Lee Byung-sang of Materials Engineering, Development of Analyzing technology to form of Nanomaterials by Utilizing AI
- 공과대학
- Hit4543
- 2021-01-24
Lee Byung-sang (Professor Lee Jung-heon's research team) of the Department of New Materials Engineering,
Development of a technology that can statistically analyze the form of nanomaterials by utilizing AI-based image analysis technology
- It can be used in various fields where nanomaterials are utilized. -
Lee Byung-sang, a Ph.D. researcher at the Department of New Materials Engineering, and Professor Lee Jung-hun developed an algorithm that can analyze the morphological properties of nanoparticles with high accuracy by applying AI technology to electron microscopic image analysis technology.
※ Authors: Lee Byung-sang (first author, new material engineering), Yoon Seok-young (participating author, Sungkyunkwan University), Dr. Lee Jin-woong (participating author, new material engineering), Jang Jun-hyuk (participating author, new material engineering), Yoon Jae-sup (participating philosopher, system).
Although the form of nano-new materials used in various fields such as renewable energy, nanopharmaceuticals, catalysts, and sensors has a significant impact on physical and chemical characteristics, no technology has existed so far to accurately read and quantitatively analyze its morphological information in large quantities.
Lee Byung-sang and Lee Jung-hun's team developed a technique for analyzing nanoparticles with 99.75% higher accuracy and 0.25% lower misperception rate by automatically optimizing various methods and variables used for image analysis using a machine learning technique called genetic algorithm. In particular, this algorithm can significantly increase the density and analysis accuracy of nanoparticles at the same time by finding attached or enclosed nanoparticles in the electron microscope image, separating them by themselves, and removing them if they are not separated.
Using this technique, the researchers conducted various statistical analyses of the morphological properties of nanoparticles. First of all, using the Markov chain Monte Carlo performed based on Bayesian statistics, we estimated the distribution of a large number of nanoparticle forms, which are 160,000 and found the number of nanoparticles needed to be representative at the confidence level. Furthermore, we have developed algorithms to automatically classify and cluster various forms of nanoparticles present in the samples, and we can confirm that the optical properties of nanoparticles can be calculated with considerable accuracy when utilizing this statistical information.
The results of this study were published in the online edition of ACS Nano (IF=14.58) and adopted as an ACS Editors' Choice paper recommended and selected by more than 400 editors of 44 journals published by the American Chemical Society.
※ 논문명 : Statistical characterization of the morphologies of nanoparticles through machine learning based electron microscopy image analysis
The study is the first to precisely analyze and statistical analysis of the morphological properties of hundreds of thousands of nanoparticles by applying machine learning techniques to electron microscopic image analysis, and is expected to be one of the key technologies needed to build big data. In particular, since the shape of individual nanomaterials has a significant impact on physical and chemical properties such as optical properties and surface reactions, this technology can be used not only to evaluate the reliability of nanomaterials but also to develop new nanomaterials.