一种反演散射障碍的循环神经网络模型

A Recurrent Neural Network Model for Inverse Scattering Obstacles

针对通过相关的远场数据反演障碍物形状的反散射问题,构建一种基于门控思想的循环神经网络模型。本文将远场数据和障碍物边界曲线方程的傅里叶系数分别作为网络模型的输入和输出,利用网络模型中的门控思想有选择地提取远场数据中障碍物的特征信息,并通过Adam优化算法更新网络模型的权重和偏置,进而反演障碍物的形状参数。最后,数值实验说明该循环神经网络模型的有效性。

Aiming at the inverse scattering problem of recovering the shape of obstacles, a recurrent neural network model is constructed which is based on the gated thought. This paper considers the far-field data and the Fourier parameters of obstacle boundary curve equation as the input and output of the network model, respectively. The characteristic information of obstacles in the far-field data is selectively extracted through the gated thought of the proposed network model. The Adam optimization algorithm is applied to update weight matrices and biases of the network model in order to inverse the obstacle shape. Finally, the numerical experiments demonstrate the effec-tiveness of this recurrent neural network model.