基于SVM的光谱信息分析

电气自动化论文编号:ZD905    论文字数:13711,页数:33  
 目录
摘要 II
1   绪论 2
1.1几种测量法简介 2
1.2与化学方法相比光谱分析法的优缺点 3
1.3定量校正方法 4
 1.3.1  主成分回归 6
1.3.2  偏最小二乘 7
1.3.3 人工神经网路 7
1.3.4、支持向量机 9
2   SVM原理具体介绍 11
2.1 SVM背景介绍 11
2.2支持向量机方法的优点 11
2.3 SVM原理介绍 11
2.3.1　线性可分情况 12
2.3.2　线性不可分的情况 14
2.3.3  内积核函数 14
2.4支持向量回归原理 15
2.5核函数的意义 15
3   实验过程及相关数据 16
3.1实验平台 16
3.2实验参数选择 17
3.3 SVM回归分析 18
3.3实验结果 18
4   结论 22
致谢 23
参考文献 24

摘要
 近红外光谱分析技术是一种新的化学分析方法，具有无污染、无破坏、分析速度快、效率高、成本低及可以实现在线分析等特点，在食品、医药、化工、石油等领域获得了空前的发展，并且其涉及的领域愈来愈广。为此，本文对近红外光谱定性分析技术作了深入研究，具体包括以下内容:
 1.通过阅读大量的中英文文献，对近红外光谱分析技术的概念、原理与应用作了较为系统完整的阐述，对模式识别技术和定量校正方法在近红外分析中的应用作了细致地描述。
 2.本文将支持向量机(SVM)分类算法成功应用到汽油牌号快速识别中。与其他分类算法如K一近邻法(KNN)、相似分析法(SIMCA)的比较结果表明，SVM法分类的效果最好，其最大分类错误率很低，运行结果很稳定，受样本变化、参数变化等影响很小，具有广泛的应用价值。
 3.为了提高定量分析的精度，本文提出了混合偏最小二乘  (PartialLeastsquares，PLS)法。首先用svM法对测试样本进行分类，然后选用与待测样本性质相近的同类部分校正集样品建模来预测待测属性值。为了克服分类错误样本的影响，混合PLS法包含了基于分类的局部PLS法和基于全部训练样本集的局部PLS法两种算法。混合PLS法通过比较两种局部PLS法的输出，计算测试样本的待测属性值。针对一批汽油样本的实验结果表明，该算法对汽油各项参数的预测效果达到了较高的精度。
 4.基于上述理论成果，针对已研制的汽油各项参数近红外光谱分析仪，并把提出的混合PLS近红外光谱定量分析法应用于汽油各项参数的定量预测中，提高了仪器的性能。
 

关键词  近红外光谱，支持向量机，偏最小二乘, 定量校正，分类。

Abstract
 Near infrared（NIR） spectroscopy is a new chemical analysis method, with no pollution, no damage, rapid analysis, high efficiency, low cost and so on-line analysis can be achieved, in food, medicine, chemical industry, oil fields obtained unprecedented development, and its increasingly wide area involved. Therefore, this article on qualitative analysis by near infrared spectroscopy depth research, specifically including the following:
By a lot of reading in English and Chinese literature, the concept of near infrared spectroscopy, principles and application made a more systematic and complete exposition of the correction pattern recognition technology and quantitative methods in the near infrared analysis were careful to describe .
This will support vector machine (SVM) classification algorithm is successfully applied to the fast recognition of brands of gasoline. With other classification algorithms such as K-Nearest Neighbor (KNN), similar analysis (SIMCA) in the comparison shows that, SVM classification method was the best, the maximum classification error rate is very low, the results are stable and changes by the sample, the parameter changes have little effect, with a wide range of application.
In order to improve the accuracy of quantitative analysis, this paper presents a hybrid partial least squares (PartialLeastsquares, PLS) method. First, the test samples with svM method to classify, and then use a similar nature and applied to the sample part of the calibration samples similar model to predict property values tested. In order to overcome the impact of classification error samples, mixed PLS method contains the local PLS-based classification method and all the training set based on the local PLS method two algorithms. Mixed PLS method by comparing the output of two local PLS method to calculate the value of the tested samples tested property. Number of gas samples for experimental results show that the algorithm of the parameters of gasoline reached a high predictive accuracy.
Based on the above theoretical results, for the parameters of gasoline has been developed near-infrared spectrometer, and the proposed hybrid PLS Near Infrared Spectroscopy Quantitative Analysis of the parameters used in gasoline in the quantitative prediction to improve the performance of the instrument.

Keywords: NearinfraredsPeetroseoPy，quantitativeealibration，Partialleastsquares
suPPortveetormaehine，  elassifieation，