Trapezoidal pile-up nuclear pulse parameter identification method based on deep learning transformer model. (December 2022)

• Record Type: Journal Article
• Authors: Wang, Qingtai; Huang, Hongquan; Ma, Xingke; Shen, Zhiwen; Zhong, Chenglin; Ding, Weicheng; Zhou, Wei; Zhou, Jianbin
• Is Part Of: Applied radiation and isotopes. Volume 190(2022)
• Publication Date: 2022-12
• Subjects: Nuclear pulse -- Attention mechanism -- Transformer; Radiology -- Periodicals; Radiation -- Industrial applications -- Periodicals; Nuclear chemistry -- Periodicals; Internet resource; Periodical; 660.298
• DOI: 10.1016/j.apradiso.2022.110515
• Languages: English
• ISSNs: 0969-8043

Abstract

Abstract: Pile-up between adjacent nuclear pulses is unavoidable in the actual detection process. Some scholars have tried to apply deep learning techniques to identify pile-up nuclear pulse parameters. However, traditional deep learning recurrent neural networks (RNNs) suffer from inefficient pulse recognition and poor recognition of pile-up nuclear pulses with short intervals between adjacent pulses. In this paper, a Transformer model with an attention mechanism as the core to recognize pile-up nuclear pulses is innovatively applied, aiming to provide a more accurate and efficient method for pile-up nuclear pulse recognition. Thus, it gives a better help for the spectrum correction with a high count rate. Highlights: Using the more advanced Transformer model in the field of deep learning. Short intervals of pile-up pulses can be identified. The recognition accuracy has been improved compared to previous studies.