AI

Abstract Neural Network/ Photo By Sergey Soldatov via 123RF

 

In actual data processing, devices recognize images and objects by relying on a piece of equipment that can “see” them, like a camera sensor. But a team of engineers developed a device capable of recognizing objects at the speed of light without relying on another device.

The device is called a diffractive deep neural network that uses light bouncing from the target object to identify it. Compared to a standard computer, the device can perform the identification process in a short period of time due to its lightning speed. Also, the equipment does not consume energy to recognize objects, since it uses the diffusion of light.

“This optical artificial neural network device is intuitively modeled on how the brain processes information. It could be scaled up to enable new camera designs and unique optical components that work passively in medical technologies, robotics, security or any application where image and video data are essential.” Said Aydogan Ozcan, the principal investigator of the study.

The engineers patterned the device based on a computer-simulated design. To build it, they used a 3D printer to produce several eight-centimeter square polymer wafers. The wafers had bumpy surfaces that allowed incoming light from an object to diffract in different directions. 

Also, the layers consisted of tens of thousands of tiny pixels where the light could travel through. About submillimeters of wavelength, terahertz light frequencies could travel through the wafers.

They aligned the wafers to create a series of pixelated layers to form an optical network that shaped the manner of traveling light bounced by the object. As a network, the wafers could recognize an object from the light diffracted to a pixel assigned on a particular type of object.

To train the network on different types of objects, the team used a computer and allowed the device to learn the patterns of light diffraction of every object in front of it. A deep learning mechanism was included in the network so it can improve its learning capability.

The engineers said that the innovation could be applied in microscopic imaging to view and sort millions of cells for any signs of a disease.