A Lightweight State Space Model With Multiscale Morphology and Low‐Rank Head for Hyperspectral Image Classification
Annals of the New York Academy of Sciences · 2026
DOI: 10.1111/nyas.70271 ↗Kurz gefasst
Fachartikel in der Zeitschrift Annals of the New York Academy of Sciences, 2026. Verfasst von Shanglei Chai und 5 weiteren Autor:innen. Beteiligte Einrichtung: College of Mechatronics and Control Engineering & State Key Lab of Radio Frequency Heterogeneous Integration, Shenzhen University, Shenzhen, China. Hauptthemen laut MEDLINE: Hyperspectral Imaging, Image Processing, Computer-Assisted. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 55 Literaturangaben. Gefördert von: Guangdong Basic and Applied Basic Research Major Project, Shenzhen Talent Startup Funds, Shenzhen University.
Abstract (Originalsprache)
State space models (SSMs) have advanced hyperspectral image (HSI) classification, yet existing approaches have limitations. They typically rely on single‐scale feature extractors, limiting their ability to model various spatial geometries, and their standard backbones can exhibit training instability when processing complex HSI data. To overcome these challenges, this paper proposes a novel lightweight multiscale morphology‐enhanced low‐rank head residual state space network (MMLH‐RSSN), built on a synergistic framework developed for robust feature representation and efficient modeling. Specifically, we first designed a multiscale morphological module to explicitly capture hierarchical spatial features, a crucial step for distinguishing spectrally similar classes with varying scales. To effectively encode these complex features, we then introduced an enhanced Residual SSM, which integrates residual connections and layer normalization to significantly improve model stability and learning capacity. An end‐to‐end lightweight design was ensured by a parameter‐efficient low‐rank decomposition head. Extensive experiments on four benchmark datasets show that MMLH‐RSSN achieves state‐of‐the‐art performance, with overall accuracies of 98.51% and 99.69% on the Pavia University and Botswana datasets, using only 0.063 M parameters. This work demonstrates that a synergistic combination of multiscale priors and a stabilized SSM backbone offers a highly accurate and efficient solution for HSI classification, particularly for resource‐constrained scenarios.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Autor:innen und Institutionen
- College of Mechatronics and Control Engineering & State Key Lab of Radio Frequency Heterogeneous Integration, Shenzhen University, Shenzhen, China.
- School of Computing and Information Systems, Singapore Management University, Singapore, Singapore.
- School of Computer and Information Science, Chongqing Normal University, Chongqing, China.
ORCID-Profile: Shanglei Chai, Zhenpeng Zhang, Zhiyuan Zhang, Zhi Zeng, Dajiang Lu, Yibin Tian
Förderung
- Guangdong Basic and Applied Basic Research Major Project
- Shenzhen Talent Startup Funds
- Shenzhen University
- Shenzhen Science and Technology Innovation Program
Zitieren
APA
Chai, S., Zhang, Z., Zhang, Z., Zeng, Z., Lu, D., & Tian, Y. (2026). A Lightweight State Space Model With Multiscale Morphology and Low‐Rank Head for Hyperspectral Image Classification. Annals of the New York Academy of Sciences. https://doi.org/10.1111/nyas.70271
Vancouver
Chai S, Zhang Z, Zhang Z, Zeng Z, Lu D, Tian Y. A Lightweight State Space Model With Multiscale Morphology and Low‐Rank Head for Hyperspectral Image Classification. Annals of the New York Academy of Sciences. 2026. doi:10.1111/nyas.70271
BibTeX
@article{Chai2026,
author = {Chai, Shanglei and Zhang, Zhenpeng and Zhang, Zhiyuan and Zeng, Zhi and Lu, Dajiang and Tian, Yibin},
title = {A Lightweight State Space Model With Multiscale Morphology and Low‐Rank Head for Hyperspectral Image Classification},
journal = {Annals of the New York Academy of Sciences},
year = {2026},
doi = {10.1111/nyas.70271}
}
EUVIMED