PLANETARY SCIENTIFIC TARGET DETECTION VIA DEEP LEARNING: A CASE STUDY FOR FINDING SHATTERCONES IN MARS ROVER IMAGES

Robotic Mars missions contain op-tical instruments for various mission-related and sci-ence tasks, such as 2D and 3D mapping, geologic characterization, the characterization of scientific con-text, and the identification of scientific targets of in-terest. The considerable variability of appearance of potential scientific targets calls for well-adapted yet flexible techniques, one of them being Deep Learning (DL). Our “Mars-DL” (Planetary Scientific Target Detection via Deep Learning) approach focuses on training for visual DL by virtual placement of known targets in a true context environment. The 3D context environment is taken from reconstructions using Mars rover imagery. Objects of scientific interest, such as impact-characteristic shatter cones (SCs) from terres-trial impact craters, and/or meteorites, are captured and 3D reconstructed using photogrammetric tech-niques, providing a 3D data base of high resolution mesh and albedo map models. By placing the models randomly into realistic scenes using image rendering methods (Fig. 1), we create an artificial training data set, which is used to train the Deep Learning solution.