Photonics-based 3D radar imaging with CNN-assisted fast and noise-resistant image construction

Area of Science:

• Photonics
• Radar Imaging
• Artificial Intelligence

Background:

• High-resolution 3D radar imaging is crucial for various applications.
• Traditional back projection (BP) methods for image construction can be computationally intensive and susceptible to noise.
• Developing faster and more robust imaging techniques is an ongoing challenge.

Purpose of the Study:

• To demonstrate a photonics-based high-resolution 3D radar imaging system.
• To implement a fast and noise-resistant image construction method using a convolutional neural network (CNN)-assisted back projection (BP).
• To enhance imaging speed and suppress noise in 3D radar imaging.

Main Methods:

• Utilized a 2D radar array with broadband radar transceivers based on microwave photonic frequency multiplication and mixing.
• Employed a pre-trained 3D convolutional neural network (CNN) for mapping low-resolution to high-resolution images, forming the CNN-assisted BP imaging method.
• Trained the CNN using noise-free or low-noise ground truth images to enable noise suppression.

Main Results:

• Achieved a significant enhancement in imaging speed by a factor of approximately 55.3 using the CNN-assisted BP method compared to basic BP.
• Demonstrated effective noise suppression, leading to the generation of high-quality images.
• Verified the noise-resistant capability of the CNN-assisted BP method through peak signal-to-noise ratio (PSNR) comparisons.

Conclusions:

• The CNN-assisted BP imaging method offers a substantial improvement in speed and noise resistance for photonics-based 3D radar imaging.
• This approach reduces computational complexity, making high-resolution 3D radar imaging more efficient.
• The developed technique holds promise for applications requiring fast and high-quality radar imaging.