Gamma-ray anomaly detector for airborne surveys based on a machine learning methodology

Area of Science:

• Nuclear geophysics
• Machine learning applications
• Radiation detection

Background:

• Airborne gamma-ray spectrometry is crucial for environmental monitoring and radiation source detection.
• Automated analysis of large spectral datasets is needed for efficient screening.
• Distinguishing anomalous signatures from background noise is a significant challenge.

Purpose of the Study:

• To develop and validate an automated classification model for detecting anomalous gamma-ray spectra.
• To enhance the capability for identifying non-background radiation signatures in airborne survey data.
• To support first responders in locating radioactive materials and managing environmental incidents.

Main Methods:

• Machine learning, specifically piecewise linear discriminant analysis (PLDA), was employed.
• Spectra underwent altitude-based normalization and digital filtering for noise reduction.
• A training set comprised radioisotope signatures and background spectra for model construction.

Main Results:

• The developed spectral anomaly detector successfully located known radioisotope sources in 17 airborne surveys with high confidence.
• A false detection rate of 3.7% was observed, with many near actual sources or attributable to background variations.
• The model demonstrated effectiveness in automated screening of large spectral datasets.

Conclusions:

• The automated spectral anomaly detection model is a valuable tool for airborne survey data analysis.
• This methodology can significantly aid in the rapid identification of radioactive sources and environmental releases.
• Further refinement could improve discrimination between true anomalies and complex background signals.