ImageOP: The Image Dataset with Religious Buildings in the World Heritage Town of Ouro Preto for Deep Learning Classification

A new image dataset, ImageOP, comprising 1613 images of religious buildings from Ouro Preto, the first Brazilian UNESCO World Heritage Site, achieves high accuracy in deep learning classification. Using EfficientNet B0 and MobileNet V2 architectures, the dataset demonstrated exceptional performance with an accuracy of 94.5%, precision, recall, and F-score of 96.0% in simulations. Real-time mobile device testing confirmed its effectiveness, achieving 88.0% average accuracy across nine monuments. This advancement addresses a critical gap in datasets for Brazilian religious building classification, offering potential for heritage digitization and conservation applications.

Ottoni, A. L. C., & Ottoni, L. T. C. (2024). ImageOP: The Image Dataset with Religious Buildings in the World Heritage Town of Ouro Preto for Deep Learning Classification. Heritage, 7(11), 6499–6525. https://doi.org/10.3390/heritage7110302