AI-Powered Archaeology: Determining the Origin Culture of Various Ancient Artifacts Using Machine Learning

Abstract

This paper attempts to determine the degree of accuracy to which we can originate and classify ancient artifacts using AI. It compares the performance of two CNN models (InceptionResNetV2, and a VGG-19) on a specially curated dataset of 55,000 ancient artifact images belonging to over 343 different cultures. The dataset consists of samples from different regions all over the world, and is built using the British Museum’s online collection of over 3 million artifact images. For training the CNNs, an NVIDIA RTX A6000 GPU was used on a cloud computing platform, achieving high accuracy for both models. Two trials were run for the InceptionResNetV2 model, which achieved a final validation accuracy of 65 percent for its best trial. The VGG-19 performed poorly on the dataset, only achieving a 12.95 percent validation accuracy. Afterwards, the models obtained from both InceptionResNetV2 trials were evaluated on a set of artifacts of unknown origin, coming up with surprising results and providing insight into how effective image recognition can be in the field of archaeology.