摘要
针对GNSS-R进行海面风速反演过程中时频域相关物理量较多,数据耦合性强等问题,提出了基于反向传播(BP)神经网络反演海面风速的方法。建立反演过程中相关观测量与风速的对应关系,选取多观测量作为输入,对输入数据进行处理,设置神经元与激励函数,使用BP神经网络自适应调整拟合参数,将风速作为神经网络输出端的特征量提取。反演结果,风速≤20m/s时,反演均方根误差RMSE=1.21m/s,风速>20m/s时反演均方根误差RMSE=2.54m/s,反演结果优于使用时延相关曲线前沿斜率(LES)和时延多普勒相关功率均值(DDMA)方法得到的反演结果,且迭代次数较少,复杂度较低,证明该方法可以应用于GNSS-R海面风速反演。
In the sea surface wind speed inversion of GNSS-R,the time-frequency domain related physical quantity is large,and the data coupling is strong.A method of inversion of sea surface wind speed based on Back-Propagation(BP)neural network is proposed.This paper establishes the corresponding relationship between the correlation observation and the wind speed in the inversion process.Selecting the multi-view measurement as the input,the input data are processed,the neuron and the excitation function are set.using the BP neural network,the fitting parameters are adaptively adjusted.And the wind speed is used as the extraction feature in the neural network.The inversion results show that when the wind speed is≤20 m/s,the inversion Root Mean Square Error(RMSE)is 1.21 m/s,and the inversion RMSE is 2.54 m/s when the wind speed is>20 m/s.The result is better than the inversion results obtained by the Delay Leading Edge Slope and Delay-Doppler Average methods,and the number of iterations is small and the complexity is low,proving that the method can be applied to GNSS-R sea surface wind speed inversion.
作者
高涵
白照广
范东栋
GAO Han;BAI Zhaoguang;FAN Dongdong(DFH Satellite Co.,L td.,Beijing 100094,China)
出处
《航空学报》
EI
CAS
CSCD
北大核心
2019年第12期193-201,共9页
Acta Aeronautica et Astronautica Sinica
