Three-dimensional Image Reconstruction Based On The Hollow Convolutional Neural Network ECT

The Electrical Capacitance Topography(ECT)technology,an engineering imaging based on capacitance-sensitive mechanism,has the advantages of non-radiation,high temperature resistance,non-invasive,fast speed response,low price,simple structure,and wide range of application.This paper conducts an all-round in-depth study of key technical issues such as sensor design,distribution of sensitive fields,and image reconstruction algorithms.Firstly,this paper deeply studies the working principle,system composi tion and technical characteristics of ECT system,analyzes the basic principle of ECT system from theory,adopts finite element method to model and analyze ECT system,and uses PyCharm and ANSYS18.2 software programming to obtain computer experimental data and calculate the sensitivity distribution respectively,because the sensitive field of ECT system is distributed in three-dimensional space.The results of the two-dimensional image reconstruction of the conventional ECT system cannot reflect the spatial medium distribution information.The structural parameters of the ECT sensor are studied on the basis of the simulation model,and the structural parameters of the ECT sensor are determined experimentally,and the three-ring eight-electrode ECT capacitive sensor is designed to increase the number of capacitance values,and the performance of the ECT sensor is significantly improved.Secondly,several classical image reconstruction algorithms currently exist are studied in depth,and the optimized BP neural network algorithm is applied to the image reconstruction algorithm of ECT system for the problems of soft field characteristics and poor imaging stability of ECT system.Firstly,the improved K-means algorithm is used to cluster a large amount of training data to extract feature data samples,which greatly reduces the scale of training samples,and according to Gauss’ theorem and ECT system boundary equation is pushed to the class sensitivity equation,and the class sensitivity equation is used as the evaluation function of ECT image reconstruction,and the BP neural network weights are adjusted and optimized,and the results show that the algorithm is simple,fast and has high quality of reconstructed images.To address the underqualification problem of image reconstruction algorithm,this paper proposes an improved Hybrid Dilated Convolutional NeuralNetwork image reconstruction algorithm(IHDC),which combines the structural characteristics of AlexNet and Lenet-5 networks,improves the convolutional network structure,introduces the dilation rate in the convolutional kernel(Dilation rate)in the convolutional kernel,expanding the perceptual field to obtain multi-scale information by setting different Dilation rates,and using the adaptive Adam gradient descent algorithm in back propagation to avoid the network from falling into local optimum and improve the training speed.Finally,this paper adopts linear inverse projection algorithm,optimized BP neural network algorithm and improved hybrid cavity convolutional neural network algorithm for 3D image reconstruction for six pipeline flow types respectively,and the experimental results show that the improved hybrid cavity convolutional neural network algorithm significantly improves both the accuracy and sp eed of reconstruction.