해상 이미지를 활용한 3D 합성곱 신경망과합성곱 장단기 메모리 신경망 기반의 유의 파고 추정Significant Wave Height Regression from a Raw Ocean Image with Convolutional LSTM and 3D Convolutional Networks
- Other Titles
- Significant Wave Height Regression from a Raw Ocean Image with Convolutional LSTM and 3D Convolutional Networks
- Authors
- 손규빈; 최희정; 이정호; 노영빈; 강필성
- Issue Date
- 2020
- Publisher
- 한국경영과학회
- Keywords
- Significant Wave Height; 3D Convolution; Convolutional LSTM; Image Processing
- Citation
- 한국경영과학회지, v.45, no.1, pp.11 - 24
- Indexed
- KCI
- Journal Title
- 한국경영과학회지
- Volume
- 45
- Number
- 1
- Start Page
- 11
- End Page
- 24
- URI
- https://scholar.korea.ac.kr/handle/2021.sw.korea/59684
- ISSN
- 1225-1119
- Abstract
- One of the most common measures implemented in the operation of large vessels is to find the route that takes the least fuel consumption based on marine conditions, such as wave height. The model that predicts wave height can roughly be categorized into two methods, namely, a numerical method that calculates by physical formula and a soft-computing method that collects weather information and learns the machine learning algorithm. These models are difficult to apply in the real world because of their high computational complexity and the use of expensive radar equipment. In this study, we propose to estimate the wave height in real time using the images of the ocean. We used the image data consisting of four consecutive images instead of a single image and applied the combination of convolutional LSTM and 3D CNN networks that can best handle the data structure as a regression model. In this way of prediction, existing methods are not only outperformed but are also more robust to outliers. We used data from the “Weather 1st” ship provided by Daewoo Shipbuilding & Marine Engineering and confirmed that the mean absolute error is 1.59 cm, and the mean absolute percentage error is as low as 1.61% based on the test set.
- Files in This Item
- There are no files associated with this item.
- Appears in
Collections - College of Engineering > School of Industrial and Management Engineering > 1. Journal Articles
Items in ScholarWorks are protected by copyright, with all rights reserved, unless otherwise indicated.