马氏距离与浓度残差剔除近红外异常样品研究

Research for eliminating outerlier samples in near-infrared spectroscopy by method of mahalanob distance-concentration residual

以新疆南疆灰枣为研究对象,采用马氏距离法对近红外校正集中异常光谱样品进行剔除,并运用浓度残差法对异常化学值样品进行剔除,从校正集中的100个红枣样品剔除了1个异常光谱样品和23个异常化学值样品,用剩下的76个样品建立红枣水分校正模型,并对预测集的51个红枣样品进行预测分析,用预测相关系数(RP)、预测标准偏差(SEP)、平均相对误差(Er)来作为评价指标。结果表明:RP为0.9258,RMSEP为1.6197,Er为0.0333,与剔除前校正集所建模型相比,模型的稳定性和预测精度得到显著的提高。

With Hui jujube in Southern region of Xinjiang as research object,outerlier samples often strongly influence the stability of the model in near-infrared（NIR）spectrum.By using the method of mahalanob distance,1outliers were detected and eliminated,and concentration residual method was employed to detect and eliminate 23 outliers chemical value sample from the calibration set of 100 jujube samples.With the remaining 76 samples to establish a calibration model of red dates moisture,prediction and analysis of 51 samples of red dates of the prediction set.The results showed the predicted correlation coefficients（RP）of the predicted set were0.9258;the standard error of validation（SEP）of the prediction were 1.6197;the average relative error of the prediction were 0.0333.Compared to the original calibration model the predicted accuracy and stability of this model were greatly improved.