论文部分内容阅读
为解决近红外光谱快速检测乳品成分及含量时光谱数据的预处理问题,提出一种基于直方图分层映射技术的近红外光谱主成分得分重置(SR)预处理方法。以葡萄糖氯化钠水溶液三组分样品中的葡萄糖含量、鲜牛奶样品中的乳糖含量为定量检测目标,进行散射光谱主成分得分累计贡献率的分层分段规定化映射预处理,利用偏最小二乘(PLS)回归分析建模手段,对相应近红外光谱中的糖含量信息进行测试及分析。结果表明,经过SR预处理后,牛奶中乳糖含量PLS模型的校正集样品交互验证预测偏差降低23.9%,实际预测偏差降低27.8%;验证集实际预测偏差降低16.7%。该SR光谱预处理方法兼顾光谱、参考值及组分相关性等多尺度信息,以实现光谱信息增强去噪,能避免有用信息误删,防止不充分拟合及过拟合。
In order to solve the problem of preprocessing spectral data of dairy products and their content by near infrared spectroscopy (FTIR), a method of principal component fractional reset (SR) preprocessing based on histogram delamination mapping was proposed. Taking the glucose content in the three components of glucose and sodium chloride aqueous solution and the lactose content in the fresh milk sample as the quantitative detection target, the stratified and segmented prescribed mapping preprocessing of the accumulated contribution rate of the main components of the scattering spectrum was performed, (PLS) regression analysis of modeling tools, the corresponding near infrared spectroscopy sugar content information testing and analysis. The results showed that after SR pretreatment, the calibration set of PLS model for milk lactose decreased by 23.9% and the actual prediction error decreased by 27.8%. The actual prediction bias of validation set decreased by 16.7%. The SR spectrum preprocessing method takes into account the multi-scale information such as spectra, reference values and component correlations so as to achieve enhanced de-noising of spectral information, avoids accidental deletion of useful information, and prevents inadequate fitting and over fitting.