Terahertz image super-resolution based on a deep convolutional neural network

Area of Science:

• Optics and Photonics
• Computer Vision
• Artificial Intelligence

Background:

• Terahertz (THz) imaging offers unique capabilities but often suffers from low resolution and noise.
• Improving THz image quality is crucial for various scientific and industrial applications.

Purpose of the Study:

• To propose an effective and robust method for terahertz image super-resolution.
• To enhance the resolution and reduce noise in THz images using deep learning.

Main Methods:

• A deep convolutional neural network (CNN) model was designed for end-to-end learning.
• The CNN was trained using synthetic THz images with varied blur kernels and noise levels.
• The method was quantitatively and qualitatively compared against existing super-resolution techniques.

Main Results:

• The proposed CNN method demonstrated superior accuracy and visual improvements over other super-resolution techniques on synthetic data.
• Experimental results on real THz images showed significant enhancement in resolution and noise reduction.
• The method proved practical and accurate for improving terahertz image quality.

Conclusions:

• The developed deep CNN method is effective and robust for terahertz image super-resolution.
• The approach successfully addresses the challenges of low resolution and noise in THz imaging.
• This technique offers a valuable tool for advancing THz imaging applications.