MSFANet: A Multi-Scale Feature Fusion Transformer with Hybrid Attention for Remote Sensing Image Super-Resolution.

Area of Science:

• Remote Sensing
• Computer Vision
• Image Processing

Background:

• Remote sensing image resolution is limited by sensors and transmission.
• Super-resolution reconstruction is crucial for detailed analysis.

Purpose of the Study:

• To propose MSFANet, a multi-scale feature fusion network for remote sensing image super-resolution.
• To improve the quality and efficiency of super-resolution reconstruction.

Main Methods:

• Developed MSFANet based on Swin Transformer architecture.
• Incorporated Feature Refinement Augmentation (FRA), Local Structure Optimization (LSO), and Residual Fusion Network (RFN) for deep feature extraction.
• Employed shallow feature extraction and high-quality image reconstruction modules.

Main Results:

• MSFANet outperformed state-of-the-art models (HSENet, TransENet) on RSSCN7, AID, and WHU-RS19 datasets.
• Achieved superior performance across ×2, ×3, and ×4 super-resolution tasks based on five metrics.
• Demonstrated reduced computational overhead compared to existing methods.

Conclusions:

• MSFANet offers an effective solution for remote sensing image super-resolution.
• The model balances reconstruction quality with computational efficiency.
• Highlights the potential of Swin Transformer architectures in remote sensing image enhancement.