Research On EIT Image Reconstruction Algorithm Based On Deep Learning

Electrical Impedance Tomography(EIT)has been investigated extensively during the past decades as a visualization and measurement technique.Due to the advantages of being non-radiant and non-intrusive,rapid response,simple structure and low cost,EIT has wide foreground for application in the fields of medical imaging,industrial imaging and geological exploration.The problem of EIT image reconstruction is a nonlinear and ill-posed problem.The images reconstructed by the traditional methods always have artifacts because of the noise in the measure system.An intelligent learning method of the shallow network was proposed in this thesis.In order to overcome the shortcomings of the shallow neural networks,Deep Learning method was proposed further more.The main algorithms were as follows:1.EIT image reconstruction algorithm based on Hopfield network was proposed.The establishment of energy function to the network,the solution of the energy function and the process of image reconstruction were introduced in detail.2.EIT image reconstruction algorithm based on steepest descent BP network was proposed.The process of the forward propagation and error back propagation to the network and the process of using the steepest descent method to establish the EIT nonlinear relationship between the measured boundary voltages and conductivity were introduced in detail.The effectiveness of the algorithm was verified by the simulation and system experiments.3.EIT image reconstruction algorithm based on Deep-learning model was proposed.The process of establishment,initialization and training of EIT deep learning network model were introduced in detail.Then the effectiveness of the algorithm was verified by the simulation and system experiments.It was proved that this method can effectively overcome the problems of easy fitting,strong dependence on the parameters and the limitation of representation to the complex functions in shallow network learning.4.A sparse imaging method based on deep dictionary was proposed.The sparse coding and the process of dictionary learning with Deep Learning were introduced in detail.A good edge-preserving characteristic of the algorithm was verified through the simulation and system experiments.Compared with traditional methods,image reconstruction algorithms proposed in this thesis can effectively suppress the artifacts in the reconstructed image and have an edge-preserving characteristic.Thus,these algorithm can further improve the quality of the reconstructed images of EIT.