Architectural cultural heritage embodies historical, artistic, and cultural values that require careful preservation while promoting sustainable development. The integration of advanced computational methods offers new opportunities for informed protection strategies. However, existing heritage protection approaches often rely on manual inspection, limited datasets, and subjective evaluations, leading to inefficiencies, delayed risk detection, and inadequate resource allocation. To address these limitations, this study proposes a Deep Learning-Assisted Heritage Risk Prediction (DL-HRP) framework, which combines convolutional neural networks (CNNs) for structural image analysis with data mining techniques for extracting hidden risk patterns from environmental, structural, and socio-economic datasets. The framework enables automated heritage condition assessment, risk level prediction, and adaptive preservation planning. The proposed method is used for real-time monitoring, predictive maintenance scheduling, and policy support, ensuring both immediate protection and long-term sustainability of heritage sites. Experimental results were obtained on the Architectural Heritage Elements Image64 Dataset (Kaggle) and compared with traditional heritage protection methods demonstrate that DL-HRP improves prediction accuracy by 18% compared to traditional methods, optimizes resource allocation, and supports data-driven decision-making. This approach provides a robust, scalable, and intelligent solution for heritage conservation, balancing preservation needs with sustainable urban development. The proposed method gradually improves the prediction accuracy by 93%, environmental risk prediction by 91.4%, false alarm rate by 6%, resource allocation efficiency by 91.2%, inference speed by 23.8 FPS, and sustainability impact score by 86.5%.
Fanwei Meng (Tue,) studied this question.