By Jay Kakade 17 Nov, 2024

Collected at: https://www.techexplorist.com/ai-enhanced-metalens-achieves-high-resolution-full-color-imaging/92969/

The electrical engineers at the Pohang University of Science and Technology, Korea have made advances in Metalenses that rival the traditional image-capturing lenses. This small and efficient package could revolutionize a wide range of industries.

With the widespread use of imaging in camerasvirtual reality, and augmented reality, imaging systems are evolving to become more efficient and compact. The foremost problem with traditional imaging systems is their reliance on bulky glass lenses.

Along with their large size, the low efficiency at multiple wavelengths poses challenges in developing efficient systems. To produce high-quality images in thin designs, researchers have developed metasurface lenses (Metalenses) with great potential in compact imaging.

Metalenses are ultra-thin lenses made from tiny nanostructures that can manipulate light at the nanoscale. They can solve core imaging issues like chromatic and angular aberrations. Despite the strides, Metalenses faces hindrances in focusing efficiency, lens diameter, and spectral bandwidth.

In a study published in Advanced Photonics, researchers have used deep learning systems to overcome most of these limitations. This innovative Metalense imaging system uses an image framework powered by deep learning. By integrating Artificial Intelligence (AI) with the Metalense optical system, the team achieved high-resolution and aberration-free images.

Though Metalenses are efficient in focusing light even at the nanoscale, sometimes the system fails to focus all colors to the same point, resulting in distorted or blurry images. The deep learning is specialized to recognize and correct the color aberrations.

More importantly, the AI is already trained with a large dataset of images and, it keeps learning from the captured images. This further elevates the efficient photo capturing.

The image restoration framework uses two neural networks. While one network is trained to correct images, the second network assesses the quality of the image. The combined training has significantly increased color accuracy and sharpness from different views.

Metalens imaging with deep learning
(a) Ground truth images, (b) metalens images, and (c) images reconstructed by our model. The images are affiliated with the test set data. The central (red) and outer (yellow) regions of the images are enlarged to access the restoration of the metalens image at high and low viewing angle, respectively. The outer regions of the metalens images (yellow box) are successfully restored, even though those are more severely degraded than the inner region (red box) due to the angular aberration under high viewing angle.

We propose a comprehensive imaging solution poised to replace conventional geometric lens-based systems. The proposed system effectively addresses the distortions mentioned above, to leverage the inherent strengths of metalenses to make a significant step toward high-quality, aberration-free images,” says the study.

Journal Reference

  1. Seo, J., Jo, J., Kim, J., Kang, J., Kang, C., Moon, S., Lee, E., Hong, J., Rho, J., & Chung, H. (2023). Deep-learning-driven end-to-end metalens imaging. Advanced Photonics, Vol. 6, Issue 6, 066002 (November 2024). DOI: 10.1117/1.AP.6.6.066002

Leave a Reply

Your email address will not be published. Required fields are marked *

0 0 votes
Article Rating
Subscribe
Notify of
guest
0 Comments
Inline Feedbacks
View all comments