【摘要】 介绍采用近红外光谱分析方法快速检测牛奶主要成分含量的测量原理,探讨研究直接正交(DO)校正的基本方法。利用牛奶成分近红外光谱测量系统分别采集牛奶样品和葡萄糖白蛋白两成分溶液样品的近红外光谱,采用DO法进行光谱数据预处理,并采用偏最小二乘(PLS)法分别建立其相应的数学模型。实验及数据处理结果表明:经DO法预处理后,滤除了原始光谱中的部分噪声信息,但保留了原始光谱中的主要信息。PLS校正模型采纳的最佳因子数随着DO因子的依次滤除相应减少。牛奶中脂肪和蛋白质校正模型在原始光谱分别被滤除3和4个主成分时达到性能最佳,校正标准偏差SEC分别为0.3204和0.2727,预测标准偏差SEP为0.7316和0.4460,两成分溶液样品中白蛋白和葡萄糖校正模型在原始光谱被滤除1个因子时达到性能最佳。校正标准偏差SEC分别为0.2513和0.2780,预测标准偏差SEP为0.5169和0.7870,单位(g/dL),与DO法预处理之前的PLS模型相比,预测标准偏差相应降低,采纳的主成分数减少,模型得到简化。更多还原

【Abstract】 The basic principles of the near infrared spectra (NIR) of milk measurement were introduced and basic methods of direct orthogonal correction (DO) were studied. The near infrared(NIR) of milk measurement system was used to collect two home-made components (glucose, albumin) samples’ and milk’s near infrared spectra, to preprocess spectral data and to establish the calibration model by partial least squares (PLS). This indicated that the major information in milk spectrum could be reserved while part noise was removed by DO method. The number of optimal factors of PLS model used to predict the main components fat and protein content against milk spectra would be reduced in accordance with DO factors reduction. In this study, the optimum PLS calibration model was obtained when 3 and 4 DO factors were respectively filtered from fat and protein, to have obtained the standard error of calibration(SEC) of 0.3204 and 0.2727 and the standard error of prediction (SEP) of 0.7316 and 0.4460 and when 1 DO factors were respectively filtered from albumin and glucose, to have obtained the standard error of calibration(SEC) of 0.2513 and 0.2780 and the standard error of prediction (SEP) of 0.5169 and 0.7870. Although this model could not improve precise to a great extent, but in comparison on with the model before DO pretreatment,less reduction factors would be necessary and the model would be simpler.更多还原