Spatial-Temporal Analysis of Ship Collision Risk

Abstract

The ongoing increase in maritime traffic has caused an impact on the occurrence of safety accidents at sea. Consequently, finding out the causal relationship to these accidents has become an important field of research to ensure safe navigation. Since the scale and the cost of the damage caused by ship collisions are tremendous, research efforts are currently being actively conducted to avoid catastrophes by analyzing their characteristics, which are mostly influenced by human factors. This paper presents an approach that enables the investigation of ship collisions at sea in terms of both space and time. By utilizing the AIS data, a map is created that represents the relative traffic density, which indicates the relative proximity between ships. Then, using the maps created, a Convolutional Neural Network (CNN) model is trained for the classification of ship collision. The approach presented in this study aims to enhance maritime safety through the utilization of spatial-temporal analysis of collision occurrences. First, from the perspective of spatial analysis, Grad-CAM is utilized to investigate a specific region of the density map that has a significant impact on collision occurrences. Second, from a temporal perspective, the study also analyzes changes in marine traffic and the timing of collisions using transition analysis in relation to collision events. From the maps created from AIS data, we showed the difference between cases where ship accidents occurred and cases where accidents did not occur in influence from the perspectives of time and location using the maps proposed in this study.