.png){kind=icon}
.png){kind=icon}
Return
High-resolution lunar crater detection and chronology estimation through generative enhancement and redundancy filtering
.png){kind=icon}
PRE
AI
Q
T
Y
L
S
W
A
DOI：10.1016/j.image.2026.117573.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Lunar impact craters are key indicators of the Moon's geological evolution, yet weak-texture surfaces, illumination gradients, and densely distributed small craters make automatic detection extremely challenging. This study presents a task-oriented detection-chronology pipeline that combines local enhancement, morphology-preserving detection, and overlap-aware fusion to improve multi-scale crater analysis. A lightweight GAN is applied to restore rim contrast and shadow gradients in weak-texture regions, while the ADown module is integrated into the YOLOv8 backbone to retain fine lunar morphological cues that are easily lost during conventional downsampling. An 8 & times; 8 tile-global two-stage pipeline further improves the recovery of small craters while maintaining contextual consistency for large structures spanning multiple tiles. To merge heterogeneous detections, a Priority-Based Redundancy Filtering (PB-RF) strategy is used to reconcile tile duplication and suppress boundary artefacts more effectively than standard NMS, Soft-NMS, and DIoU-NMS under the evaluated settings. The overall system achieves precision 0.925, recall 0.902, and mAP50 0.946 with only 2.95 M parameters, 7.92 GFLOPs, 5.9 MB model size, 203.5 MB peak memory, and 3.96 ms latency (252.2 FPS). Experiments across four representative lunar regions show that the fused detector captures both weak and overlapping craters and provides stable size-frequency distributions for subsequent chronological estimation. Ages derived from Neukum and Chang'e-5 models remain comparatively consistent in regions with abundant 0-2 km craters, supporting the utility of the recovered crater populations for regional analysis. Overall, the proposed pipeline serves as a practical framework for high-resolution lunar crater mapping, while future work may further explore illumination-aware enhancement, DEM-guided shadow correction, and contour-aware transformer architectures. Code is available at:https://github.com/miluaaaaa/Full-Scale-Detection-and-Chronological-Analysis-of-Lunar-Impact-Craters/tree/master
Keywords:
Lunar impact crater identification
Segmentation strategy
Deep learning
Chronometric analysis
Journal
S
IF:
2.7
Papers:
18
Citations:
0
Organization
Cited Papers