Tackling Small Lunar Impact Crater Classification: A Novel Augmented Data Set and Enhanced Deep Learning Framework

Abstract Small lunar impact craters, exhibiting diverse morphological characteristics, represent the most abundant geological features on the Moon's surface. These structures serve as crucial indicators for understanding regolith thickness distribution and surface modification processes. While automated crater detection has seen remarkable progress, classification of small crater morphotypes remains underexplored, particularly due to inherent class imbalance in existing data sets. This study presents an optimized YOLOv7‐based classification framework specifically designed for multi‐type small crater recognition. We develop the Small Craters Augmented Data set (SCAD) incorporating advanced data augmentation strategies to address type imbalance, particularly enhancing representation of rare morphologies: flat‐bottomed, central mound, and concentric craters. The proposed model integrates Focal Loss for class imbalance mitigation and Efficient‐IoU optimization for improved boundary detection. Experimental results demonstrate superior performance with 90.70% precision, 71.95% recall, and 80.21% F1, outperforming both DeepLabv3+ (F1: 73.54%) and SegFormer (F1: 73.97%) in comparative analysis. Validation against ground truth confirms the model's effectiveness in detecting sub‐kilometer scale features, advancing automated lunar surface analysis capabilities for planetary geological studies.