Research On Open Electrical Impedance Tomography Algorithm Based On Deep Learnin

Electrical impedance tomography(EIT)is an attractive,non-radiation imaging method that aims to visualize the electrical conductivity distribution inside the detected object from the feedback induced voltage.Open electrical impedance tomography(OEIT)method concentrates the excitation electrode and the measurement electrode on the boundary of the imaging region,which overcomes the limitations of traditional imaging methods.Due to the limitation of the electrode position,the number of input signals is reduced,so that the ill-conditionedness of the OEIT inverse problem is magnified.Therefore,the resolution of reconstructed image cannot meet the requirements,which also restricts the measurement depth of image reconstruction algorithm and increases the sensitivity of the algorithm.In order to overcome the problems of OEIT inverse problem such as low imaging accuracy,sensitivity to noise,and large artifacts area of reconstructed images,this thesis first proposes to use a shallow neural network to solve the OEIT inverse problem.The non-linear ability of BP neural network is used to reconstruct the complex relationship between boundary voltage and corresponding conductivity distribution.The simulation data and actual experimental data show that this OEIT algorithm can solve the OEIT inverse problem in a more reliable way.Secondly,open electrical impedance tomography algorithms based on the deep network model were proposed.The OEIT algorithm based on the convolutional neural network hybrid model uses the convolutional neural network to complete the feature extraction of the voltage sequence.The spatial pyramid pooling layer performs secondary extraction for the output features of the last convolutional layer,which is used to deepen the network’s understanding of voltage sequence and avoid over-fitting phenomenon of network.The OEIT algorithm based on the multi-scale residual network uses stacked multi-scale residual modules to complete the feature extraction of the voltage sequence.The multi-scale residual module composed of convolutions of different sizes can mine the multi-scale information in the data.In addition,the model also combines the characteristics of the residual network and combines the shallow features with the deep features to help the deep network converge.The effectiveness and anti-interference ability of the OEIT algorithm based on the deep network are verified through simulation data sets and actual experimental data.It also proves that the multi-scale residual network algorithm can effectively suppress artifacts and accurately reconstruct the location and conductivity distribution of the detected target.Finally,the design of the OEIT system is realized.An OEIT data acquisition system is designed to measure the actual voltage signal with NI6281 data acquisition card as the core.The current excitation and voltage acquisition of the imaging region were carried out by using data acquisition card.The AC620 FPGA development board is used to control the path switching of the multi-switch module,to measure the voltage between adjacent electrodes.In the end,the collected voltage sequence is input into the trained model to obtain the reconstructed image.The experimental data show that the OEIT algorithm based on deep learning has good reconstruction accuracy and anti noise ability,which provides a reference direction for the optimization of OEIT algorithm.