Nuclide Identification Algorithm for the Large-Size Plastic Detectors Based on Artificial Neural Network

Hiep, Cao Van; Anh, Nguyen Ngoc; Khang, Pham Dinh; Hải, Nguyễn Xuân; Nguyen, Kien-Cuong; Hung, Dinh Tien; Chuan, Phan Van

doi:10.1109/tns.2022.3173371

Cited by 8 publications

(1 citation statement)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Intelligent nuclide identification methods have received more and more attention. Bayesian inference [3], support vector machines [4], and artificial neural networks [5][6][7][8][9][10][11] are applied to nuclide identification and quantification. Based on artificial neural networks, some variants of neural networks, such as convolutional neural network (CNN) [12][13][14][15], hybrid neural network [16], and generative adversarial network [17], have been applied to nuclide identification.…”

Section: Introductionmentioning

confidence: 99%

Multiple radionuclide identification using deep learning with channel attention module and visual explanation

et al. 2022

View full text Add to dashboard Cite

As a rapid and automatic method, multiple radionuclide identification using deep learning has drawn wide interest from researchers in the field of nuclear safety and nuclear security. However, the network model in deep learning often appears in the form of a black box, which makes it difficult for people to understand its decision-making basis. It is necessary to develop an interpretable deep learning model for multiple nuclide identification. In the work on nuclide identification using deep learning, very few interpretable studies have been conducted. In this paper, channel attention weights are used for interpretable radionuclide identification for the first time. We propose a multiple radionuclide identification method using deep learning with channel attention module and visual explanation. A dataset of gamma spectra simulated by Geant4 was created, containing 256 combinations of 8 radionuclides. These gamma spectra were used to train using a convolutional neural network (CNN) with a channel attention module. The obtained accuracies on training, validation, and test sets are 97.8%, 97.6%, and 97.1%, respectively. The result of interpretation of spectral features show that based on the channel attention module, the CNN can make full use of the feature information of the photoelectric peak and Compton edge and suppress the background and noise interference. In addition, the t-distributed stochastic neighbor embedding (t-SNE) method was used to visualize the inner working process of the CNN and visually illustrate the correctness of feature extraction. This research will promote the application of artificial intelligence algorithms in nuclide identification instruments.

show abstract