FEASIBILITY STUDY ON THE FUSION OF PHITS SIMULATIONS AND THE DLNN ALGORITHM FOR A NEW QUANTITATIVE METHOD OF IN-SITU MULTIPLE-CHANNEL DEPTH DISTRIBUTION SPECTROMETRY

Abstract We have recently have developed an in-situ multiple-channel depth distribution spectrometer (DDS) that can easily acquire on-site measurements of the depth distribution of specific radioactivities of Cs-134 and Cs-137 underground. Despite considerable improvements in the hardware developed for this device, the quantitative method for determining of radioactivities with this DDS device cannot yet achieve satisfactory performance for practical use. For example, this method cannot discriminate each γ-ray spectra of Cs-134 and Cs-137 acquired by the 20 thallium-doped caesium iodine CsI(Tl) scintillation crystal detectors of the DDS device from corresponding depth levels of underground soil. Therefore, we have applied deep learning neural network (DLNN) as a novel radiation measurement technique to discriminate the spectra and to determine the specific radioactivities of Cs-134 and Cs-137. We have developed model soil layers on a virtual space in Monte-Carlo based PHITS simulations and transported γ-ray radiation generated from a particular single soil layer or multiple layers as radiation sources; next, we performed PHITS calculations of those specific radioactivity measurements for each soil layer using DDS device based on machine learning via the DLNN algorithm. In this study, we obtained informative results regarding the feasibility of the proposal innovative radiation measurement method for further practical use in on-site applications.