A Study on Ship Detection and Classification Using KOMPSAT Optical and SAR Images

The ocean, which occupies approximately 70% of the Earth's surface, is where numerous ships navigate. Most maritime countries use various information systems for the systematic management of the maritime domain. The use of satellites to monitor the oceans is expanding, because they can acquire periodic images over wide areas. Particularly high-resolution satellites-such as the Korea Multipurpose Satellite (KOMPSAT) series-can effectively detect and identify various seagoing ships. In this study, ship detection and classification were performed using high-resolution optical and synthetic aperture radar (SAR) images from the KOMPSAT. To utilize deep learning (DL) technology which has recently been widely used in image processing, the generation of high-quality training data should be prioritized. Therefore, training data were produced for each ship type by combining automatic identification system (AIS) and fishing ship positioning system (V-Pass) information to allow for the differentiation between ship types in KOMPSAT images. In addition, a labeling tool was developed to increase the effectiveness of such training data generation. Subsequently, various DL models were applied to detect and classify ship targets in the KOMPSAT optical and SAR images. These models were developed as quantum geographic information system (QGIS) plugin modules, facilitating ship detection and classification via the QGIS platform and allowing a visualization of results. This paper presents the results of ship detection and classification based on DL models using KOMPSAT optical and SAR imagery.