Supervised Image Processing Learning for Wall MARFE Detection Prior to Disruption in JET With a Carbon Wall

In the last years, several diagnostic systems have been installed on Joint European Torus (JET) providing new information that may be potentially useful for disruption prediction. The fast visible camera can deliver information about the occurrence of multifaceted asymmetric radiation from the edge (MARFE) instabilities that precede disruptions in density limit discharges. Two image processing methods—the sparse image representation using overcomplete dictionaries and the Histogram of oriented gradients (HOGs)—have been used for developing MARFE classifiers with supervised learning. The methods have been tested with JET experimental data and a good prediction rate has been obtained. The HOG method is able to provide predictions useful for online disruption prediction.