Segmentation Optimization in Trajectory-Based Ship Classification

Abstract

An analysis over trajectory segmentation techniques is carried out by the study of the different algorithms and the experimentation over a ship classification problem, which use a data preparation and classification system used in previous works. With the data preparation, the system handles real-world Automatic Identification System (AIS) data, cleaning wrong measurements and smoothening the trajectories by the application of an Interacting Multiple Model (IMM) filter. Also applies some balancing algorithms to address the lack of an equal distribution among classes. To correctly evaluate the classification with the imbalanced data a multiple objective analysis is proposed to consider the minority class and the global accuracy. Over that multi-objective analysis, different segmentation algorithms and its variations are tested to analyze the influence of them into the classification problem. The results show a Pareto front with different viable solutions for the proposed multi-objective problem, without a dominant algorithm over rest of the tested segmentation algorithms.