Comparative analysis of manifold learning algorithms for tomographic sensor processing

Electrical Impedance Tomography sensors are non-intrusive sensors used to estimate conductivity fields. These estimates are based on a set of measured induced voltages generated by some currents injected to the process. Processing the measurements of an EIT sensor to estimate the underlying changes in the conductivity field requires the use of high dimensional models and solve a nonlinear optimization problem. In some applications the changes in the conductivity are due to changes in just a couple of factors, and therefore the sensor output can be described by a set of variables lying in a lower dimensional space. Manifold Learning Algorithms can learn these low dimensional spaces embedded in the measurement space. In this work, three popular MLA are analyzed as tools to discover manifolds in the EIT measuring space. Several simulations and experimental results show that Laplacian eigenmap algorithm is a suitable MLA for this type of applications.