Flat-Bottom Lunar Craters Selection using Deep Learning-Based Identification and Depth Estimation

The selection of optimal lunar landing sites is a critical aspect of mission planning for future space exploration. Flatbottom craters present promising locations due to their stable terrain and relative safety for spacecraft touchdown. This paper introduces a deep learning approach employing Convolutional Neural Networks (CNNs) for the automated identification of flatbottom lunar craters and the estimation of their depths. The methodology involves two specialized CNN models: one for binary classification to identify flat-bottom craters and another for regression to estimate crater depths using high-resolution lunar imagery and Digital Elevation Models (DEMs). The first model utilizes transfer learning with a modified ResNet50 architecture to classify craters, achieving high accuracy and robustness through data augmentation and regularization techniques. The second model adapts the ResNet50 for regression tasks to predict crater depths, incorporating Global Average Pooling and dropout layers to prevent overfitting. The models were trained and validated using annotated datasets, ensuring geological accuracy. Experimental results demonstrate the effectiveness of the proposed models, with significant implications for enhancing lunar landing site selection processes.