基于多尺度学习的电介质目标定位与重构

Dielectric target localization and reconstruction based on multi-scale learning

利用神经网络将电磁逆散射问题与多尺度方法相结合,通过将散射场的场强数值输入多尺度融合模型中进行不断训练,实现目标的定位与重构.对于目标区域内的手写数字散射体,首先利用Lenet网络模型定位目标散射体所在的区域;然后将散射体所在的区域进一步通过SmaAt-UNet神经网络学习,训练重构散射体的形状,进而确定该数字,不同的模型负责提取不同的特征;最后将特征融合在一起,以增强最终结果的表征能力.

The electromagnetic inverse scattering problem was combined with multi-scale method by using neural network in this paper.The target location and reconstruction were realized by inputting the field strength value of scattering field into multi-scale fusion model for continuous training.Firstly,for the handwritten digital scatterer in the target area,the Lenet network model was adopted to locate the area where the target scatterer was.Secondly,the area where the scatterer located was further learned by SmaAt-UNet neural network,and the shape of the reconstructed scatterer was trained to determine the number.Different models were responsible for extracting different features respectively.Finally,these features were integrated to enhance the characterization ability of the final result.