Edge-Distilled and Local-Global Feature Selection Network for Hyperspectral Image Super-Resolution

Area of Science:

• Computer Vision
• Machine Learning
• Remote Sensing

Background:

• Convolutional neural networks show progress in hyperspectral image super-resolution.
• Existing methods struggle with extracting edge details and capturing both local and global features.

Purpose of the Study:

• To propose an Edge-Distilled and Local-Global Feature Selection network (EDLGFS) for hyperspectral image super-resolution.
• To enhance super-resolution reconstruction quality by leveraging edge details and local-global features.

Main Methods:

• An edge-guided super-resolution network using knowledge distillation to transfer edge information.
• A Local-Global Feature Selection mechanism (LGFS) integrating multi-size convolutions and self-attention.
• A dynamic loss mechanism to balance loss term contributions.

Main Results:

• The proposed EDLGFS network demonstrates superior super-resolution reconstruction quality.
• Experiments were conducted on three public datasets, validating the method's effectiveness.

Conclusions:

• The EDLGFS network effectively addresses limitations in hyperspectral image super-resolution.
• The integration of edge distillation and local-global feature selection significantly improves reconstruction quality.