Ocean Surface Wind Speed Estimation from X-Band Marine Radar Data Using Wavelet Scattering Transform-CNN

Accurate estimation of ocean surface wind speed is critical for oceanographic applications. In this study, a wavelet scattering transform (WST)-convolutional neural network (CNN)-based method is proposed for wind speed estimation from X-band marine radar images. The proposed approach begins with applying WST to radar images to extract hierarchical and multi-scale features that capture the underlying patterns of wind-driven sea surface motion. WST effectively retains critical structural information while reducing the sensitivity to noise. The extracted features are then processed using a CNN which is trained to map these features to the corresponding wind speed values. The proposed method is validated using real radar data collected from a ship-borne Decca radar in a sea area approximately 300 km from Halifax, Canada, in 2008. Comparative experiments were conducted using a classic CNN model and the proposed WST-CNN method. The result shows that WST-CNN achieves significant improvement, with a reduction in root mean square error (RMSE) by 0.14 m/s compared to CNN. Moreover, the correlation coefficient (CC) of the proposed method reaches 0.92, demonstrating superior accuracy and robustness.