Self-supervised spectral super-resolution for a fast hyperspectral and multispectral image fusion

Area of Science:

• Remote Sensing
• Computer Vision
• Deep Learning

Background:

• Hyperspectral-multispectral image fusion (HSI-MSI Fusion) aims to enhance hyperspectral image resolution.
• Deep learning methods are popular for HSI-MSI Fusion but face challenges like data scarcity and poor generalization.
• High computational costs are also a significant issue in current deep learning fusion techniques.

Purpose of the Study:

• To propose an innovative deep learning technique for HSI-MSI Fusion.
• To address the challenges of data scarcity, poor generalization, and high computational costs in HSI-MSI Fusion.
• To reconstruct high-resolution hyperspectral images using spectral super-resolution.

Main Methods:

• A tiny deep neural network is trained for spectral super-resolution.
• High-resolution training data is artificially generated using a spatial degradation model.
• The method reconstructs high-resolution hyperspectral images from high-resolution multispectral images.

Main Results:

• The proposed method overcomes data scarcity and poor generalization issues.
• Computational burden is significantly reduced compared to existing methods.
• Experimental results demonstrate promising performance for HSI-MSI Fusion.

Conclusions:

• The developed tiny deep neural network offers an effective solution for HSI-MSI Fusion.
• The approach provides a computationally efficient and generalizable method for enhancing hyperspectral image resolution.
• The study contributes a novel technique to the field of remote sensing image fusion.