Improved CycleGAN-based shadow estimation for ocean wave height inversion from marine X-band radar images

Abstract

A novel algorithm is developed to estimate the shadowing ratio for the significant wave height (SWH) inversion of the ocean wave fields imaged by horizontal polarized X-band nautical radar intelligently and conveniently. To solve the problem that the accuracy of the calculated ratio of shadowing in local image areas is not ideal, and the high resolution radar images will lead to time-consuming in estimation of root mean square slope and angle-blurred for sea surface image edge detection, a shadow estimation model from marine X-band radar images based on Convolutional Neural Network (CNN) is established. The model applies the improved CycleGAN to SWH estimation using the geometric shadow effect, which is visible on the marine X-band radar sea surface images due to the presence of the modulation effect of the rough surface. The neural network model can be successfully trained from simulation-based data and then applied to real measured data, and the algorithm does not require any reference measurements. Compared with the traditional shadow-based method, the SWH derived by using this proposed method matches well with that measured by an in-situ buoy nearby, which indicates the goodness of our proposal.