Automatic lunar dome detection methods based on deep learning

Abstract

Lunar domes are common structures on the lunar surface and are important for studying the geological evolution of the moon. The distribution of spatial frequencies of lunar domes provides significant evidence for the evolution of lunar volcanoes. In recent years, deep learning methods have been rapidly developing in many fields. However, most of the existing dome detection algorithms use manual or semi-automatic traditional methods. In this paper, we propose an automatic deep learning recognition method to simplify the traditional dome identification process, which is an end-to-end detection method. We built a lunar dome dataset using digital elevation model data and compared eleven advanced deep learning target detection algorithms, which include three types of detection architecture. The region of Marius Hills was selected for validation to evaluate method performance. By comparing the results with manual identification, the proposed method has an identification precision of 88.7%. In addition, we detected 12 unrecorded potential domes/cones. The morphological characterization and visualization results indicate that the detected features may be domes/cones and our method may provide novel dome detection.