Detecting shielded explosives by coupling prompt gamma neutron activation analysis and deep neural networks

Area of Science:

• Nuclear Physics
• Analytical Chemistry
• Artificial Intelligence

Background:

• Prompt Gamma Neutron Activation Analysis (PGNAA) is a nuclear technique for material composition analysis.
• PGNAA is utilized for explosives detection by analyzing gamma spectra from neutron-bombarded samples.
• Sample shielding can distort PGNAA spectra, leading to inaccurate material identification.

Purpose of the Study:

• To address the challenge of shielding in PGNAA for explosives detection.
• To develop a deep neural network framework to improve PGNAA accuracy.
• To enable reliable explosives identification irrespective of shielding conditions.

Main Methods:

• Utilized deep neural networks (DNNs) as an end-to-end framework.
• Trained DNNs on gamma spectrum data from Prompt Gamma Neutron Activation Analysis.
• Evaluated the DNN model's performance on known and previously unseen explosive samples.

Main Results:

• The DNN framework achieved 95% accuracy in differentiating explosives from non-explosives in the training dataset.
• The model demonstrated 80% accuracy in generalizing to explosives not included during development.
• The approach effectively mitigated the impact of shielding on PGNAA analysis.

Conclusions:

• Coupling PGNAA with deep neural networks offers a robust solution for explosives detection.
• This integrated approach shows significant potential for high-accuracy identification, even with unknown shielding.
• DNNs enhance the reliability and applicability of PGNAA in security applications.