Real-time radionuclide identification in γ-emitter mixtures based on spiking neural network

Area of Science:

• Nuclear Physics
• Radiation Detection and Measurement
• Artificial Intelligence in Security

Background:

• International borders require rapid detection of illicit nuclear materials.
• Traditional off-line gamma-ray spectrometry is time-consuming for real-time threat assessment.
• Existing methods struggle with complex gamma-emitter mixtures and low-statistics spectra.

Purpose of the Study:

• To develop and validate a real-time radionuclide identification algorithm for portal radiation monitors.
• To enable faster detection of radiological threats at international borders.
• To overcome limitations of traditional off-line spectrum analysis.

Main Methods:

• Development of a real-time processing algorithm utilizing artificial neural networks and Bayes' rule.
• Validation using gamma-emitter spectra from Am-241, Ba-133, Bi-207, Co-60, and Cs-137.
• Utilizing a high-efficiency well-type NaI(Tl) detector for spectral acquisition.
• Employing an iterative, on-line analysis approach requiring low-statistics spectra.

Main Results:

• The proposed algorithm successfully achieved fast identification of individual gamma-emitting radionuclides.
• The method effectively utilizes information from the entire spectrum for identification.
• The real-time approach demonstrated efficacy even with low-statistics spectra and without energy calibration.

Conclusions:

• The artificial neural network and Bayes' rule-based real-time analysis offers a viable alternative for rapid radionuclide identification.
• This method significantly enhances the speed of radiological threat assessment at borders.
• The algorithm's ability to work with low-statistics spectra makes it practical for on-site, real-time monitoring.