In a laboratory in southern China, a peach stone the size of a thumbnail sits beside a computer screen. This tiny pit traveled through more than 5,000 years of history to reach this moment, and now artificial intelligence is helping archaeologists figure out exactly where it came from and what it means.
Researchers at Lingnan University in Hong Kong and Shandong University have built the world's first AI system designed to identify ancient plant seeds from archaeological digs. The system, called APSNet, learned to recognize seeds by studying photographs of 8,340 seeds collected from 18 dig sites across China. These seeds date back as far as 5,400 years before the common era, covering a span of human civilization longer than written history itself.
Archaeobotanists — scientists who study ancient plant remains — traditionally identify seeds by examining each one individually under a microscope. This work is extremely slow and requires years of specialized training. "It usually takes several years to train an archaeobotanical expert to identify ancient plant seeds independently," said Sam Kwong Tak-wu, a professor at Lingnan University who helped lead the project. The new AI system can do in seconds what takes humans hours.
The team collected images of 17 different types of seeds, including staple crops like barley, wheat, foxtail millet, and broomcorn millet, as well as fruit remains like peach stones. They then taught the AI to think like an expert, first looking at a seed's overall size and shape before examining smaller details like the position of the embryo and the texture of the seed's surface. When tested, APSNet correctly identified seeds 90.2 percent of the time, outperforming the best existing image-recognition AI by 5.3 percentage points.
The system is not meant to replace human experts entirely. Researchers can upload seed photos or connect a digital microscope to get quick results, which are automatically stored in a shared database. This lets specialists spend less time on routine identification and more time on deeper analysis. "This helps reduce their workload and allows them to devote more time to in-depth archaeobotanical analysis," Kwong explained.
Ancient seeds tell stories about what ancient people ate, how they farmed, and how civilizations grew and spread. By making identification faster and easier, this technology could help archaeologists around the world study more sites in less time. The research was published in the journal Heritage Science.