摘要
利用可见/近红外光谱技术研究了一种液态纯牛奶品牌鉴别模型。首先利用近红外分析仪InfraXactTM获得3种品牌共90个样本的漫反射光谱,分别对其赋值,采用偏最小二乘法(PLS)建立鉴别模型;其次比较典型光谱散射预处理和数学预处理对建模的影响,探索到建模优化预处理方法为SNV+Detrending、5点间隙的一阶导数处理、5点平滑波长;实验结果表明,模型的RMSEC,R2c,RMSEP,Bias和R2p分别为0.515,0.937,0.561,0.296和0.912,对57个校正样本和30个独立预测样本进行鉴别,其准确率分别达到了100%和96.7%。
A model for brand discrimination of liquid pure milk by Vis/NIR was developed. Firstly, the diffuse reflectance spectrum of 90 samples were measured by the InfraXactTM spectrometer, and PLS was used to build the discriminating model. Secondly, the calibration results were compared by different pretreatment methods, and the preferable method（SNV＋ Detrending, 1st Derivative, 5, 5） was found out. The experimental results show that the RMSEC, Rc2, Rp2, RMSEP, Bias and of the model are 0. 515, 0. 937, 0. 561, 0. 296 and 0. 912, and the discriminating rate of the 57 training samples and 30 predictive samples are 100% and 96.7%, respectively.
出处
《现代电子技术》
2011年第17期166-168,共3页
Modern Electronics Technique
基金
西安文理学院中青年科研基金项目资助(kyc201024)
