남권우, 노명일, 이혜원, 이원재, "자율 운항 선박을 위한 딥 러닝 기반 선박 이미지 분류 방법", 한국CDE학회 논문집, Vol. 26, No. 2, pp. 144-153, 2021.06.01
Domestic Journal
2021.03.24 09:50
남권우, 노명일, 이혜원, 이원재, "자율 운항 선박을 위한 딥 러닝 기반 선박 이미지 분류 방법", 한국CDE학회 논문집, Vol. 26, No. 2, pp. 144-153, 2021.06.01
조회 수 269
| 첨부 '1' |
|---|
| Abstract | In the last few years, researches on autonomous ships have attracted attention. One of the essential techniques required for autonomous ships is the awareness of surroundings, including detection and classification of objects. Although researches on computer vision regarding the classification of ship images are still making progress, it is challenging to encounter a lack of enough database that was adequately labeled for ship classification. In this study, data obtained from Singapore Maritime Dataset (SMD) and public datasets such as MARVEL, FleetMon, and VesselFinder were labeled and integrated into a unified dataset for further study for ship classification. The ship image dataset was classified into seven classes, including bulk carrier, container ship, cruise ship, naval surface ship, tanker, tug boat, and buoy. Subsequently, Convolutional Neural Networks (CNNs) based on GoogleNet, VGG16, and ResNet were implemented for ship image classification, and a comparative test was done. As a result, the CNNs which were trained with the unified dataset showed high accuracy. The classification results were analyzed by the heatmap visualization with Grad-CAM, which indicates critical features best activating each class of ships, and further discussion was made. |
|---|---|
| Publication Date | 2021-06-01 |
| Role | Corresponding Author |
| Category | KCI |
