AI-Powered heritage conservation: comparative studies of techniques for structural decay detection in historic Asian architecture

The preservation of historic Asian architecture holds immense cultural and societal significance, as these structures embody centuries of heritage, craftsmanship, and identity. However, structural decay caused by environmental factors, material aging, and anthropogenic influences poses a critical threat to their longevity. Existing decay detection approaches, such as conventional image analysis or handcrafted feature-based models, often suffer from low generalizability, limited segmentation precision, and inadequate robustness in handling complex multimodal heritage datasets. To overcome these limitations, this study introduces a novel artificial intelligence-powered framework for heritage conservation that integrates advanced deep learning techniques for accurate structural decay detection and risk assessment. The proposed framework introduces two major contributions. First, Segmentation via DecayoNet leverages a combination of Modified SegNet, Point-Wise Convolutions, and a Cross-Task Attention Mechanism to achieve fine-grained and context-aware boundary delineation of decayed regions. Second, the Decay Detection Engine via DeFusion-Net, a Decay-Aware Fusion Network, incorporates multi-scale feature extraction through a Hierarchical Vision Transformer and Graph Convolutional Networks, followed by an attention-based fusion and a multi-head classification module. Proposed framework which is reliable and scalable solution for safeguarding historic Asian architecture, offering a pathway toward proactive conservation and sustainable cultural heritage management.