Related Experiment Video
Updated: Jul 6, 2025
Tracking Rats in Operant Conditioning Chambers Using a Versatile Homemade Video Camera and DeepLabCut
Published on: June 15, 2020
A comparative study of a traditional localization algorithm and a deep learning model for radioactive particle
Roos Sophia de Freitas Dam1, Renato Raoni Werneck Affonso2, William Luna Salgado2
1Programa de Engenharia Nuclear, Universidade Federal Do Rio de Janeiro, Avenida Horácio de Macedo 2030, Bloco G - Sala 206, Zip Code 21941-914, Cidade Universitária, RJ, Brazil; Instituto de Engenharia Nuclear, Rua Hélio de Almeida 75, Zip Code 21941-906, Cidade Universitária, RJ, Brazil.
This study compares traditional algorithms and deep neural networks for radioactive particle tracking. Deep neural networks show higher accuracy in tracking particle trajectories, especially with larger datasets.
Area of Science:
- Nuclear Engineering
- Computational Physics
- Data Science
Background:
- Radioactive particle tracking is crucial for hydrodynamics in chemical and civil engineering.
- Traditional algorithms and artificial intelligence (AI) methods can reconstruct particle trajectories.
Purpose of the Study:
- To compare the accuracy of traditional algorithms versus deep neural networks (DNNs) for radioactive particle tracking.
- To evaluate the influence of calibration dataset size on tracking performance.
Main Methods:
- Developed a simplified concrete mixer model with six NaI(Tl) detectors and a 137Cs source.
- Utilized MCNPX code for simulating measurement geometry and generating a dataset of 3615 patterns.
- Implemented a traditional C++ minimization algorithm and a DNN with hyperparameters optimized by Optuna (Python library).
Main Results:
- Traditional algorithm yielded Mean Absolute Percentage Errors (MAPE) of 20.81% (x), 10.33% (y), and 16.84% (z).
- Deep neural network achieved lower MAPE: 6.87% (x), 2.70% (y), and 22.79% (z).
- Preliminary analysis suggests larger calibration datasets improve performance for both methods.
Conclusions:
- Deep neural networks demonstrate superior accuracy for radioactive particle tracking compared to traditional algorithms.
- The choice of method and dataset size significantly impacts the precision of trajectory reconstruction.
- Further research on dataset optimization is warranted for enhanced radioactive particle tracking applications.
