摘要
针对采用多类分类方法进行白酒真假识别时存在的真酒样本和假酒样本（正类样本和异类样本）数量无法均衡以及异类样本无法全部获取的问题,提出应用单类支持向量机分别单独对每一种品牌的白酒训练单类分类器进行真假识别的方法。首先采用自主设计的电子鼻系统对不同品牌白酒进行采样测试;采样后的传感器阵列数据依次经过数据预处理、特征生成、特征选择降维处理,得到可用于分类的白酒样本;再通过格点搜索获取每种白酒单类分类器的最优参数;最后测试各个单类分类器对相应品牌白酒的真假识别效果。各单类分类器的真假识别率分布在93%~98%之间,结果表明,采用自主设计的电子鼻结合单类支持向量机可以很好地对白酒真假进行识别。
Aimed at the imbalanced number between the true and fake Chinese liquor samples（normal and abnor-mal samples）and the lack of abnormal sample categories in truecfake Chinese liquor recognition by multi-classclassification,a true-fake recognition method using one-class SVM（Support Vector Machine）to train a one-classclassifier for each brand of liquor is put forward. Firstly,a self-designed electronic nose system was used to sampledifferent brands of liquor. Secondly,after data pre-processing,feature generation and feature reduction in turn,thedata sampled from the sensor array was transformed into test samples of classification. Thirdly,the optimal parame-ters for each one-class classifier were found by grid-search. Finally,each one-class classifier was tested on the truefake recognition effect by the corresponding samples. The true-fake recognition rate of one-class classifiers rangesfrom 93% to 98%,which indicates that the self-designed electronic nose system combined with one-class SVM hasa good performance of true-fake Chinese liquor recognition.
出处
《传感技术学报》
CAS
CSCD
北大核心
2015年第12期1741-1746,共6页
Chinese Journal of Sensors and Actuators
基金
国家自然科学基金项目(61271321
61473207
61401303)
教育部博士点基金项目(20120032110068)
天津市科技支撑计划项目(14ZCZDS F00025)
关键词
电子鼻
白酒
真假识别
单类支持向量机
electronic nose
Chinese liquors
true-fake recognition
one-class SVM
