Method for Detecting Unknown Floating Objects in Maritime Environment Using Efficient GAN

In recent years, research and development of maritime object detection technology has been conducted to support onboard monitoring, with the goal of eventually realizing autonomous ship operation and remote ship maneuvering. However, in a real-world environment, there may be unknown objects not found in training data, and their detection is challenging. Conventional object detection methods often do not consider the detection of these unknown objects. In this study, we propose a new method for detecting unknown objects when areas of an image differ from expected sea surface characteristics using a GAN-based anomaly detection method. Experiments using a prototype implementation of the system confirmed that the proposed system can detect floating objects on the sea without learning specific objects.