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

Abstract

Investigating micro/nanomotors' (MNMs') motion behavior is crucial for their practical applications and fundamental understanding of propulsion mechanisms. However, existing methods focus exclusively on a single-drive system and lack discriminative capability between different driving modes. Herein, we employ deep learning methods to distinguish different driving modes of catalytic micromotors (e.g., platinum driven and enzyme driven). When using trajectories as input, the classification accuracy is as high as 70.19%, which was primarily due to short-term motion. By extracting an optical flow map from original motion videos, classification accuracy reached 97.75% through multisegment sampling with transfer learning. Crucially, single-modality analysis demonstrated that driving mode differences occur within optical flow frames spanning just 0.45 s, which aligns precisely with trajectory-based findings. This work not only overcomes limitations of traditional analysis methods on MNMs' motion behaviors but also provides insights for future elucidation of fundamental motion mechanisms of chemically powered MNMs.