| Abstract | An unmanned surface vehicle (USV) requires the ability to detect adjacent objects and correctly classify them into ships and non-ships. In this study, one of the state-of-the-art neural network based object detection algorithms is applied to detect ships from the images and videos taken on the sea. Firstly, the universally pre-trained “reference model,” which is to detect and classify objects into a range of 20 general classes such as person, dog, cat, table, car, or boat, is prepared and tested. Secondly, the “proposed model” is trained using a public maritime dataset so that it can detect all types of floating objects and classify them into ten specific classes, e.g., ship, speedboat, buoy, etc. The proposed model outperforms the reference model in detecting maritime objects. It also shows real-time speed about 30 frames per second. |
|---|---|
| Publication Date | 2018-06-13 |
Sung-Jun Lee, Myung-Il Roh, Hye-Won Lee, Jisang Ha, Il-Guk Woo, "Image-based Ship Detection and Classification for Unmanned Surface Vehicle Using Real-Time Object Detection Neural Networks", Proceedings of ISOPE 2018, Sapporo, Japan, pp. 726-730, 2018.06.10-15