基于近红外光谱的大米水分及蛋白质含量检测方法研究

发布时间：2018-08-18 10:13

【摘要】：随着人们生活质量的不断提高以及对身体健康的重视程度越来越强,消费者对大米的适口性、营养等品质方面的需求也随之提高。大米的品质不但能让消费者在感官上特别享受,直接关系着人们身体对大米的消化以及吸收,而且在大米的贸易和相关科学领域的研究方面也至关重要。在大米的品质评定中,水分和蛋白质是评价大米价值的关键因素。大米的水分含量不单影响着大米的品质,更重要的是关系到公众的食品安全问题。大米以及它的副产品同时提供人们每日所需要的能量和蛋白质,蛋白质含量的高低在大米的食味品质方面起着重要的作用。因此本文将以此作为着手点,采用大米作为研究对象并利用近红外光谱技术结合其水分和蛋白质的含量进行图谱拟合,以求探索快速检测实现的可能性。主要进行了如下研究工作:(1)从黑龙江各地收集109个不同种类的大米样品,利用Thermo Fisher公司Antaris II近红外光谱仪对其进行光谱扫描。然后采用传统的国家标准化学方法对主要组成成分水分和蛋白质的化学值进行测定,为后续的数据处理以及模型的建立做好准备工作。(2)针对大米蛋白质的定量分析模型分别采用霍特林T2统计、X-Y残差以及3D视图分析三种方法进行异常样品的剔除。经对比发现通过X-Y残差分析方法剔除异常点后建立模型的RMSECV和R2分别从最初的0.2428和0.7626提升到了0.2060和0.8364。(3)采用The DUPLEX Method将水分和蛋白质的样品划分为校正集和预测集,结果表明二者不单单在含量的范围方面十分相近,在对样品求得的平均值以及标准差上也十分相似,从而获得符合实验要求且均匀分布的校正集和预测集,达到了本研究的预期效果。(4)对大米水分的原始光谱图分别采用导数、归一化以及平滑三种方法进行去噪处理,经对比发现平滑点数采用15点时去除噪声后的建模效果最佳。对大米蛋白质的原始光谱图分别采用一阶导数+平滑、二阶导数+平滑以及正交信号校正三种去噪方法对光谱中所含的噪声进行消除,研究表明二阶导数+平滑处理具有最好的去噪效果,其所建立模型的校正集决定系数R2更趋向于1,同时有着较低的校正集均方误差根RMSECV。(5)对大米水分光谱采用MWPLS和IPLS波长选择方法进行模型验证,通过对比发现,MW-IPLS对水分光谱特征吸收波长的选取是一种有效的方法。结合MWPLS和IPLS方法进行波长选择,确定二者具有交叉的最优波长范围4108-4386cm-1。在此波段采用PLSR建立大米水分的定量模型,预测集均方误差根RMSEP为0.2753,其决定系数R2达到0.8597。(6)分别对80个校正集和24个预测集大米样品的蛋白质光谱建立PLSR和PCR模型。通过对比预测集均方根误差根RMSEP和决定系数R2可以得出,PCR模型对大米蛋白质含量的预测能力上表现更优,预测集均方根误差根RMSEP和其决定系数R2达到0.1288和0.8865。综上所述,本文利用近红外光谱技术针对大米中水分和蛋白质含量所建立的模型精度较高,具有一定的可行性,为今后大米成分的快速检测提供新方法。
[Abstract]:With the continuous improvement of people's quality of life and the increasing emphasis on health, consumers'demand for the quality of rice, such as palatability, nutrition and so on, has also increased. Water and protein are the key factors in evaluating the value of rice. The water content of rice not only affects the quality of rice, but also the food safety of the public. The energy and protein needed and the protein content of rice play an important role in the food quality. Therefore, this paper will take this as the starting point, take rice as the research object, and use near infrared spectroscopy to fit the water and protein content of rice, in order to explore the possibility of rapid detection. The main research work is as follows: (1) 109 different kinds of rice samples were collected from Heilongjiang Province and scanned by Antaris II near infrared spectrometer of Thermo Fisher Company. Then the chemical values of water and protein were determined by traditional national standard chemical method for the follow-up. (2) Hotelling T2 statistics, X-Y residuals and 3D view analysis were used to eliminate the abnormal samples for the quantitative analysis model of rice protein. By comparison, the RMSECV and R2 of the model were established after removing the abnormal points by X-Y residuals analysis. (3) The DUPLEX Method was used to divide the water and protein samples into correction set and prediction set. The results showed that the two samples were not only very similar in the content range, but also very similar in the average value and standard deviation obtained from the samples, which accorded with the experimental requirements. (4) Derivative, normalization and smoothing were used to denoise the original spectrogram of rice moisture, and the smoothing point was found to be the best when the smoothing point was 15. The original spectrogram of rice protein was collected separately. Three denoising methods, first derivative + smoothing, second derivative + smoothing and orthogonal signal correction, are used to eliminate the noise contained in the spectrum. The results show that the second derivative + smoothing method has the best denoising effect. The calibration set determinant R2 of the model tends to be 1, and has a lower root mean square error of the calibration set RMSECV. (5) The MWPLS and IPLS wavelength selection methods were used to validate the model of water spectrum of rice. It was found that MW-IPLS was an effective method to select the characteristic absorption wavelength of water spectrum. The root mean square error (RMSEP) of prediction set was 0.2753, and the determination coefficient R2 was 0.8597. (6) PLSR and PCR models were established for protein spectra of 80 correction sets and 24 prediction sets respectively. The RMSEP and its determinant R2 are 0.1288 and 0.8865. In summary, the model established by near infrared spectroscopy for moisture and protein content in rice is more accurate and feasible, which provides a new method for rapid detection of rice components in the future. Method.
【学位授予单位】：东北农业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TS210.7;O657.33

【参考文献】