Context-aware lightweight remote-sensing image super-resolution network.

Area of Science:

• Remote Sensing
• Computer Vision
• Artificial Intelligence

Background:

• Convolutional Neural Networks (CNNs) dominate remote-sensing image super-resolution (RSISR).
• CNNs struggle with limited receptive fields, hindering long-range feature capture.
• Existing RSISR models face challenges in deployment due to high computational costs and parameter counts.

Purpose of the Study:

• To develop an efficient and effective RSISR network for remote-sensing images.
• To overcome the limitations of CNNs in capturing long-range dependencies.
• To reduce computational complexity and parameter count for easier deployment.

Main Methods:

• Proposed a Context-Aware Lightweight Super-Resolution Network (CALSRN).
• Introduced Context-Aware Transformer Blocks (CATBs) with Local Context Extraction Branch (LCEB) and Global Context Extraction Branch (GCEB).
• Employed Swin Transformer for global features and CNN-based cross-attention for local features, with a Dynamic Weight Generation Branch (DWGB) for feature aggregation.

Main Results:

• CALSRN effectively captures both local and global image features.
• The proposed method achieves high-quality super-resolution reconstruction.
• Experimental results show lower parameter count and computational complexity compared to existing methods.

Conclusions:

• CALSRN offers a promising solution for remote-sensing image super-resolution.
• The network's design enhances the ability to capture multi-scale image dependencies.
• CALSRN provides a computationally efficient and high-performance alternative for RSISR applications.