基于SVM的光譜信息分析

電氣自動化論文編號:ZD905    論文字數:13711,頁數:33  
 目錄
摘要 II
1   緒論 2
1.1幾種測量法簡介 2
1.2與化學方法相比光譜分析法的優缺點 3
1.3定量校正方法 4
 1.3.1  主成分回歸 6
1.3.2  偏最小二乘 7
1.3.3 人工神經網路 7
1.3.4、支持向量機 9
2   SVM原理具體介紹 11
2.1 SVM背景介紹 11
2.2支持向量機方法的優點 11
2.3 SVM原理介紹 11
2.3.1　線性可分情況 12
2.3.2　線性不可分的情況 14
2.3.3  內積核函數 14
2.4支持向量回歸原理 15
2.5核函數的意義 15
3   實驗過程及相關數據 16
3.1實驗平臺 16
3.2實驗參數選擇 17
3.3 SVM回歸分析 18
3.3實驗結果 18
4   結論 22
致謝 23
參考文獻 24

摘要
 近紅外光譜分析技術是一種新的化學分析方法，具有無污染、無破壞、分析速度快、效率高、成本低及可以實現在線分析等特點，在食品、醫藥、化工、石油等領域獲得了空前的發展，并且其涉及的領域愈來愈廣。為此，本文對近紅外光譜定性分析技術作了深入研究，具體包括以下內容:
 1.通過閱讀大量的中英文文獻，對近紅外光譜分析技術的概念、原理與應用作了較為系統完整的闡述，對模式識別技術和定量校正方法在近紅外分析中的應用作了細致地描述。
 2.本文將支持向量機(SVM)分類算法成功應用到汽油牌號快速識別中。與其他分類算法如K一近鄰法(KNN)、相似分析法(SIMCA)的比較結果表明，SVM法分類的效果最好，其最大分類錯誤率很低，運行結果很穩定，受樣本變化、參數變化等影響很小，具有廣泛的應用價值。
 3.為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法對測試樣本進行分類，然后選用與待測樣本性質相近的同類部分校正集樣品建模來預測待測屬性值。為了克服分類錯誤樣本的影響，混合PLS法包含了基于分類的局部PLS法和基于全部訓練樣本集的局部PLS法兩種算法�；旌螾LS法通過比較兩種局部PLS法的輸出，計算測試樣本的待測屬性值。針對一批汽油樣本的實驗結果表明，該算法對汽油各項參數的預測效果達到了較高的精度。
 4.基于上述理論成果，針對已研制的汽油各項參數近紅外光譜分析儀，并把提出的混合PLS近紅外光譜定量分析法應用于汽油各項參數的定量預測中，提高了儀器的性能。
 

關鍵詞  近紅外光譜，支持向量機，偏最小二乘, 定量校正，分類。

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，