.png){kind=icon}
.png){kind=icon}
Return
HGCNet: Hierarchical Global–Local Collaborative Network for Hyperspectral Image Classification
.png){kind=icon}
PRE
AI
C
X
H
Q
W
C
Y
DOI：10.1109/tgrs.2026.3717086.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Global contextual modeling and local detail preservation are both essential for hyperspectral image (HSI) classification, as they jointly determine the discriminability of spectral–spatial representations. Effective global modeling captures long-range contextual dependencies and overall semantic structures, whereas careful use of local information is critical for preserving fine-grained details, such as edges, textures, and subtle spatial variations. Although recent global modeling techniques, such as Mamba, can efficiently capture long-range dependencies with linear computational complexity, they often overlook or weaken local structural details, leading to incomplete feature representations. To address this limitation, this article proposes a hierarchical global–local collaborative network (HGCNet), a HGCNet for HSI classification. Specifically, HGCNet is built on a hierarchical spectral–spatial collaborative framework, in which spatial and spectral representations are progressively learned and jointly optimized within a unified architecture. To enhance spatial feature learning, a hybrid spatial recalibration module (HSRM) is designed to jointly capture global contextual dependencies and local structural details. Meanwhile, a dual-path spectral recalibration module (DSRM) is developed to collaboratively model global spectral correlations and local band-wise variations, thereby improving spectral discriminability. Furthermore, a bidirectional spectral–spatial interaction module (BSSIM) is introduced to promote interaction between spatial and spectral features, thereby strengthening their complementary representation capability. By jointly exploiting global and local information and enhancing spectral–spatial interaction, HGCNet learns more discriminative and robust feature representations for HSI classification. Extensive experiments on four benchmark datasets demonstrate that the proposed method achieves superior classification performance over existing advanced methods. The code will be available online at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/YichuXu/HGCNet</uri>
Keywords:
Bidirectional interaction fusion
global–local collaborative modeling
hyperspectral image (HSI) classification
Mamba
spectral–spatial learning
Journal
IF:
8.6
Papers:
2.1W
Citations:
10.7W
Organization
Cited Papers