Monitoring of flood embankments through EIT machine ensemble learning

Abstract

The research aimed to develop an optimal way of using known machine learning techniques in electrical impedance tomography (EIT) of flood embankments. The innovative approach is based on the smart use of many machine learning techniques to allow the optimal selection of one of these techniques for each pixel of the tomographic image. An additional advantage of the presented concept is that selecting the optimal method for each pixel depends on the measurement set of a given case. This fact makes the method flexible and enables the automation of dyke monitoring using cyber-physical systems. Several machine learning methods were used during the research, including Elastic Net, Support Vector Machine, and Artificial Neural Networks. The comparison of the new concept with popular methods showed that thanks to pixel-oriented ensemble learning, the reconstructions obtained with the new approach are much better than those obtained with typical machine learning methods.