A least squares support vector regression coupled linear reconstruction algorithm for ECT

Abstract

Linearization error of the simplified linear electrical capacitance tomography (ECT) model is one of the leading causes of ECT reconstruction errors. In this paper, the least squares support vector regression (LSSVR) is used to fit the correlation between the capacitance vector and the linearization error. And it is trained by the training samples of typical phase distributions. When removing the linearization error from equations derived by the linear model, the reconstruction problem becomes an exact linear inverse problem because the nonlinearity of ECT is completely included in the linearization error. Then a reconstruction algorithm combining the LSSVR and the Landweber iteration is proposed. Numerical results show that the proposed algorithm achieves significantly better reconstruction accuracies than the linear back projection and the Landweber algorithm for both the noise-free and noisy cases. Compared with the Landweber algorithm, The image errors of the reconstructions are reduced by about 23%–68%, and the correlation coefficient increased by about 0.04–0.14. And the calculation time of the proposed algorithm for all the tested cases is about 0.4–0.6s, which makes it have the potential for real-time imaging. Static experimental results show that the reconstructions of the proposed algorithm have more accurate phase boundary shapes and fewer artifacts.