멀티모달 대규모 언어모델과 기계학습을 활용한 도시 가로 경관 쇠퇴 영향요인 분석

Abstract

본 연구에서는 가로경관 이미지를 기반한 주관적 쇠퇴경관 설문조사 자료를 활용하여 쇠퇴경관에 대한 인간의 인식과이에 영향을 미치는 물리적 환경요인을 분석하고자 한다. 또한, 본 연구를 통해 서울시 쇠퇴경관 저감과 스마트 도시계획 및 관리에 기여할 수 있는 정책적 시사점을 도출하고자 한다. 이를 위해본 연구의 흐름은 <그림 1>과 같이 진행되며, 세 가지의 분석 내용을 설정하였다. 먼저, 쇠퇴경관에 대한 인간의 주관적 인식을 정량적으로 추출하기 위해 가로경관 이미지를 활용하여 이미지 쌍별 비교 설문조사를 진행하고 Trueskill 알고리즘을 적용하였다. 다음으로 의미론적 분할(Semantic Segmentation)과 멀티모달 대규모 언어모델(Multi-Modal Large Language Model, MLLM)을통해 물리적 환경요인의 비율(Quantity)과 상태(Quality) 점수를도출하였다. 마지막으로 기계학습과 해석가능한 기계학습을 사용하여 쇠퇴경관에 대한 인간의 주관적 인식과 물리적 환경요인 간의 비선형적 관계를 분석하여 정책적 시사점을 도출하였다.

Cities are constantly evolving, with growth, vitality, decline, and shrinkage occurring as interrelated phenomena. While urban vitality and revitalization have been extensively studied, research on urban decline and shrinkage remains comparatively limited. Existing studies on urban decline have primarily focused on diagnostic indicators, such as physical aging, population decline, and reduction in the number of businesses, to assess patterns of decline. However, studies on the subjective perception of urban decline, particularly in relation to urban landscapes, remain limited. Given this gap, it is crucial to identify the causes of urban decline by analyzing the factors that influence the public perception of declining landscapes through subjective evaluations of urban scenery. This study quantitatively analyzes how people perceive declining urban landscapes and identifies the key factors that influence these perceptions, using street view images of Seoul. A survey was first conducted to assess urban landscape decline based on streetscape images. The Trueskill algorithm was applied to quantify perceived level of decline. Subsequently, machine learning was used to analyze the primary factors influencing these perceptions. The results of the analysis are as follows. First, perception of decline decreased as the proportion of physical environmental elements such as roads, green spaces, sidewalks, and cars increased. In contrast, an increased presence of elements such as buildings, bicycles, walls, and fences was associated with a heightened perception of urban decline. Second, an analysis of the importance of contributing factors indicated that roads, sidewalks, green spaces, and cars were the most influential in shaping perception, in that order. Third, the relationship between the proportion of physical environmental elements in urban landscape images and perceptions of decline was found to be non-linear. This study presents a methodology for evaluating urban landscape decline based on people's subjective perceptions and provides policy implications by identifying the streetscape features that substantially influence perceptions of urban decline.