A New Approach for Classification and Detection of World Cultural Heritages with YOLOv3

Abstract

Since the formation of the world, many civilizations have lived and these civilizations have left important information about their social and architectural structures by creating a cultural heritage infrastructure in order to transfer their experiences to future generations. In this process, civilizations have left many cultural heritage works, both tangible and intangible. The concept of digital heritage has emerged for the protection and promotion of cultural heritage sites and works that still exist today, and studies have been carried out in the literature for the digitization of the works of certain regions. In this study, apart from literature studies, the images of 200 different artifacts included in the World Heritage List, which contain images of various structures from different regions created by the United Nations Educational, Scientific and Cultural Organization (UNESCO), are manually collected, labeled and classified and necessary pre-processing is carried out. By performing the steps, it was primarily turned into a data set. Then, with the created dataset, a training process was carried out using the You Only Look Once version 3 (YOLOv3) technology, which detects objects in images using Convolutional Neural Networks, one of the deep learning techniques, and the process of modeling is explained. Our main goal in creating the model has been to classification these works, which have scientific data characteristics, in the digital environment, to identify them, to be recognized by future generations, to transfer them in a safe way, and to ensure that researchers who will work in this field can use them as a source. The performance of the model was tested by using different images to verify with the data for the designed model. As a result of the performance tests carried out, an accuracy rate of 97% was obtained. © 2023 IEEE.