A semi-supervised deep learning model for ship encounter situation classification

Maritime safety is an important issue for global shipping industries. Currently, most of collision accidents at sea are caused by the misjudgement ship’s operators. The deployment maritime autonomous surface ships (MASS) can greatly reduce ships’ reliance on human operators using automated intelligent avoidance system to replace decision-making. To successfully develop such a system, capability autonomously identifying other and evaluating their associated encountering situation paramount importance. In this paper, we aim identify encounter modes deep learning methods based upon Automatic Identification System (AIS) data. First, segmentation process developed divide each AIS data into different segments that contain only one mode. This majority studies have proposed mode classification hand-crafted features, which may not reflect actual movement states. Furthermore, number present tasks conducted substantial labelled followed supervised training paradigm, applicable our dataset as it contains large unlabelled Therefore, method called Semi-Supervised Convolutional Encoder–Decoder Network (SCEDN) ship proposed. structure network able automatically extract features from but also share parameters SCEDN uses encoder–decoder convolutional with four channels segment (distance, speed, Time Closed Point Approach (TCPA) Distance (DCPA)) been developed. performance model evaluated comparing several baselines experimental results demonstrating higher accuracy be achieved model.