Image Reconstruction Of Electrical Capacitance Tomography Based On AlexNet Convolutional Neural Network

Electrical capacitance tomography(ECT-Electrical Capacitance Tomography)is a process imaging technology based on the principle of capacitance sensitivity and using corresponding algorithms to calculate the distribution image of the flowing medium.Electrical capacitance tomography has the advantages of non-invasiveness,good application range,fast response speed,and good real-time performance.In recent years,electrical capacitance tomography technology has been widely used at home and abroad,and the current research results also show that this technology has very broad application prospects in various industrial fields such as petroleum and chemical industry in the future.However,due to the existence of soft field characteristics,the traditional electrical capacitance tomography image reconstruction algorithm is not very effective,so it is of great value to carry out relevant research.This article mainly discusses how to use convolutional neural network to optimize the accuracy and speed of ECT technology image reconstruction algorithm.The specific arrangements are as follows:First,combined with the actual situation,the background and significance of ECT technology,the recent domestic and foreign research status and its application scope are briefly introduced,and the composition and principle of the ECT system are described in detail.Then it gives an overview of Convolutional Neural Network(CNN)and explains its structure and principles.Secondly,on the basis of consulting a large number of domestic and foreign references and materials,it is found that traditional ECT technology image reconstruction algorithms,such as linear back projection(Linear Back Projection,LBP)algorithm,Fourier transform reconstruction algorithm,support vector machine,etc.,are all using the sensitivity distribution that characterizes the relationship between the measured capacitance value and the dielectric constant distribution of the measured area,the pixel gray value of the corresponding part is calculated.In principle,the sensitivity field will change with the distribution of the medium to be measured.Therefore,the sensitivity matrix should be constantly updated according to the distribution of the medium to be measured for accurate use.However,in practical applications,the dielectric constant distribution or material distribution is unknown.Therefore,the sensitivity matrix can only be continuously updated according to the reconstructed image to improve the solution of the inverse problem.According to the characteristics of the convolutional neural network,a large amount of sensitivity matrix data can be established first,and then the network model can be trained using the samples.The Landweber algorithm is used for simple processing,and then the processed value is used as the input value of the network model.Obtain the adjusted sensitivity matrix,making the sensitivity matrix closer to the actual value.Third,in order to improve the recognition efficiency,traditional convolutional neural networks generally improve their generalization performance by collecting a large number of original samples.However,the problem of over-fitting is prone to occur from this,so this article uses the AlexNet convolutional neural network structure to suppress the over-fitting problem,and simply optimizes it to better match the experimental data.Finally,image reconstruction is based on the optimized AlexNet convolutional neural network.Adjust the network structure,use a large number of different types of fluid media samples to train the network model.After simple processing by the Landweber algorithm,the network model proposed in this paper is used for image reconstruction.Finally,through simulation experiments,the feasibility of the method proposed in this paper is analyzed.