Overcoming Bottlenecks in Seed Identification: A Proposal for AI-Assisted Archaeobotanical Research on Siraya Cultural Sites

CHANG Chi-Shan1* and CHUANG Shih-Ying2

Plant seed remains are essential for reconstructing plant resource use and human–environment interactions in late prehistoric to early historic Taiwan. However, archaeobotanical research faces structural bottlenecks, including modern seed intrusion, poor preservation in humid tropical environments, and morphological degradation that obscures diagnostic traits. This study proposes an AI-assisted methodological framework to address preservation bias in Siraya cultural sites (ca. 300–500 BP) within the Tainan Science Park archaeological cluster. Drawing on curated seed collections, the project will establish a preservation grading system, digitize morphological features, and integrate stratigraphic controls to mitigate contamination. Deep learning models (e.g., YOLOv8) will be tested to assess their feasibility for family- and genus-level classification across varying preservation states. By systematically evaluating the applicability of AI image recognition, this proposal aims to explore whether computational methods can mitigate preservation bias and contribute to a replicable workflow for humid tropical archaeobotany.