessels from a plethora of blips captured in frame-by-frame radar images. First of all, the inter-frame differences or graph characteristics of blips, such as velocity, direction, and shape, are quantified and selected as nodes to construct a Directed Acyclic Graph (DAG), which is used for reasoning the probability of a blip being a moving vessel. Particularly, an unequal-distance discretisation method is proposed to reduce the intervals of a blip's characteristics for avoiding the combinatorial explosion problem. Then, the undetermined DAG structure and parameters are learned from manually verified data samples. Finally, based on the probabilities reasoned by the DAG, judgments on blips being moving vessels are determined by an appropriate threshold on a Receiver Operating Characteristic (ROC) curve. The unique strength of the proposed methodology includes laying the foundation of targets extraction on original radar images and verified records without making any unrealistic assumptions on objects' states. A real case study has been conducted to validate the effectiveness and accuracy of the proposed methodology.