Research On ERT Image Reconstruction Method Based On Compressed Sensing

Electrical resistance tomography（ERT）technology has the characteristics of anti-interference,visualization,fast imaging and non-contact measurement,so it has unique advantages in the field of two-phase flow monitoring.However,the ERT problem is underdetermined,and the high-dimensional conductivity signal must be solved according to the low-dimensional voltage measurement value,so the imaging quality of traditional algorithms is not ideal.Compressed sensing（CS）theory points out that for a signal that satisfies the sparse condition,even if the sampling process does not conform to the Nyquist theorem,the original signal can be reconstructed with high quality from a small number of sampled values.Therefore,this thesis introduces CS theory into ERT in order to solve the underdetermination of ERT and improving the imaging quality.The work of this thesis revolves around three aspects of CS theory.In terms of sparse representation,a new sparse basis based on Haar wavelet is proposed and constructed.This basis is more suitable for ERT conductivity signal,and its sparse effect is better than discrete cosine transform（DCT）basis and discrete fourier transform（DFT）basis.In addition,the transformation matrix is also constructed using the index prior information.The matrix introduces spatial information,which can not only further improve the sparsity,but also make the transformation result present a block-like structure and enhance the anti-noise performance of the sparse signal.In terms of observation matrix,in view of the problem that the ERT sensitivity matrix does not satisfy the restricted isometric property（RIP）,three classical optimization methods in CS theory are introduced to optimize the sensitivity matrix.Combined with the four evaluation indicators of matrix structure,correlation coefficient,reconstruction time and imaging quality,an optimization method for ERT sensitivity matrix is obtained.In terms of reconstruction algorithm,a classification imaging strategy is adopted.Different imaging algorithms are used for different flow patterns.For core flow and bubbly flow,the iterative soft threshold（IST）algorithm and gradient projection sparse reconstruction（GPSR）algorithm based on L₁ norm are used for imaging.For laminar flow,based on the L₀ norm-based sparsity adaptive matching pursuit（SAMP）algorithm,a series of improvements are made to obtain an imaging algorithm suitable for this flow pattern.The experimental results show that,the imaging quality of the CS algorithm is better than that of the traditional algorithm:less artifacts,more sensitive to changes in the size of the flow pattern,more sensitive to the central area of the pipeline,stronger noise immunity and meet real-time requirements.In addition,in order to provide reliable flow pattern prediction results for the classification imaging strategy,this thesis also conducts simple research on the method of flow pattern recognition based on neural network.