Answers are grounded in this paper and cite their sources.
Hyperspectral image (HSI) and LiDAR data provide complementary spectral, spatial, and structural information for land-cover classification, yet their effective fusion remains challenging due to strong modal heterogeneity and insufficient semantic guidance in existing methods. To address this issue, we propose a Semantic Query-Guided Network (SQGNet) for joint HSI-LiDAR classification. Specifically, class-level textual prompts are encoded to build semantic prototypes and transformed into class-specific query banks, which guide the retrieval of category-relevant evidence from hierarchical HSI and LiDAR features through cross-modal attention. A recurrent query refinement strategy is further introduced to progressively enhance semantic consistency between class queries and retrieved multimodal evidence. Finally, classification is performed by semantic matching rather than conventional feature-level prediction, enabling category-aware decision making. Experiments on the MUUFL and Trento datasets demonstrate that SQGNet consistently achieves superior performance over several representative methods. The results verify the effectiveness of introducing language as an explicit semantic prior for multimodal fusion and highlight the potential of semantic query-guided learning for HSI-LiDAR classification.
No takes yet. Share an insight, caveat, or question.
Wu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: