基于空间关联记忆的自然资源要素识别方法
By constructing a multimodal dataset, using an autoencoder and contrastive learning to extract embedding vectors, generating a spatial correlation graph, and combining it with a graph neural network for natural resource element identification, the problem of inaccurate identification in existing technologies is solved, achieving high-precision and stable identification results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JIANGXI PROVINCIAL LAND & SPACE SURVEY & PLANNING RES INST
- Filing Date
- 2025-09-09
- Publication Date
- 2026-07-17
AI Technical Summary
Existing methods for identifying natural resource elements are unable to fully explore the spatial relationships and semantic information in multi-source data. The identification results are easily affected by factors such as occlusion, terrain undulation, and overlapping ground features. They lack robustness and generalization ability and cannot achieve accurate identification in complex geographical scenarios.
By collecting multi-source natural resource data, a multimodal raw dataset is constructed. Image feature point registration and point cloud registration are performed to establish high-quality data blocks in a unified format. Embedded vectors are extracted using autoencoders and contrastive learning, and prototype vectors are obtained through clustering. A memory bank is constructed. A spatial correlation graph is generated and combined with graph neural networks for recognition and refined segmentation.
It achieves high-precision identification of natural resource elements in complex terrain and multi-type coverage areas, improves the accuracy and context adaptability of identification, and outputs structured expression results, which are convenient for practical applications.
Smart Images
Figure CN121095778B_ABST