Context-based behaviour modelling and classification of marine vessels in an abalone poaching situation

A decision-support system for combating abalone poaching is proposed. A dynamic Bayesian network (DBN) is used to model context-based behaviour of vessels in a maritime abalone poaching situation. The context and behaviour is informed by Expert knowledge. The model is utilised for both data generation and behaviour classification. Data generation is performed by sampling in the DBN. The result is that a set of vessels are simulated in an abalone poaching situation. Several vessel classes including poaching, patrol, fishing, tourist, and recreational vessels are modelled. The generated data is intended to model surveillance data that may have been produced by sensors such as optical, infrared or radar sensors. Classification is performed using the filtering and smoothing inference methods on the DBN. A vessel class is inferred given tracked vessel data and contextual information. The purpose is to identify vessels that exhibit poaching behaviour. The novelty of this work includes a derivation of the generalised pseudo Bayes smoothing algorithm for the classification model. This smoothing algorithm is demonstrated to provide more accurate classification results than the previously proposed filtering method.