SqUNet: An High-Performance Network for Crater Detection With DEM Data

Abstract

Identification of craters plays an essential role in planetary exploration. This article proposes a new neural network, called Square U-Net (SqUNet) for automatic detection of craters using digital elevation model (DEM) images of lunar and Mars. The SqUNet uses an embedded U-Net architecture, which includes an encoding and a decoding structure, to replace the traditional convolution module. This kind of structure can significantly improve the feature learning ability. Moreover, a skip link is added inside the embedded U-Net structure to retain the feature information of the original map. We compare our SqUNet model with six other state-of-the-art crater detection models through experiments. The experiments show that the SqUNet can effectively improve the recall rate and the precision of detection. Our model can not only discover new craters but also improve the accuracy of crater segmentation. In addition, the testing results on the Mars DEM dataset have also demonstrated the strong generalization and robustness of the SqUNet network model. It holds significant potential for various applications, providing valuable support for lunar exploration and geological research.

Keywords

• Crater detection
• deep learning
• planetary exploration
• U-Net