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
陈佳琳 1杨春夏1
作者信息
- 1. 上海师范大学 信息与机电工程学院,上海 201418
- 折叠
摘要
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.
Abstract
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.
关键词
电磁逆散射/多尺度/深度学习/Lenet/SmaAt-UNetKey words
electromagnetic inverse scattering/multiscale/deep learning/Lenet/SmaAt-UNet引用本文复制引用
出版年
2024
NETL
NSTL
万方数据