Yilin Duan

Jointly modeling multimodal LiDAR and hyper spectral image (HSI) is crucial for improving classification performance. However, existing studies still face challenges in simultaneously capturing the global spectral–structural dependencies of HSI, the fine-grained geometric variations of LiDAR, and achieving reliable cross-modal alignment. To address these issues, we propose a novel multimodal framework– Global-Local Aligned Cross-modal Network (GLAC-Net), which comprises a Global Multi-scale Graph Attention (GMGA) module, a Local Graph Attention (LGA) module, and a Late-stage Patching based Global Context Preserving (LPGCP) module. Specifically, GMGA constructs a fine-grained enhanced graph to capture long-range spectral–spatial relationships and produce globally multi-scale consistent region-level representations, while LGA models each patch as a local graph and employs attention-augmented graph convolution to extract local geometric features. The global, local, and interaction features are jointly optimized within LPGCP to yield the final classification; LPGCP module explicitly preserves long-range inter-patch contextual relationships to effectively align complementary global–local information. Extensive experiments on three benchmark datasets demonstrate that GLAC-Net consistently outperforms the recent state-of-the-art competitors in terms of OA, with the most significant gain reaching 2.13%, while also yielding smoother and more semantically coherent classification maps. These results verify that GLAC-Net effectively integrates global multi-scale contextual information into local patch-level classification, while preserving long-range spatial consistency across the scene.

State-space models (SSMs), particularly the Mamba model, have recently garnered significant attention in hyperspectral image (HSI) classification tasks due to their capability to model global dependencies with linear computational complexity. However, existing SSM-based approaches predominantly adopt patch-based strategies with single-scale information, which inherently limits their ability to effectively characterize boundary structures and shapes in complex land cover regions. We propose a novel multiscale superpixel Mamba (MSMamba) framework for HSI classification that captures long-range dependencies using a two-stage multiscale Mamba structure. In the first stage, a multiscale superpixel generation and fusion strategy (MSSGF) and a multiscale Mamba module (MMamba) are introduced. Specifically, multiscale superpixels are fused and unfolded into bidirectional sequences. The MMamba module models global dependencies through cross-scale sequence unfolding. In the second stage, we propose a multi-shape fusion mechanism to integrate the global features of MMamba under varying shape configurations, aiming to reinforce boundary-sensitive representation learning. Additionally, our MSMamba introduces a dual-branch architecture that synergistically integrates multiscale Mamba with a multiscale spectral-spatial convolutional network (SSCNN) to enhance fine-grained local spectral-spatial details. Extensive experiments conducted on four publicly available HSI datasets demonstrate that our MSMamba method significantly outperforms related methods in classification performance, particularly excelling at identifying complex terrain boundaries.

The Geographic Knowledge Graph (GeoKG) serves as an effective method for organizing geographic knowledge, playing a crucial role in facilitating semantic interoperability across heterogeneous data sources. However, existing GeoKGs are limited by a lack of hierarchical modeling and insufficient coverage of geographic knowledge (e.g., limited entity types, inadequate attributes, and insufficient spatial relationships), which hinders their effective use and representation of semantic content. This paper presents HGeoKG, a hierarchical geographic knowledge graph that comprehensively models hierarchical structures, attributes, and spatial relationships of multi-type geographic entities. Based on the concept and construction methods of HGeoKG, this paper developed a dataset named HGeoKG-MHT-670K. Statistical analysis reveals significant regional heterogeneity and long-tail distribution patterns in HGeoKG-MHT-670K. Furthermore, extensive geographic knowledge reasoning experiments on HGeoKG-MHT-670K show that most knowledge graph embedding (KGE) models fail to achieve satisfactory performance. This suggests the need to accommodate spatial heterogeneity across different regions and improve the embedding quality of long-tail geographic entities. HGeoKG serves as both a reference for GeoKG construction and a benchmark for geographic knowledge reasoning, driving the development of geographical artificial intelligence (GeoAI).