Onboard Image Processing for Autonomous Spacecraft Detection of Volcanic Plumes

Introduction: Outer planets missions have imaged active volcanic plumes at Io and Enceladus. These phenomena provide key constraints on subsurface processes and models of planetary composition [1]. To date, over 70 distinct images containing plumes have been collected. Since these plumes are transient events and cannot be anticipated in advance, capturing them requires good luck and many images. This demands a prohibitive fraction of the spacecraft’s limited cache and bandwidth, and generally precludes surveys dedicated to monitoring plume activity. Onboard processing could enable long-term plume monitoring campaigns that use very high rates of image capture. The spacecraft could search these images onboard for plumes and only save the fraction containing key events for downlink. Similar technologies have recently been demonstrated for change detection on the WATCH system aboard the Mars Exploration Rovers [2]. Here, automated processing identified images containing dust devils. Candidate image regions were enclosed in “subframe” images for preferential transmission. Similar selective downlink strategies could enable plume surveys, with event detections triggering preferential storage and transmission of coincident observations by Thermal or Visible NearInfrared imagers. In prior research, Bue et al. demonstrate an efficient plume detection algorithm [3]. This research expands on their work with a new algorithm variant (Figure 1) that shows strong performance on a test set of plume images from Enceladus and Io. Approach: We first analyze the image with the Canny edge detection algorithm [4] to identify the planetary horizon. We then fit an ellipse to the horizon using a RANSAC method [5] that samples random subsets of 6 edge points and computes the optimal interpolating ellipse using the closed-form solution of [6]. The RANSAC algorithm fits several thousand ellipses to random point subsets, scoring each according to the number of total edge pixels that lie within a small distance of this contour. We take the best-fitting ellipse as the horizon. Next we identify bright pixel regions that lie outside the horizon ellipse. We look for plumes in pixels at altitudes up to a user-defined range of the limb (in this work we use altitudes up to 10% of the planetary diameter). The set of all pixels in this annular region are taken to be the “background;” any with intensity greater than τ standard deviations above the mean are detected as plumes and grouped into contiguous connected regions. For this work, we favor τ = 5.0 as our threshold; more lenient values would result in higher detection rates at the risk of false positive detections.