Research On Image Reconstruction Algorithms Based On Electrical Impedance Imaging Technology

The special environment at sea causes certain damage to the health of personnel working at sea for a long time,and at present,traditional physiological index monitoring methods are mainly used to monitor physical health,while the commonly used CT(Computed Tomography)imaging equipment is difficult to be used on ships due to its large size,complex equipment and high price.Therefore,it is especially important to study a medical detection technology that can be easily applied at sea.Electrical Impedance Tomography(EIT)technology reconstructs the resistivity distribution of the medium in the sensitive field by imaging algorithms,which not only has the advantages of non-radiation,non-invasive and fast response,but also has the advantages of portability,low price and wide range of use,and has received the attention of many scientists.The pathological nature of the EIT inverse problem as well as its severe discomfort make its solution accuracy unsatisfactory and the reconstructed image resolution low,so a new imaging method needs to be found to improve the image reconstruction quality.With the aim of improving the image reconstruction quality,this paper conducts an in-depth study on the EIT image reconstruction algorithm as follows:1.Finite element dissection and solution of the EIT positive problem are performed to establish a field simulation model.Different targets are placed in the field and their shapes,positions and conductivities are changed to generate the data set required for neural network training.2.A PSO-TR algorithm combining Particle Swarm Optimization(PSO)and Tikhonov Regularization(TR)is proposed to effectively reduce the instability of the solution by adding a regular term to the objective function to constrain the inverse problem.The accuracy and robustness of the reconstructed images are improved.3.The Full Connect Neural Network(FCNN)model is designed,and a Fully-Connected Residual Neural Network(FCRN)algorithm is proposed based on its improvement.The algorithm is to add cross-layer connections and shortcut branches to the FCNN to build the residual structure,deepening the network so that the network can better learn the mapping relationship between boundary voltage and impedance in the EIT problem,achieving deeper network training and higher accuracy result prediction.Meanwhile,the total loss function combining Focal Loss and L1 Loss is proposed for the category imbalance problem of EIT dataset,which makes the network more inclined to the learning of target samples,thus enabling the network to better localize the targets and improve the generalization ability and robustness of the model.4.The PSO-TR algorithm and FCNN algorithm are simulated and compared with the classical image reconstruction algorithm.The experimental results show that the PSO-TR algorithm can roughly identify the target object,and FCRN can locate the target area more accurately compared with other algorithms,and the imaging is clear without artifacts,which improves the imaging accuracy.