Using unique 2D materials and machine learning, CV “sees” millions of colors like a human

Human eyes can see millions of colors, and now artificial intelligence can too.

Recently, an interdisciplinary research team from Northeastern University used new artificial intelligence technology to build a new device A-Eye that can recognize millions of colors. The field of machine vision has taken a big step forward and will be widely used in a range of technologies such as self-driving cars, agricultural sorting and remote satellite imaging.

The research paper was published in Materials Today.

Paper address: https://www.sciencedirect.com/science/article/abs/pii /S1369702122002255

Swastik Kar, corresponding author of the study and associate professor of physics at Northeastern University, said: "As automation becomes more widespread, machines are becoming more and more capable of identifying the color and shape of objects. is becoming more and more important."

The research designed a 2D material with special quantum properties. By embedding light into the optical window of the A-Eye machine, it can achieve "very high Process a rich variety of colors with unprecedented precision.

Furthermore, A-Eye is able to accurately identify and reproduce "seen" colors with zero deviation from the original spectrum. This is possible thanks to machine learning algorithms developed by an AI research team led by Sarah Ostadabbas, assistant professor of electrical and computer engineering at Northeastern University.

The main technology of the entire research focuses on the quantum and optical properties of a class of materials called transition metal dichalcogenides. This unique material has been considered to have Unlimited potential, especially in sensing and energy storage applications.

Research Overview

When identifying colors, machines typically use traditional RGB (red, green and blue) filters to break down the color into its component parts and then use that The information essentially guesses and reproduces the three primary colors of light. When you point your digital camera at a colored object and take a picture, the light from the object passes through a set of detectors with filters in front of them that separate the light into these raw RGB colors.

Kar said, “You can think of color filters as funnels that send visual information or data to separate boxes, and then those funnels assign artificial numbers to natural colors. ." But if you just break down color into its three components (red, green, and blue), there are some limitations.

However, instead of using color filters, Kar and his team used "transmissive windows" made of a unique 2D material.

Kar says they are getting the machine, known as the A-Eye, to recognize colors in a completely different way. When colored light hits a detector, instead of breaking it down into its main red, green and blue components or just looking for those components, the researchers used the entire spectrum of information.

Most importantly, the researchers used techniques to modify and encode these ingredients and store them in different ways. So they were given a set of numbers and were able to identify the primary colors in a very different way than conventionally.

The upper left of the picture below shows the 2D materials used to build A-Eye, the upper right shows the workflow of A-Eye, and the bottom of the picture shows the color comparison between the test color and A-Eye’s estimate.

Another author, Sarah Ostadabbas, said that when light passes through these transmission windows, A-Eye processes color into data. And built-in machine learning models look for patterns to better identify the corresponding colors analyzed by A-Eye.

At the same time, A-Eye can also continuously improve the color estimation results by adding any correct guesses to its training data set.

Introduction to the first author

The first author of this study, Davoud Hejazi, is currently a senior data scientist at Titan Advanced Energy Solutions, focusing on statistical modeling, machine learning, signal processing, image processing, cloud computing and data Visualization and other fields.

In May of this year, he obtained a Ph.D. in physics from Northeastern University. His graduation thesis was "Dispersion-Free Accurate Color Estimation using Layered Excitonic 2D Materials and Machine Learning."

#Thesis address: https://repository.library.northeastern.edu/files/neu:4f171c96c

The above is the detailed content of Using unique 2D materials and machine learning, CV “sees” millions of colors like a human. For more information, please follow other related articles on the PHP Chinese website!