Combined feed-forward neural network and iterative linear back projection for Electrical Capacitance Volume Tomography

Abstract

In Electrical Capacitance Volume Tomography, the internal permittivity distribution of a region of interest has a nonlinear relationship with the measured capacitance. Most image reconstruction algorithms neglects the nonlinear characteristic and use instead a linearized sensitivity approach to solve the non-linear problem, affecting the accuracy of the reconstructed image. In this study, we used feed-forward neural network to solve the non-linear forward problem to replace the linearized sensitivity matrix. The reconstruction process uses an iterative linear back projection technique. Comparison results showed considerable improvement on the image reconstruction of the proposed technique.