Towards a practical implementation of EIT-based sensors using artificial neural networks

Electrical Impedance Tomography (EIT) is an imaging technique used to display conductivity changes inside a region of an electrically conductive body. EIT-based sensors are low-cost, can be applied over 3D surfaces, and show multi-touch sensing capabilities, but their drawback is their low spatial resolution. In this paper we propose to solve the EIT reconstruction problem with Artificial Neural Networks (ANNs). We present our study using a Finite Element (FE) model of our EIT-based sensor. Results using different target dimensions show that the ANN can learn to reconstruct the target placement over the skin. This is performed with an improved detected size of the objet and position accuracy for small objects of up to 37% and 14.7% respectively and 4.3% and 15.2% for larger objects. By improving the performance of the EIT-based sensitive skin through ANNs, we want to contribute to the literature by surmounting their major limitations that so far have been hindering their integration in several applications. To the best of our knowledge, this work is the first one introducing ANNs for target placement recognition in EIT-based sensitive skins.