Preprocessing Energy Intervals on Spectrum for Real-Time Radionuclide Identification

In this study, we present a preprocess method using radiation energy intervals on a gamma-ray spectrum based on a deep learning algorithm to achieve real-time radionuclide identification. Data preprocessing is performed by classifying energy intervals, distinctly corresponding to pulse amplitudes of each radiation measurement system. Since the energy intervals are distinguished with noise, backscatter area, Compton edge, and photopeaks depending on radionuclides, raw data are sorted in each interval in preprocessed dataset using a deep learning algorithm. Using ⁶⁰ Co, ¹³⁷ Cs, and the energy interval preprocessing, the multi-source identification shows 100% accuracy in 2000 measured data compared with 70% accuracy for those without the preprocessing method. The measured time is 72 s for 2000 test data, dramatically reduced from the conventional data collection time of 60 min for 100 000 data. The proposed approach reduces the minimum number of data to identify radionuclides before visualizing the spectrum. With the preprocess method, radionuclide identification is completed in tens of seconds, applicable for low radiation activity areas such as decommissioning reactor sites.