Autonomous Detection of Eruptions, Plumes, and Other Transient Events in the Outer Solar System

The outer solar system abounds with visually stunning examples of dynamic processes such as eruptive events that jettison materials from satellites and small bodies into space. The most notable examples of such events are the prominent volcanic plumes of Io, the wispy water jets of Enceladus, and the outgassing of comet nuclei. We are investigating techniques that will allow a spacecraft to autonomously detect those events in visible images. This technique will allow future outer planet missions to conduct sustained event monitoring and automate prioritization of data for downlink. Our technique detects plumes by searching for concentrations of large local gradients in images. Applying a Scale Invariant Feature Transform (SIFT) to either raw or calibrated images identifies interest points for further investigation based on the magnitude and orientation of local gradients in pixel values. The interest points are classified as possible transient geophysical events when they share characteristics with similar features in user-classified images. A nearest neighbor classification scheme assesses the similarity of all interest points within a threshold Euclidean distance and classifies each according to the majority classification of other interest points. Thus, features marked by multiple interest points are more likely to be classified positively as events; isolated large plumes or multiple small jets are easily distinguished from a textured background surface due to the higher magnitude gradient of the plume or jet when compared with the small, randomly oriented gradients of the textured surface. We have applied this method to images of Io, Enceladus, and comet Hartley 2 from the Voyager, Galileo, New Horizons, Cassini, and Deep Impact EPOXI missions, where appropriate, and have successfully detected up to 95% of manually identifiable events that our method was able to distinguish from the background surface and surface features of a body. Dozens of distinct features are identifiable under a variety of viewing conditions and hundreds of detections are made in each of the aforementioned datasets. In this presentation, we explore the controlling factors in detecting transient events and discuss causes of success or failure due to distinct data characteristics. These include the level of calibration of images, the ability to differentiate an event from artifacts, and the variety of event appearances in user-classified images. Other important factors include the physical characteristics of the events themselves: albedo, size as a function of image resolution, and proximity to other events (as in the case of multiple small jets which feed into the overall plume at the south pole of Enceladus). A notable strength of this method is the ability to detect events that do not extend beyond the limb of a planetary body or are adjacent to the terminator or other strong edges in the image. The former scenario strongly influences the success rate of detecting eruptive events in nadir views.