Ship Trajectory Classification using CNN-Based Deep Learning Method: A Detailed Analysis

Abstract

Translate Article

Trajectory classification is a well-known class of problems in movement data analysis. The technical advancement in GPS enabled devices and autonomous systems has made it possible to collect movement data of objects and events in real time. This leads to massive and complex trajectory data that has associated movement patterns. Trajectory analysis becomes challenging due to its voluminous size and large dimensionality. Furthermore, traditional methods fail to handle this massive data for efficient processing. So, the focus is shifted to the automation of trajectory data analysis by researchers and practitioners. Deep Learning (DL) based techniques are efficient in the automatic handling of trajectory analysis. It is quite promising in solving many trajectory-related tasks, such as classification, prediction, segmentation, etc.The current paper delves into the trajectory classification activity using deep learning techniques. It specifically investigates Convolutional Neural Networks (CNNs) based methodologies to solve classification tasks. Due to the increased accuracy and reliability of CNN models, their applicability for trajectory classification is widely being investigated. Our work brought a deep study of trajectory classification and implemented a few notable CNN architectures, viz. ResNet152, NasNet, EfficientNet, DenseNet, and Xception. It presents a comparative analysis of these CNN methods to classify trajectory data. The methods are tested on the Automated Information System (AIS) dataset, which is part of the ship’s onboard Automated Information System (AIS). The AIS datasets are preprocessed and made ready for the implemented system.