Blur-Resistant Hyperspectral Image Super-Resolution via Dual-Degradation Fusion Model

Area of Science:

• Computer Vision
• Remote Sensing
• Image Processing

Background:

• Deep unfolding networks show promise for hyperspectral image super-resolution (HSI-SR).
• Existing methods often ignore multispectral image (MSI) degradation, limiting real-world applicability.
• Addressing MSI blurring is crucial for advancing HSI-SR.

Purpose of the Study:

• To propose a novel Dual-Degradation Fusion (D2-Fusion) model for HSI-SR.
• To develop an end-to-end trainable network, BHSR-Net, incorporating both HSI and MSI degradation.
• To enhance the performance and practical utility of HSI-SR by accounting for MSI blurring.

Main Methods:

• Developed the Dual-Degradation Fusion (D2-Fusion) model.
• Applied a second-order semismooth Newton algorithm to solve the D2-Fusion optimization problem.
• Mapped the solution steps into the end-to-end trainable BHSR-Net architecture.

Main Results:

• BHSR-Net is the first network to explicitly consider MSI blurring artifacts in HSI-SR.
• The network demonstrates superior SR performance and generalization across diverse datasets.
• A lightweight variant achieves comparable results with only 38K parameters, showing excellent learning efficiency.

Conclusions:

• The proposed BHSR-Net effectively models dual degradation (HSI and MSI blurring) for improved HSI-SR.
• The network's structure ensures physical interpretability and strong mathematical correspondence with the optimization algorithm.
• BHSR-Net offers a significant advancement in HSI-SR, particularly for real-world applications.