Using cutting-edge artificial intelligence techniques, a multidisciplinary team of researchers at Northeastern University has created a tool that can identify “millions of colors,” marking a significant advancement in the field of machine vision, a highly specialized area with numerous potential uses for a variety of technologies.
According to an article describing the findings published in Materials Today, the device, dubbed “A-Eye,” is capable of monitoring and processing color far more precisely than previous devices. According to Swastik Kar, a Northeastern associate professor of physics and a co-author of the study, as society and industry grow more automated, the capacity for machines to recognize or “see” color becomes more crucial.
According to Kar, the most popular criteria by which a machine may distinguish objects in the field of automation are their forms and colors.
The innovation has two components. Researchers have created a two-dimensional material with unique quantum properties that, when incorporated into an optical window that admits light into the device, can process a wide range of color variations with “very high accuracy”—something that experts in the field have not previously been able to accomplish.
Additionally, Sarah Ostadabbas, an assistant professor of electrical and computer engineering at Northeastern University, and her team of AI researchers developed machine-learning algorithms that allow A-Eye to “accurately recognize and reproduce’seen’ colors with zero deviation from their original spectra.” The study is the outcome of a special partnership between the Augmented Cognition and quantum materials laboratories at Northeastern.
The fundamental components of the technological advancement are the transition metal dichalcogenides class of materials’ quantum and optical characteristics. Long praised by scientists as having “almost infinite potential,” the special materials offer a wide range of “electronic, optoelectronic, sensing, and energy storage applications.”
The topic at hand, according to Kar, is what happens to light when it travels through quantum matter. When these materials are grown on a certain surface and light is allowed to travel through it, an electrical signal that [Ostadabbas’s] team can use as data emerges from the other end when it hits a sensor.
According to Kar, there are several commercial uses for this study in machine vision, including distant satellite photography, agricultural sorting, and autonomous cars.
Color is employed as one of the key elements in differentiating “good” from “bad,” “go” from “no-go,” hence there is a significant industrial use here, according to Kar.
Machines often identify color by dissecting it into its component parts using traditional RGB (red, green, and blue) filters, and then utilizing that information to basically guess at and duplicate the original color. The light from a colorful item passes through a collection of detectors with filters in front of them when you point a digital camera at it and snap a picture, differentiating the light into those main RGB hues.
These color filters can be compared to funnels that divide visual information into several compartments before assigning “artificial numbers to natural colors,” according to Kar.
There are certain restrictions if you only divide it into the three colors red, green, and blue, Kar explains.
Kar and his team employed “transmissive windows” formed of the special two-dimensional material as an alternative to filters.
We are redefining how a machine recognizes color, claims Kar. “When a colored light, let’s example, comes on a detector, we are utilizing the complete spectrum information rather than separating it into its main red, green, and blue components. Additionally, we are employing several methods to change, encode, and save them in various ways. Therefore, it gives us a set of numbers that make it lot easier for us to identify the original hue than it would otherwise.”
According to Ostadabbas, as the light passes through these windows, the system analyzes the color as data and has machine learning models built in to hunt for trends that would help it more accurately identify the associated hues.
The researchers concluded that “A-Eye can constantly improve color estimation by adding any corrected estimates to its training database.”