EUVIMEDEUVIMED entdecken
Crossref

A UNIFIED FRAMEWORK FOR SPECTRAL CONTEXT-AWARE META-LEARNING IN HYPERSPECTRAL IMAGE CLASSIFICATION

Shahnas P, S.Malathy

Boletim Paranaense de Geociências · 27.02.2026 · 84

DOI: 10.5380/bpg.v84i1.103552 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Boletim Paranaense de Geociências, online erschienen am 27.02.2026. Verfasst von Shahnas P und 1 weiteren Autor:innen. Die Arbeit nennt 37 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Hyperspectral imaging (HSI) enables pixel-level land-cover classification by exploiting rich spectral information across hundreds of narrow bands, making it highly valuable for remote sensing applications. Despite recent advances, HSI classification remains challenged by high spectral dimensionality, spectral redundancy and noise, patch-based sampling that induces semantic inconsistency near class boundaries, and poor generalization caused by limited labeled data particularly in heterogeneous and mixed-class regions. To address these limitations, this study proposes a Context-Aware HyperNet (CAHN), a unified and modular spectral–spatial learning framework designed to enhance robustness and generalization under limited supervision. CAHN integrates adaptive spectral refinement, spatial representation learning, multi-scale attention mechanisms, and meta-learning–based classification within a context-aware pipeline. The spectral refinement module effectively suppresses noise and redundant spectral information while preserving class-discriminative features. Spatial representations are adaptively constructed to capture homogeneous land- cover patterns and reduce semantic ambiguity near boundaries. Furthermore, the attention- based multi-scale learning strategy models complex, non-local spectral–spatial dependencies, while the meta-learning classifier enables rapid adaptation to novel or under-represented classes, improving performance in low-label and imbalanced scenarios. Extensive experiments conducted on the Pavia University and Kennedy Space Center benchmark datasets demonstrate that CAHN consistently outperforms state-of-the-art methods. The proposed framework achieves superior Overall Accuracy, Average Accuracy, and Kappa coefficients (approximately 99.9%), while maintaining robustness under class imbalance and sparse training conditions. Overall, CAHN provides an effective and generalizable solution to spectral redundancy, semantic inconsistency, and limited supervision, offering reliable high- accuracy HSI classification for real-world remote sensing applications.

Zitieren

APA

P, S., & S.Malathy (2026). A UNIFIED FRAMEWORK FOR SPECTRAL CONTEXT-AWARE META-LEARNING IN HYPERSPECTRAL IMAGE CLASSIFICATION. Boletim Paranaense de Geociências, 84. https://doi.org/10.5380/bpg.v84i1.103552

Vancouver

P S, S.Malathy. A UNIFIED FRAMEWORK FOR SPECTRAL CONTEXT-AWARE META-LEARNING IN HYPERSPECTRAL IMAGE CLASSIFICATION. Boletim Paranaense de Geociências. 2026;84. doi:10.5380/bpg.v84i1.103552

BibTeX
@article{P2026,
  author = {P, Shahnas and S.Malathy},
  title = {A UNIFIED FRAMEWORK FOR SPECTRAL CONTEXT-AWARE META-LEARNING IN HYPERSPECTRAL IMAGE CLASSIFICATION},
  journal = {Boletim Paranaense de Geociências},
  year = {2026},
  volume = {84},
  doi = {10.5380/bpg.v84i1.103552}
}
Kostenlos starten →