Multi-scale Convolutional Neural Network Image Reconstruction And Hu-ELM Flow Pattern Recognition Of ECT System

Electrical Capacitance Tomography(ECT)is a technology developed on the basis of Process Tomography(PT).The parameters of multiphase flow system can be detected online and in real time.It is of great significance to accelerate the pace of industrial development.At present,ECT system has the advantages of simple structure and easy operation,which is widely used.But ECT technology still has some problems to be solved.In this paper,image reconstruction algorithm and flow pattern recognition of ECT system are studied.The main work and achievements are as follows:1.The image reconstruction method based on multi-scale dual channel convolutional neural network is studied.In order to solve the problems of edge missing and obvious artifacts in ECT reconstruction algorithm,an image reconstruction method based on multi-scale Convolution Neural Network(CNN)is proposed.The image reconstructed by Landweber algorithm is input into CNN.A dual channel is designed to divide the frequency of the input image.Four scales are selected for feature extraction,and the feature images are fused by using jump connection between the scales.Simulation experiments are carried out for four typical flow patterns.Compared with other algorithms,the algorithm in this paper reduces the error,improves the correlation coefficient,effectively solves the problem of missing edges in ECT image reconstruction,and reduces the degree of image edge artifacts.2.The Hu-ELM flow pattern recognition algorithm based on reconstructed image is studied.The current ECT flow pattern recognition method based on image reconstruction has the problem of slow speed and poor recognition effect.This paper proposes a Hu-ELM flow pattern recognition algorithm based on reconstructed image.First,the image is segmented by the super-pixel method.After the pre-processed image is obtained,the Hu moment is used to extract the features of the high-permittivity connected regions.The three feature parameters Hu moment,area ratio,and number of high-permittivity connected regions are used.Combining the ELM method to identify the four typical flow patterns,the recognition rate reached 98.13%.Compared with other algorithms,the algorithm in this paper has a high flow pattern recognition rate and a shorter time,which is an effective recognition method.