基于SVM的白酒红外光谱分析方法研究  

Research on SVM-based liquor infrared spectrum analysis method

作　　者：姜安[1] 彭江涛[1] 彭思龙[1] 魏纪平[2] 李长文[2] 

机构地区：[1]中国科学院自动化研究所,北京100190 [2]天津天士力集团食品研究所,天津300410

出　　处：《计算机与应用化学》2010年第2期233-236,共4页Computers and Applied Chemistry

基　　金：国家科技支撑计划重点项目(2006BAK07B04)

摘　　要：为实现白酒品评自动化,采集了297个不同香型、86个不同等级、60个不同年份的白酒样品红外光谱图,共计443个。针对这些红外光谱图,采用3次多项式插值拟合的方法进行基线漂移校正,并用小波软阈值法去除光谱噪声,然后用标准归一化的方法消除散射效应。对于白酒的香型、等级和年份这3种不同的分类问题,分别选择样本的75%为训练集,余下25%为测试集,利用支持向量机(SVM)方法建立对应的香型、等级和年份分类模型,并在测试集上验证了模型的分类性能。实验结果表明该方法行之有效,香型分类正确率达到98%以上,等级分类正确率达到92%以上,年份分类正确率达到100%。In order to taste liquor automatically, spectra of 443 liquor samples were collected, consisted of 297 samples of different flavor, 86 samples of different grade, and 60 samples of different year. Cubic polynomial fitting method is used to correct baseline drift, wavelet soft threshold based method is applied to reduce spectral noise, and standard normalization method is adopted to eliminate seattering effect, respectively. Seventy five percent of the samples are selected as the training set and the left as testing set. SVM-based liquor flavor, grade and year classification models are established on the training sets and used to detect classification performance on testing sets, respectively. Experimental results show the proposed method is effective, and high classification accuracy is achieved, which is 98% to different flavor liquor, 92% to different grade liquor, and 100% to different year liquor.

关 键 词：红外光谱 支持向量机 白酒香型 白酒等级 白酒年份 

分 类 号：O657.3[理学—分析化学] TQ015.9[理学—化学]