基于街景与卫星图像的存量空间识别方法、装置及介质
By using a heterogeneous dual-branch multimodal deep neural network model and Grad-CAM heatmap, the problems of data timeliness, multimodal fusion, and model black boxing in urban stock space identification are solved, achieving accurate and interpretable stock space identification, which is applicable to urban planning.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHONGQING UNIV
- Filing Date
- 2026-05-20
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies for identifying existing urban spaces suffer from problems such as poor data timeliness, simplistic multimodal fusion, black-box modeling, coarse recognition edges, and data imbalance, resulting in low recognition accuracy and difficulty in guiding urban planning.
A heterogeneous dual-branch multimodal deep neural network model is adopted, which is combined with Grad-CAM heatmap for feature fusion and model interpretability analysis. High-resolution satellite imagery and panoramic street view images are used for spatiotemporal verification and feature extraction. Focal Loss and contrast loss are used for iterative training, and CRF edge-preserving algorithm and spatial interpolation technology are combined for post-processing to achieve accurate identification of existing space.
It enables rapid, accurate, and interpretable identification of existing spaces, improves identification accuracy and reliability, reduces manual annotation costs, is applicable to large-scale urban planning, and provides highly interpretable and efficient data support.
Smart Images
Figure CN122223582B_ABST