FCNNs Show Promise, Challenges in Neutron Resonance Detection

Nataly R. Panczyk, Athanasios Stamatopoulos, Josef Svoboda, Majdi I. Radaideh· August 6, 2026 View original

Key takeaways

  • FCNNs can achieve high accuracy in detecting neutron resonances in specific datasets.
  • Generalization to unseen isotopes remains a significant challenge for current models.
  • Larger, more diverse training data is crucial for improving model robustness.
  • Integrating physical characteristics into models could enhance generalization capabilities.

Who benefits

Nuclear EnergyScientific ResearchMaterials ScienceDefense

Summary

This study investigates using fully convolutional neural networks (FCNNs) to automatically detect neutron resonances in complex and noisy transmission spectra, aiming to augment traditional R-Matrix codes. While achieving 93% classification accuracy, the model currently struggles to generalize reliably to previously unseen isotopes, highlighting the need for more diverse training data and integration of physical characteristics.

Analyzing neutron transmission data to identify resonances is a complex, manual process often relying on traditional R-Matrix codes and prior evaluations. This research explores the potential of fully convolutional neural networks (FCNNs) to automate and accelerate this post-experimental processing, aiming to reduce bias and manual effort. The study applied an FCNN to classify points in seven transmission spectra as either resonance or non-resonance regions, achieving a classification accuracy of 93%. However, further analysis revealed that this metric overstates the model's ability to generalize. Despite including additional training data, the FCNN did not reliably perform on previously unseen isotopes. Future work is recommended to address these limitations. This includes evaluating whether a significantly larger and more diverse training dataset can produce a generalizable model, and crucially, incorporating known physical characteristics of neutron resonances to improve the model's performance and robustness across different isotopes.

Why it matters

Automating neutron resonance detection could significantly accelerate nuclear data analysis, reduce human bias, and improve the efficiency of research in nuclear physics and related applications.

How to implement this in your domain

  1. 1Explore applying FCNNs or similar deep learning models to other complex signal processing tasks in your domain.
  2. 2Prioritize collecting larger and more diverse datasets when developing AI models for scientific applications.
  3. 3Investigate methods for incorporating domain-specific physical knowledge into neural network architectures to improve generalization.
  4. 4Conduct thorough generalization tests beyond simple accuracy metrics to ensure model robustness for real-world deployment.

Original post by Nataly R. Panczyk, Athanasios Stamatopoulos, Josef Svoboda, Majdi I. Radaideh

"arXiv:2608.04027v1 Announce Type: new Abstract: This work investigates the feasibility of augmenting traditional R-Matrix codes with a robust machine learning framework for automatically detecting neutron resonances in transmission spectra. Neutron transmission data are often com…"

View on X

Originally posted by Nataly R. Panczyk, Athanasios Stamatopoulos, Josef Svoboda, Majdi I. Radaideh on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses