Recognizing past shapes: sex differentiation through deep learning on European Upper Palaeolithic hand stencils

Abstract

This study explores the application of advanced deep learning techniques in analyzing Upper Palaeolithic hand stencil representations, focusing on sex classification of individuals involved in prehistoric rock art activity. The research highlights the effectiveness of deep learning models, particularly EfficientNetV2-S, which achieved an accuracy rate of 81.03 % for experimental blown hand stencils and 95.08 % in delineated contemporary hand image samples for sex identification, surpassing traditional morphometric methods. The study demonstrates that deep learning can differentiate male and female hand stencils with high precision, suggesting a mixed-sexual participation in creating prehistoric art, with a slight prevalence of male hand representations in the studied caves. The integration of user-friendly platforms, such as Google Colab, facilitates the reproducibility and validation of these findings, promoting methodological transparency. However, the accuracy of deep learning models is contingent on the quality and preservation of the images, presenting challenges when working with deteriorated or incomplete samples. This work highlights the potential of advanced technologies in archaeological research, opening new avenues for investigating the creation of prehistoric graphic expressions and their social implications.