Reconstruction of Compton Edges in Plastic Gamma Spectra Using Deep Autoencoder

Area of Science:

• Nuclear Instrumentation
• Spectroscopy
• Artificial Intelligence

Background:

• Plastic scintillation detectors are valuable for radiation measurement but limited to counting applications due to energy broadening and lack of photo peaks.
• Existing pseudo spectroscopy methods struggle with direct identification of incident gamma ray energies.
• Overcoming these limitations is crucial for enhancing the utility of plastic scintillators in spectroscopy.

Purpose of the Study:

• To propose and validate a method for direct pseudo gamma spectroscopy using plastic scintillators.
• To reconstruct Compton edges in plastic gamma spectra using an artificial neural network (ANN).
• To enable accurate identification of incident gamma ray energies despite detector limitations and poor counting statistics.

Main Methods:

• Developed an ANN model trained on MCNP 6.2 simulated spectra to reconstruct Compton edges.
• Designed a specific dataset generation procedure to ensure model performance with low counting statistics.
• Evaluated minimum reconstructible counts for isotopes like 60Co, 137Cs, 22Na, and 133Ba using mean averaged percentage error.

Main Results:

• The ANN model successfully reconstructed Compton edges in simulated plastic gamma spectra.
• The model demonstrated capability in reconstructing Compton edges even for spectra with poor counting statistics.
• Performance was validated using simulated spectra from a PVT detector, showing successful reconstruction on measured data.

Conclusions:

• The proposed ANN method enables direct pseudo gamma spectroscopy with plastic scintillators by reconstructing Compton edges.
• This approach overcomes the inherent limitations of plastic scintillators, allowing for incident gamma ray energy identification.
• The method shows promise for applications requiring sensitive radiation detection and spectral analysis, even in low-count scenarios.