Research on Urban Landscape Image Processing and Feature Extraction Method Based on Computer Vision

With the growing demand for smart city development, accurate interpretation of urban landscape imagery is essential for spatial analysis. This paper presents a computer vision-based approach for urban landscape image processing, supported by IoT-enabled data collection. A lightweight segmentation framework is proposed, combining image preprocessing, backbone perception, multi-scale context modeling, and local semantic enhancement. Key components include ASPP for multi-scale feature extraction and a localized Transformer for improved edge and small-object detection. Experiments on Cityscapes and a custom UrbanScene-BJ dataset—captured via urban IoT cameras—show that the method achieves higher mIoU and F1 scores compared to existing baselines. The results highlight its effectiveness for real-world urban analysis and its feasibility for practical deployment.