中文摘要：

对近红外光谱分析青贮玉米中性洗涤纤维（NDF）中异常光谱的判别进行了研究。该试验通过将马氏距离阈值分别设定为3（固定值）、2倍马氏距离平均值和马氏距离平均值＋2倍马氏距离标准差三种不同值,分别判别和剔出建模过程中的异常光谱,比较不同的阈值设定对模型效果的影响。结果表明,当马氏距离阈值设为3（固定值）时,异常光谱剔出不收敛,可靠性不高。当马氏距离阈值设为2倍马氏距离平均值时,剔出后模型的相关系数、标准差、决定系数及综合得分均低于剔出前,可靠性不高。只有当马氏距离阈值设定为马氏距离平均值＋2倍马氏距离标准差时,预测模型的最高相关系数r达到0.97,标准差为2.456,模型预测效果最佳。因此,将马氏距离阈值设定为马氏距离平均值＋2倍马氏距离标准差对近红外光谱分析青贮玉米NDF中异常光谱判别的可行性较高,在这种情况下剔出异常光谱后模型效果最佳。

英文摘要：

An experiment was conducted to study the outlier diagnosis on the near infrared spectroscopy （NIRS） analysis of NDF content in corn silage feeds. Various Mahalanobis＇ distances including 3 and 2 × Mahalanobis＇ distances and the average of Mahalanobis＇ distance＋2 SD （AV＋2 SD） were set to diagnose the spectral outliers during the model development, and their effects on the calibration models were compared respectively. The results showed that it was feasible to diagnose the spectral outliers for NIRS analysis of NDF content in corn silage feeds when the Mahalanobis＇ distance was AV＋2 SD, the r is 0. 97 and the SEC is 2. 456. The calibration model optimized under such conditions was the best.