Archaeologists have a new tool for sorting through thousands of years of plant remains: an AI system that can classify ancient seeds with 90.2 percent accuracy. APSNet, developed by researchers at Lingnan University and Shandong University, was trained on 8,340 images from 18 archaeological sites across China, with samples dating from 5400 BCE to 220 CE. For research scientists, the system promises to cut months of painstaking laboratory work down to minutes while maintaining specialist oversight.
The research, published in npj Heritage Science, provides both a large standardized image collection of archaeological seeds and an AI system designed specifically to help archaeobotanists identify them.
The bottleneck in archaeobotany
Seed identification is foundational work, but it moves slowly. Archaeologists recover plant remains through flotation, where excavated soil is processed in water to separate lightweight carbonized seeds and other organic material. Specialists then examine each fragment under a microscope, comparing size, shape, and surface texture. Becoming proficient takes years of training.
Large excavations produce substantial quantities of material, each fragment requiring individual examination. The researchers argue that this creates a bottleneck when archaeobotanical evidence must be processed on a large scale. Agricultural practices, diets, and how communities adapted to changing environments all depend on the ability to identify what was grown and eaten - but the identification itself often slows research down.
To address this, the team assembled the Ancient Plant Seed Image Classification dataset, bringing together 8,340 images of ancient plant remains across 17 categories. The collection includes barley, wheat, foxtail millet, broomcorn millet, and peach stones. The broad chronological and geographical range matters: charred, fragmented, or distorted seeds rarely resemble textbook modern reference specimens. Seeds of the same species can look different; seeds from different species can look nearly identical.
How APSNet works
APSNet was designed around the same visual clues specialists use when examining archaeological plant remains. Instead of relying only on general image patterns, the system processes information about seed size while analyzing finer morphological characteristics. The research team describes this as a way of guiding the model toward features useful for distinguishing specimens that may otherwise look extremely similar.
In practice, a researcher can upload microscope images of seeds to the system and receive a preliminary classification. Identification records are stored digitally, creating a standardized workflow for comparing large collections. Tests comparing APSNet with 28 existing image-classification methods showed it achieved the strongest performance on the archaeological seed dataset.
Researchers emphasize that APSNet is intended as an assistive tool, not a replacement for archaeological expertise. Specialists still verify classifications and interpret each specimen in context - whether it reflects cultivation, storage, food preparation, trade, or accidental deposition.
For professionals in AI for Science & Research, the approach offers a use case in specialized tasks where small sample sizes and high variability, not just data quantity, present the real challenge. It also works with a model that works with messy real-world data, not pristine lab specimens.
Potential for large-scale investigations
Faster classification changes what questions archaeologists and archaeobotany professionals can realistically ask. A faster method for preliminary classification could allow specialists to process larger assemblages of plant remains while freeing more time and other resources for the detailed interpretation of what those remains reveal about past communities, diets, and agricultural practices.
Those interpretations address fundamental questions: when particular crops appeared in a region, which plants became staple foods, and how cultivation strategies adapted to local ecological conditions. Standardized identification may also make it easier to compare plant remains across different sites and periods with less inconsistency.
In China, where a long archaeological record captures major developments in millet and wheat agriculture, this capacity could prove particularly useful for tracing changing relationships between communities, crops, and landscapes.
Why this matters for science and research professionals
AI Learning Path for Research Scientists professionals must often weigh the value of automation against the risk of losing domain expertise. APSNet solves a practical problem rather than an academic one. The bottleneck in archaeobotany is not detection - it is volume and specialization. A system that reduces the time spent sorting thousands of charred fragments, while still requiring human verification, aligns with how archaeological research works. The model is not trying to replace the specialist; it is designed to handle the drudge work so the specialist can spend more time on the interpretation that only a domain expert can provide. For science professionals evaluating AI applications, this signals a shift: the technology's ability to bootstrap its own training data from a mix of reinforced applications.
Your membership also unlocks:
