Deep Learning Assisted Motion Behavior Analysis of Catalytic Micromotors Based on Trajectory and Optical Flow

Area of Science:

• * Materials Science and Engineering
• * Chemical Engineering and Nanotechnology
• * Artificial Intelligence and Machine Learning

Background:

• * Understanding micro/nanomotor (MNM) motion is vital for applications and fundamental propulsion studies.
• * Existing analysis methods struggle to differentiate between various MNM driving modes.

Purpose of the Study:

• * To develop deep learning methods for distinguishing different driving modes of catalytic micromotors.
• * To overcome limitations of traditional MNM motion analysis techniques.

Main Methods:

• * Utilized deep learning algorithms to classify MNM motion behaviors.
• * Extracted optical flow maps from motion videos for enhanced analysis.
• * Employed multisegment sampling and transfer learning techniques.

Main Results:

• * Trajectory-based classification achieved 70.19% accuracy, limited by short-term motion.
• * Optical flow map analysis with transfer learning reached a classification accuracy of 97.75%.
• * Driving mode differences were identified within 0.45-second optical flow frames.

Conclusions:

• * Deep learning, particularly with optical flow maps, significantly improves the ability to distinguish MNM driving modes.
• * This approach overcomes limitations of traditional methods for analyzing MNM motion.
• * Provides foundational insights for understanding the propulsion mechanisms of chemically powered MNMs.