Sea surface wind and wave parameter estimation from X-band marine radar images with rain detection and mitigation

Chen, Xinwei (2021) Sea surface wind and wave parameter estimation from X-band marine radar images with rain detection and mitigation. Doctoral (PhD) thesis, Memorial University of Newfoundland.

[img] [English] PDF - Accepted Version
Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission.

Download (24MB)


In this research, the application of X-band marine radar backscatter images for sea surface wind and wave parameter estimation with rain detection and mitigation is investigated. In the presence of rain, the rain echoes in the radar image blur the wave signatures and negatively affect estimation accuracy. Hence, in order to improve estimation accuracy, it is meaningful to detect the presence of those rain echoes and mitigate their influence on estimation results. Since rain alters radar backscatter intensity distribution, features are extracted from the normalized histogram of each radar image. Then, a support vector machine (SVM)-based rain detection model is proposed to classify radar images obtained between rainless and rainy conditions. The classification accuracy shows significant improvement compared to the existing threshold-based method. By further observing images obtained under rainy conditions, it is found that many of them are only partially contaminated by rain echoes. Therefore, in order to segment between rain-contaminated regions and those that are less or unaffected by rain, two types of methods are developed based on unsupervised learning techniques and convolutional neural network (CNN), respectively. Specifically, for the unsupervised learning-based method, texture features are first extracted from each pixel and then trained using a self organizing map (SOM)-based clustering model, which is able to conduct pixel-based identification of rain-contaminated regions. As for the CNN-based method, a SegNet-based semantic segmentation CNN is �rst designed and then trained using images with manually annotated labels. Both shipborne and shore-based marine radar data are used to train and validate the proposed methods and high classification accuracies of around 90% are obtained. Due to the similarities between how haze affects terrestrial images and how rain affects marine radar images, a type of CNN for image dehazing purposes, i.e., DehazeNet, is applied to rain-contaminated regions in radar images for correcting the in uence of rain, which reduces the estimation error of wind direction significantly. Besides, after extracting histogram and texture features from rain-corrected radar images, a support vector regression (SVR)-based model, which achieves high estimation accuracy, is trained for wind speed estimation. Finally, a convolutional gated recurrent unit (CGRU) network is designed and trained for significant wave height (SWH) estimation. As an end-to-end system, the proposed network is able to generate estimation results directly from radar image sequences by extracting multi-scale spatial and temporal features in radar image sequences automatically. Compared to the classic signal-to-noise (SNR)-based method, the CGRU-based model shows significant improvement in both estimation accuracy (under both rainless and rainy conditions) and computational efficiency.

Item Type: Thesis (Doctoral (PhD))
Item ID: 15263
Additional Information: Includes bibliographical references (pages 138-152).
Keywords: X-band marine radar, ocean, wind and wave parameter, rain, machine learning, image processing
Department(s): Engineering and Applied Science, Faculty of
Date: October 2021
Date Type: Submission
Digital Object Identifier (DOI):
Library of Congress Subject Heading: Radar; Radar in navigation; Ocean-atmosphere interaction; Ocean waves; Support vector machines; Precipitation (Meteorology)--Measurement; Remote sensing; Rain and rainfall.

Actions (login required)

View Item View Item


Downloads per month over the past year

View more statistics