基于遗传神经网络算法的海面风速反演新方法

针对海面运动的复杂性、海面电磁散射理论模型的局限性以及利用SAR图像反演海面风速存在的非线性现象，基于遗传神经网络的方法，以业务化的CMOD4模式函数数据为基础，采用Fletcher-Reeves算法的变梯度反向传播算法，建立一种SAR风速反演的新模型。试验结果表明，利用遗传神经网络方法反演海面风速是可行的，当随机误差小于10%时，模型的抗噪能力较强，风速反演的精度较为理想。比较不同风速下的反演结果可以发现，在中、小风速的情况下，模型的抗噪能力较强，模型学习拟合和预测检验的精度相对较高；在大风速的情况下，模型的反演能力有待于进一步提高。

New method of sea surface wind speed retrieval using genetic algorithms and neural network

There exist the complexity of sea surface and the limitation of theoretical model of electromagnetic scattering, and nonlinear phenomena in the retrieval of sea surface wind speed, which is based on synthetic aperture radar (SAR) images. With the method of genetic neural network and Fletcher Reeves, this paper established a new model of retrieving wind speed based on operational data of CMOD4 model function. The result shows that this model is available in retrieving ocean surface wind. When random error is less than 10%, this model has high denoising ability and the accuracy of the retrieved ocean surface wind speed is ideal. Comparing the results of different wind speed, shows that in the case of low or middle wind, the fitness of learning model and the accuracy of predicted tests have both ideal accuracy, and that in the case of strong wind, the inversion result of this model is comparatively poor.