基于语义拓扑修正与原型学习的遥感图像场景图生成方法
By employing semantic topology correction and prototype learning methods, the problems of error cascading and long-tailed distribution in remote sensing image scene graph generation are solved. This achieves decoupling of visual perception and semantic reasoning and accurate identification of rare relationships, thereby improving the accuracy and robustness of remote sensing image scene graph generation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANKAI UNIV
- Filing Date
- 2026-06-22
- Publication Date
- 2026-07-17
AI Technical Summary
Existing methods for generating scene graphs from remote sensing images suffer from problems such as cascading error propagation, rigid use of prior knowledge, and chaotic feature space distribution. In particular, during the target detection stage, error propagation caused by noise and long-tailed distribution, as well as the difficulty in identifying rare relationships, are particularly problematic.
We adopt a method based on semantic topology correction and prototype learning. By combining global context and semantic prior through an additive fusion strategy, we actively correct topological connections and feature representations. We also use prototype learning to standardize the feature space distribution, design parallel global context paths and semantic prior paths, and dynamically adjust message passing weights to achieve decoupling between visual perception and semantic reasoning.
It effectively blocks the accumulation and diffusion of detection noise into the relation prediction stage, improves the accuracy of relation reasoning and the ability to identify rare relations, and significantly improves the accuracy and robustness of scene graph generation.
Smart Images
Figure CN122416280A_ABST