以添加不同辅料奶粉的近红外图谱作为聚类分析的对象,化学成分的含量不同,红外的谱图就会有差异。在建立混合成分模型的基础上,采用SIMCA聚类分析法对添加不同辅料奶粉进行了快速的分类研究。对市场25０种不同种类奶粉进行近红外扫描,并做了重现性测定。含有乳糖或蔗糖的不同辅料的奶粉的近红外谱图,从表观上难以进行指纹特征提取,需采用模式识别技术来进行智能识别。从中随机抽取160张谱图进行聚类分析,并对未知样品进行预测。结果表明:尽管各样品的近红外谱图很相似,难以直接区别,但采用近红外光谱和聚类分析相结合的方法可以鉴定添加不同辅料的奶粉。从模型图上看,两个不同辅料奶粉之间没有重叠,互不干扰,各样品之间光谱聚类的结果较理想,盲样检测的正确率可达90％以上。如果能够获得足够多的样本,增加训练集样本数和采样的代表性,加强操作的标准程度,该法的准确率将会大大提高。近红外光谱与聚类分析法相结合可以快速、无损识别奶粉中添加的不同辅料,是一种有效的奶粉质量检测技术。

The near infrared spectrum atlases of milk powders mingled with different adjuvant are the object for cluster analysis.Drawing assistance from the disparity in infrared fingerprint atlas that change according to the contents of chemical constituent,and making mingled component models,the milk powders mingled with different adjuvant were taken for a rapid sorting test using SIMCA clustering analytical method.In the experiment,two hundred fifty sorts of milk powders in the markets from different manufacturers were scanned by near infrared ray,and were tested with reproducibility determination.It was found difficult to extract fingerprint characters just from the external appearance of the near infrared spectrum atlases from milk powders mingled with different adjuvant,and it is needed to adopt pattern recognition technique to determine intelligently.One hundred sixty atlases were drawn out randomly for cluster analysis,and unknown samples were pretested.Results showed that the milk powders mingled with different adjuvant can be identified by near infrared spectrum analysis associated with cluster analysis methods,notwithstanding the similar near infrared spectrum atlases of different sample were difficult to identify directly.No overlapping phenomenon was found among milk powders mingled with different adjuvant,and they did not interfere with each other.The results from clustering spectra between samples were satisfactory,and the correct ratios of blind detections were over 90﹪.In addition,the correct ratios of this method may be elevated remarkably with sufficient number of samples,increasing training set sample quantity and sampling representation,and strengthening the standard degree of manipulation.It is concluded that the designed model to determine milk powders mingled with different adjuvant is rational,and the determination capability is fine.