A Comparable Study of CNN-Based Single Image Super-Resolution for Space-Based Imaging Sensors

Area of Science:

• Computer Vision
• Astrodynamics
• Image Processing

Background:

• Space-based space surveillance (SBSS) systems face challenges with low-resolution images due to vast sensor-target distances.
• Image super-resolution techniques are crucial for extracting detailed information from these low-resolution images.

Purpose of the Study:

• To comparatively analyze four popular single image super-resolution models based on convolutional neural networks (CNNs) for space applications.
• To fine-tune natural image super-resolution models using simulated space object imagery.
• To assess the performance of different CNN models under conditions relevant to SBSS.

Main Methods:

• Utilized four recent CNN-based single image super-resolution models.
• Fine-tuned models with simulated space object images.
• Evaluated model performance under various SBSS-specific conditions.

Main Results:

• Demonstrated the varying advantages and drawbacks of different CNN models for space object super-resolution.
• Identified model performance differences based on specific imaging conditions relevant to SBSS.
• Provided empirical data on the effectiveness of fine-tuning natural image models for space applications.

Conclusions:

• The study offers valuable insights for selecting appropriate CNN-based super-resolution methods for space object imagery.
• Findings can guide the development and application of super-resolution techniques in SBSS.
• Highlights the importance of model fine-tuning and condition-specific testing for optimal performance.