Real-time analysis and classification of observational data collected within synoptic sky surveys is a huge challenge due to constant growth of data volumes. Machine learning techniques are often applied in order to perform this task automatically. The current bottleneck of transients detection in most surveys is the process of filtering numerous artifacts from candidate detection. We present a new method for automated artifact filtering based on hierarchical unsupervised classifier employing Self-Organizing Maps (SOMs). The system accepts 97 % of real transients and removes 97.5 % of artifacts when tested on the OGLE-IV Transient Detection System. The improvement of the artifacts filtering allowed for single-frame based rapid detections of transients within OGLE-IV, which now alerts on transient discoveries in less than 15 minutes from the image acquisition.
↧