DTEA: Degradation-Aware Taylor Expansion Approximation Network for Pansharpening

Area of Science:

• Remote Sensing
• Image Processing
• Computer Vision

Background:

• Pansharpening aims to enhance spatial resolution of multispectral images using a high-resolution panchromatic image.
• Current pansharpening methods struggle with degraded panchromatic (PAN) images, limiting performance in real-world scenarios.
• PAN image degradation includes noise and sensor limitations, impacting the quality of the generated high-resolution multispectral (HRMS) image.

Purpose of the Study:

• To develop a novel pansharpening network that is aware of and compensates for PAN image degradation.
• To improve the spectral integrity and spatial resolution of HRMS images generated from degraded PAN data.
• To enhance the generalization capability of pansharpening methods in complex real-world conditions.

Main Methods:

• A degradation-aware Taylor expansion approximation (DTEA) network is proposed for pansharpening.
• The PAN image is decomposed into feature maps using a Taylor expansion approximation network (TEANet) to capture degradation information.
• A multi-level information fusion network (MIFNet) integrates these features with LRMS images, followed by inverse Taylor expansion to synthesize the HRMS image.

Main Results:

• The DTEA network demonstrates superior performance compared to state-of-the-art methods across diverse PAN image qualities.
• Extensive quantitative and qualitative experiments on three datasets validate the effectiveness of the proposed approach.
• The method shows excellent generalization capability, particularly in real-world scenarios with degraded PAN images.

Conclusions:

• The DTEA network effectively addresses the challenge of PAN image degradation in pansharpening.
• The proposed method achieves state-of-the-art results, preserving spectral integrity and enhancing spatial resolution.
• The DTEA network offers a robust solution for pansharpening applications dealing with imperfect input data.