A pan-sharpening network using multi-resolution transformer and two-stage feature fusion.

Area of Science:

• Remote Sensing
• Image Processing
• Computer Vision

Background:

• Pan-sharpening fuses low-resolution multispectral and high-resolution panchromatic images to create enhanced remote sensing data.
• Existing deep learning methods often fail to fully extract and utilize diverse features, leading to information loss.

Purpose of the Study:

• To propose a novel pan-sharpening method using a multi-resolution transformer and a two-stage feature fusion strategy.
• To address the limitations of existing methods in feature extraction and exploitation for improved pan-sharpening results.

Main Methods:

• A transformer-based multi-resolution feature extractor is designed to capture diverse image features.
• A two-stage feature fusion strategy is employed, including multi-resolution fusion and shallow-deep fusion modules.

Main Results:

• The proposed method achieves superior visual and quantitative performance compared to state-of-the-art approaches on QuickBird and WorldView-3 datasets.
• The method demonstrates effectiveness with fewer parameters, indicating improved efficiency.

Conclusions:

• The novel transformer-based feature extractor and two-stage fusion scheme significantly enhance pan-sharpening performance.
• The proposed method offers a more effective solution for generating high-resolution multispectral remote sensing images.