Detection and Thickness Estimation of Oil under Saline Ice Using Machine Learning

Abstract

Electrical Capacitance Tomography (ECT) for coplanar electrodes has been used in very few applications with non-destructive single side access to detect changes in the sensing domain. In such applications, the image reconstruction algorithms are usually applied to image a horizontal cross section of the sensing domain and are not always accurate. In this paper, we propose performing image reconstruction for a vertical cross section of the sensing domain. Inspired by ECT solutions applied to pipes, we use machine-learning to estimate the presence (classification) and thickness (regression) of oil layers between saline ice layer and seawater. Results on simulated data demonstrated good performance in classification with an f1 score exceeding 90%, as well as in regression with a mean percentage error of -8.148%, a mean squared error of 14.952, and a mean absolute error (MAE) of 2.933 mm for a sensing domain 45 mm deep.