Spectral analysis was a method of identifying substances, determining their chemical composition and calculating their content based on their spectral characteristics. This paper mainly discussed the application of va...Spectral analysis was a method of identifying substances, determining their chemical composition and calculating their content based on their spectral characteristics. This paper mainly discussed the application of various spectroscopic techniques, mainly including atomic absorption spectrometry (AAS) inductively coupled plasma emission spectrometry (ICP-AES) X-ray fluorescence spectroscopy (XRF) atomic fluorescence spectroscopy (AFS) direct reading spectroscopy (OES) glow discharge emission spectroscopy (GD-OSE) laser-induced breakdown spectroscopy (LIBS), in the formulation of non-ferrous metal standards in China. The AAS method was the most widely used single-element microanalysis method among the non-ferrous metal standards. The ICP-AES method was good at significant advantages in the simultaneous detection of multiple elements. The XRF method was increasingly used in the determination of primary and secondary trace elements due to its simple sample preparation and high efficiency. The AFS was mostly detected by single-element trace analysis. OES GD-OES and LIBS were playing an increasingly important role in the new demand area for non-ferrous metals. This paper discussed matrix elimination, sample digestion, sample preparation, instrument categories and other aspects of some standards, and summarized the advantages of spectral analysis and traditional chemical analysis methods. The new methods of future spectroscopic technology had been illustrated in the process of developing non-ferrous metal standards.展开更多
在X荧光光谱分析中,为了解决传统谱分析方法中存在的特征峰计数率损失以及影子峰的问题,本文拟采用一种基于深度学习的长短期记忆(Long and Short Term Memory,LSTM)神经网络模型,该模型对核脉冲幅度时间序列具有较好的适用性,通过对样...在X荧光光谱分析中,为了解决传统谱分析方法中存在的特征峰计数率损失以及影子峰的问题,本文拟采用一种基于深度学习的长短期记忆(Long and Short Term Memory,LSTM)神经网络模型,该模型对核脉冲幅度时间序列具有较好的适用性,通过对样本的学习能够对核脉冲信号的幅度进行准确估计。鉴于核脉冲信号样本较大,模型训练效率低,特引入卷积神经网络(Convolutional Neural Network,CNN),利用其特有的卷积核结构逐层提取样本特征,能够有效减少样本数量,降低模型训练复杂度。使用粉末铁矿样品测量得到的一系列离线核脉冲序列产生模型训练所需的数据集,该数据集的64000个条目中,44800个用作训练集,12800个用作验证集,余下6400个用作测试集。实验结果表明:训练好的CNN-LSTM模型能够极大地节省训练时间,克服传统方法局部收敛的缺陷,也能够对不同程度畸变的输入脉冲进行准确的参数估计,在训练集和验证集上得到的准确率都高于99%。进一步分析计数修复结果,得到三个影子峰校正比例的平均值为91.52%,表明训练的CNN-LSTM模型对畸变脉冲产生的计数损失的校正比例约为91.52%。该模型能够有效校正因畸变脉冲幅度损失造成的影子峰,改善X射线荧光光谱中特征峰计数率精度,在X射线荧光光谱领域具有较高的应用价值。展开更多
文摘Spectral analysis was a method of identifying substances, determining their chemical composition and calculating their content based on their spectral characteristics. This paper mainly discussed the application of various spectroscopic techniques, mainly including atomic absorption spectrometry (AAS) inductively coupled plasma emission spectrometry (ICP-AES) X-ray fluorescence spectroscopy (XRF) atomic fluorescence spectroscopy (AFS) direct reading spectroscopy (OES) glow discharge emission spectroscopy (GD-OSE) laser-induced breakdown spectroscopy (LIBS), in the formulation of non-ferrous metal standards in China. The AAS method was the most widely used single-element microanalysis method among the non-ferrous metal standards. The ICP-AES method was good at significant advantages in the simultaneous detection of multiple elements. The XRF method was increasingly used in the determination of primary and secondary trace elements due to its simple sample preparation and high efficiency. The AFS was mostly detected by single-element trace analysis. OES GD-OES and LIBS were playing an increasingly important role in the new demand area for non-ferrous metals. This paper discussed matrix elimination, sample digestion, sample preparation, instrument categories and other aspects of some standards, and summarized the advantages of spectral analysis and traditional chemical analysis methods. The new methods of future spectroscopic technology had been illustrated in the process of developing non-ferrous metal standards.
文摘在X荧光光谱分析中,为了解决传统谱分析方法中存在的特征峰计数率损失以及影子峰的问题,本文拟采用一种基于深度学习的长短期记忆(Long and Short Term Memory,LSTM)神经网络模型,该模型对核脉冲幅度时间序列具有较好的适用性,通过对样本的学习能够对核脉冲信号的幅度进行准确估计。鉴于核脉冲信号样本较大,模型训练效率低,特引入卷积神经网络(Convolutional Neural Network,CNN),利用其特有的卷积核结构逐层提取样本特征,能够有效减少样本数量,降低模型训练复杂度。使用粉末铁矿样品测量得到的一系列离线核脉冲序列产生模型训练所需的数据集,该数据集的64000个条目中,44800个用作训练集,12800个用作验证集,余下6400个用作测试集。实验结果表明:训练好的CNN-LSTM模型能够极大地节省训练时间,克服传统方法局部收敛的缺陷,也能够对不同程度畸变的输入脉冲进行准确的参数估计,在训练集和验证集上得到的准确率都高于99%。进一步分析计数修复结果,得到三个影子峰校正比例的平均值为91.52%,表明训练的CNN-LSTM模型对畸变脉冲产生的计数损失的校正比例约为91.52%。该模型能够有效校正因畸变脉冲幅度损失造成的影子峰,改善X射线荧光光谱中特征峰计数率精度,在X射线荧光光谱领域具有较高的应用价值。
