Research On EMT Image Reconstruction Algorithm Based On Auto Encoder Neural Network

Electromagnetic tomography The main principle of this technology is to extract the state information of the object to be measured through the electromagnetic induction principle for image reconstruction,to obtain the distribution and state information of the object to be measured.It has the advantages of simple hardware structure,low cost,no contact,no intrusion,wide application range and so on.However,there are still some technical problems in the application of EMT,such as the poor sensitivity field performance and the quality of reconstructed image can not meet the requirements.Therefore,it is of great significance to study the EMT image reconstruction algorithm.In this paper,based on a large number of references at home and abroad,combined with the EMT hardware system,the EMT induction coil is optimized to improve the sensitivity field performance.Based on the basic theory and mathematical model of EMT image reconstruction,combined with the idea of auto encoder neural network,a new image reconstruction algorithm is proposed.The main work of this paper is as follows:1.Firstly,the research background and significance of EMT technology and the development status of EMT system at home and abroad are analyzed,and several traditional image reconstruction algorithms widely used at present are summarized and classified,and their advantages and disadvantages are analyzed and compared.2.Based on the study of the working principle of EMT system,the forward and inverse problems of EMT system are analyzed.The mathematical model of electromagnetic detection is derived by using the Maxwell equation of electromagnetics,and then the detection voltage value is solved.At last,the geometry model of 8-coil sensor is built by COMSOL.3.Aiming at the non-linear problem in the sensitivity field of EMT system,this paper proposes a method to improve the non-linear problem of the sensitivity field based on the quadratic approximation boundary optimization(BOBYQA)algorithm,which is an optimization method for the complex objective function problem without calculating the derivative of the objective function.Firstly,according to the characteristics of the sensitivity field of EMT system,the optimization objective function of the sensitivity field is established,and then the structural parameters of the sensor are optimized by using the BOBYQA algorithm.Finally,the experimental verification is carried out.The results show that the sensitivity field indexes of the optimized EMT system are improved,and the nonlinear problem of the sensitivity field is obviously improved.4.Aiming at the problem of low accuracy of nonlinear mapping in encoder network,a method based on LM Neural Network is proposed to enhance the nonlinear mapping ability of encoder.LM algorithm is a kind of algorithm which uses least square method to fit nonlinear mapping.Firstly,the structure of encoder is introduced and the principle of encoder dimensionality reduction is analyzed.Secondly,the parameters of encoder nonlinear mapping process are adjusted by the ability of LM Neural Network fitting nonlinear mapping.Finally,the experimental verification is carried out,and the results show that the improved encoder structure has strong nonlinear fitting ability.5.Aiming at the problem of low precision of traditional image reconstruction algorithm,an EMT image reconstruction algorithm based on auto encoder neural network is proposed.Auto encoder neural network is a kind of deep learning model,which consists of encoder and decoder.Using the network structure of the model,the encoder coding process corresponds to the electromagnetic detection process in the EMT system,and the decoder decoding process corresponds to the image reconstruction process.Firstly,the image information of different field and its detection voltage matrix are obtained by using COMSOL software,and the field image and detection voltage matrix are used as training data set,then the auto encoder neural network is trained by using the training data set.Finally,the detection voltage output by EMT system is input into decoder network,and the reconstructed image of EMT is obtained.6.COMSOL is used to build the field flow pattern,and the EMT image reconstruction algorithm based on the auto encoder neural network is used to reconstruct the image respectively with the traditional LBP and Landweber algorithm.The image evaluation parameters are introduced for comparison.The experimental data shows that the reconstruction results of this algorithm are smaller in image error,larger in correlation coefficient and better in reconstruction image quality compared with the traditional algorithm,which proves that The effectiveness of the combination of auto encoder neural network and EMT system in image reconstruction.