Precision particle tracking of discretized light spots in images using machine learning Open Access

Abstract

Particle tracking is a broad field of software with myriad applications, including use in the study of dusty plasma, a system in which colloidal “dust” particles in the μm regime are suspended in a plasma. Dusty plasma can be created experimentally in a lab or found in space and the Earth’s upper atmosphere, but attempts to study the dynamics of such microscopic particles are often plagued by pixel-locking, a phenomenon under which estimations of positions for randomly positioned particles in digital images are unnaturally biased towards integer values. This bias is due to the discretization of light detected by digital camera sensors to produce pixelized images, as well as due to the feature detection method used to estimate particle positions. Accordingly, we attempted to train machine learning models on a data set containing simulated images of individual dusty plasma particles with known positions to better estimate the positions and dynamics of simulated test images. We found that the error between the predicted positions found by the rudimentary models, as well as the presence of pixel-locking, and their ability to track a moving particle was comparable to if not better than the extant standard in Python-based particle tracking with TrackPy. Our findings imply that trained machine learning models are capable of becoming more accurate at simulated particle position estimation than pixel-intensity weighted algorithms, and this performance increase possibly extends to testing on actual experimental data. So long as the relationships between the accuracy metrics we employ to quantify method performance persist on experimental data, machine learning models could provide faster and more accurate dynamical measurements for small particles with possible benefits of great biological, astronomical, and physical interest.

Table of Contents

1 Introduction 1

1.1 What is Particle Tracking? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1.1 Broad Applications of Particle Tracking . . . . . . . . . . . . . . . . 4

1.1.2 Pixel-Locking and Limitations of Particle Tracking Methods . . . . . 5

1.2 Dusty Plasma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.3 Machine Learning: A Better Solution for Tracking Dusty Plasma Particles . 12

2 Methods 14

2.1 Image Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.2 Machine Learning Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.3 Particle as a Stochastic Harmonic Oscillator . . . . . . . . . . . . . . . . . . 23

3 Results and Discussion 26

3.1 Parameters and Pixel-Locking . . . . . . . . . . . . . . . . . . . . . . . . . . 26

3.2 Machine Learning Performance and Pixel-Locking . . . . . . . . . . . . . . . 27

3.3 Velocity Distributions and Estimated p Fit . . . . . . . . . . . . . . . . . . . 32

4 Conclusion 38

4.1 Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

Bibliography 40

About this Honors Thesis

Rights statement

• Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.

School	Emory College
Department	Physics
Degree	B.S.
Submission	Honors Thesis
Language	English
Research Field	Physics, Fluid and Plasma Physics, General
Keyword	Simulation Dusty Plasma Machine Learning Imaging
Committee Chair / Thesis Advisor	Burton, Justin C. , Emory University
Committee Members	Bans, Alissa, Emory University Brody, Jed, Emory University

Last modified

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	Precision particle tracking of discretized light spots in images using machine learning ()	2023-04-10 16:26:56 -0400	Press to Select an action Download

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions