Enhanced Robust Vortex Detection

We propose to leverage methods of machine learning to enhance robustness of feature detection algorithm. First, we use semi-supervised learning to develop strategies for guiding the selective refinement process based on training with the domain expert. Second, we propose to combine several local feature detection algorithm into a single, more robust compound classifier using AdaBoost that produces validated feature detection. The compound classifier would combine the best of all local classifiers as they respond to the underlying physical signal. The specific application of interest is vortex detection in turbulent flows. We applied our algorithms to fluid datasets to illustrate the efficacy of our approach.