Electrical Capacitance Tomography Image Reconstruction Based On Deep Learning

Electrical capacitance tomography(ECT)is the earliest imaging technology in the field of process tomography,which is mainly used for real-time monitoring of dielectric constant distribution in industrial pipelines.ECT technology is favored because of its low cost,non invasion and easy operation.However,the ill condition of ECT Image reconstruction results in poor imaging accuracy,which also restricts the application and development of ECT technology.To solve this problem,this paper studies ect image reconstruction algorithm.1.ECT Image Reconstruction Algorithm Based on improved Le Net-5convolutional neural network is studied.Because of the good performance of deep learning in solving nonlinear problems,the ill conditioned ect image reconstruction can be solved by deep learning algorithm.The finite element simulation package developed by the research group is used to randomly make training network samples,and the Le Net-5 algorithm is improved to make it suitable for ECT Image Reconstruction and complete training.The capacitance data of simulation and static test are used to test the algorithm,and the imaging results are compared with LBP and Landweber iterative algorithm.The results of simulation expe riment and static experiment show that the reconstruction quality of ECT Image Reconstruction Algorithm Based on improved Le Net-5 convolutional neural network is improved obviously,and the shape and size of objects can be distinguished obviously.2.ECT Image Reconstruction Algorithm Based on 1D CNN is studied.Using 1D CNN can make more effective use of a small amount of capacitance value in ECT system.An ECT Image Reconstruction Algorithm Based on one-dimensional convolutional neural network is proposed.Convolution of one-dimensional data directly can not only increase the convolution layer,enhance the ability of data feature extraction,but also reduce the invalid data in the input data.Finite element simulation is used to make training samples randomly to train 1D CNN.By comparing the trained algorithm with LBP and Landweber iterative algorithm,the simulation and static experimental results show that the ECT Image Reconstruction Algorithm Based on one-dimensional convolutional neural network not only meets the requirements of on-line monitoring,but also improves the imaging accuracy,and the image of the object edge is clear.