Marine vessels or ships have been considered one of the primary vehicles used for sea transportation, which can also be used as an intermediary tool to serve numerous other marine-related activities. In tracking and monitoring the activities of these ships, automatic vessel object detection is undoubtedly challenging to extract the number and position of the vessels from complex seawater backgrounds. In this study, we build a one-stage network of YOLOv5x6 based deep learning model on ShipRSImageNet large-scale dataset. With 50 ship categories, our model obtained a promising performance with a mean average precision of 75.18%. Our findings are potentially beneficial to support maritime security enforcement policy including counter-measuring illegal fisheries and managing seawater traffic surveillance.
Marine Vessels Detection on Very High-Resolution Remote Sensing Optical Satellites using Object-Based Deep Learning
2022-11-03
776410 byte
Conference paper
Electronic Resource
English