Research On ERT Image Reconstruction Algorithm Based On Attention Mechanism And Deep Neural Network

Electrical Resistance Tomography(ERT)is a non-invasive visual imaging technology,which is widely used in chemical,thermal,petroleum,and other industrial production fields.ERT technology applies current excitation to the electrode sensor outside the measured field and reconstructs the internal conductivity distribution images of the measured field by using the image reconstruction algorithms and the measured boundary voltage values.However,due to the inherent’soft field’ characteristics of ERT,the image reconstruction process is highly nonlinear,ill-posed,and morbid.The traditional ERT image reconstruction algorithms have many problems such as more artifacts and blurred edges.In this paper,the nonlinear mapping ability of deep learning algorithms is used to solve the positive and inverse problems of ERT,and the image reconstruction quality is further improved based on the classical Landweber iteration algorithm.The main works are as follows:(1)To better simulate the actual gas-liquid two-phase flow pattern and solve the problem that the ERT image reconstruction algorithm based on deep l earning only aims at the typical flow pattern,a simulation sample generation method for irregular flow pattern is proposed,which effectively expands the sample types and quantities of the simulation data set.(2)In the iterative process of Landweber algorithm,the error is increased to solve the forward problem through the linearization method.Therefore,the forward problem-solving method of ERT based on long short-term memory network(LSTM)is proposed.Combined with the Landweber algorithm,the LSTM-Landweber iterative algorithm is formed,and the optimal threshold is used to optimize the imaging results.The simulation results show that the imaging quality of LSTM-Landweber is better than Landweber and ELM-Landweber under the same iteration number.(3)Aiming at the problem of the limited accuracy of classical ERT image reconstruction algorithm,an ERT image reconstruction algorithm based on deep residual attention neural network(DRAN)is proposed,which realizes the end-to-end application from boundary voltage measurement sequence to conductivity distribution image.The simulation results show that DRAN can effectively restore the detailed information of conductivity distribution image,and has good generalization and anti-noise performance.In addition,DRAN is used to post-process the conductivity distribution image obtained by Landweber iteration for 50 times.The static experimental results show that this method can effectively improve th e accuracy of image reconstruction.