Research On Deep Learning Electrical Impedance Imaging Method Based On Electric Field Feature Space Mapping

Electrical Impedance Tomography(EIT)is a non-destructive functional image reconstruction technology,which has broad application prospects in many fields such as medicine and industry.The EIT technology mainly implements image reconstruction of the medium distribution inside the measured object by applying a certain electrical excitation to the measured object,and then using the measured boundary measurement values.Since the end of the 1980 s,it has attracted people's attention and has developed rapidly.The EIT system for human lung detection must have strong sensitivity to changes in lung function status,and it has a very high resolution and quality of reconstructed images.So how to further improve the quality of reconstructed images is the key to applying EIT technology to lung function detection.Because the amount of boundary voltage measurement data is much smaller than the number of unknown conductivity distributions to be reconstructed,EIT technology has inherent "soft-field" characteristics,which leads to ill-posed ill-conditioned problems in image reconstruction.At present,traditional image reconstruction algorithms usually map the inverse EIT problem to a specific space and then solve it,but the linearization process loses a lot of specific information,resulting in the reconstruction image cannot reach high accuracy,and the process is usually very expensive Time.Existing artificial neural network methods are easy to fall into local extreme points during training,have slow convergence speed,and have limited ability to represent complex non-linear functions,resulting in insufficient reconstruction of the image.The main research contents of this article are as follows:1.A new method for deep learning EIT image reconstruction based on electric field feature space mapping is proposed.By establishing a mapping model and finding the non-linear relationship between the two based on large data samples,it avoids problems such as the accuracy reduction caused by linearization of the inverse EIT problem.2.In order to overcome the ill-posed problem in the solution process of the EIT inverse problem as much as possible,and improve the accuracy of the reconstructed image,the boundary voltage measurement value was preprocessed based on the equipotential theory,so that its voltage value was mapped to the pixel block in the sensitive field,forming The sensitive field feature space sequence of the target object distribution information,the big data samples required to construct the network.3.A convolutional neural network method for EIT image reconstruction is proposed,which establishes a one-to-one mapping model of sensitive field feature space sequence to reconstructed conductivity distribution.The network structure is based on Le Net,and the stability of the parameters is enhanced by using a random inactivation layer and a moving average model to improve the generalization ability of the network model.4.In this paper,the corresponding simulation experiments,noise robustness experiments and actual EIT system experiments are carried out to verify the effectiveness of the proposed method.The results show that the method proposed in this paper has good noise robustness and generalization ability.Compared with the results obtained by other traditional imaging algorithms,the reconstructed image has higher accuracy,and can meet the specific requirements for the resolution of the reconstructed image.