A Hybrid CNN-Transformer Network for Global Ocean Wind Speed Retrieval from GNSS-R Data

Global navigation satellite system reflectometry (GNSS-R) has emerged as a promising technique for global sea surface wind speed measurement. Currently, a lot of methods based on convolutional neural networks (CNNs) have been developed to extract features from delay Doppler maps (DDMs) and retrieve wind speed from GNSS-R data. However, CNNs emphasize local features and cannot capture global features due to their inherent locality. In order to address this limitation, a hybrid CNN-Transformer Network (CTN) is proposed in this study. Specifically, the hybrid CTN consists of a CNN branch tailored for capturing local features and a Transformer branch for global features. By combining these distinct features, the designed hybrid CTN facilitates a comprehensive representation of DDMs. Experiments conducted on Cyclone GNSS (CYGNSS) data demonstrate improved accuracy and robustness of the proposed method in wind speed retrieval from GNSS-R data.