On the autonomous detection of coronal mass ejections in heliospheric imager data

We report on the development of an Automatic Coronal Mass Ejection (CME) Detection tool (AICMED) for the Solar Mass Ejection Imager (SMEI). CMEs observed with heliospheric imagers are much more difficult to detect than those observed by coronagraphs as they have a lower contrast compared with the background light, have a larger range of intensity variation and are easily confused with other transient activity. CMEs appear in SMEI images as very faint often-fragmented arcs amongst a much brighter and often variable background. AICMED operates along the same lines as Computer Aided CME Tracking (CACTus), using the Hough Transform on elongation-time J-maps to extract straight lines from the data set. We compare AICMED results with manually measured CMEs on almost three years of data from early in SMEI operations. AICMED identified 83 verifiable events. Of these 46 could be matched with manually identified events, the majority of the non-detections can be explained. The remaining 37 AICMED events were newly discovered CMEs. The proportion of false identification was high, at 71% of the autonomously detected events. We find that AICMED is very effective as a region of interest highlighter, and is a promising first step in autonomous heliospheric imager CME detection, but the SMEI data are too noisy for the tool to be completely automated.