A Study on Acquiring of Dataset for Training AI Model of Maritime Weapon Systems

Abstract

Recently, the defense field has been actively adopting innovative technologies to enhance the performances of weapon systems and strengthen combat capabilities. Among these technologies, Artificial Intelligence technology plays an important role in enabling autonomous operation and accurate decision-making of weapon systems. However, since AI models must be learned relying on using large amounts of high-quality dataset, acquisition of dataset for training weapon systems is recognized as an important task in the defense field. This study aims to construct highly available dataset for robustly training maritime weapon systems and apply them to an artificial intelligence model. To this end, efforts were made to overcome the limitations of existing maritime dataset(challenges): lack of information obtained from a ship perspective and low diversity/training effect. The maritime dataset in this study are constructed focusing on original data obtained from an actual ship's perspective. To achieve this, the fusion of day and night images and location information was carried out using not only color cameras and infrared cameras but also distance measuring devices and radar. The dataset constructed in this way are completed as processed data including metadata through labeling and annotation. Furthermore, a suitable object tracking AI algorithm was selected considering the characteristics of maritime objects such as dynamic and frequent occlusion. Compared to the performances of the AI tracking models trained with existing dataset, this study confirmed that performances are improved when trained with newly constructed dataset. From the results, it is clear that this study contributes to select an AI model suitable for actual maritime situations and the acquisition of dataset for training it. It is also anticipated that these dataset can be used as a base material for training maritime weapon systems in the future. The integration of these dataset and AI model is expected to make a significant contribution to promoting the application of artificial intelligence technology in the defense field and supporting the efficient operation of weapon systems.