Deep-Learning-Assisted Simultaneous Target Sensing and Super-Resolution Imaging.

Area of Science:

• Electromagnetics
• Metasurface Technology
• Deep Learning Applications

Background:

• Metasurfaces enable manipulation of electromagnetic waves for sensing and imaging.
• Metasurface integration increases complexity in retrieving target information.
• Existing deep learning methods often focus on single functions, limiting versatility.

Purpose of the Study:

• To develop a multifunctional deep learning network for metasurface-target interactive systems.
• To demonstrate the network's capability in sensing target properties and super-resolution imaging.
• To explore deep learning for both inverse reconstruction and forward electromagnetic prediction.

Main Methods:

• A preliminary experiment validated noise tolerance in metasurface scenarios.
• Captured electric field distributions were processed by a multifunctional deep learning network.
• A separate network was developed for forward electromagnetic prediction.

Main Results:

• The network accurately sensed target quantity and relative permittivity.
• Precise super-resolution images of targets were generated.
• The proposed deep learning approach demonstrated effectiveness in inverse reconstruction and forward prediction.

Conclusions:

• The multifunctional deep learning network offers a versatile approach for target information recovery in metasurface systems.
• This methodology accelerates advancements in target sensing and superimaging.
• Deep learning shows significant promise for diverse electromagnetic applications, including inverse and forward problems.