Embedding architectural prototype knowledge in interpretable deep learning for morphological similarity evaluation of traditional villages

Prototypes are inherited spatial rules that allow village morphology to persist through migration while adapting to local environments. Conventional studies of village morphology remain limited in large-scale comparison. Although remote-sensing-based deep learning enables large-scale quantification of morphological features, inherited features and adaptive changes remain difficult to distinguish without an explicit prototype-based reference. This study addresses how prototype knowledge from architectural typology can be embedded into deep learning to quantify morphological similarity in traditional villages at scale. We propose a prototype-driven interpretable deep learning framework, situated within the broader field of explainable AI, that compares real village samples with ideal prototypes. Evidence from Guangdong shows that villages with high prototype similarity largely coincide with known cultural core areas and extend outward in patterns partly corresponding to reported migration directions. This framework provides a scalable and interpretable method for studying traditional village morphology, with potential applications in migration history and heritage screening.