Reproduce “Deep learning framework for crater detection and identification on the Moon and Mars” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yihan Ma, Zeyang Yu, Rohitash Chandra
Researchers from UNSW Sydney developed a deep learning framework evaluating CNN, YOLOv11, and ResNet-50 for automated crater detection and identification on the Moon and Mars, providing annotated datasets and insights into model performance across different crater sizes. The CNN model achieved an F1-score of 0.99 for small craters on the Moon, while YOLOv11 demonstrated more balanced detection for medium and large craters.
Key insights
- Class imbalance significantly impacts model performance across all evaluated architectures, particularly affecting the detection of less frequent medium and large craters.
- Different deep learning models exhibit varied strengths: CNNs are highly proficient in detecting small craters, while YOLOv11 offers more balanced detection performance across medium and large crater sizes.
- A hybrid prediction strategy combining direct image input with a sliding window approach and Non-Maximum Suppression (NMS) effectively handles the wide range of crater sizes, improving detection accuracy for smaller craters.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.