ERT Image Reconstruction Algorithm And Image Quality Assessment Based On Deep Learning

As a visual inspection technology,electrical Resistance Tomography(ERT)technology can display the actual situation inside the closed pipeline or process container equipment in the form of images,so it has been developed rapidly in the field of visual detection.However,the imaging accuracy of existing image reconstruction algorithms is not very high,and the imaging accuracy of ERT reconstructed images should be further improved.The ERT reconstructed image will have some distortion under the influence of various factors,such as soft field characteristics,algorithm accuracy,noise,etc.,and the judgments of subsequent process control based on the distorted image will have certain errors.Therefore,it is more important to assess the quality of ERT reconstructed images for obtain reliable image information,but in actual situations,reference images are usually not be obtained.Therefore,our study is dedicated to exploring a new ERT image reconstruction algorithm and a no-reference ERT image quality assessment method that does not rely on reference images.The main work contents are as follows:(1)An ERT image reconstruction algorithm combining long short-term memory network and convolutional neural network is proposed.Firstly,input the ERT simulation data into the long short-term memory network for flow patten recognition and classification;then,implemented the image reconstruction through the corresponding convolutional neural network model;finally,compare the imaging results of the algorithm with the results of the classic image reconstruction algorithm,the ERT image reconstruction algorithm proposed in this study has higher accuracy.(2)A no-reference ERT image quality assessment method based on improved VGG network is proposed.Simulation software has been used as established the ERT reconstruction image library.The relative error and correlation coefficient of the image are selected as the quality assessment score.Firstly,divide each ERT image into blocks to increase the sample size,perform preprocessing and extract the information entropy of each block image;then,normalize the information entropy as a weight into the loss function,and train an optimal model through the improved VGG network;finally,this model is used to estimate the quality fraction of ERT reconstructed image.By comparing with common no-reference image quality assessment methods,the method proposed in this study has higher accuracy.(3)In order to verify the effectiveness of the algorithm proposed in this research in the actual measurement of ERT reconstruction image quality assessment,the algorithm model obtained from the study was applied to the visual inspection platform of mine filling pipeline based on ERT for actual measurement verification.The experiment shows simulation images can be used as training samples for image quality assessment.