Abstract: Under the background of stock renewal, urban village renovation faces challenges such as neglecting the spatial needs of vulnerable groups and difficulties in quantifying urban perception. To reveal the perception patterns of left-behind elderly in renovation-oriented urban villages and optimize age-friendly spaces, this paper takes urban villages in Dong'an Subdistrict, Changsha, as the empirical area and constructs a deep learningbased perception prediction method and spatial optimization strategy. Specifically, the DeepLabV3+ model is employed to extract semantic features from street view images; the TrueSkill ranking algorithm is utilized to quantify six-dimensional perceptions and train an urban perception prediction model for large-scale prediction; and the CRITIC weighting method is combined to establish a spatial environment evaluation index system for comprehensive analysis. The results indicate that the comprehensive perception of the study area exhibits spatial differentiation, characterized by 'optimal performance in the central core area and relatively weak performance in the southern and riverside marginal areas'. High values of positive perception are concentrated in the northern and central living areas, whereas high values of negative perception are distributed in the southern, eastern, and riverside marginal zones. Furthermore, the sense of depression, sky ratio, and green view index are identified as key factors influencing perception levels. Accordingly, three age-friendly spatial optimization strategies are proposed: mitigating negative perception, consolidating positive perception, and promoting holistic regional enhancement. These strategies provide an empirical basis for inclusive urban village renewal that prioritizes bottom-line fairness.