A new MarsRock-FasterRcnn method to extract Mars rocks using HiRISE images at Utopia Plainitia

Rocks on Mars' surface are an important feature of its geomorphology, and rock abundance is a key consideration for the selection of landing sites, as well as the engineering safety of the landing. In this study, we proposed a Faster R-CNN-based intelligent extraction method (i.e., MarsRock-FasterRcnn) for detecting Mars rocks from images (0.25 m/pixel) acquired by the High-Resolution Imaging Camera (HiRISE) onboard the MRO orbiter. MarsRock-FasterRcnn method can extract rocks with diameter of 0.5–2 m on HiRISE images. We built a rock sample set containing 2000 images of size 416 × 416 to train the MarsRock-FasterRcnn model, and finally evaluated the prediction effect of the model using assessment metrics including precision, recall, and F1 score. The results showed that the recall and precision were both about 80 %, and the maximum confidence was above 99 %, indicating excellent model performance. The extraction and validation of rocks in Utopia Plainitia show that the rock abundance is below 7 % in most of the region, and the overall rock abundance in the southern part of the region is below 5 %, indicating weaker effects by the rocky obstacles when landing. We also found a strong relationship between the rock abundance and topography, where a higher rock abundance relates to a greater topographic undulation. This study can provide scientific clues for the investigation of the topography and geology of Mars, as well as engineering constraints for the selection of landing sites for exploration missions.