241Am-9Be 数据集和 238Pu-9Be 数据集
收藏资源简介:
本文介绍了用于辐射检测的脉冲形状鉴别(PSD)算法的综合调查和基准测试,将近六十种方法分为统计(时域、频域、基于神经网络的)和先验知识(机器学习、深度学习)两大类。论文在两个标准化数据集上实现了所有算法并进行评估:一个来自 241Am-9Be 源的无标签数据集和一个来自 238Pu-9Be 源的时间飞行标签数据集。评估指标包括性能指标(FOM)、F1 分数、ROC-AUC 和方法间相关性。研究结果表明,深度学习模型,特别是多层感知器(MLPs)和结合统计特征与神经回归的混合方法,通常优于传统方法。
This paper presents a comprehensive survey and benchmarking of pulse shape discrimination (PSD) algorithms for radiation detection. Nearly sixty methods are categorized into two major classes: statistical methods (including time-domain, frequency-domain, and neural network-based approaches) and prior knowledge-based methods (covering machine learning and deep learning techniques). All algorithms are implemented and evaluated on two standardized datasets: an unlabeled dataset from a 241Am-9Be source, and a time-of-flight labeled dataset from a 238Pu-9Be source. Evaluation metrics include figure of merit (FOM), F1-score, ROC-AUC, and inter-method correlation. The study's findings demonstrate that deep learning models, particularly multi-layer perceptrons (MLPs) and hybrid methods combining statistical features and neural regression, generally outperform traditional approaches.

- 1Pulse Shape Discrimination Algorithms: Survey and Benchmark成都理工大学 · 2025年



