Abstract: From May to September 2022, 165 photos were took along Qingshanhu Lake waterfront greenway, Zhejiang province. Based on the pyramid scene analytic network (PSPNet) model, the image semantic segmentation model was trained. The results showed that the image semantic segmentation model could accurately identify landscape elements of greenway image after training. Combining questionnaire survey and SPSS statistical method, relationship between landscape characteristics of waterfront greenway and aesthetic perception was analyzed. It showed that there was correlation between landscape feature element indicators and aesthetic perception indicators. Among landscape feature element indicators, green vision rate, sky visual field and blue visual field rate had significant relationship with beauty in aesthetic perception indicators (P＜0.05). Green vision rate had extremely significant relation with naturalness (P＜0.01). There was a significant relationship between interference factor index and aesthetic interference (P＜0.05).