A Novel Scheme for Extracting Sea Surface Wind Information from Rain-Contaminated X-Band Marine Radar Images

Abstract

The presence of rain degrades the performance of sea surface parameter estimation using X-band marine radar. In this article, a novel scheme is proposed to improve wind measurement accuracy from rain-contaminated X-band marine radar data. After extracting texture features from each image pixel, the rain-contaminated regions with blurry wave signatures are first identified using a self-organizing map (SOM)-based clustering model. Then, a convolutional neural network used for image haze removal, i.e., DehazeNet is introduced and incorporated into the proposed scheme for correcting the influence of rain on radar images. In order to obtain wind direction information, curve fitting is conducted on the average azimuthal intensities of rain-corrected radar images. On the other hand, wind speed is estimated from rain-corrected images by training a support vector regression-based model. Experiments conducted using datasets from both shipborne and onshore marine radar show that compared to results obtained from images without rain correction, the proposed method achieves relatively high estimation accuracy by reducing measurement errors significantly.