Perfect vision with metal lenses backed by artificial intelligence

Modern image capture systems, such as those used in smartphones, virtual reality and augmented reality devices, are subject to continuous improvements, as they are sought to be more compact, more efficient and with greater performance. But the road is difficult.

Traditional very small optical systems are based on bulky glass lenses, and have limitations such as chromatic aberrations, low efficiency at multiple wavelengths, and larger than desired physical sizes. These drawbacks pose problems when designing smaller, lighter systems that still produce high-quality images.

To overcome these problems, metallenses were invented, ultrathin lenses composed of tiny nanostructures capable of manipulating light on a nanometer scale. Metal lenses offer enormous potential to miniaturize optical systems, but they are not without difficulties, especially when it comes to capturing full-color images without distortions.

Now, a team led by Joonhyuk Seo, from Hanyang University in Seoul, South Korea, has created an innovative system that combines metalente with artificial intelligence, managing to overcome many of those limitations of metalentes. The resulting images are high-resolution, aberration-free, and perfectly colored, and the system remains tiny.

The metallens is fabricated using nanoimprint lithography, a cost-effective method suitable for large-scale production, followed by atomic layer deposition, enabling large-scale production of these lenses.

The metallens, manufactured using a cheap process suitable for industrial-scale production, is designed to focus light efficiently but, like most metallenses, suffers from chromatic aberration and other distortions due to its interaction with light of different lengths. wave. To solve this, the artificial intelligence system (specifically the modality known as “deep learning”) is trained so that it is capable of recognizing and correcting chromatic distortions and blurring caused by the metal lens. The system learns from a large set of image data and applies these corrections to future images captured by it.

The artificial intelligence system consists of two neural networks, which are trained together. One network generates corrected images and the other evaluates their quality, pushing the system to continually improve. Additionally, advanced techniques help the system understand how image distortions change depending on the viewing angle. The result is significant improvements in the processed images, especially in terms of color accuracy and sharpness throughout the field of view.

On the left, a metallens, composed of an array of nanostructures with arbitrary rotation angles, acquires an image. On the right, this image is processed using artificial intelligence and the final result is a very faithful reproduction of what has been photographed. (Illustration: Seo et al., doi 10.1117/1.AP.6.6.066002.)

The system produces images that rival the quality of bulky traditional camera lenses.

This innovation has the potential to revolutionize a wide range of sectors, from smartphones to digital cameras, virtual reality and augmented reality.

Joonhyuk Seo and his colleagues present the technical details of their innovative artificial intelligence metalens in the academic journal Advanced Photonics, under the title “Deep-learning-driven end-to-end metalens imaging.” (Fountain: NCYT by Amazings)