利用近红外光谱技术检测掺假豆浆

为了对豆乳内在营养指标及掺假豆乳进行快速检测,试验运用近红外光谱技术,利用偏最小二乘法进行回归分析,分别建立83个真伪豆浆样品的蛋白质和总固形物含量定标模型,并对模型的预测性能进行分析。结果表明:选取主成分数为12和14,蛋白质和总固形物含量的近红外光谱预测值与化学实测值之间的相关系数R分别为0.9756和0.9489,校正均方根误差分别为0.186和0.175,预测集样品的预测值和实测值之间的残差值均较小、接近零,残差之和分别为-0.074和-1.191,说明建立的定标模型可以准确预测豆浆中蛋白质和总固形物含量,且预测性能较好;通过对预测集样品的预测值与豆浆行业标准规定值相比较,确定预测集样品中掺假豆浆的正确判别率为100%,说明建立的蛋白质和总固形物含量定标模型可以应用于掺假豆浆的判别检测,且判别结果准确率高。本试验表明利用近红外光谱技术可实现对豆浆主要品质指标的快速无损检测,也可准确进行真伪豆浆的快速判别,本检测方法可为豆乳行业健康持续发展提供一定的技术支撑。

Adulteration detection of soymilk based on near-infrared spectroscopy

Abstract: In order to rapidly detect the internal nutritive index and discriminate adulteration soymilk, the near infrared transmission spectrometer such as Purespect was used to obtain spectrums for 83 unadulterated and adulterated soymilk samples. The spectral scanning procedure was conducted in dark room, 643.26-954.15 nm wavelength range was chosen, scanning wavelength interval was 1.29 nm. Each soymilk sample was scanned three times. Pure soymilks were made according to the regulations in soymilk products industry standard SB/T 10633-2011. A lot of water, essence, thickening agent, food sunset yellow pigment were added to unadulterated soymilk samples in order to obtain adulterated soymilk samples. In this study, 31 adulteration samples and 62 unadulterated samples were prepared in the processing laboratories. 14 soymilk samples were gathered from the market. All samples were used to scan the spectrum and determining chemical composition. The experimental results indicated that smooth lines and clear spectrogram were obtained using Savitzky-Golay and the second derivative method. Chemometrics method of partial least squares (PLS) was used to the model calibration for protein and total solid content in samples. The correlation coefficient of predicted value and measured value of protein and total solid content for soymilk calibration samples were 0.9756 and 0.9489 respectively. The correction of root mean square error for soymilk calibration samples were 0.186 and 0.175 respectively. 12 and 14 was selected for principal component number respectively. 24 prediction samples were prepared for analyzing predictive capability. The results indicated that the residual values of predicted value and measured value for soymilk prediction samples were small and close to zero. The distribution of residual was uniform for both sides of zero line. The residual sums between predicted and measured protein and total solid content values were -0.074 and -1.191 respectively. The results verified that the calibration models could accurately predict the protein and total solid content for soymilk samples. According to the standard of soymilk industry, the internal nutrition of sample was satisfied and the sample was disqualified when the protein mass fraction in soymilk samples was less than 2%, the total solids mass fraction in soymilk samples was less than 6%. Through comparing the predicted values of prediction samples with the regulation values by soymilk industry standard, No. 6-22 samples were disqualified. The resolution of individually adulterated soymilk from all prediction samples was 100% based on the practical sample collection conditions and measured value by chemical methods,. The results verified that the protein and total solid content calibration models were capable of discriminating the adulteration soymilk. This experiment indicated that rapidly detect the major quality index and discriminate adulteration soymilk were achieved based on NIR spectra. This detection method can be used to support for the healthy and abidingly development of soymilk industry.