Kim Jinhwan, Park Kyeongjin, Cho Gyuseong
Department of Nuclear & Quantum Engineering, Korea Advanced Institute of Science and Technology, Republic of Korea.
Department of Nuclear & Quantum Engineering, Korea Advanced Institute of Science and Technology, Republic of Korea.
Appl Radiat Isot. 2019 May;147:83-90. doi: 10.1016/j.apradiso.2019.01.005. Epub 2019 Feb 10.
Radioisotope identification using a plastic scintillation detector has been a challenging issue because of the poor spectral resolution and low cross-sections of these types of detectors when used for photoelectric absorption. In this paper, we propose an algorithm that identifies a single radioisotope and multiple radioisotopes from the gamma spectrum of a plastic scintillator using an artificial neural network. The spectra were simulated using Monte Carlo N-Particle Transport Code 6 to formulate the training set, and the spectra were measured by a two-inch EJ-200 to create the test set (1440 spectra in total). The ANN-based algorithm presented here ensures an identification accuracy of 98.9% for a single radioisotope and 99.1% for multiple radioisotopes. Even if the spectra were intentionally shifted by 36 keV for low and high energies, the trained ANN predicts radioisotopes with high accuracy. In addition, we have determined the minimal required number of detected counts to identify the radioisotope with 5% false negative and false positive.
使用塑料闪烁探测器进行放射性同位素识别一直是一个具有挑战性的问题，因为这类探测器在用于光电吸收时光谱分辨率较差且截面较低。在本文中，我们提出了一种算法，该算法使用人工神经网络从塑料闪烁体的伽马能谱中识别单一放射性同位素和多种放射性同位素。使用蒙特卡罗N粒子输运代码6模拟能谱以形成训练集，并通过两英寸的EJ - 200测量能谱以创建测试集（总共1440个能谱）。这里提出的基于人工神经网络的算法确保单一放射性同位素的识别准确率为98.9%，多种放射性同位素的识别准确率为99.1%。即使能谱在低能和高能情况下故意偏移36 keV，经过训练的人工神经网络仍能高精度地预测放射性同位素。此外，我们已经确定了以5%的假阴性和假阳性识别放射性同位素所需的最少检测计数。
