The Three Dimensional Image Reconstruction Of Electrical Capacitance Tomography Based On AlexNet Convolution Neural Network

Electrical capacitance chromatography(ECT)is a kind of detection technology with simple structure,no radiation and low cost in process tomography technology,which is widely used in two-phase flow and multiphase flow.The main principle of this technique is to change the measurement capacitance value caused by the change of the dielectric constant of the medium in the area under test,and to transform the dielectric constant distribution of the medium into the medium distribution by the image reconstruction algorithm.Electrical capacitive tomography technology has broad industrial prospects in oil and gas transportation,and its development is of great significance.This paper introduced the composition and working mechanism of capacitive tomography system,analyzed the relationship between electrical capacitive value,sensitive field and grayscale value in the idea of finite element,and provided a theoretical basis for image reconstruction of capacitive tomography system.Because the sensitive field of electrical capacitive tomography system is distributed in three-dimensional space,the traditional ECT technology takes the form of two-dimensional tomography as the result of reconstruction,and can not reflect the spatial information of the media distribution.This paper used the direct three-dimensional ECT technology to study the influence of the main structural parameters on the basis of the simulation model,the value of the structural parameters has been established by experiment.A different three electrode excitation mode has been designed on the sensor model of three-layers eight-electrodes,and the experiment showed that the single excitation mode had a larger range of changes in capacitance value than the multi-excitation mode,but the three-electrodes excitation mode was more accurate in image reconstruction than the other two modes,and the image reconstruction results were stable when there was noise interference.Because the convolutional neural network is faced with the problem of slow conver speed and low reconstruction accuracy in The ECT system,this paper makes an in-depth study of the AlexNet convolution neural network model,improves the capacitance value data of the ECT system,optimizes the maximum pooled operator in the pooled layer,and uses the weighted average-maximum pooled operator.The data is guaranteed to be panning and robust after feature extraction.In reverse propagation,Adam gradient descent algorithm of adaptive learning rate is used instead of random gradient drop algorithm,which avoids the neural network from falling into local optimal solution and improves convergence speed.The experiment used linear anti-projection algorithm(Linear Back Projection,LBP)and AlexNet convolutional neural network for image reconstruction for different flow types,and the results showed that the improved AlexNet neural network image reconstruction accuracy was higher.