Centrality-Aware Coordinate Attention: A Lightweight Spectral-Spatial Network for Hyperspectral Image Classification

Read full text Summarise with VIRAT
Authors: Yong Shi, Hao F. Zhang
Cited by: 0
References: 36

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Hyperspectral Image (HSI) classification is a pivotal task in remote sensing.Patch-based methods aim to integrate spatial and spectral information for accurate classification of the central pixel.However, the inherent high dimensionality of HSI data and complex spectralspatial correlations often lead to substantial parameter counts and high computational costs in existing models.To address these challenges, this paper proposes a novel Lightweight Central pixel-guided Coordinate Attention Network (Lite-C2ANet).The framework introduces three key innovations: First, a Spectral Group Reduction and Interaction (SGRI) module is designed to group adjacent bands and facilitate inter-group information exchange.This effectively reduces spectral redundancy and enables efficient encoding of highdimensional spectral data.Second, a Central pixel-guided Coordinate Attention (CPCA) module is developed to capture long-range spatial dependencies.It separately aggregates features along horizontal and vertical directions using lightweight shared convolutions, explicitly incorporating spatial coordinate information.Finally, within the CPCA, an Adaptive Central Similarity Guidance (ACSG) mechanism is proposed with negligible parameter overhead.This mechanism adaptively fuses spatial similarity metrics centered on the pivotal central pixel of the patch, guiding the model to focus on the most classificationrelevant contextual regions and enhancing the spatial discriminability of class-related attention.Extensive experiments on three public HSI benchmark datasets demonstrate that Lite-C2ANet achieves highly competitive classification accuracy.Crucially, these results are attained with significantly fewer parameters and lower computational overhead compared to state-of-the-art methods, underscoring the effectiveness and efficiency of the proposed lightweight design.

Fields of Study & Subjects

Remote-Sensing Image Classification Advanced Neural Network Applications Advanced Image Fusion Techniques

Publication details

Journal / source
Traitement du signal
Volume
43
Issue
1
Pages
75–89
Publication date
28 February 2026
Language
EN