3D ECT sensor interior recognition using neural networks and normalized capacitance data
Robert Banasiak1, Radosław Wajman1, Adam Dyjas1, Dominik Sankowski1 1Institute of Applied Computer Science, Lodz University of Technology. corresponding author: email@example.com
This work discusses a novel idea of recognition of 3-dimensional objects using normalized ECT spatial capacitance data and Kohonen’s neural network (as known as Self-Organizing Maps). To achieve this goal the authors developed and implemented an algorithm that is able to detect and recognize shapes of objects with dielectric properties located in a multi-plane capacitance sensor interior. The object type and shape evaluation is performed by pre-trained self-organized map on the basis of randomly selected 3D ECT experimental normalized data. The preliminary results demonstrated that proposed algorithm is able to recognize tested shape phantoms with more than 80% certainty. The current implementation is able to work offline however the next step will be a development of this method towards a real-time sensor interior identification task and its features extraction.
Keywords 3D ECT imaging, spatial permittivity, shapes recognition, self-organized maps
Copyright © International Society for Industrial Process Tomography, 2016. All rights reserved.