摘要
电阻抗层析成像因其无辐射、非侵入的测量方式备受青睐,但边界电压与被测场域内电导率分布间固有的非线性和逆问题的病态性导致成像质量欠佳。针对此问题,该文将RBF神经网络与深度神经网络U-Net结合,形成一种新的、称之为RBF-V-Net的V型网络结构。用RBF神经网络实现预映射,用U-Net实现特征提取和深度重建,用残差连接解决网络退化问题,从而实现由边界电压值到电导率分布的更高精度映射。仿真重建实验结果表明,50 dB信噪比下重建图像相对误差和相关系数的平均值分别为0.0224和0.9991;与传统算法TR、Landweber以及神经网络6L-FCDN相比,该文方法的重建图像更接近真实分布。
Electrical impedance tomography is favored for its non-radiation and non-invasive measu-rement methods.However,the inherent nonlinearity between the boundary voltage and the conductivity distribution in the measured field and the ill-posedness of the inverse problem lead to poor imaging quality.Aiming at this problem,this paper combines RBF neural network with deep neural network U-Net to form a new V-shaped network structure called RBF-V-Net.RBF neural network is used to realize pre-mapping,U-Net is used to realize feature extraction and deep reconstruction,and residual connection is used to solve the problem of network degradation,so as to realize higher precision mapping from boundary voltage value to conductivity distribution.The simulation results show that the average relative error and correlation coefficient of the reconstructed image under 50 dB SNR are 0.0224 and 0.9991,respectively.Compared with the traditional algorithms TR,Landweber and neural network 6L-FCDN,the reconstructed image of this method is closer to the real distribution.
作者
蔡婷婷
颜华
CAI Tingting;YAN Hua(School of Information Science and Engineering,Shenyang University of Technology,Shenyang 110870,China)
出处
《电子设计工程》
2024年第20期84-88,共5页
Electronic Design Engineering
关键词
电阻抗层析成像
深度学习
图像重建
重建算法
electricalimpedancetomography
deeplearning
imagereconstruction
reconstructionalgorithm