傅里叶近红外光谱仪模型传递及药品鉴别方法研究

发布时间：2018-07-01 15:43

本文选题：傅里叶近红外光谱仪 + 模型传递　；参考：《北京邮电大学》2017年硕士论文

【摘要】：药品关系人民健康,真假药鉴别和药品种类鉴别,在药品监督中有强烈的应用需求。傅里叶近红外光谱仪是一种光机电结合的精密测量装置,具有现场、快速、无损检测等优点,结合统计学或化学计量学方法,常用于各类物理化学值的测量,也于近年成为我国药品流动检测车中的必配装备。在药品鉴别应用中,上百台仪器常同时使用,因此本文研究台间差产生的原因并给出模型传递方法,并重点研究两类和多类的药品鉴别。本文首先介绍近红外光谱仪的分类和傅里叶变换的工作原理,以及近红外光谱分析应用的基本流程,然后介绍了小波变换光谱预处理方法,以及自编码网络等光谱特征提取方法的基本原理。本文接着介绍了傅里叶近红外光谱仪的核心——迈克尔逊干涉仪的机械结构,并分析了光谱检测误差产生的机械和环境因素。研究了将小波变换光谱预处理方法与一元线性回归直接标准化算法(SLRDS)结合的模型传递方法,实验结果表明,引入小波变换可更好地消除仪器机械和环境因素带来的测量误差,提升模型传递效果。本文提出一种稀疏降噪自编码结合高斯过程的药品鉴别二分类算法wSDAGsM。该算法首先对光谱数据进行一维小波连续变换,然后应用稀疏降噪自编码结合高斯过程进行二分类。实验结果表明,本文提出的建模方法wSDAGsM,对比BP神经网络等算法,在分类准确率及稳定性方面,均取得了更优的结果。同时,实验也表明小波变换可以较好地消除光谱噪声。本文提出一种稀疏降噪自编码结合支持向量机(SVM)的药品鉴别二分类和多分类算法wSDAMRBF。该算法首先对光谱数据进行一维小波连续变换,然后用稀疏降噪自编码结合SVM进行二分类和多分类。本文对wSDAGSM和wSDAMRBF算法开展了对比实验研究,结果表明,两个算法都能较好地用于药品鉴别,相对而言,wSDAMRBF算法在分类准确率和结果稳定性更优。
[Abstract]:Drugs are closely related to people's health, genuine and false drugs and drug types, which have a strong demand for application in drug supervision. Fourier near Infrared Spectrometer (FNIR) is a kind of precision measuring device combined with light and electromechanical. It has the advantages of field, fast and nondestructive testing. It is often used in the measurement of various physical and chemical values combined with statistics or chemometrics. In recent years, it has become the necessary equipment in the mobile drug testing vehicle in China. In drug identification applications, hundreds of instruments are often used at the same time, so this paper studies the causes of the difference between stations and gives the method of model transfer, and focuses on the identification of two or more kinds of drugs. This paper first introduces the classification of near infrared spectrometer and the working principle of Fourier transform, and the basic flow of near infrared spectrum analysis and application, then introduces the pretreatment method of wavelet transform spectrum. And the basic principle of spectral feature extraction method such as self-coding network. In this paper, the mechanical structure of Michelson interferometer, which is the core of Fourier near infrared spectrometer, is introduced, and the mechanical and environmental factors of spectrum detection error are analyzed. The model transfer method which combines wavelet transform spectral pretreatment method with linear regression direct standardization algorithm (SLRDS) is studied. The experimental results show that wavelet transform can better eliminate the measurement errors caused by mechanical and environmental factors. Improved model delivery effect. In this paper, a novel two-classification algorithm for drug identification, wSDAGsMbased on sparse noise reduction self-coding and Gao Si process, is proposed. The algorithm firstly performs one-dimensional wavelet continuous transform for spectral data and then uses sparse noise reduction self-coding and Gao Si process to classify the spectral data. The experimental results show that the proposed modeling method wSDAGsM, compared with BP neural network, has better results in classification accuracy and stability. At the same time, the experiment also shows that wavelet transform can eliminate spectral noise. In this paper, a sparse denoising self-coding and support vector machine (SVM) algorithm for drug identification is proposed. The algorithm firstly performs one-dimensional wavelet continuous transform for spectral data, and then uses sparse denoising self-coding and SVM to carry out two-classification and multi-classification. In this paper, a comparative study of wSDAGSM and wSDAMRBF algorithms is carried out. The results show that both algorithms can be used for drug identification, and the classification accuracy and stability of wSDAMRBF algorithm are better than that of wSDAMRBF algorithm.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：O657.33;TQ460.72

【参考文献】