AI 이미지 분석모델을 활용한 국가정원 경관 특성 비교 연구† - 순천만, 태화강국가정원 소셜미디어 이미지를 중심으로 -

초록

National gardens function not merely as green spaces but as landscape assets representing their regions, yet the landscapes that visitors perceive have been understood, evaluated, and managed only to a limited extent. This study combined BLIP-2 image captioning with CLIP image embedding to analyze differences in the landscape characteristics of Suncheonman and Taehwagang National Gardens. Using 1,552 Suncheonman and 1,015 Taehwagang images from Naver Map reviews, landscape keywords were derived with BLIP-2, and visual characteristics were examined with CLIP embeddings and UMAP. Per 1,000 images, garden appeared 108.2 times in Suncheonman, 4.4 times the 24.6 in Taehwagang, whereas Taehwagang was concentrated on specific natural resources, with bamboo (216.7) and forest (154.7) predominating. Visual diversity was similar (0.2120 and 0.2119), and the number of clusters remained comparable after sample balancing. However, the landscape types forming each garden’s dominant clusters contrasted: planned elements such as themed gardens and greenhouses in Suncheonman, and the bamboo forest, seasonal flowers, and urban scenery in Taehwagang. This indicates that the two gardens share common elements of park landscapes but differ in the landscape elements that compose each garden. AI-based image analysis combining textual and visual information provides baseline information for comparing the landscape characteristics of national gardens and reviewing management directions.

키워드