Green quality perception of urban streets: a study combining semantic segmentation and multimodal noise enhancement techniques

In recent years, the quantitative analysis of the green quality of urban streets has become a hot spot in the cross-research of computer vision and urban environment. In this paper, we propose a semantic segmentation method based on a dynamic multimodal noise addition technique combined with the SegFormer model to improve the robustness of street view image segmentation in complex environments. The study takes five major roads (West Beijing Road, North Taiping Road, Changjiang Road, Yihe Road, and Central Road) in Nanjing as objects, and based on the streetscape and remote sensing data of the vegetation growth season from 2017 to 2022, the Gaussian noise, motion blur, and weather effects are simulated by the multimodal noise addition technique, which effectively improves the adaptability of the model to the dynamic interference. The experiments showed that the Green View Index analysis revealed a general upward trend of street greening (average increase of 0.94%). Furthermore, the Kendall rank correlation analysis reveals the positive correlation between environmental factors and green visualization rate, and the influence law of environmental factors on green perception is verified. This study provides a high-precision computer vision analysis framework for urban greening planning and promotes the application of semantic segmentation models in complex urban scenes.