YOLOv8-LCNET: An Improved YOLOv8 Automatic Crater Detection Algorithm and Application in the Chang'e-6 Landing Area

Area of Science:

• Planetary Science
• Artificial Intelligence
• Remote Sensing

Background:

• The Chang'e-6 mission targets the South Pole-Aitken basin on the Moon's far side, necessitating precise impact crater analysis for safe landing and geological research.
• Existing automated methods for lunar crater identification face challenges including complex backgrounds, low accuracy, and high computational demands.

Purpose of the Study:

• To develop an efficient and accurate automatic impact crater detection model for the Chang'e-6 landing area.
• To enhance existing deep learning models for improved lunar crater identification, addressing limitations in perception and feature fusion.

Main Methods:

• Proposed YOLOv8-LCNET, an enhanced YOLOv8 model incorporating a Partial Self-Attention (PSA) mechanism and a Gather-and-Distribute (GD) mechanism.
• Integrated PSA into the Backbone to improve global perception and reduce missed detections with low computational cost.
• Integrated GD into the Neck for multi-level feature fusion and enhanced detection of various crater sizes.

Main Results:

• YOLOv8-LCNET achieved 87.7% Precision, 84.3% Recall, and 92% AP, significantly outperforming the original YOLOv8 model.
• The model demonstrated substantial improvements over the baseline YOLOv8, with Precision, Recall, and AP increases of 24.7%, 32.7%, and 37.3%, respectively.
• Automatic detection in the CE-6 landing area identified 770,671 craters ranging from 13 m to 19,882 m in diameter.

Conclusions:

• The YOLOv8-LCNET model offers a highly effective solution for automatic lunar impact crater detection.
• The generated catalogue of craters provides critical data supporting the Chang'e-6 mission's landing site selection and characterization.
• This research establishes a foundation for advanced lunar geological studies and exploration missions.