重点：多光谱成像用于检测碎肉造假，PLS-DA和LDA鉴别分类有效率可达98.48%

测量中使用了多个批次，采用了额外批次验证，PLS-DA与LDA相比精度较高（外置验证）

摘要

本研究目的是调查多光谱成像应用潜力，通过多变量分析法检测碎牛肉（掺杂猪肉）或猪肉中掺杂牛肉。对来自4个独立试验的220个样本（每个试验55个样品）进行了18个波段的多光谱成像。

将适量的牛肉和猪肉碎肉混合来获得9种不同比例的掺假混合以及共2种纯猪肉和牛肉分类样品。图像处理后，前3个试验的数据包括来自纯牛肉和纯羊肉样品数据用于偏较小二乘判别分析（PLS-DA） ，来区分所有掺假分类。结果显示在纯肉和掺假样品之间具有良好的区分度。另外，使用LDA和PLS-DA法，98.48%和96.97% 的样品分类精度在掺假类别（± 10%）内。较后，利用第4个实验进行独立检测来验证模型，用PLS-DA做的所有纯肉样品和掺假样品分类正确，而用LDA法精度不太高。

Multispectral image analysis approach to detect adulteration of beef and pork in raw meats

A.I. Ropodia, 1, D.E. Pavlidisa, 1,F. Moharebb,E.Z. Panagoua, G.-J.E. Nychasa, , 

Highlights

Multispectral imaging was used for the detection of adulteration of minced meat.

PLS-DA and LDA yielded 98.48% overall correct classification.

Multiple batches of meat were used.

External batch validation was employed.

PLS-DA was more accurate compared to LDA in the case of external validation.

Abstract

The aim of this study was to investigate the potential of multispectral imaging supported by multivariate data analysis for the detection of minced beef fraudulently substituted with pork and vice versa. Multispectral images in 18 different wavelengths of 220 meat samples in total from four independent experiments (55 samples per experiment) were acquired for this work. 

The appropriate amount of beef and pork-minced meat was mixed in order to achieve nine different proportions of adulteration and two categories of pure pork and beef. After an image processing step, data from the first three experiments were used for partial least squares-discriminant analysis (PLS-DA) and linear discriminant analysis (LDA) so as to discriminate among all adulteration classes, as well as among adulterated, pure beef and pure pork samples. Results showed very good discrimination between pure and adulterated samples, for PLS-DA and LDA, yielding 98.48% overall correct classification. Additionally, 98.48% and 96.97% of the samples were classified within a ± 10% category of adulteration for LDA and PLS-DA respectively. Lastly, the models were further validated using the data of the fourth experiment for independent testing, where all pure and adulterated samples were classified correctly in the case of PLS-DA, while LDA was proved to be less accurate.

Keywords

Meat adulteration; Multispectral image analysis; Discriminant Analysis; Minced beef/pork; External validation