Deep learning approach for airborne alpha radioactivity monitoring in atypical atmospheric conditions

In nuclear facilities, the mandatory monitoring of airborne alpha radioactivity contamination is carried out by dedicated instruments that collect aerosols on a filter, measure the deposited radioactivity and trigger an alarm when a predetermined activity threshold is exceeded. The radioactivity measurement is highly influenced by variations in aerosol size and concentration on the filter, leading to numerous false alarms. In order to overcome this difficulty, we are interested in using artificial intelligence to automatically compensate the background noise and hence obtain precise information on the presence of artificial alpha emitters based on the alpha-particle spectrum. The ultimate aim is to reduce the false alarm rate. • Continuous Air Monitors cause numerous false alarm in dusty atmospheric conditions. • A convolutionnal neural network is trained to classify simulated data. • The neural network significantly improve classification performances of the monitor. • The application of deep learning demonstrate its usefulness and is set to be refined.