Research On Magnetic Detection Electrical Impedance Imaging Algorithm Based On Deep Learning

In recent years,Electrical Impedance Tomography(EIT)has been developed rapidly as a non-invasive functional detection technology.As an important branch of EIT,magnetic detection electrical impedance tomography(MDEIT)improves the shortcomings of EIT in detecting the surface area of the imagingbody,and the internal conductivity of the imaged body is reconstructed by the magnetic induction intensity distribution generated around the imaging body with the current injection.Just like conventional EIT,MDEIT is also non-invasive,easy to measure,and inexpensive.In addition,MDEIT avoids the effect of contact impedance on imaging accuracy through non-contact measurement.However,there are nonlinear,ill-conditioned and ill-posed problems in the MDEIT image reconstruction process.The traditional reconstruction algorithms solve the nonlinear problem by approximate linearization,but it will lead to the problem of low precision and poor noise resistance.In view of the shortcomings of the above reconstruction algorithms,in order to further improve the reconstruction accuracy,this thesis first proposes the back propagation(BP)neural network algorithm as a shallow neural network model for MDEIT image reconstruction.The process of establishing the weight matrix by means of forward propagation and updating the weight matrix by means of backpropagation is introduced in detail.The BP neural network model is established for the nonlinear relationship between the magnetic induction intensity distribution and the conductivity distribution.The comparison between simulation experiments and traditional MDEIT reconstruction algorithms verifies the effectiveness of BP neural network algorithm.Furthermore,based on the shallow neural network,the stacked auto-encoder(SAE)deep learning algorithm is applied to MDEIT image reconstruction.The structure of SAE network is introduced and the training process of deep learning model is analyzed in detail.The deep learning model is established.The simulation experiment proves that SAE deep learning model is effective for solving the nonlinear problem of MDEIT inverse problem.Secondly,to verify the validity of the algorithm,this thesis proposes several evaluation indicators for objectively evaluating the quality of image reconstruction.The center position of the anomaly evaluates the ability of the algorithm to locate the anomaly.The cross-sectional view evaluates the ability of the algorithm to reconstruct the shape of the anomaly.Finally,to evaluate the problems of noises in the measurement system,a certain amount of Gaussian white noises is added to the simulation experiment.The anti-noise performance of the algorithm is evaluated by the reconstruction result.At last,in order to enrich the magnetic field detection information,a double excitation model is proposed,and the effectiveness of the model is proved by simulation experiments.The algorithm proposed in this thesis is applied to the phantom experiment.The reconstruction results verify the stability of the algorithm.The MDEIT image reconstruction algorithm based on stacked auto-encoder deep learning significantly improves the reconstruction accuracy and anti-noise performance of conductivity images,and promotes the development of magnetic detection electrical impedance tomography.