Using a image classification model in deep learning (CNN), we aim to automatically detect masses of rock referred to as boulders on the Moon’s surface. Boulders on the Moon’s surface are obstacles that make it hard for probes to land and rovers to drive. On the other hand, locations with uneven distributions of boulders are correlated with places of scientific interest such as uneven distribution areas of crust and mantle materials under the Moon. Automatic detection of boulders contributes to the selection of landing and exploration sites.

Scientists detect craters and boulders from images of the Moon’s surface for planning exploration missions and scientific research . This work is time-consuming, so various ideas have previously been devised to deal with it. Together with universities and research institutions, JLPEDA is also investigating methods that utilize machine learning.

Figure 1 shows an example of automated detection results. Yellow dots are spots that exist boulders (visually detected). Red squares are spots among those indicated by yellow dots for which automated detection of boulders through the recognition model was successful. Blue squares are spots where detection was unsuccessful.