Memory and skip variational autoencoder: a novel approach for ship trajectory anomaly detection

Reconstruction-based methods serve as an important role in ship trajectory anomaly detection. Owing to down-sampling during the reconstruction process, detailed information from the original images is lost in potential encoding representations, resulting in blurred reconstructed images; Simultaneously, the process of selecting thresholds is quite complex to judge abnormal trajectories, with low efficiency and credibility. In view of this, an anomaly detection model for ship trajectory data based on the Memory and Skip Variational Autoencoder (MSVAE) is proposed by using an unsupervised method. Initially, we input ship trajectory images into encoder and decoder. Then, feature extraction and classification of reconstructed trajectory data are carried out. In addition, we introduce an adversarial network that assists to identify anomaly by comparing the features of the original and the reconstructed. Finally, we calculate the anomaly score to judge whether the trajectory data is abnormal. The research results indicate that the detection rate of the proposed model is 94.5%, and the false alarm rate is 0.468%, which is better than the current models. This research can provide technical support for ship trajectory data analysis and risk management of maritime transportation.