基于卷积神经网络的色散介质电磁参数重构

色散介质的经验模型适合描述等离子体、水、生物肌体组织等媒介。为了反演色散媒介的电磁参数,提出一种基于卷积神经网络的色散介质电磁参数反演方法。在电磁参数反演的过程中,利用前向算法获得色散介质的散射电场,反演算法通过卷积神经网络将原逆散射问题转化为一个回归估计问题。提取不同频率TM波照射下色散介质的散射电场值的实部和虚部作为样本信息并作为卷积神经网络的输入,色散介质电磁参数作为输出,经过适当的训练,重构出自由空间中色散介质圆柱体电磁参数。经过与BP神经网络反演结果的比较,验证了该方法的有效性及准确性。

Reconstruction of Electromagnetic Parameters of DispersiveMedia Based on Convolutional Neural Network

The empirical models of dispersive media are suitable for describing media such as plasma, water, and biologicaltissues. In order to reconstruct the electromagnetic parameters of the dispersive media, this paper proposes a methodfor inversion of electromagnetic parameters of dispersive media based on convolutional neural networks. In the process of electromagneticparameters inversion, the forward algorithm is used to obtain the scattered electric field values of the dispersivemedia, and the backward algorithm transforms the original inverse scattering problem into a regression estimation problemthrough the convolutional neural network. The real and imaginary parts of the scattered electric field values of the dispersivemedia radiated by TM waves of different frequencies are extracted as the sample information and the input of the convolutionalneural network, and the electromagnetic parameters of the dispersive media are taken as the output. After proper training,the electromagnetic parameters of the dispersive medium cylinder in free space are reconstructed. The comparison with the inversionresults of BP neural network verifies the effectiveness and accuracy of the method in this paper.