FPW-YOLO11n: A Lightweight Frequency-Perception Framework for Lunar Impact Crater Detection

Area of Science:

• Planetary Science
• Geoinformatics
• Computer Vision

Background:

• Automated lunar impact crater detection from digital elevation model (DEM) data is vital for geological analysis, landing site selection, and catalog updates.
• Challenges include variations in crater scale, degraded rims, ambiguous boundaries, and complex topographic backgrounds, demanding efficient and accurate detection models.
• Existing methods struggle to balance accuracy, model complexity, and inference speed for large-scale lunar remote sensing applications.

Purpose of the Study:

• To propose FPW-YOLO11n, a novel frequency-perception crater detection method based on YOLO11n, to address the challenges in automated lunar crater identification from DEM data.
• To enhance the representation of crater features and improve detection accuracy and efficiency through novel module integration and loss function optimization.

Main Methods:

• Introduced a Frequency-Directional Attention Module (FDA-Module) in the backbone to enhance representations of crater rim structures and topographic cues using frequency and direction-aware attention.
• Designed a C2PSA-LRSA module by embedding Local Region Self-Attention into C2PSA to improve local contextual feature interaction while reducing computational cost.
• Replaced the original CIoU loss with Inner-WIoU, combining auxiliary-box mechanisms and sample-quality-aware weighting for more flexible bounding-box regression on challenging crater features.

Main Results:

• FPW-YOLO11n achieved 78.3% Precision, 66.2% Recall, 75.1% mAP@0.5, and 50.2% mAP@0.5:0.95 on a constructed DEM-based lunar crater dataset, outperforming the YOLO11n baseline.
• Consistent performance improvements were observed even with geographically disjoint data splitting, indicating the robustness of the proposed structural enhancements.
• Despite an increase in computational cost (6.3 to 24.0 GFLOPs), FPW-YOLO11n maintained a compact parameter size (2.59 M) and high inference speed, showing an improved accuracy-efficiency trade-off.

Conclusions:

• The proposed FPW-YOLO11n method effectively enhances lunar crater detection from DEM data by improving feature representation and bounding-box regression.
• The integration of FDA-Module, C2PSA-LRSA, and Inner-WIoU loss function offers a superior accuracy-efficiency trade-off compared to the baseline YOLO11n.
• FPW-YOLO11n demonstrates significant potential for large-scale lunar remote sensing applications, including geological analysis and landing-site selection.