Nan Jing, Wang Yexin, Di Kaichang, Xie Bin, Zhao Chenxu, Wang Biao, Sun Shujuan, Deng Xiangjin, Zhang Hong, Sheng Ruiqing
State Key Laboratory of Remote Sensing Science, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100101, China.
University of Chinese Academy of Sciences, Beijing 100101, China.
Sensors (Basel). 2025 Jan 3;25(1):243. doi: 10.3390/s25010243.
The Chang'e-6 (CE-6) landing area on the far side of the Moon is located in the southern part of the Apollo basin within the South Pole-Aitken (SPA) basin. The statistical analysis of impact craters in this region is crucial for ensuring a safe landing and supporting geological research. Aiming at existing impact crater identification problems such as complex background, low identification accuracy, and high computational costs, an efficient impact crater automatic detection model named YOLOv8-LCNET (YOLOv8-Lunar Crater Net) based on the YOLOv8 network is proposed. The model first incorporated a Partial Self-Attention (PSA) mechanism at the end of the Backbone, allowing the model to enhance global perception and reduce missed detections with a low computational cost. Then, a Gather-and-Distribute mechanism (GD) was integrated into the Neck, enabling the model to fully fuse multi-level feature information and capture global information, enhancing the model's ability to detect impact craters of various sizes. The experimental results showed that the YOLOv8-LCNET model performs well in the impact crater detection task, achieving 87.7% Precision, 84.3% Recall, and 92% AP, which were 24.7%, 32.7%, and 37.3% higher than the original YOLOv8 model. The improved YOLOv8 model was then used for automatic crater detection in the CE-6 landing area (246 km × 135 km, with a DOM resolution of 3 m/pixel), resulting in a total of 770,671 craters, ranging from 13 m to 19,882 m in diameter. The analysis of this impact crater catalogue has provided critical support for landing site selection and characterization of the CE-6 mission and lays the foundation for future lunar geological studies.
嫦娥六号（CE-6）在月球背面的着陆区位于南极-艾特肯（SPA）盆地内的阿波罗盆地南部。对该区域撞击坑进行统计分析，对于确保安全着陆和支持地质研究至关重要。针对现有撞击坑识别存在的背景复杂、识别精度低、计算成本高等问题，提出了一种基于YOLOv8网络的高效撞击坑自动检测模型YOLOv8-LCNET（YOLOv8-月球撞击坑网络）。该模型首先在主干网络末端引入了部分自注意力（PSA）机制，使模型能够以较低的计算成本增强全局感知并减少漏检。然后，在颈部集成了聚集-分布机制（GD），使模型能够充分融合多级特征信息并捕捉全局信息，增强了模型检测各种大小撞击坑的能力。实验结果表明，YOLOv8-LCNET模型在撞击坑检测任务中表现良好，精确率达到87.7%，召回率达到84.3%，平均精度达到92%，分别比原始YOLOv8模型高出24.7%、32.7%和37.3%。随后，将改进后的YOLOv8模型用于嫦娥六号着陆区（246千米×135千米，数字正射影像图分辨率为3米/像素）的撞击坑自动检测，共检测到770671个撞击坑，直径从13米到19882米不等。对该撞击坑目录的分析为嫦娥六号任务的着陆点选择和特征描述提供了关键支持，并为未来的月球地质研究奠定了基础。
