A GAN-based fast focusing method for circular SAR images.

Area of Science:

• Remote Sensing
• Signal Processing
• Artificial Intelligence

Background:

• Circular Synthetic Aperture Radar (CSAR) imaging is susceptible to atmospheric and motion errors, causing phase center offsets and motion errors.
• Traditional time-domain phase compensation methods like Auto-regressive Back-projection (ARBP) are computationally intensive, requiring direction-by-direction processing and significant resources.

Purpose of the Study:

• To develop a more efficient and accurate method for focusing CSAR images.
• To address the computational and memory limitations of existing CSAR focusing techniques.

Main Methods:

• A novel approach utilizing a Generative Adversarial Network (GAN) for direct focusing of CSAR sub-aperture images.
• Introduction of Auto-focus Frequency Loss (AFFL) to mitigate the network's low-frequency bias.
• Proposal of Focus Position Feature Attention (FPFA) to enhance focus position extraction accuracy.
• Implementation of a new fusion strategy for post-focusing sub-aperture images.

Main Results:

• Experimental validation demonstrated significant improvements in CSAR image focusing efficiency.
• The proposed method achieved enhanced accuracy in CSAR image focusing.
• The GAN-based approach effectively overcomes the limitations of traditional time-domain methods.

Conclusions:

• The novel GAN-based method offers a superior alternative for CSAR image focusing.
• The integration of AFFL and FPFA contributes to improved focus accuracy and robustness.
• This approach presents a promising advancement in synthetic aperture radar imaging processing.