Scholarlink

<p>Detecting and cataloguing spacecraft hardware on the lunar surface remains challenging even after six decades of exploration. We present a lightweight computer vision system, YOLO-ETA (You-Only-Look-Once – Extraterrestrial Artefact), adapted from TinyYOLOv2 for identifying anthropogenic objects in high-resolution Lunar Reconnaissance Orbiter Camera (LROC) imagery. Trained on Apollo landing-site data, YOLO-ETA achieved balanced precision–recall (F1 ≈ 0.60) and an 80% mean confidence score for lander detections in previously unseen images and correctly localised the Luna 16 spacecraft. Applying the model to a 5 × 5 km region surrounding the historically uncertain Luna 9 landing area yielded several high-confidence detections of artificial objects near 7.03° N, –64.33° E. Topographic analysis indicates that the candidate site’s horizon geometry is potentially consistent with Luna 9 surface panoramas. These findings identify promising locations for follow-up imaging and demonstrate that compact, edge-deployable machine-learning models can support future orbital surveys of lunar artefacts and surface assets.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Possible identification of the Luna 9 Moon landing site using a novel machine learning algorithm

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月21日

查看全文
  • Lewis J. Pinault,
  • Ian A. Crawford,
  • Hajime Yano

摘要

Detecting and cataloguing spacecraft hardware on the lunar surface remains challenging even after six decades of exploration. We present a lightweight computer vision system, YOLO-ETA (You-Only-Look-Once – Extraterrestrial Artefact), adapted from TinyYOLOv2 for identifying anthropogenic objects in high-resolution Lunar Reconnaissance Orbiter Camera (LROC) imagery. Trained on Apollo landing-site data, YOLO-ETA achieved balanced precision–recall (F1 ≈ 0.60) and an 80% mean confidence score for lander detections in previously unseen images and correctly localised the Luna 16 spacecraft. Applying the model to a 5 × 5 km region surrounding the historically uncertain Luna 9 landing area yielded several high-confidence detections of artificial objects near 7.03° N, –64.33° E. Topographic analysis indicates that the candidate site’s horizon geometry is potentially consistent with Luna 9 surface panoramas. These findings identify promising locations for follow-up imaging and demonstrate that compact, edge-deployable machine-learning models can support future orbital surveys of lunar artefacts and surface assets.