Building Models to Assess Streetscape Quality Preferences Using Deep Learning Techniques: Comparative Analysis of Different Street Types in Taichung City

Amid escalating urban environmental challenges, digital evaluation of street environments becomes vital. Deep Learning (DL) and related technologies have been increasingly applied to analyze streetscapes. Existing research predominantly focuses on urban street elements, neglecting comparative analyses across diverse contexts like urban, suburban, and distinctive street types. This research, conducted in Taichung City, Taiwan, employs DL and other methods for a comprehensive assessment of urban street elements, including building facades, dimensions, furniture, greening, and openness, creating a thorough streetscape quality evaluation framework. Utilizing the Delphi method and Analytic Hierarchy Process (AHP), 5 main categories and 16 factors were identified. Street greening emerged as the most critical, accounting for 33.39%, while street proportions were the least at 12%. Visible sky extent and building color harmony were prioritized. The study applied DL to assess landscape quality preferences near the National Taichung Theater, Shuren Business District, and Audit New Village, revealing the highest landscape quality score around the National Taichung Theater (82.7%), the lowest in Shuren Business District (59.6%), and a mid-level at Audit New Village (63.2%). Pearson correlation analysis confirmed a positive correlation between questionnaire and DL-evaluated scores. Multivariable linear regression showed these five categories significantly impact streetscape quality preferences, providing a scientific basis for urban development and enhancing resident satisfaction.