2026· IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing· Vol 19, pp. 34423-34440· 0 citations· 51 references
TL;DR
A dynamic gating cross-attention network with frequency-domain spectral for HSI and LiDAR classification and a structure-preserving alignment strategy that facilitates bidirectional information exchange while preserving the intrinsic semantic organization of each modality are proposed.
Abstract
Joint exploitationof hyperspectral image (HSI) and light detection and ranging (LiDAR) data has emerged as a promising approach for land-cover classification. However, existing methods are often limited by the inadequate exploitation of discriminative frequency cues and the presence of unreliable cross-modal interactions. To address these issues, we propose a dynamic gating cross-attention network (DGCANet) with frequency-domain spectral for HSI and LiDAR classification. The architecture adopts a dual-stream parallel framework with frequency-domain enhancement and dynamic-gating interaction. Specifically, a symmetric multiscale attention inception (MSAI) extractor is designed to capture fine-grained spectral–spatial attributes, providing a stable foundation for semantic alignment. Moreover, the FFT-based multiscale feature extractor (FFMS) integrates a frequency modulation branch with a spatial multiscale branch, utilizing a learnable frequency-domain filter to capture global structural features and thereby achieving a unified full-frequency representation of local details and global semantics. To achieve robust cross-modal complementarity, a dynamic gating cross-attention (DGCA) module is designed. It employs a structure-preserving alignment strategy that facilitates bidirectional information exchange while preserving the intrinsic semantic organization of each modality. In particular, it introduces adaptive context-aware gating kernels to dynamically evaluate the credibility of cross-modal features, effectively filtering unreliable responses and suppressing noise propagation. This mechanism significantly enhances feature discriminability and accurately characterizes complex land-cover distributions. Overall, MSAI and FFMS first extract complementary features, which are transformed into semantic tokens and subsequently processed by DGCA for adaptive cross-modal interaction before final classification. Extensive experiments on four public benchmark datasets demonstrate that DGCANet achieves significant improvements over state-of-the-art methods.
Hyperspectral images (HSIs), with their hundreds of contiguous narrow bands, capture detailed physicochemical properties of land covers; LiDAR provides high-precision 3-D spatial and geometric structure, and the two modalities are highly complementary. However, differences in imaging mechanisms lead to insufficient cro...
Shu-Fang Xu, Wei-Wen Xu, Qi-Yuan Xue et al.· IEEE Transactions on Geoscie...· 0 citations
The proposed TriFusion framework consistently achieves the best overall accuracy among recent CNN-based, Transformer-based, and hybrid methods, with gains of 0.26–1.42 percentage points over the strongest competing baseline across the four datasets.
Fei-Long Fan, An-Ming Dong, En Li et al.· IEEE Journal of Selected Top...· 0 citations
The integration of multisource information from hyperspectral imagery (HSI) and light detection and ranging (LiDAR) provides reliable data support for land cover classification. The heterogeneity between HSI and LiDAR data poses significant challenges for cross-modal feature alignment, while the rich structural priors...
Yu-Qing Zhao, Yu Song, Chao Liang et al.· IEEE Journal of Selected Top...· 0 citations
A local–global cross-modal fusion network (LGCF-Net), a hybrid fusion framework that explicitly models local–global feature interactions for HSI–LiDAR classification and achieves the best overall accuracy on MUUFL, Trento, and Augsburg, while attaining the highest average accuracy and Kappa on Houston.
M. Z. ur Rehman, S. Islam, Anwaar Ulhaq et al.· IEEE Transactions on Geoscie...· 0 citations
Semantic segmentation of high-resolution remote sensing images faces challenges due to intraclass heterogeneity and interclass spectral similarity. Traditional single-domain methods struggle to balance computational efficiency, high-frequency details, and a global receptive field. To address this, we propose the spatia...
Li-Jian Xie, Bo-Hua Zhao, Ai-Hua Shen et al.· IEEE Transactions on Geoscie...· 1 citation
Results confirm that BMCA effectively preserves fine-grained spatial details, promotes semantic consistency across modalities, and provides a robust and generalizable solution for feature-level SAR–MS fusion in remote-sensing semantic segmentation.
Hao-Tian Xu, Xi-Min Yuan, Xiu-Jie Wang et al.· Remote Sensing· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.