Scholarlink

<p>Effective identification of starch grain botanical origins constitutes a critical aspect of archaeological research. This study proposes a quantitative starch grain morphology analysis framework integrating geographic information systems (GIS) and machine learning (ML), addressing limitations in traditional archaeobotanical analysis—specifically high subjectivity, low efficiency, and inconsistent data standards. Leveraging the ArcGIS platform, we developed a standardized morphological acquisition system that extracts 44 quantitative indices spanning three dimensions: size, shape, and feature. These indices were integrated with multivariate statistical analysis and ML algorithms (Random Forest, XGBoost, Artificial Neural Networks, and Naive Bayes) to achieve automated starch grain classification and identification. Using 15 common Chinese root and tuber species as reference samples, we constructed a high-accuracy ML classification model through microscopic imaging, GIS-based processing, and morphological quantification. The XGBoost model achieved 94.4% overall classification accuracy, significantly outperforming traditional clustering methods (52.1%). SHAP interpretability analysis further identified ARE (area) and Hu3 as pivotal discriminative morphological indicators, highlighting the classification potential of size and feature-class parameters. This work establishes a unified analytical standard for archaeobotanical starch grain morphology and advances AI applications in archaeological science.</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

Integrating GIS and machine learning for automated starch grain morphometric analysis: a novel framework for standardized archaeobotanical classification

  • 期刊论文

  • 发表日期：

    2026年2月20日

查看全文
  • Zhiwei Wan,
  • Bin Yuan,
  • Chenghao Zhang,
  • Xuewen He,
  • Lingyue Liu,
  • Ji Zeng,
  • Zhikun Ma,
  • Shifan Qiu,
  • Wen Lai,
  • Xiujia Huan

摘要

Effective identification of starch grain botanical origins constitutes a critical aspect of archaeological research. This study proposes a quantitative starch grain morphology analysis framework integrating geographic information systems (GIS) and machine learning (ML), addressing limitations in traditional archaeobotanical analysis—specifically high subjectivity, low efficiency, and inconsistent data standards. Leveraging the ArcGIS platform, we developed a standardized morphological acquisition system that extracts 44 quantitative indices spanning three dimensions: size, shape, and feature. These indices were integrated with multivariate statistical analysis and ML algorithms (Random Forest, XGBoost, Artificial Neural Networks, and Naive Bayes) to achieve automated starch grain classification and identification. Using 15 common Chinese root and tuber species as reference samples, we constructed a high-accuracy ML classification model through microscopic imaging, GIS-based processing, and morphological quantification. The XGBoost model achieved 94.4% overall classification accuracy, significantly outperforming traditional clustering methods (52.1%). SHAP interpretability analysis further identified ARE (area) and Hu3 as pivotal discriminative morphological indicators, highlighting the classification potential of size and feature-class parameters. This work establishes a unified analytical standard for archaeobotanical starch grain morphology and advances AI applications in archaeological science.