A machine learning approach to analyze the structural formation of soft matter via image recognition

ABSTRACT

A novel method is developed to analyze the structural formation of colloidal particles based on image recognition via a convolutional neural network (CNN). This makes it possible to analyze various complex structures that are difficult to study using a traditional bond-order parameter analysis. Molecular dynamics simulations on soft colloidal systems are performed in quasi two-dimensional and three-dimensional systems, and the efficiency of the proposed method is demonstrated.

KEYWORDS:

• Convolutional neural networks
• colloidal crystals
• structural formation

Acknowledgments

The author acknowledges the Information Technology Center at the University of Tokyo for the use of their facilities.

Additional information

Funding

This work was supported by JSPS KAKENHI (Grants-in-Aid for Scientific Research by JSPS) Grant Number 16K05038.