How to achieve auto-identification in Raman analysis by spectral feature extraction & Adaptive Hypergraph
How to achieve auto-identification in Raman analysis by spectral feature extraction & Adaptive Hypergraph
复制标题
如何通过光谱特征提取实现拉曼分析中的自动识别
DOI：
10.1016/j.saa.2019.04.078
复制
发表时间：
2019
期刊：
影响因子：
--
通讯作者：
Yongming Zeng
中科院分区：
文献类型：
--
作者：
Yi Xie;Qiaobei You;Pingyang Dai;Shuyi Wang;Peiyi Hong;Guokun Liu;Jun Yu;Xilong Sun;Yongming Zeng
随着拉曼光谱仪的小型化，拉曼光谱（包括表面增强拉曼光谱）已被广泛应用于各个领域，特别是朝向快速检测应用。为了处理伴随而来的海量数据库，需要以有效和自动的方式处理和识别大量的拉曼光谱。提出了一种利用机器学习方法对拉曼光谱进行分析的材料自动识别算法。首先，设计了一种通用的光谱特征提取方法，对背景扣除后的拉曼光谱进行自动处理。然后，利用机器学习领域中的一种高效分类器--自适应超图（AH），将提取的特征向量用于目标材料的自动分类识别，准确率达到99%。与支持向量机（SVM）和随机森林（RF）两种典型的分类方法相比，AH分类器具有更好的性能，无需针对不同的目标调整任何参数。第三，引入三次样条插值，以提高算法在不同厂家、不同型号的拉曼光谱仪数据库之间的通用性。分别以高频采样谱作为学习谱和低频采样谱作为测试谱，识别正确率达到98%。
With the miniaturization of Raman spectrometers, Raman spectroscopy (including Surface-enhanced Raman spectroscopy) has been widely applied to various fields, especially towards rapid detection applications. In order to deal with the accompanied massive databases, large numbers of Raman spectra require to be handled and identified in an effective and automatic manner. This paper proposes an algorithm of material auto-identification, which makes use of machine learning methods to analyze Raman spectra. Firstly, a universal method of spectral feature extraction is designed to automatically process Raman spectra after the background subtraction. Secondly, the extracted feature vectors are used to classify and identify target materials by Adaptive Hypergraph (AH), an efficient classifier in the field of machine learning, in a manner of automation with an accuracy rate of ~99%. Compared with Support Vector Machine (SVM) and Random Forest (RF), two typical methods of classification, the AH classifier provides better performance free of tuning any parameter facing different targets. Thirdly, Cubic Spline Interpolation is introduced to enhance the universal of the proposed algorithm between different databases from different Raman spectrometers with variant vendors. The identification accuracy rate is up to 98% using the high frequency sampling spectra as the learning and the low frequency sampling ones as the testing, respectively.