Research Data - Artificial intelligence achieves stylistic classification of painted pottery: Identifying vessel shape and decoration in one model
Description
Recent advances in artificial intelligence modeling have begun to assist in the archaeological analysis of the patterns and images on pottery corpuses. Current automated classification models for archaeological images, however, only consider single variables, i.e., vessel shape or ornamentation, in a single function. We report on a convolutional neural network classification model trained by the authors to analyze and consider multiple parameters in pottery style, including shape, texture, ornamentation and pattern placement. The model, trained on a photographic image dataset of Chinese color-painted pottery, achieved 91.3% accuracy in matching images of pots to corresponding archaeological cultures. Based on feature vectors obtained from the training procedure, similarity coefficients for the given images were calculated so that assigned images could be matched with the most similar examples in the dataset, allowing for automated comparison between vessels. This work highlights the potential of convolutional neural network (CNN) approaches in analyses of pottery style, providing a new tool for studying chronology, craft standardization and specialization, interregional technical and stylistic inter action, and formation of decorative design.