Research on the Spatial Configuration and Layout of Chinese Hakka Architecture Based on the CNN-LSTM-Attention Method

This study pioneers the application of a deep learning-based CNN-LSTM-ATTENTION hybrid model for the quantitative analysis of the spatial layout of Chinese Hakka architecture and its relationship with traditional cultural philosophies. Focusing on 13 representative Hakka buildings (including dwellings and public buildings) and one ancient city in the Gannan region, their floor plans were used as research objects. Using the Later Heaven Eight Trigrams diagram as the spatial reference, each plan was divided into nine directional zones. The constructed enhanced CNN-LSTM-ATTENTION model extracted and fused features from three dimensions: local image features (CNN), spatial layout structure, and building density distribution, to quantitatively assess the spatial matching degree between the architectural layouts and the traditional Chinese Five Elements and Eight Trigrams philosophy. Experimental results show that the model achieved a 100% directional matching rate across all 126 data points, with an average comprehensive similarity score as high as 0.9904, demonstrating excellent accuracy, stability, and robustness. The analysis reveals that Hakka architectural layouts strictly adhere to the correspondence between trigram directions and functional zones, evolving from a pattern of "dense core, sparse periphery" in dwellings to one of "globally high-density equilibrium" in significant public buildings, with the "Wanshou Palace" achieving the highest congruence (similarity 0.9927). This research validates the effectiveness of the CNN-LSTM-ATTENTION model in the digital analysis of traditional architectural cultural heritage, providing quantitative evidence for understanding the construction wisdom of Hakka architecture characterized by "invariant core layout and adaptive functional details," and offering a novel technical approach for the digital preservation and study of traditional architecture.