Three-Dimensional Terahertz Coded-Aperture Imaging Based on Matched Filtering and Convolutional Neural Network.

Area of Science:

• Electromagnetics and Signal Processing
• Computational Imaging and Machine Learning

Background:

• Terahertz coded-aperture imaging (TCAI) offers high-resolution, forward-looking, and staring imaging capabilities.
• Existing 3D TCAI methods face challenges with heavy computational load due to large reference-signal matrices and poor low signal-to-noise ratio (SNR) target resolution.

Purpose of the Study:

• To develop an efficient and high-resolution 3D TCAI method for low SNR targets.
• To reduce the computational burden associated with traditional 3D TCAI.

Main Methods:

• A novel 3D TCAI model based on matched filtering (MF) was developed for frequency-hopping (FH) signals.
• MF processes the original echo, and extracted echoes from different spike pulses enable simultaneous reconstruction of targets in different planes, decomposing computational complexity.
• A convolutional neural network (CNN) was designed to improve the MF-based TCAI's ability to reconstruct low SNR targets.

Main Results:

• The proposed MF-TCAI method significantly reduces computational load by downsizing the reference-signal matrix.
• Experimental results demonstrate impressive imaging performance, achieving high resolution for low SNR 3D targets.
• The integration of CNN further enhances the resolution and accuracy of low SNR target reconstruction.

Conclusions:

• The MF-TCAI method provides efficient and high-resolution 3D imaging, particularly effective under low SNR conditions.
• The combination of MF and CNN offers intelligent imaging capabilities, overcoming limitations of conventional TCAI.
• This advanced TCAI technique holds significant potential for applications in security screening, nondestructive detection, and medical diagnosis.