基于支持向量机的近红外光谱技术鉴别掺假牛肉

Identification of adulteration of minced beef with near infrared spectroscopy based on support vector machine

【目的】利用近红外光谱技术对掺入鸭肉的掺假牛肉进行判别研究,为实现掺假牛肉的快速、准确检测提供参考。【方法】市购冷冻储存的牛里脊肉和鸭肉,在牛里脊肉糜中掺入质量分数为0,5%,10%,15%,20%,25%的鸭肉糜（对应m（鸭肉糜）∶m（牛肉糜）分别为0∶20,1∶19,2∶18,3∶17,4∶16和5∶15）,制备供试牛肉和掺假牛肉,采用近红外漫反射方式在波数为10 000~4 000cm-1时采集牛里脊肉、掺假牛里脊肉和鸭肉的近红外光谱,对原始光谱经多元散射校正（Multiplicative scatter correction,MSC）、标准正态变量变换（Standard normal variate transformation,SNV）、面积归一化（Area normalization）、15点平滑处理和一阶导数处理等方法预处理后,对所建支持向量机（nu-SVM）判别模型的预测效果进行比较。【结果】经MSC预处理后所建的牛里脊肉、掺假牛里脊肉和鸭肉的nu-SVM判别模型判别性能稳定,对建模集的正确判别率为97.09%,对检验集的正确判别率为94.00%。【结论】近红外光谱技术结合MSC预处理后所建的nu-SVM模型可用于掺杂鸭肉的牛里脊肉的鉴别。

【Objective】A near infrared（NIR）spectroscopy was employed to analyze the adulteration of duck meat in minced beef to realize the nondestructive and quick determination of adulterated beef.【Method】Frozen beef and duck were purchased,and adulterated beef samples were obtained by interfusing different amounts（0,5%,10%,15%,20%,25%）of duck meat（mass ratios of duck to beef were 0∶20,1∶19,2∶18,3∶17,4∶16and 5∶15）.The original spectra of beef,adulterated beef and duck meat in the wavenumber range of 10 000-4 000cm-1 were obtained by spread reflection NIR.Then,the predicted effects of different support vector machine（SVM）discriminate models that established by original spectra,multiplicative scatter correction（MSC）,standard normal variate transformation（SNV）,normalization,15 point smooth and 1st derivative were evaluated.【Result】The established model with MSC preprocessing methodhad the best accuracies of 97.09% and 94.00%for the training set and prediction set,respectively.【Conclusion】The nu-SVM model built by NIR combined with MSC preprocessing method was proved to be a feasible method to identify duck meat in minced beef.