Artificial Intelligence-Based Deep Fusion Model for Pan-Sharpening of Remote Sensing Images

Area of Science:

• Remote Sensing
• Image Processing
• Artificial Intelligence

Background:

• Remote sensing image fusion aims to enhance spatial resolution of multispectral (MS) images using high-spatial-resolution panchromatic (PAN) images.
• Existing deep learning fusion models often neglect the distinct image distributions of MS and PAN data, leading to color and gradient distortions.
• Improvements in spatial and spectral information are crucial for accurate remote sensing applications.

Purpose of the Study:

• To propose an efficient artificial intelligence-based deep transfer learning model for remote sensing image fusion.
• To address the color and gradient distortion issues prevalent in current fusion techniques.
• To enhance the quality of fused remote sensing images by preserving spatial and spectral information.

Main Methods:

• An Inception-ResNet-v2 deep learning model was adapted and enhanced with a color-aware perceptual loss (CPL).
• A gradient channel prior was employed as a postprocessing step to further refine the fused images.
• The proposed model was evaluated using benchmark datasets for remote sensing image fusion.

Main Results:

• The developed model demonstrated superior performance in preserving color and gradient information compared to existing methods.
• Experimental results confirmed the effectiveness of the color-aware perceptual loss and gradient channel prior.
• Fused images exhibited improved spatial detail and spectral fidelity.

Conclusions:

• The proposed deep transfer learning approach offers a significant advancement in remote sensing image fusion.
• The integration of CPL and gradient channel prior effectively mitigates common fusion artifacts.
• This method provides a robust solution for generating high-quality fused remote sensing images.