The Application of Machine Learning to At-Risk Cultural Heritage Image Data

Abstract

This project investigates the application of Convolutional Neural Network (CNN) methods and technologies to problems related to At-Risk cultural heritage object recognition. The primary aim for this work is the use of developmental software combining the disciplines of computer vision and artefact studies, developing applications in the field of heritage protection specifically related to the illegal antiquities market. To accomplish this digital image data provided by the Durham University Oriental Museum was used in conjunction with several different implementations of pre-trained CNN software models, for the purposes of artefact Classification and Identification. Testing focused on data capture using a variety of digital recording devices, guided by the developmental needs of a heritage programme seeking to create software solutions to heritage threats in the Middle East and North Africa (MENA) region. Quantitative data results using information retrieval metrics is reported for all model and test sets, and has been used to evaluate the models predictive results.