Fast and Accurate Crater Detection on Martian Surface Using SUN et 3+

Abstract

Crater detection algorithms (CDAs), playing a key role in deep space exploration and planetary science, could help researchers determine the age and depth of the planetary surface by identifying the location, size and amounts of impact craters on planetary surface. In this paper, we propose a simplified full-scale skip connected UN et termed SUN et 3+ to achieve fast and accurate crater detection on Mars surface with digital elevation model (DEM) data. The full-scale connection in SUNet 3+ could incorporate low-scale and large-scale features of craters to improve detection accuracy. And the skip connection could improve the computation efficiency with less network parameters than other UN et-series models. The experimental results show that our proposed architecture achieve a 84.97% accuracy with 13.6 MB size which make it possible to deploy the algorithms on aerospace crafts with high computation efficiency.