광학 이미지를 이용한 딥러닝 기반 해상 상태 추정 방법의 선박 초기 적용

Abstract

In this study, we utilized a deep learning-based sea state now-casting network to classify the actual sea state. The learning model was constructed as a combined one, which is Convolutional Neural Network(CNN) and Long-Short Term Memory(LSTM). Data was obtained from the southern coastal region of South Korea. With this model, we previously found a significant level of accuracy when performing now-casting on data for lower sea state ranges, which was obtained at a fixed offshore plant. Next, we checked the now-casting performance against measured data from a real ship. We found that it provides a relatively good estimate of the sea state, although the accuracy is lower than that for a fixed structure. Based on our findings, we have summarized a roadmap for expanding training data collection, a roadmap for validation, and a plan for applying the latest deep learning techniques.