Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Spectral Graph Reasoning Network for Hyperspectral Image Classification

Huiling Wang

TL;DR

The paper tackles hyperspectral image classification by addressing the limited spectral receptive field in traditional CNNs. It introduces the spectral graph reasoning network (SGR), which casts spectral embeddings as nodes in a graph and employs a spectral decoupling module to build a multi-scale spectral embedding hierarchy, followed by a spectral ensembling module that uses recurrent graph propagation to fuse information across levels. Key contributions include a KNN-based graph construction with a normalized Laplacian for spectral convolutions, a graph pyramid for multi-scale spectral reasoning, and a recurrent ensembling mechanism that integrates higher level spectral context into lower level graphs before final classification. Experiments on two public HSI datasets show that SGR achieves state-of-the-art performance with notable margins over competitive methods, highlighting the value of explicit spectral channel interactions and multi-context spectral embeddings for robust HSI classification. The approach provides a practical framework for leveraging spectral domain structure in remote sensing, with potential generalization to other high-dimensional spectral data tasks.

Abstract

Convolutional neural networks (CNNs) have achieved remarkable performance in hyperspectral image (HSI) classification over the last few years. Despite the progress that has been made, rich and informative spectral information of HSI has been largely underutilized by existing methods which employ convolutional kernels with limited size of receptive field in the spectral domain. To address this issue, we propose a spectral graph reasoning network (SGR) learning framework comprising two crucial modules: 1) a spectral decoupling module which unpacks and casts multiple spectral embeddings into a unified graph whose node corresponds to an individual spectral feature channel in the embedding space; the graph performs interpretable reasoning to aggregate and align spectral information to guide learning spectral-specific graph embeddings at multiple contextual levels 2) a spectral ensembling module explores the interactions and interdependencies across graph embedding hierarchy via a novel recurrent graph propagation mechanism. Experiments on two HSI datasets demonstrate that the proposed architecture can significantly improve the classification accuracy compared with the existing methods with a sizable margin.

Spectral Graph Reasoning Network for Hyperspectral Image Classification

TL;DR

The paper tackles hyperspectral image classification by addressing the limited spectral receptive field in traditional CNNs. It introduces the spectral graph reasoning network (SGR), which casts spectral embeddings as nodes in a graph and employs a spectral decoupling module to build a multi-scale spectral embedding hierarchy, followed by a spectral ensembling module that uses recurrent graph propagation to fuse information across levels. Key contributions include a KNN-based graph construction with a normalized Laplacian for spectral convolutions, a graph pyramid for multi-scale spectral reasoning, and a recurrent ensembling mechanism that integrates higher level spectral context into lower level graphs before final classification. Experiments on two public HSI datasets show that SGR achieves state-of-the-art performance with notable margins over competitive methods, highlighting the value of explicit spectral channel interactions and multi-context spectral embeddings for robust HSI classification. The approach provides a practical framework for leveraging spectral domain structure in remote sensing, with potential generalization to other high-dimensional spectral data tasks.

Abstract

Convolutional neural networks (CNNs) have achieved remarkable performance in hyperspectral image (HSI) classification over the last few years. Despite the progress that has been made, rich and informative spectral information of HSI has been largely underutilized by existing methods which employ convolutional kernels with limited size of receptive field in the spectral domain. To address this issue, we propose a spectral graph reasoning network (SGR) learning framework comprising two crucial modules: 1) a spectral decoupling module which unpacks and casts multiple spectral embeddings into a unified graph whose node corresponds to an individual spectral feature channel in the embedding space; the graph performs interpretable reasoning to aggregate and align spectral information to guide learning spectral-specific graph embeddings at multiple contextual levels 2) a spectral ensembling module explores the interactions and interdependencies across graph embedding hierarchy via a novel recurrent graph propagation mechanism. Experiments on two HSI datasets demonstrate that the proposed architecture can significantly improve the classification accuracy compared with the existing methods with a sizable margin.
Paper Structure (13 sections, 5 equations, 4 figures, 5 tables)

This paper contains 13 sections, 5 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Graph Construction
  4. Spectral Decoupling
  5. Spectral Ensembling
  6. Encoder Network
  7. Experimental Results
  8. Datasets
  9. Classification Results
  10. The University of Pavia Dataset
  11. The Indian Pines Dataset
  12. Ablation Study
  13. Conclusions

Figures (4)

  • Figure 1: Illustration of our proposed network architecture.
  • Figure 2: Illustration of the encoder network.
  • Figure 3: The university of Pavia dataset: (a) False color image (b) Groundtruth map (c) Prediction of SSLSTMs zhou2019 (d) Prediction of the proposed SPGAT
  • Figure 4: Indian Pines dataset: (a) False color image (b) Groundtruth map (c) Prediction of LBMSELM CaoYRCHS19 (d) Prediction of the proposed SPGAT