APPROACH TO CLASSIFYING SHIP TYPES FROM AIS DATA USING DNN AND CNN

Abstract

Busan Port in Korea ranked third to firth busiest port of the world during the last few years. Hence, there are high risks of ship accidents due to the complex marine traffic environment which is comprised of ships engaged in several types of activities like maritime trades, passenger transportation, different methods of fishing, as well as various port activities. Therefore, it is necessary to develop a method for identifying types of cargo ships. In this regard deep neural network (DNN) based classification of vessel types on AIS data on 2021.02.05 was accomplished for the location of Busan Port area. The AIS data were first quality controlled for removing data outliers and then 80% of the data were used for training the models. Finally, the rest 20% data were used for testing purpose. DNN based ship classification results showed 71% accuracy. Classification of the ships similarly with CNN model is in progress and thus as a comparative study CNN based results of classification will be compared with those of DNN. Thus, based on the current DNN results could be suggested to classify ships from AIS data for the research area, and this research will be continued for further improvements in order to achieve increased accuracy.