Building and evaluation of an Algerian Cultural Heritage dataset using convolutional neural networks

Abstract

Preservation of cultural heritage is a field of high importance. Recently people are sharing architectural monument images on social media. In this paper, we try to recognize architectural monuments in digital photographs of Algerian cultural heritage using a convolutional neural network (CNN). As no datasets support the diversity of Algerian monuments, adapted for CNN training, we built a new dataset and made available to the public. AlgHeritage dataset consists of 20,000 images that can serve as a benchmark for various research fields, as it combines numerous real-world challenges. We evaluate our dataset with three CNN models MobileNetV3, InceptionV3 and InceptionResNetV2, and compare it with two other datasets. MobileNetV3 with fine-tuning produces a good accuracy of 93.29% on the AlgHeritage dataset in comparison with other datasets. The AlgHeritage dataset is available at https://bit.ly/3O38FOe.