Abstract
In extraterrestrial exploration missions, stable and repeatedly observable navigation landmarks are critical for achieving high-precision autonomous visual navigation and safe landing. However, significant differences in geomorphology and illumination conditions across celestial bodies result in pronounced variations in the scale, shape, and shadow-texture appearance of landmarks, making it difficult for intelligent recognition models trained on historical mission data to be directly transferred across celestial bodies. Meanwhile, obtaining high-quality pixel-level annotations for a target celestial body is costly and time-consuming, which further increases the difficulty of cross-celestial-body landmark recognition. To address this issue, this paper proposes an intelligent method for cross-celestial-body navigation landmark recognition. First, a multi-scale prototype aggregation module jointly captures global morphological structures and local boundary and texture details across feature hierarchies, producing more complete and discriminative prototypes under scale variations and weak-texture conditions. Second, an inter-class prototype adaptive learning module combines saliency-guided spatial consistency enhancement with momentum-based prototype updating. This module strengthens class separation, stabilizes decision boundaries, and reduces prediction fragmentation and missegmentation during cross-celestial-body transfer. Experimental results on the evaluated source–target celestial-body combinations show that the proposed method achieves better overall landmark segmentation performance than the selected comparison methods under the few-shot setting.
| Original language | English |
|---|---|
| Journal | IEEE Transactions on Aerospace and Electronic Systems |
| DOIs | |
| Publication status | Accepted/In press - 2026 |
| Externally published | Yes |
Keywords
- Cross-celestial-body adaptation
- extraterrestrial exploration
- landmark recognition
- visual navigation
Fingerprint
Dive into the research topics of 'An Intelligent Cross-Celestial-Body Landmark Recognition Method for Autonomous Navigation With Minimal Annotations'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver