Sea Surface Wind Direction Retrieval Based on Convolution Neural Network and Wavelet Analysis

Sea surface wind streak is one of many geophysical phenomena in synthetic aperture radar (SAR) images, which often used to obtain sea direction. At present, the recognition streaks mainly depends on artificial experience, and efficiency accuracy are not high. In this study, transfer learning based convolutional neural network architecture Inception v3 was introduced streaks. Four categories imaged by gaofen-3 (GF-3) SAR from 2019 2020 were chosen for retraining full pre-retrained model. Then, we use retrained model identify GF-3 2018 it retrieve The results show that method effective. can reach 92.0% 95.2% after data augmented. Compared with reanalysis european centre medium-range weather forecasts (ECMWF), root mean square error retrieved direction 9.12, further verifies ability training