Quantifying homeliness: Bridging the semantic gap in China's elderly care facility design

In response to global aging, creating “homeliness”-oriented elderly care facilities has become a core goal to improve the elderly’s quality of life and emotional attachment to the environment. However, “homeliness” remains a subjective concept, lacking objective metrics to guide spatial design. This study introduces a novel methodology combining OneFormer-based image semantic segmentation and machine learning, to quantify the perceptual quality of spatial environments, and translate the findings into actionable design strategies for elderly care facilities. Through field surveys and interviews across 24 home care and 20 institutional care facilities in Xi'an, combined with a systematic literature review, we first established a quantifiable spatial assessment system based on elderly living patterns and emotional needs. Using OneFormer for spatial image semantic segmentation, we extracted and analyzed environmental indicators from spatial imagery, identifying key qualitative and quantitative differences between the home care and institutional care settings. Multiple machine learning models, interpreted via SHAP values, were then applied to elucidate the relationships between spatial elements and perceived “homeliness.” Based on “homeliness” characteristics, the study proposes spatial reconstruction strategies (unit types, layout) and targeted design advice for residential spaces, communal activity spaces and corridors, to build elderly care environments that enhance the well-being of the elderly.