A DEEP IMAGE CLASSIFICATION APPROACH FOR FISHING ACTIVITY DETECTION FROM AIS DATA

Abstract

Maritime transport and vessel activities on the ocean have a significant impact on marine life, which consequently affects human life. Therefore analyzing and monitoring the fishing activities using the vast amount of the data that Satellite-based Automatic Information Systems (S-AIS) provides, using machine learning methods, has become more popular than before. Most of the works on S-AIS data try to detect fishing activities using point-based methods that require significant preprocessing steps to extract meaningful features for the samples. In this work, we use a different perspective toward trajectory data. Human brain cannot understand the type of activity by looking at the point-based data, while experts can easily recognize fishing activity by looking at the movement of a ship on the map. In addition, the significant advances in the field of computer vision made us convert the problem to an image classification task. Informative parts of the trajectories that can contain points where the vessels have been doing an activity are extracted as sub-trajectories using DBSCAN. These sub-trajectories are depicted by drawing the lines between points and saved as images. With the new created image dataset, different CNN models are trained. Our method, unlike other methods, does not need prior information about the movement and can be used for all types of fishing vessels. Our results on the images created from the trajectories of different regions of the world show excellent performance that can be applied for detecting fishing activity from trajectory data.