Research on turbulence-removal optical imaging based on multi-scale GAN and sequential images.

Area of Science:

• Optical imaging
• Computer vision
• Artificial intelligence

Background:

• Atmospheric turbulence causes geometric distortions and blurring in optical images.
• This degradation limits applications like military reconnaissance and disaster monitoring.
• Existing methods struggle with dynamic distortions and fine details.

Purpose of the Study:

• To develop a novel Generative Adversarial Network (GAN) for effective atmospheric turbulence mitigation.
• To address geometric distortions and spatial blurring in optical image sequences.
• To enhance image clarity and detail preservation in turbulent environments.

Main Methods:

• Proposed MS-TS-GAN (Multi-Scale Spatio-Temporal GAN) with a generator featuring multi-scale convolutional structures.
• Integrated a spatio-temporal feature extraction module for dynamic distortion correction.
• Utilized a VGG-based discriminator with global and local adversarial mechanisms for fidelity.

Main Results:

• MS-TS-GAN significantly outperformed traditional methods (Wiener, inverse filtering) and other GANs (TS-GAN, SF-GAN).
• Achieved superior performance in Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) on simulated and real-world data.
• Demonstrated substantial improvements in Entropy (EN) and Average Gradient (AG), indicating enhanced clarity and robustness.

Conclusions:

• MS-TS-GAN effectively corrects geometric distortions and spatial blurring caused by atmospheric turbulence.
• The proposed network offers a robust and efficient solution for improving optical imaging in challenging environments.
• This work holds significant theoretical and practical value for overcoming limitations in current optical imaging systems.