SAR Image Generation Method Using DH-GAN for Automatic Target Recognition.

Area of Science:

• Computer Vision
• Remote Sensing
• Artificial Intelligence

Background:

• Synthetic Aperture Radar (SAR) image target recognition has advanced with Convolutional Neural Networks (CNNs).
• Acquiring real SAR images is resource-intensive (time and cost).
• SAR images suffer from speckle noise, a high-frequency interference.

Purpose of the Study:

• To introduce DH-GAN, a Generative Adversarial Network (GAN) with a dual discriminator and high-frequency pass filter.
• To generate high-fidelity simulated SAR images that capture real-world high-frequency characteristics.
• To evaluate the effectiveness of DH-GAN generated images for training CNNs in target recognition tasks.

Main Methods:

• Development of DH-GAN, incorporating a dual discriminator and high-frequency pass filter.
• Generation of simulated SAR images using the proposed DH-GAN model.
• Validation through Power Spectral Density (PSD) analysis and comparative experiments.

Main Results:

• DH-GAN successfully generates simulated SAR images that emulate the high-frequency components of real SAR data.
• Power Spectral Density (PSD) analysis confirms the fidelity of the generated images.
• CNN models trained on DH-GAN generated images show enhanced target recognition proficiency.

Conclusions:

• DH-GAN provides a viable solution for generating realistic SAR images, overcoming limitations of real data acquisition.
• The generated images effectively preserve high-frequency characteristics crucial for target recognition.
• Utilizing DH-GAN simulated data significantly improves the performance of CNN-based SAR target recognition systems.