Scholarlink

<p>Identification of archaeological wood species is critical for decoding the biogeographical, historical, and cultural narratives of artifacts, but remains a challenge in practice. To address this challenge, we developed a deep learning framework integrating computer vision and archaeological wood species analysis. Images of modern and waterlogged archaeological wood specimens were collected and augmented through rotation, scaling, and contrast adjustments. The deep learning model was established through embedded self-attention mechanisms, optimizing feature extraction capabilities. The model yielded an overall accuracy of 94.20% on 12 modern woods and 97.83% on 2 archaeological woods. The binary classifier was further evaluated on three batches of archaeological woods and achieved the accuracy of 98.31%, 96.19% and 91.82%, respectively. The deep learning approach proposed in this study achieved good performance on waterlogged archaeological wood identification while requiring less wood anatomy expertise and minimal intervention of cultural artifacts, providing a critical advancement for archaeology applications.</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

Inferring ancient from modern: a deep learning approach for species identification of archaeological wood

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月22日

查看全文
  • Jiajun Wang,
  • Aikun Lin,
  • Ling Fang,
  • Yang Lu,
  • Zhiyuan Zou,
  • Zhiguo Zhang,
  • Yafang Yin

摘要

Identification of archaeological wood species is critical for decoding the biogeographical, historical, and cultural narratives of artifacts, but remains a challenge in practice. To address this challenge, we developed a deep learning framework integrating computer vision and archaeological wood species analysis. Images of modern and waterlogged archaeological wood specimens were collected and augmented through rotation, scaling, and contrast adjustments. The deep learning model was established through embedded self-attention mechanisms, optimizing feature extraction capabilities. The model yielded an overall accuracy of 94.20% on 12 modern woods and 97.83% on 2 archaeological woods. The binary classifier was further evaluated on three batches of archaeological woods and achieved the accuracy of 98.31%, 96.19% and 91.82%, respectively. The deep learning approach proposed in this study achieved good performance on waterlogged archaeological wood identification while requiring less wood anatomy expertise and minimal intervention of cultural artifacts, providing a critical advancement for archaeology applications.