A new radionuclide identification method for low-count energy spectra with multiple radionuclides

Area of Science:

• Nuclear Physics
• Spectroscopy
• Machine Learning

Background:

• Radionuclide identification is crucial for nuclear security, relying on energy spectrum analysis.
• Traditional methods struggle with low-count and mixed nuclide spectra due to background subtraction complexities.

Purpose of the Study:

• To develop an advanced radionuclide identification method overcoming limitations of traditional techniques.
• To improve accuracy and reliability in complex environmental and mixed nuclide scenarios.

Main Methods:

• Proposed a new method employing a feature enhancer and a one-dimensional neural network.
• Utilized Geant4-simulated data for training the neural network model.
• Preprocessed energy spectrum data to extract nonlinear information.

Main Results:

• The method demonstrated strong performance on experimental low-count energy spectra.
• Achieved high recognition accuracy and minimal misjudgments for mixed radionuclide spectra.
• Successfully deployed in portable instruments for real-time radionuclide identification.

Conclusions:

• The proposed feature enhancer and neural network approach offers a robust solution for radionuclide identification.
• This method significantly enhances capabilities in challenging low-count and mixed nuclide detection scenarios.
• Real-time deployment in portable instruments highlights its practical applicability in nuclear security.