Identification and velocity measurement of microplastics based on machine learning

The settling velocity of microplastics (MPs) is a critical parameter for understanding their migration and behavior in aquatic environments. Conventional methods typically focus on tracking individual MPs and often face significant challenges in capturing the interactions and dynamics of multiple particles in a single experiment. This study introduces a novel machine learning framework to simultaneously track multiple MPs and accurately measure their individual terminal settling velocities within a square glass sedimentary column. Our approach integrates an enhanced YOLOv5-CA object detection model with the DeepSort tracking algorithm, achieves high-throughput tracking success rate of 99 % and the mean accuracy of 85.3 %. Compared to conventional methods, the maximum error observed using this framework was 1.7 %, well within acceptable limits. Furthermore, this method facilitates the study of hydrodynamic particle-particle interactions during multi-particle sedimentation. This work successfully achieves simultaneous multi-particle tracking in sedimentation processes and efficient measurement of MPs settling velocities, thereby enabling systematic investigation of particle-particle interactions in MPs transport research.