Enhanced Estimation of Significant Wave Height From Rain-contaminated X-Band Radar Image Sequences

As a result of the rain contamination on radar images, the accuracy of significant wave height (SWH) estimation from the radar images collected under rainy conditions is adversely affected. This paper presents a novel approach for enhancing SWH estimation from rain-contaminated radar images by combining DehazeNet and convolutional gated recurrent unit (CGRU) networks. Firstly, a CNN-based dehazing algorithm, i.e., DehazeNet, is employed to correct the visual degradation caused by rain in the radar images. Subsequently, the CGRU network is utilized to perform further SWH measurements from the dehazed radar images. The training and testing data used in this study were acquired by a shipborne radar in 2008 from a maritime area off the southeastern coast of Halifax, Canada. Besides, floating buoys deployed around the ship provided real-time SWH measurements, serving as the ground truth for the experiment. The experimental result demonstrates that the estimation accuracy, with a root-mean-square error (RMSE) of 0.47 m, obtained from the dehazed images surpasses the RMSE of 0.53 m achieved by applying regression networks directly to the original radar images.