【摘要】 为解决电磁层析成像（electromagnetic tomography,EMT）传统成像算法由于逆问题的不适定性和病态性导致重建图像质量差的问题，提出了一种基于改进U-Net深度网络模型的新型电磁层析成像方法。首先，以UNet深度网络模型为基础，加入残差模块使网络提取更多特征信息并避免网络训练时梯度消失的问题；其次在此结构上引入注意力机制来提升重要特征信息，抑制无用的特征信息，加强对缺陷边缘和形状特征的权重分配。通过仿真和金属缺陷检测实验评估了本文所提出算法的性能，并与线性反投影算法和共轭梯度算法进行了对比。仿真实验和金属缺陷检测实验结果表明：本文提出的算法在精确率、召回率和F1-Score分别达到88.41%、90.38%和89.38%，重建图像对于缺陷位置和形状的预测更为准确。更多还原

【Abstract】 In order to solve the problem of poor quality of reconstructed images due to the ill-posed and ill-posed inverse problem of traditional electromagnetic tomography(EMT) imaging algorithms. In this paper, a novel electromagnetic tomography method based on the improved U-Net deep network model is proposed. First, based on the UNet deep network model, a residual module is added to enable the network to extract more feature information and avoid gradients during network training. Secondly, an attention mechanism is introduced into this structure to enhance important feature information, suppress useless feature information, and strengthen the weight distribution of defect edge and shape features. The performance of the proposed algorithm is evaluated by simulation and metal defect detection experiments, and compared with the linear backprojection algorithm and the conjugate gradient algorithm. The results of simulation experiments and metal defect detection experiments show that the algorithm proposed in this paper achieves 88.41%, 90.38%, and 89.38% in precision, recall and F1-Score, respectively,and the reconstructed image is more accurate in predicting the location and shape of defects.更多还原