Fine-Grained Urban Greenery Assessment for Sustainable Urban Planning Using Semantic Segmentation: Tree–Grass–Plant Recognition and Green View Index Estimation

Urban greenery is an important component of sustainable urban environments, yet conventional Green View Index (GVI) assessment aggregates visible vegetation and does not distinguish greenery structures. This study develops a fine-grained assessment framework integrating semantic segmentation with structure-specific and total GVI estimation. Vegetation-related annotations in ADE20K were remapped into background, tree, grass, and plant categories, and SegNet, Pyramid Scene Parsing Network (PSPNet), DeepLabV3+, and SegFormer-B0 were evaluated under a unified protocol. In the representative run with seed 42, DeepLabV3+ produced the numerically highest mIoU (66.564%) and Green mIoU (57.938%). Across three random seeds, SegFormer-B0 achieved the highest mean mIoU (67.120±0.721%) and Green mIoU (58.583±0.967%), demonstrating generally consistent segmentation performance across repeated runs. Tree was recognized more accurately than grass, while plant showed the lowest IoU. Under the primary vegetation-only definition, direct evaluation on Cityscapes without fine-tuning showed that SegNet achieved a Green IoU of 73.826% and a GVI R2 of 0.897. A sensitivity analysis additionally including the Cityscapes terrain class retained SegNet as the best-performing model, with a Green IoU of 75.267% and a GVI R2 of 0.920. A supplementary application-oriented case study was also conducted using 2000 street-view images from 500 four-direction sampling points in Hangzhou. The full Hangzhou dataset was used to examine agreement with automated reference estimates, while an additional subset of 80 manually annotated images provided independent pixel-level validation of green/non-green segmentation. These results provide a structure-aware measure of visible greenery that can complement total GVI in monitoring street-level greenery and identifying locations for more detailed assessment in sustainable urban planning.