CraterID-Loc: End-to-End Crater Identification for Lunar Image Localization

Lunar positioning and navigation is a fundamental capability for autonomous operations in deep space missions. Existing methods typically follow a two-stage pipeline of feature detection and identification, which suffers from poor robustness, low efficiency, and complex deployment under variations in illumination, scale, and viewpoint. This paper proposes CraterIDLoc, a novel end-to-end framework that, for the first time, performs crater detection and identification using a single neural network and retrieves geographic coordinates directly via table lookup, significantly improving accuracy and system simplicity. To address the challenges posed by real imaging conditions, we construct a digital elevation model (DEM)-rendered dataset covering diverse illumination, viewpoints, and scales. A lightweight detection network, Lightweight Context-Aware YOLO (LCA-YOLO), is designed with structural enhancements in feature extraction, contextual modeling, and classification. Experimental results demonstrate that the proposed method outperforms existing approaches in accuracy, efficiency, and stability, showing strong potential for practical deployment and task adaptability.