March 12, 2017 23:43:23

Solar Images – Sparse Coding

This work introduces the use of an appearance model based on sparse coding. This model is used to classify solar event detections as either the same detected event at a later time or an entirely different solar event of the same type. The output of such classification tasks can be adapted to be a component of the tracking algorithm known as multiple hypothesis tracking. The advantage of the presented model is that it learns the appearance of the known event online and then generates the matching likelihood for each of the choices presented to it. This online learning is crucial to eliminate reliance on previous tracking information, which does not exist for many solar event types. We show that this online method preforms as well, or better in the task of differentiation than the offline learning of appearance on solar events.