GAFnet: Group Attention Fusion Network for PAN and MS Image High-Resolution Classification

Area of Science:

• Remote Sensing
• Computer Vision
• Image Processing

Background:

• Panchromatic (PAN) and multispectral (MS) images offer complementary spatial and spectral information crucial for accurate image interpretation.
• Existing methods may not fully leverage the synergistic potential of PAN and MS data for classification tasks.

Purpose of the Study:

• To propose a novel deep learning classification method for fused PAN and MS images.
• To enhance image interpretation by effectively integrating spatial and spectral features from both image types.

Main Methods:

• A deep group spatial-spectral attention fusion network was developed.
• Multispectral images were upsampled to match PAN image resolution.
• Group spatial and spectral attention modules extracted features, followed by an attention fusion module to integrate multilevel features.

Main Results:

• The proposed network successfully fused low-level and high-level features, preserving both global and local information.
• Pixel-level classification maps were generated from the fused features.
• Experiments on four datasets demonstrated comparable results to existing methods.

Conclusions:

• The deep group spatial-spectral attention fusion network offers a promising approach for PAN and MS image classification.
• The method effectively utilizes the complementary nature of PAN and MS data for enhanced image analysis.