一种基于CNN-SVR多模态信息融合的GNSS-R风速反演研究

Research on GNSS-R wind speed retrieval based on CNN-SVR multimodal information fusion

针对在全球卫星导航系统反射信号(GNSS-R)技术应用中如何提升海面风速反演总体精度的问题,该文提出一种基于卷积神经网络(CNN)和支持向量机回归(SVR)组合的海面风速反演模型。实验利用气旋全球导航卫星系统(CYGNSS)的L1级数据得到预测风速与参考风速数据相关系数达到了0.89,风速反演的总体均方根误差(RMSE)为1.23m/s,较传统方法的RMSE降低了12.7%~45.3%。

In view of the problem that how to improve the overall accuracy of sea surface wind speed retrieval in the application of global navigation satellite system reflectometry(GNSS-R),a combined sea surface wind speed retrieval model based on convolutional neural networks(CNN)and support vector regression(SVR)was proposed.The L1level data of cyclone global navigation satellite system(CYGNSS)was used in the experiment to obtain a correlation coefficient of 0.89between the predicted wind speed and the reference wind speed data,and the overall root mean square error(RMSE)of wind speed retrieval was 1.23m/s,which was 12.7%~45.3%lower than the RMSE of the traditional method.