品质性状的化学测定操作繁琐且存在破坏性和耗时较长等不足的问题, 光谱测定具有高效、 快速、 成本低等优点, 但测定准确度受到不同仪器以及不同机型的影响。 为了建立和优化快速测定苜蓿样品的粗蛋白（CP）、 粗脂肪（EE）、 酸性洗涤纤维（ADF）和中性洗涤纤维（NDF）近红外漫反射光谱的模型, 更好的测定苜蓿品质性状。 选取了25份苜蓿材料147份试验样品, 采用傅里叶变换近红外光谱技术（NIRS）扫描, 获得扫描光谱范围4 000~10 000 cm-1的光谱值, 软件TQ Analyst v9选用偏最小二乘法（PLS）和OPUS7.0选用定量2方法建立定量模型并优化, 并进一步交叉验证和外部检验评估模型效果。 结果表明利用2种软件建立的模型都能很好的预测CP的含量, 建模决定系数（R2cal）分别达到0.999 9和0.984 8, 交叉验证的均方根误差（RMSECV）分别为2.121和0.471, 外部验证决定系数（R2）都大于0.97, 残留预测偏差（RPD）值大于6.0。 EE应用TQ Analyst v9所建立的模型效果更好, R2cal为0.999 7, RMSECV为1.502, 外部验证的R2为0.9293, RPD值为3.89; ADF和NDF利用OPUS7.0建立的模型效果更好, R2cal分别为0.944 1和0.978 8, RMSECV分别为1.040和0.514, 外部验证的R2依次为0.914 5和0.911 8, RPD值分别为3.66和3.43。 4种品质性状建模效果表明, 相对分子结构相对简单的蛋白质和脂肪, 利用TQ Analyst v9更准确, 而对于分子结构更复杂的纤维素, OPUS7.0的预测效果更好。

The chemical determination method of quality traits is cumbersome, destructive and time-consuming. The spectral determination method has the advantages of high efficiency, speed and low cost, but the accuracy is affected by different instruments and models. In order to establish and optimize the model for rapid determination of crude protein (CP), ether extract (EE), acidic detergent fiber (ADF) and neutral detergent fiber (NDF) using near-infrared diffuse reflectance spectra of alfalfa samples, and better determine the quality traits of alfalfa. A total of 147 samples of 25 alfalfa materialswere selected. The scanning spectral values of the spectral range of 4 000～10 000 cm-1 are obtained by scanning with Fourier transform near-infrared spectroscopy (NIRS). The software TQ Analyst V9 adopts partial least squares (PLS), and OPUS 7.0 adopts the quantitative 2 methods to establish and optimize the quantitative model and further carry out cross-validation and external test to evaluate the effect of the model. The results showed that the models for determining CP content were also through two software. Two modeling coefficient of determination (R2cal) were 0.999 and 0.984 8, the root mean square error (RMSECV) of cross-validation was 2.121 and 0.471, respectively. The coefficient of determination (R2) of external validation is greater than 0.97, and the ratio of standard deviation to SEP (RPD) was greater than 6.0. The model established by TQ analyst V9 was better for EE with R2cal of 0.999 7, RMSECV of 1.502, R2 of external verification of 0.929 3 and RPD value of 3.89. The models established by OPUS 7.0 were better for ADF and NDF with R2cal of 0.944 1 and 0.978 8, RMSECV of 1.040 and 0.514, R2 of external verification of 0.914 5 and 0.911 8, and RPD of 3.66 and 3.43, respectively. The modeling results of four quality traits showed that the models of TQ Analyst V9 are more accurate for CP and EE with relatively simple molecules structure, while the models of OPUS 7.0 are more accurate for ADF and NDF with relatively complex molecular structures.