Sustainable heritage management requires understanding visitors’ perceptions beyond technological approaches. This study integrates deep learning (DL), data mining, and spatial grid analysis to investigate tourist impressions of Chinese classical gardens, using 4980 images from Dianping.com and 4024 photographs from TripAdvisor. A DL-based semantic segmentation method combined with spatial grid analysis identifies key garden elements and their spatial distributions. Relationships among compositional features (i.e., component combinations, composition layouts, and composition hotspots) are then examined. Perceptual differences between Chinese and Western tourists across northern and southern gardens are also compared. Results show that the presence, spatial coverage, and compositional prominence of garden elements vary by garden type and cultural background, reflecting distinct visitor preferences. Pixel-level analysis of photographic composition provides an objective, fine-grained understanding of landscape preferences, offering practical insights for heritage planning, visitor experience design, preservation policy, and cross-cultural value co-creation.
Chai et al. (Tue,) studied this question.