一种星载GNSS-R海面风向反演方法

针对目前海面风场反演中风向反演困难，准确性低的问题，提出一种利用深度网络反演海面风向的新方法，该方法采用星载GNSS-R信号作为遥感源，根据海面反射信号特性与风向相关关系选择反演观测量，建立海面风向反演的深度网络模型。引入ReLU函数作为网络激活函数，并基于遗传算法(GA)优化了网络模型，对输入数据进行时空匹配与归一化处理，通过机器学习建立风向与输入变量的对应关系，利用深度网络模型反演得到海面风向，该方法反演结果的均方根误差εRMSE=18.14°，满足海面风向测量精度均方误差小于20°的一般测量要求，并通过两组对比试验验证了该方法的有效性和鲁棒性。

A Method of Spaceborne GNSS-R Sea Surface Wind Direction Inversion

Aimed at the difficult wind direction inversion with a low accuracy in current sea surface wind field retrieval, a new method using deep network for sea surface wind direction inversion is proposed. The method adopts the spaceborne GNSS R signal as the source of remote sensing, and selects the inversion observations according to the correlation between the surface reflected signal characteristics and the wind direction, so as to establish the deep network model for sea surface wind direction inversion. The ReLU function is introduced as the network activation function, the network model is optimized based on the genetic algorithm (GA), the temporal spatial matching and the normalized processing are conducted for the input data, the corresponding relation between the wind direction and the input variables is established through machine learning, and thus the sea surface wind direction is obtained by using the deep network model inversion. The root mean square error of the inversion results of this method is 18.14°, meeting the general requirement of sea surface wind measurement accuracy with a mean square error less than 20°. The effectiveness and robustness of this method are verified by two groups of contrast tests.