基于卷积神经网络及有限元仿真的电容层析成像图像重建

为求解电容层析成像（electrical capacitance tomography，ECT）图像重建非线性病态逆问题，提出基于一维卷积神经网络（one-dimensional convolutional neural network，1D CNN）的电容层析成像图像重建算法。通过1D CNN建立了ECT系统独立测量值与重建图像灰度值之间的非线性映射关系。采用有限元仿真软件获得6种典型流型的随机分布样本，完成了1D CNN网络的训练及测试，进行了静态实验，并与常用的线性反投影、Landweber迭代算法进行比较。仿真及静态实验结果均表明：基于1D CNN算法的ECT重建图像质量得到明显提高，具有较好的泛化能力及实时性。

Image Reconstruction of Electrical Capacitance Tomography Based on Convolutional Neural Network and Finite Element Simulation

In order to resolve the nonlinear and ill-posed inverse problem of the image reconstruction of electrical capacitance tomography (ECT), an image reconstruction algorithm based on one-dimensional convolutional neural network (1D CNN) is presented. The nonlinear mapping relationship between the independent measurement value of ECT system and the gray value of reconstructed image is established by 1D CNN. Six typical flow regimes with random distribution are obtained by the finite element simulation software and a 1D CNN is successfully trained. Simulation and static experiments are carried out and the reconstructed images using linear back projection, Landweber iterative algorithm and 1D CNN are compared. Experimental results show that the 1D CNN algorithm has good generalization ability and the quality of reconstructed images by 1D CNN is obviously improved compared with the other two algorithms.