Remote sensing image super-resolution using multi-scale convolutional sparse coding network.

Area of Science:

• Computer Vision
• Remote Sensing
• Image Processing

Background:

• Convolutional Neural Networks (CNNs) have advanced remote sensing image super-resolution (SR).
• Existing methods struggle with detail extraction in remote sensing images compared to natural images, leading to unsatisfactory SR reconstruction.
• A need exists for improved SR techniques that can recover fine details in remote sensing imagery.

Purpose of the Study:

• To develop a novel Multiscale Convolutional Sparse Coding Network (MCSCN) for remote sensing image super-resolution.
• To enhance the extraction of high-frequency features and improve the richness of reconstructed image details.
• To achieve superior performance in peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) for remote sensing SR.

Main Methods:

• The proposed MCSCN utilizes a Multiscale Convolutional Sparse Coding Module (MCSCM) with dictionary convolution units.
• Multiple sparse feature sizes are combined to enrich feature information extraction.
• A sub-pixel convolution layer integrates global and local features for the final reconstruction block.

Main Results:

• The MCSCN demonstrates improved extraction of high-frequency features crucial for detail recovery.
• Combining multiple sparse feature sizes leads to more plentiful feature information.
• Experimental results show the MCSCN outperforms several state-of-the-art methods.

Conclusions:

• The MCSCN effectively addresses the detail deficiency in remote sensing image super-resolution.
• The proposed network architecture significantly enhances the quality of SR reconstruction.
• MCSCN offers a promising approach for high-fidelity remote sensing image SR.