Sustainable artificial intelligence for historic building components classification: a case of UNESCO world heritage site in Congonhas, Brazil

Purpose This study aims to propose a sustainable artificial intelligence (AI) approach for the historic building components classification. To this end, the energy efficiency of deep learning models is investigated in the detection of key architectural elements of churches in the Historic Town of Congonhas, Brazil. In addition, practical tests are conducted at the Sanctuary of Bom Jesus, a significant Brazilian monument and UNESCO World Heritage Site. Design/methodology/approach The methodology proposed in this paper consists of five main stages: (1) data collection through the capture of photographs of cultural heritage buildings,(2) dataset organization for training, validation, and testing experiments, (3) selection of six traditional deep learning models from the literature, (4) design of experiments for simulation and real-world testing and (5) sustainable artificial intelligence calculations to assess the energy efficiency of the deep learning models. Findings The results demonstrate that it is possible to conduct experiments for historic building component classification using more energy-efficient computational models, such as MobileNet and MobileNetV2. In other words, these models require less energy for training the artificial intelligence. Furthermore, the sustainable AI models achieved accuracy levels comparable to those of more energy-intensive structures. Originality/value This study presents an innovative contribution through a comprehensive analysis of energy consumption for historic building component detection using computer vision. Additionally, the case study involving sustainable artificial intelligence applied to a UNESCO World Heritage Site in Congonhas represents a novel approach in the recent literature.