Cross-Scope Spatial-Spectral Information Aggregation for Hyperspectral Image Super-Resolution.

Area of Science:

• Computer Vision
• Remote Sensing
• Signal Processing

Background:

• Hyperspectral image super-resolution aims to enhance spatial detail.
• Convolutional methods struggle with global spatial-spectral information.
• Existing transformer methods have limitations in capturing long-range dependencies.

Purpose of the Study:

• To propose a novel Cross-Scope Spatial-Spectral Transformer (CST) for single hyperspectral image super-resolution.
• To efficiently investigate long-range spatial and spectral similarities.
• To overcome limitations of existing methods in capturing global spatial-spectral information.

Main Methods:

• Devised cross-attention mechanisms in spatial and spectral dimensions.
• Designed a cross-scope spatial self-attention integrating global information.
• Constructed a cross-scope spectral self-attention to capture spectral band correlations.

Main Results:

• The proposed CST method demonstrates superior performance over state-of-the-art methods.
• Quantitative and visual experimental results validate the effectiveness of CST.
• The method successfully enhances spatial resolution in hyperspectral images.

Conclusions:

• The CST model effectively addresses the challenge of capturing long-range spatial-spectral dependencies.
• CST offers a significant advancement in hyperspectral image super-resolution.
• The proposed approach provides a robust solution for enhancing hyperspectral data quality.