수리모형실험 영상의 딥러닝 분석을 통한 연안류 정량화

Abstract

In this study, we propose a deep learning method to quantify nearshore currents from video imagery acquired from wave flume experiments. A framework of convolutional neural network with skip connection is implemented for deep learning model and the model is trained with video imagery by deriving optimal vector fields for spatial transformation between two successive frames. By applying the model to video, the velocity for 10 minutes is computed with 1/3 seconds time interval along the wave flume.