摘要
食品的品种不同则其含有营养成分和功效存在差异,得到的傅里叶变换红外光谱也存在差异。为了准确的实现品种分类,设计了一种将傅里叶变换红外光谱与模糊聚类分析方法相结合的品种鉴别方法。在模糊Kohonen聚类网络(FKCN)基础上将模糊K调和聚类(FKHM)引入到Kohonen聚类网络的学习速率和更新策略中,提出了模糊K-Harmonic-Kohonen网络(FKHKCN)算法。FKHKCN利用模糊C均值(FCM)聚类的模糊隶属度计算其学习速率,以FKHM的聚类中心为基础通过推导计算得到FKHKCN的聚类中心,可以解决模糊Kohonen聚类网络方法对于初始类中心敏感而导致聚类结果不稳定的问题。FKHKCN作为一种模糊聚类算法,可实现傅里叶变换红外光谱数据的聚类分析。采用三种数据集:(1)采集产自四川的三种茶叶(优质和劣质的乐山竹叶青以及峨眉山毛峰)作为实验样本,样本总数为96。(2)两个品种(robusta和arabica)的咖啡样本。(3)三个品种(鸡肉、猪肉和火鸡)的肉类样本。首先对三个光谱数据集进行预处理,利用多元散射校正降低茶叶样本原始光谱数据集的散射影响,使用Savitzky-Golay减少噪声对肉类和咖啡这两个光谱数据集的影响。再利用主成分分析将高维的三种光谱数据集压缩至低维。然后采用线性判别分析进行特征提取,将光谱数据投影到求得的鉴别向量上。最后分别采用FCM,FKCN和FKHKCN对茶叶、肉类和咖啡进行判别。最终结果如下:FCM,FKCN和FKHKCN对茶叶品种的聚类准确率分别为90.91%,90.91%和93.94%;对肉类品种的聚类准确率分别为90.83%,0.00%和92.50%;对咖啡品种的聚类准确率分别为89.17%,89.17%和90.83%。以上实验结果表明:采用傅里叶红外光谱技术结合主成分分析、线性判别分析和FKHKCN的方法能够较有效地对食品的品种进行鉴别,且鉴别准确率比FCM和FKCN更高,聚类结果更稳定。
Different foods contain different nutrients and effectiveness,and there are differences in their Fourier transform infrared spectra.In order to classify varieties of foods correctly,this paper presented the way to classify varieties by combining Fourier transform infrared spectroscopy(FTIR)with fuzzy clustering analysis.Fuzzy K-harmonic Kohonen clustering network(FKHKCN)was proposed by introducing fuzzy K-harmonic means(FKHM)clustering into the learning rate and update strategy of the Kohonen clustering network.The learning rate of FKHKCN is computed by fuzzy membership values of fuzzy C-means(FCM)clustering,and the cluster centers of FKHKCN can be derived from the cluster centers of FKHM.Therefore,FKHKCN can solve the problem that the Fuzzy Kohonen clustering network(FKCN)is sensitive to the initial cluster centers,and the clustering result is unstable.FKHKCN can achieve the clustering analysis of FTIR data as a fuzzy clustering algorithm.This experiment involves three datasets:(1)Three kinds of tea samples(Emeishan Maofeng,good and poor Leshan trimeresurus)were obtained from Sichuan,China as experimental samples with a total number of 96.(2)Two kinds of coffee samples(robusta and arabica).(3)Three meat samples(chicken,pork and turkey).To start with,three datasets were preprocessed.Scattering effects in the original spectra data of tea samples were reduced by multiple scattering correction(MSC).Savitzky-Golay was used to reduce noise in FTIR spectra of coffee and meat samples.Secondly,the high dimensional FTIR data of three datasets were reduced to by the low dimensionaldata by principal component analysis(PCA).Thirdly,tea data after PCA were extracted by linear discriminant analysis(LDA)and the spectral data were projected into the obtained discriminant vectors.Finally,FCM,FKCN and FKHKCN were used to classify the three datasets,respectively.The experimental results showed that FCM,FKCN and FKHKCN achieved the clustering accuracies for the tea varieties with the values:90.91%,90.91%and 93.94%,respectively;the clustering accuracies for the meat varieties with the values:90.83%,0.00%and 92.50%,respectively;the clustering accuracies for the coffee varieties with the values:89.17%,89.17%and 90.83%,respectively.The above experimental results indicated that FTIR technology coupled with PCA,LDA and FKHKCN was an effective method for classifying food varieties,and its clustering accuracy was higher than FCM and FKCN,and its clustering result was stable.
作者
陈勇
郭云柱
王威
武小红
贾红雯
武斌
CHEN Yong;GUO Yun-zhu;WANG Wei;WU Xiao-hong;JIA Hong-wen;WU Bin(School of Electrical and Information Engineering,Jiangsu University,Zhenjiang 212013,China;High-tech Key Laboratory of Agricultural Equipment and Intelligence of Jiangsu Province,Jiangsu University,Zhenjiang 212013,China;Research Institute of Zhejiang University-Taizhou,Taizhou 317700,China;Department of Information Engineering,Chuzhou Polytechnic,Chuzhou 239000,China)
出处
《光谱学与光谱分析》
SCIE
EI
CAS
CSCD
北大核心
2023年第1期268-272,共5页
Spectroscopy and Spectral Analysis
基金
国家自然科学基金项目(31471413)
江苏高校优势学科建设工程项目(PAPD)
滁州职业技术学院校级自科重点项目(YJZ-2020-12)
滁州职业技术学院院级人才项目“优秀骨干教师”项目(YG2019026,YG2019024)
江苏大学大学生科研立项课题(19A083)资助。
