Encoding PGAA Spectra as Images for Material Classification with Convolutional Neural Networks.

Area of Science:

• Nuclear Physics
• Materials Science
• Artificial Intelligence

Background:

• Prompt Gamma Ray Activation Analysis (PGAA) is a nuclear technique for elemental analysis.
• Deep Convolutional Neural Networks (CNNs) have shown success in various pattern recognition tasks.
• Material identification often relies on spectral data analysis, which can be complex.

Purpose of the Study:

• To investigate the efficacy of deep convolutional neural networks (CNNs) for material classification using PGAA spectra.
• To explore the use of transfer learning with pre-trained 2D CNN models to reduce trainable parameters.
• To assess the explainability and out-of-distribution detection capabilities of CNNs for material identification.

Main Methods:

• Training 2D CNN models on PGAA spectral data for material classification.
• Utilizing transfer learning from open-source computer vision models.
• Employing class activation maps for model interpretability.
• Implementing out-of-distribution detection methods to identify novel materials.

Main Results:

• CNNs successfully classified materials based on their PGAA spectra.
• Transfer learning enabled model development with fewer trainable parameters.
• Class activation maps provided insights into the decision-making process of the CNNs.
• Out-of-distribution tests demonstrated the ability to flag spectra from un Dseen materials.

Conclusions:

• CNNs are effective tools for automated material identification using PGAA spectra.
• The combination of transfer learning, explainability, and out-of-distribution detection makes CNNs suitable for real-world applications.
• This approach offers a promising direction for advancing material analysis and identification technologies.