Content-Adaptive Unfolding Wavelet Transformer for Hyperspectral Image Super-Resolution.

Area of Science:

• Remote Sensing
• Computer Vision
• Signal Processing

Background:

• Hyperspectral image super-resolution (HSI-SR) commonly fuses high-resolution multispectral images (HR-MSIs) with low-resolution hyperspectral images (LR-HSIs).
• Deep unfolding frameworks offer a structured approach but face limitations in data adaptivity and high-frequency information capture.
• Existing HSI-SR methods struggle with fixed parameters and inadequate transformer capabilities for detailed spectral-spatial information.

Purpose of the Study:

• To address the limitations of current HSI-SR methods by proposing a novel deep unfolding framework.
• To enhance the adaptivity of the data module and improve high-frequency information extraction in the prior module.
• To achieve superior HSI-SR performance with reduced computational overhead.

Main Methods:

• Proposed Content-Adaptive Unfolding Wavelet Transformer (CAUWT) with iteration-adaptive parameter learning.
• Introduced Wavelet-Assisted Transformer (WAT) integrating Discrete Wavelet Transform (DWT) and Hybrid Spectral-Spatial Attention Block (HSSAB).
• DWT captures multi-scale, multi-frequency details; HSSAB models correlations within spectral-spatial sub-bands.

Main Results:

• CAUWT demonstrated significant improvements in HSI-SR on both simulated and real-world datasets.
• The proposed Wavelet-Assisted Transformer effectively enhances high-frequency information quality without additional network complexity.
• Experimental results show superior performance compared to mainstream HSI-SR methods.

Conclusions:

• The proposed CAUWT method effectively resolves key issues in deep unfolding HSI-SR.
• Adaptive parameter learning and the novel WAT significantly boost HSI-SR performance.
• CAUWT achieves state-of-the-art results with improved efficiency.