建立良好的蛋白饲料近红外光谱定量分析模型及实现在不同仪器间的模型共享, 能极大提高模型的利用效率, 满足饲料行业快速发展的需要。 针对蛋白饲料原料粗蛋白含量近红外分析模型适用性问题, 首次采用光谱差值转移、 直接校正和分段直接校正法进行了三台不同类型的近红外光谱仪之间的模型转移研究。 实验样品为四种蛋白饲料原料： 玉米蛋白粉、 菜粕、 酒糟和鱼粉。 实验仪器包括MATRIX-Ⅰ傅里叶变换型近红外光谱仪（主仪器）, Spectrum 400傅里叶变换型近红外光谱仪（从仪器1）和SupNIR-2750光栅扫描型近红外光谱仪（从仪器2）。 研究表明, 同一样品体系在主仪器和从仪器2上所得光谱数据的差异性相对较小, 且均与从仪器1所得光谱数据的差异性相对较大。 除分段直接校正法对玉米蛋白粉从仪器2的预测结果无促进作用之外, 其他模型的预测均方根误差和系统偏差均明显低于转移前。 玉米蛋白粉、 菜粕和酒糟样品采用三种方法转移后的模型预测相对分析误差（RPD）均大于3.0, 预测效果良好。 鱼粉样品模型转移后的预测RPD均大于2.5, 预测效果较好。 三种方法对于蛋白饲料原料不同仪器间的光谱差异进行了有效校正。 该研究结果对于蛋白饲料品质近红外快速分析模型的广泛应用具有重要意义。

The near infrared spectrometric quantitative model of protein feed and its sharing in different instruments can greatly improve the utilization efficiency of the model and meet the needs of rapid development of feed industry. Considering the issue of applicability of near infrared spectrometric models for crude protein of protein feed materials, calibration transfer was explored among three types of instruments using spectral subtraction correction, direct standardization and piecewise directs standardization methods for the first time. Four kinds of protein feed raw materials were involved in the present study, corn protein powder, rapeseed meal, fish meal and distillers dried grains with soluble. The experimental instruments included MATRIX-I Fourier transform near infrared instrument (master instrument), Spectrum 400 Fourier transform near infrared instrument (slave 1 instrument), and SupNIR-2750 grating near infrared instrument (slave 2 instrument). Results showed that the spectral data difference for all the samples between the master and slave 2 instrument was relatively small, and the difference between the master and slave 1 instrument, and slave 1 and slave 2 instrument were relatively large. All the root mean square error of prediction and bias values after calibration transfer were lower than the values before calibration transfer, except that no improvement was found for the prediction of corn protein powder of slave 2 instrument corrected by piecewise direct standardization method. The relative prediction deviation (RPD) of corn protein powder, rapeseed meal and distillers dried grains with soluble transferred by all three methods were higher than 3, which indicated good predictions, while the RPD of fish meal were all higher than 2.5, which indicated relative good predictions. All three techniques used in the study were effective in the correction of the difference between different instruments for protein feed materials. This study is of important practical significance for the application of near infrared spectrometric models for crude protein of protein feed materials.