Towards smart optical focusing: deep learning-empowered dynamic wavefront shaping through nonstationary scattering media

Yunqi Luo; Suxia Yan; Huanhao Li; Puxiang Lai; Yuanjin Zheng

doi:10.1364/PRJ.415590

Photonics Research, Volume. 9, Issue 8, B262(2021)

Towards smart optical focusing: deep learning-empowered dynamic wavefront shaping through nonstationary scattering media

Yunqi Luo^1、†, Suxia Yan^1、†, Huanhao Li^2,3、†, Puxiang Lai^2,3,4、*, and Yuanjin Zheng^1,5、*

Author Affiliations

¹School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 639798, Singapore

²Department of Biomedical Engineering, The Hong Kong Polytechnic University, Hong Kong SAR, China

³The Hong Kong Polytechnic University Shenzhen Research Institute, Shenzhen 518034, China

⁴e-mail: puxiang.lai@polyu.edu.hk

⁵e-mail: yjzheng@ntu.edu.sg

show less

Supplementary Materials

Visualization 1 | (a) Simulation results. All figures in (a) use the same colorbar and scale. The playback speed of the video is slowed down to 1/3 of that in simulation. The speckle decorrelation time (SDT) of each stationary sub-process is 11.5, 13, 5, and 14.8 s, respectively. The green arrow indicates the change of media SDT. (b) Experimental results. All figures in (b) use the same colorbar and scale, and images are interpolated to 253 × 253. The SDT of each stationary sub-process is 140 and 160 s, respectively. The green arrow indicates the change of media SDT. First of all, light is scattered. After training, with TFOTNet, light can be focused through the scattering medium (t = 0:00:01), while pre-trained conventional single-input-single-output ConvLSTM or CNN only results in a poor focus (simulation), or even cannot focus light to a point (experiment). In the nonstationary medium, with adaptive recursive fine-tuning of TFOTNet, focusing degradation can be recovered in time, and then the focused speckle can be retained all the time in both simulation and experiment. In contrast, focusing performance of ConvLSTM and CNN is becoming worse over time even though fine-tuning is recursively conducted, and from t = 0:00:10, they cannot make light focused any more (simulation). In experiments, ConvLSTM or CNN never makes light focused.

Download

Tools

Get Citation

Copy Citation Text

Yunqi Luo, Suxia Yan, Huanhao Li, Puxiang Lai, Yuanjin Zheng, "Towards smart optical focusing: deep learning-empowered dynamic wavefront shaping through nonstationary scattering media," Photonics Res. 9, B262 (2021)

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Special Issue: DEEP LEARNING IN PHOTONICS

Received: Nov. 20, 2020

Accepted: Jan. 21, 2021

Published Online: Jul. 22, 2021

The Author Email: Puxiang Lai (puxiang.lai@polyu.edu.hk), Yuanjin Zheng (yjzheng@ntu.edu.sg)

DOI:10.1364/PRJ.415590

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology