Unsupervised Deep Embedding Clustering for AIS Trajectory

Cluster analysis of ship trajectory data collected by Automatic Identification System (AIS) is an important method to study ship behavior patterns and discover traffic laws. Traditional clustering algorithms face the problems of trajectory similarity measurement, feature extraction and clustering parameter setting when dealing with a large number of AIS data. In this paper, an AIS trajectory clustering method based on unsupervised deep embedding is proposed. Using automatic encoder and deep clustering network, the data feature representation and clustering allocation are carried out simultaneously in low dimensional feature space, and iteratively optimizes clustering by minimizing Kullback - Leibler (KL) divergence. The experimental results show that the proposed algorithm can effectively cluster AIS trajectories and accurately extract the main route of ships in the water area.