Multi-Particle Tracking in Complex Plasmas Using a Simplified and Compact U-Net.

Area of Science:

• Plasma physics
• Complex plasma analysis
• Particle detection

Background:

• Accurate detection of micron-sized particles is crucial for complex plasma analysis.
• Machine learning algorithms show promise in outperforming classical particle detection methods.
• Resource-constrained environments necessitate efficient and effective particle tracking solutions.

Purpose of the Study:

• To present a U-Net based approach for tracking micron-sized particles in dense dusty plasmas.
• To evaluate the performance of various U-Net architectures against established methods like StarDist and trackpy.
• To identify compact U-Net models that balance efficiency and effectiveness for real-time applications.

Main Methods:

• Utilized a U-Net convolutional neural network architecture for image segmentation.
• Compared a full-size U-Net, three optimized U-Net variants, StarDist, and trackpy on artificial data.
• Applied the most effective U-Net models to experimental data from the Plasmakristall-Experiment 4 (PK-4).

Main Results:

• Optimized U-Net architectures demonstrated competitive accuracy compared to full-size U-Net, StarDist, and trackpy.
• Specific compact U-Net models were identified as providing a superior balance of efficiency and effectiveness.
• Successful application of U-Net models to real-world dusty plasma experimental data.

Conclusions:

• U-Net architectures are effective for micron-sized particle tracking in complex dusty plasmas.
• Compact U-Net models offer a viable solution for real-time particle detection in resource-limited experimental settings.
• This research contributes to advancing particle analysis techniques in plasma physics research.