Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Dual-Domain Convolutional Network for Hyperspectral Single-Image Super-Resolution

Murat Karayaka, Usman Muhammad, Jorma Laaksonen, Md Ziaul Hoque, Tapio Seppänen

TL;DR

The paper tackles hyperspectral single-image super-resolution by introducing DDSRNet, a lightweight dual-domain network that combines a Spatial-Net for spatial feature learning with a Haar wavelet-based frequency refinement (DWT) and a shared high-frequency branch. A hybrid Huber loss couples image-domain reconstruction with spatial and frequency-domain supervision to enhance edges and textures while preserving spectral integrity. Experimental results on PaviaC, PaviaU, and Chikusei show competitive performance at multiple scales with a very small parameter count (~0.07M), and ablation studies demonstrate the complementary contributions of spatial and wavelet components. The approach offers a practical, efficient solution for resource-constrained hyperspectral SR and suggests avenues for extending the framework with transformer or diffusion-based dual-domain architectures.

Abstract

This study presents a lightweight dual-domain super-resolution network (DDSRNet) that combines Spatial-Net with the discrete wavelet transform (DWT). Specifically, our proposed model comprises three main components: (1) a shallow feature extraction module, termed Spatial-Net, which performs residual learning and bilinear interpolation; (2) a low-frequency enhancement branch based on the DWT that refines coarse image structures; and (3) a shared high-frequency refinement branch that simultaneously enhances the LH (horizontal), HL (vertical), and HH (diagonal) wavelet subbands using a single CNN with shared weights. As a result, the DWT enables subband decomposition, while the inverse DWT reconstructs the final high-resolution output. By doing so, the integration of spatial- and frequency-domain learning enables DDSRNet to achieve highly competitive performance with low computational cost on three hyperspectral image datasets, demonstrating its effectiveness for hyperspectral image super-resolution.

A Dual-Domain Convolutional Network for Hyperspectral Single-Image Super-Resolution

TL;DR

The paper tackles hyperspectral single-image super-resolution by introducing DDSRNet, a lightweight dual-domain network that combines a Spatial-Net for spatial feature learning with a Haar wavelet-based frequency refinement (DWT) and a shared high-frequency branch. A hybrid Huber loss couples image-domain reconstruction with spatial and frequency-domain supervision to enhance edges and textures while preserving spectral integrity. Experimental results on PaviaC, PaviaU, and Chikusei show competitive performance at multiple scales with a very small parameter count (~0.07M), and ablation studies demonstrate the complementary contributions of spatial and wavelet components. The approach offers a practical, efficient solution for resource-constrained hyperspectral SR and suggests avenues for extending the framework with transformer or diffusion-based dual-domain architectures.

Abstract

This study presents a lightweight dual-domain super-resolution network (DDSRNet) that combines Spatial-Net with the discrete wavelet transform (DWT). Specifically, our proposed model comprises three main components: (1) a shallow feature extraction module, termed Spatial-Net, which performs residual learning and bilinear interpolation; (2) a low-frequency enhancement branch based on the DWT that refines coarse image structures; and (3) a shared high-frequency refinement branch that simultaneously enhances the LH (horizontal), HL (vertical), and HH (diagonal) wavelet subbands using a single CNN with shared weights. As a result, the DWT enables subband decomposition, while the inverse DWT reconstructs the final high-resolution output. By doing so, the integration of spatial- and frequency-domain learning enables DDSRNet to achieve highly competitive performance with low computational cost on three hyperspectral image datasets, demonstrating its effectiveness for hyperspectral image super-resolution.

Paper Structure

This paper contains 11 sections, 9 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Spatial-Net
  4. Discrete Wavelet Decomposition (DWT)
  5. Hybrid Loss Function
  6. Experimental setup
  7. Datasets
  8. Implementation and Evaluation Metrics
  9. Ablation Study
  10. Comparison with State-of-the-Art Methods
  11. Conclusion

Figures (2)

  • Figure 1: An overview of the proposed DDSRNet model. The left block represents the first domain, i.e., the Spatial-Net, while the right block illustrates the procedure of the Discrete Wavelet Transform (DWT).
  • Figure 2: Qualitative comparison on the PaviaU test image (false-color composite) at scaling factors of 2×, 4×, and 8×.