A Knowledge-Guided Graph Reasoning Framework for Multi-Modal Detection of Mining-Induced Land Degradation
Fuwen Hu, Xuefei Wu
IEEE Access
Problems Identified (4)
Localized Geohazard Assessment: Automated assessment of localized geohazards such as mining-induced land degradation remains challenging.
Relational Context Modeling Limitation: Existing remote sensing paradigms fail to capture spatial, semantic, and functional relations among land parcels needed for interpretation.
Localized Geohazard Assessment: Automated assessment of localized geohazards such as mining-induced land degradation remains challenging.
Relational Context Modeling Limitation: Existing remote sensing paradigms fail to capture spatial, semantic, and functional relations among land parcels needed for interpretation.
Proposed Solutions (5)
Venagnosis Graph Reasoning Framework: The paper proposes Venagnosis, a knowledge-guided graph reasoning framework for context-aware relational reasoning over multi-modal geospatial data.
Synergistic Parcel Characterization: An SPC module uses deep learning models to produce dense embeddings from DEM, multispectral, and morphological inputs.
Knowledge-Guided Multi-Relational Graph: The framework constructs a domain-knowledge-guided multi-relational graph using spatial, semantic, and functional corridor edge schemas.
Progressive Fusion Graph Network: PF-GNN is a graph attention architecture with adaptive input enhancement, hierarchical dynamic gating, and stateful cross-layer fusion for reasoning.
Venagnosis Graph Reasoning Framework: The paper proposes Venagnosis, a knowledge-guided graph reasoning framework for context-aware relational reasoning over multi-modal geospatial data.
Results (3)
State-of-the-Art Detection Performance:
High F1 And AUPRC:
Ablation Component Validation:
Research Domain
Remote sensing and geospatial environmental surveillance for mining-induced land degradation detection