In Process: The Development of an Automatic Deep-Learning Phytolith Analysis Workflow

Summary

This is an abstract from the "Advances in Macrobotanical and Microbotanical Archaeobotany Part 1" session, at the 89th annual meeting of the Society for American Archaeology.

In this paper we present our lab's latest results using deep-learning (DL) to identify and analyze phytoliths, robust inorganic silica ‘casts’ of plant-cells. This use of DL technology will revolutionize phytolith analysis transforming the possibilities of this paleoethnobotanical method. Previous studies carried out in relation to the automated detection and classification of objects such as mounds or potsherds in images have demonstrated the potential of DL in the field of archaeology. Studies with pollen or cells have also shown DL’s potential in bioarcheology. Regarding phytoliths, previous works have shown the potential of automated classifications, but these studies focused on single-cell phytoliths. Our lab recently demonstrated that it is possible, using a pretrained DL model, to identify and classify multi-cell phytoliths, specifically three key grass husk multi-cell genera: wheat, barley and oat. Focused on the wave pattern, the negative space between the phytoliths, our DL algorithms produced an identification accuracy of 93.68%. Building on this success, we have continued to develop our algorithms for Near Eastern plant taxa with positive results. This research ultimately aims to facilitate paleoenvironmental analyses at a landscape scale, making the prehistoric environment and human impacts in those environments visible at a scale otherwise not possible.

This Resource is Part of the Following Collections

• Advances in Macrobotanical and Microbotanical Archaeobotany Part 1   •
• Society for American Archaeology 89th Annual Meeting, New Orleans, LA (2024)

Keywords

Record Identifiers

Abstract Id(s): 39694.0