ERT Image Reconstruction Algorithm And Image Quality Assessment

In recent years,electrical resistance tomography(Electrical Resistance Tomography,ERT)is a relatively fast-developing technology.It has the characteristics of low cost,no radioactivity,visualization,and non-invasiveness.It is mainly based on the quasi-steady field of electromagnetic fields as a theoretical basis in many fields.Has a wide range of applications.Therefore,ERT has become a research hotspot in the field of visual inspection.In practical applications,since the real situation inside the pipeline cannot be obtained,the actual situation inside the pipeline can only be obtained by image reconstruction.In order to reduce the trouble in practical applications,the ERT image reconstruction algorithm with high imaging accuracy is crucial.Therefore,this paper proposes a long-and short-term memory neural network(Long Short-Term Memory,LSTM)+fully connected neural network ERT image reconstruction algorithm.First,the LSTM algorithm is used to classify the measured voltage of the ERT image to eliminate the interference caused by the different types of flow patterns.Then,the fully connected neural network is used to implement For the reconstruction of ERT images,208 measured voltage values are used as input,the pipeline is divided into 1024 imaging pixels as the output of the network,and the imaging results are combined with linear back projection(Local Binary Patterns,LBP)and radial basis(Radial Basis Function,RBF)neural networks And other classic algorithm imaging results.Experimental results show that the ERT image reconstructed by the algorithm proposed in this paper has higher accuracy.It is the bottleneck of the current ERT technology development to evaluate the image quality of ERT reconstructed images without reference.As mentioned above,the internal situation of the pipeline can only be obtained by image reconstruction.However,since the ERT image is reconstructed by the corresponding algorithm,there will be some distortion,so it is necessary to rely on the image evaluation algorithm to evaluate the image' quality.However,there is a certain difference between ERT image and natural light image.Therefore,this paper proposes a reference-free quality evaluation method for ERT reconstructed image based on joint features and sparse representation.In this method,the extracted joint feature vectors are combined into a dictionary according to different flow types.Finally,the K-svd sparse representation model is used to predict the quality score of the ERT reconstructed image.In order to verify the effectiveness of the proposed ERT image evaluation method,this paper uses image reconstruction errors and correlation coefficients as the ERT image quality evaluation scores.The ERT reconstruction of three typical flow patterns of gas-liquid two-phase bubbly flow,circulation and laminar flow Image evaluation.The experimental results show that,compared with other classic non-reference image evaluation methods,the proposed method shows higher accuracy for the quality prediction of ERT reconstructed images.