| 摘要: |
| 电磁逆散射问题是非线性和病态的，传统的求解方法无法兼顾成像精度和计算效率，而基于深度学习的直接重构方法缺失先验信息，导致学习过程较困难.采用结合衍射层析成像（DT）算法和融合注意力机制的U-Net混合电磁重构方案，求解电磁逆散射问题，将基于Born近似的DT算法重建的粗糙图像作为U-Net的输入，有效利用先验信息，提高逆散射问题求解的效率和精度.此外，在U-Net每次的下采样过程中加入注意力机制，进一步提高了目标散射体相对介电常数和位置的重建精度.实验结果表明，相比未引入注意力机制的方案，融合注意力机制的U-Net混合电磁重构方案重建误差较小，可实现相对介电常数分布的高精度重建. |
| 关键词: 电磁逆散射 U-Net 注意力机制 衍射层析成像(DT)算法 |
| DOI：10.3969/J.ISSN.1000-5137.2023.02.008 |
| 分类号:TP391.4 |
| 基金项目: |
|
| Hybrid electromagnetic reconstruction scheme based on U-Net with attention mechanism |
|
YANG Xirui, YANG Chunxia
|
|
College of Information, Mechanical and Electrical Engineering, Shanghai Normal University, Shanghai 201418, China
|
| Abstract: |
| Electromagnetic inverse scattering problems were nonlinear and ill-conditioned. The traditional method could not take into account the imaging accuracy and computational efficiency simultaneously. Meanwhile, the direct reconstruction method based on deep learning lacked prior information, which made the learning process difficult. A hybrid electromagnetic reconstruction scheme combining diffraction tomography (DT) algorithm and U-Net with attention mechanism was proposed to solve the electromagnetic inverse scattering problem. The rough image reconstructed by DT algorithm based on Born approximation was used as the input of U-Net, so that the prior information was effectively used to improve the efficiency and accuracy of inverse scattering problem. In addition, by adding the attention mechanism during each U-Net subsampling, the reconstruction accuracy of the relative dielectric constant and position of the target scatterer was further improved. The experimental results showed that the reconstruction error of U-Net hybrid electromagnetic reconstruction scheme with attention mechanism was much smaller than that without attention mechanism, which could achieve high precision reconstruction of relative dielectric constant distribution. |
| Key words: electromagnetic inverse scattering U-Net attention mechanism diffraction tomography (DT) algorithm |
