A multi-scale semantic segmentation approach for architectural pattern recognition in traditional cantonese ancestral halls

Traditional Cantonese ancestral halls are essential components of Chinese cultural heritage, functioning as symbolic systems that embody rich cultural values. Drawing from architectural semiotics, we propose the DualPath-WavNet_Canton model to decode these symbolic features through multi-scale semantic segmentation. This model integrates a dual-path structure, a wavelet transform convolution mechanism, and a multi-scale edge enhancement module to efficiently capture decorative details. Our research suggests that the proposed model shows significant advantages over existing technologies across most evaluation metrics. We also established a specialized dataset that, for the first time, connects visual architectural elements with their semantic and cultural interpretations across 11 key categories. This innovative approach offers new pathways for digital documentation and preservation, supporting both the technical recording of structures and the preservation of their embedded cultural meanings. This has significant implications for future multimodal analysis of this precious architectural heritage.