WAVE ESTIMATION FROM COASTAL IMAGES BY DEEP LEARNING

Abstract

 For long-term coastal monitoring, deep learning was applied to images from fixed cameras installed on the coast to estimate wave conditions. Datasets created from images of the Shichiri Mihama Ida coast in Mie Prefecture and the Hasaki coast in Ibaraki Prefecture and wave information obtained from NOWPHAS were used to compare wave estimation accuracy. Results showed that the machine provided good estimates of significant wave heights and the angle of the image suitable for estimating significant wave heights differed from coast to coast. Although the machine tended to underestimate the significant wave height, it was able to reduce this tendency by capturing images in a different direction of the incident wave. The accuracy of the estimation was further improved by using images taken from multiple shooting directions.