Deep learning detects entire multiple-size lunar craters driven by elevation data and topographic knowledge

Lunar craters are important geomorphological features, that provide valuable insights into lunar morphology, geology, and impact processes. However, the current understanding of lunar craters of different sizes, especially smaller craters (diameter <5 km), is still incomplete. The lack of understanding of small lunar craters affects our understanding of the lunar surface and its geological history. Therefore, in this study, we propose a deep learning Crater Detection Algorithms (CDA), called Lunar Topographic Knowledge Attention U-Net (LTKAU-Net) that integrates a Digital Elevation Model (DEM) and topographic knowledge. This CDA combines DEM data and topographic knowledge (Slope of Slope) through additional input channels to enhance the recognition ability for craters of different sizes, especially for small craters. The experimental results show that LTKAU-Net not only improves the extraction of craters larger than 5 km in diameter but also identifies many craters less than 5 km in diameter. In addition to detecting more craters, LTKAU-Net achieves high accuracy for both the validation and test datasets. The distribution of craters exhibits a heterogeneous spatial pattern, with lower kernel density values in the southern hemisphere than in the northern hemisphere, and lower values in the western hemisphere than in the eastern hemisphere. In addition, the Nectarian and Imbrian periods are the periods with the highest number of craters. Our research results can supplement existing manually annotated datasets, providing comprehensive evidence for further studies of the distribution and geological evolution of lunar craters.