VOGTNet: Variational Optimization-Guided Two-Stage Network for Multispectral and Panchromatic Image Fusion

Area of Science:

• Remote Sensing
• Image Processing
• Computer Vision

Background:

• Multispectral pansharpening aims to fuse multispectral (MS) and panchromatic (PAN) images for high spatial and spectral resolution.
• Existing deep learning methods often fail with noisy or blurred data due to neglecting imaging artifacts.

Purpose of the Study:

• To develop a robust multispectral pansharpening method that addresses noise and blur.
• To improve the performance and generalizability of deep learning-based pansharpening techniques.

Main Methods:

• Proposed a variational optimization-guided two-stage network (VOGTNet).
• Employed a dual-branch fusion network (DBFN) for supervised learning on noisy/blurred data to generate prior fusion results.
• Utilized estimated spectral response function (SRF) and point spread function (PSF) for unsupervised learning to restore image details.

Main Results:

• VOGTNet demonstrated improved pansharpening performance.
• The method showed strong robustness against noise and blur in datasets.
• The proposed framework can enhance other supervised learning-based pansharpening methods.

Conclusions:

• VOGTNet effectively overcomes limitations of existing methods in handling noisy and blurred multispectral pansharpening data.
• The VOGTNet framework offers a versatile approach to improve noise and blur resistance in various pansharpening applications.