Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Hybrid Deep Learning for Hyperspectral Single Image Super-Resolution

Usman Muhammad, Jorma Laaksonen

TL;DR

This work tackles hyperspectral single-image super-resolution by introducing Hybrid Deep Learning (HDL), which integrates Spectral–Spatial Unmixing Fusion (SSUF) into standard 2D CNNs to jointly enhance spatial detail and spectral fidelity. A two-branch SSUF module merges spectral unmixing with spectral–spatial feature learning, followed by residual refinement and a custom Spatial–Spectral Gradient Loss to balance pixel accuracy with spectral consistency. Across three public datasets (Chikusei, PaviaC, PaviaU) and multiple downsampling scales (2×, 4×, 8×), HDL achieves competitive or superior performance with substantially lower model complexity. The proposed approach demonstrates strong practical potential for efficient, high-quality hyperspectral SR in real-world remote sensing tasks.

Abstract

Hyperspectral single image super-resolution (SISR) is a challenging task due to the difficulty of restoring fine spatial details while preserving spectral fidelity across a wide range of wavelengths, which limits the performance of conventional deep learning models. To address this challenge, we introduce Spectral-Spatial Unmixing Fusion (SSUF), a novel module that can be seamlessly integrated into standard 2D convolutional architectures to enhance both spatial resolution and spectral integrity. The SSUF combines spectral unmixing with spectral--spatial feature extraction and guides a ResNet-based convolutional neural network for improved reconstruction. In addition, we propose a custom Spatial-Spectral Gradient Loss function that integrates mean squared error with spatial and spectral gradient components, encouraging accurate reconstruction of both spatial and spectral features. Experiments on three public remote sensing hyperspectral datasets demonstrate that the proposed hybrid deep learning model achieves competitive performance while reducing model complexity.

Hybrid Deep Learning for Hyperspectral Single Image Super-Resolution

TL;DR

This work tackles hyperspectral single-image super-resolution by introducing Hybrid Deep Learning (HDL), which integrates Spectral–Spatial Unmixing Fusion (SSUF) into standard 2D CNNs to jointly enhance spatial detail and spectral fidelity. A two-branch SSUF module merges spectral unmixing with spectral–spatial feature learning, followed by residual refinement and a custom Spatial–Spectral Gradient Loss to balance pixel accuracy with spectral consistency. Across three public datasets (Chikusei, PaviaC, PaviaU) and multiple downsampling scales (2×, 4×, 8×), HDL achieves competitive or superior performance with substantially lower model complexity. The proposed approach demonstrates strong practical potential for efficient, high-quality hyperspectral SR in real-world remote sensing tasks.

Abstract

Hyperspectral single image super-resolution (SISR) is a challenging task due to the difficulty of restoring fine spatial details while preserving spectral fidelity across a wide range of wavelengths, which limits the performance of conventional deep learning models. To address this challenge, we introduce Spectral-Spatial Unmixing Fusion (SSUF), a novel module that can be seamlessly integrated into standard 2D convolutional architectures to enhance both spatial resolution and spectral integrity. The SSUF combines spectral unmixing with spectral--spatial feature extraction and guides a ResNet-based convolutional neural network for improved reconstruction. In addition, we propose a custom Spatial-Spectral Gradient Loss function that integrates mean squared error with spatial and spectral gradient components, encouraging accurate reconstruction of both spatial and spectral features. Experiments on three public remote sensing hyperspectral datasets demonstrate that the proposed hybrid deep learning model achieves competitive performance while reducing model complexity.

Paper Structure

This paper contains 12 sections, 10 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Spectral Unmixing
  4. Spectral-Spatial Learning
  5. Residual Learning
  6. Spatial-Spectral Gradient Loss
  7. Experimental setup
  8. Datasets
  9. Implementation and Evaluation Metrics
  10. Ablation Study
  11. Comparison with State-of-the-Art Methods
  12. Conclusion

Figures (2)

  • Figure 1: Overview of the proposed hybrid deep learning (HDL) model. The gray block represents the Spectral Unmixing module, while the light green block denotes the Spectral–Spatial Learning module, which jointly extracts and fuses spatial and spectral features. Together, they form the Spectral–Spatial Unmixing Fusion (SSUF) block. The output is passed through a convolutional layer and then processed by the Residual Learning blocks (shown in white).
  • Figure 2: Qualitative comparison on the PaviaU test image, representing a false-color composite, at scaling factors $2\times$, $4\times$, and $8\times$.