Street interface color is an important factor affecting spatial quality and crowd experience. Traditional research methods are low cost efficient and difficult to carry out large-scale research, while machine learning technology is highly efficient and adaptable to carry out large-scale research. Firstly, street color aesthetics evaluation indexes are selected through literature review, including 2 indexes of color richness and color harmony. Then Gulou District of Xuzhou City is selected as the study area, utilizing Baidu Street View and machine learning technology to capture 18,000 street view images as samples and evaluating street color aesthetics in terms of color richness and color harmony. Subsequently, 50 streets are randomly sampled, and their Street view images are integrated into 50 videos respectively, rated by 100 volunteers for emotional perception. The Random Forest algorithm is used to perform correlation analysis and model creation. Finally, the training model is applied to predict the emotion perception scores of other street view images and visual analysis in ArcGIS. Results show that the street colors in Gulou District are dominated by yellow tones; the color richness and harmony are higher in the central and western areas, while lower in the north. In addition, when the color richness of the street interface is 80-100, residents feel depressed and irritated; when it is 50-80, residents feel relaxed and happy; and when it is 0-50, residents feel monotonous and bored. The higher the color harmony of the street interface, the more positive the residents' emotional perception. The study will provide a data acquisition and analysis model for urban street color research, which in turn will guide human-centered street renewal design.
 

street interface; color evaluation; street view images; machine learning; urban renewal.