Remote sensing road extraction method based on dynamic graph topology reasoning and frequency domain structure constraint
By employing dynamic graph topological reasoning and frequency domain structural constraints, the challenges of topological modeling and intersection identification in remote sensing road extraction are solved, thereby improving the accuracy of road extraction in complex environments and making it suitable for traffic planning and map updates.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- LANZHOU UNIV
- Filing Date
- 2026-04-29
- Publication Date
- 2026-07-17
AI Technical Summary
Existing remote sensing road extraction methods suffer from problems such as poor topology modeling and connectivity in complex environments, as well as low intersection recognition accuracy. In particular, they are difficult to accurately identify road structures when obscured by vegetation or shadows.
We adopt a method based on dynamic graph topology reasoning and frequency domain structural constraints. We extract multi-level semantic feature maps through deep residual networks, construct sparse dynamic topology graphs, and use graph convolutional networks for iterative reasoning. Combined with frequency domain topology constraints and intersection geometry self-supervision enhancement, we construct a topological road network for multi-task training.
It effectively improves the topological integrity and structural accuracy of road extraction in complex environments, and can simultaneously output road segmentation results, road skeleton structure and road width estimation, making it suitable for traffic planning, disaster emergency response and map updating.
Smart Images
Figure CN122416263A_ABST