TOWARDS AUTOMATIC DETECTION OF SHATTER CONES ON PLANETARY ROVER IMAGES

Impact cratering is one of the most important geological processes in our solar system, observable on surfaces of most objects that have solid surfaces. From studies of terrestrial impact structures, shatter cones (SCs) are known as the only unambiguous macroscopic evidence of shock metamorphism (e.g., French and Koeberl 2010). Yet, until now, SCs have not been found on other planets, with the exception of a possible shatter cone reported from an image at Gale crater on Mars (Newsom et al. 2015). Due to the insufficient image resolution, it is not clear if the object really is a shatter cone. Unmanned rover missions, such as ExoMars 2022 with its HRC (High Resolution Camera), are capable of producing panoramic image data in substantially higher surface resolution. Combined with automatic detection enabled by machine learning, this can help speed up the selection of scientific interesting targets in the vicinity of the rover and, therefore, enhance the tactical and strategic decision-making of planetary surfaces missions.

The focus of our research Mars-DL is on the automatic detection of shatter cones in realistic Mars rover image scenes. Several SCs from the collection of the Natural History Museum Vienna and the University of Vienna were selected for three-dimensional (3D) image capturing. They originate from different terrestrial impact structures and cover a wide range of lithologies, including sedimentary, plutonic, and volcanic rocks. Typical features of SCs, such as striations and horsetail structures, are well developed in all selected specimens and allow a clear identification. Terrestrial analog studies in Ethiopia supported our lab work. Using PRo3D (Barnes et al. 2018), a viewer for the interactive exploration and geologic analysis of high-resolution planetary surface reconstructions, the SCs are virtually placed in 3D background scenes processed from true Mars rover imagery (e.g., from MSL Mastcam). PRo3D-rendered images of such scenes are used as training data for machine learning architectures. Such training will ideally lead to a "Scientific Target Consultant" helping to automatically detect SCs in images of forthcoming planetary rover missions. The whole workflow from simulation to training and recognition has been just finalized and verified for further testing.