The efficacy of using a handheld near infrared spectrometer to predict metanil yellow (MY) adulteration levels (0-30% w/w) in dried turmeric powder was tested against a benchtop near infrared spectrometer using partial least squares regression models. The differences between near infrared instruments were resolution (i.e., 1 nm (handheld) vs. 0.5 nm (benchtop)) and sample container during scanning (plastic pouch (handheld) vs. quartz glass cup (benchtop)). Prediction performance of the calibration models developed was evaluated using number of model factors ( N F ), coefficients of determination of calibration and validation ( R 2 and r 2 , respectively), root-mean-square errors of calibration, cross-validation, and validation (RMSEC, RMSECV, and RMSEP), ratio of prediction error to standard deviation (RPD), and limits of detection (LOD) and quantification (LOQ). The best benchtop calibration models were based on spectral data preprocessed with Savitzky–Golay first derivative algorithm for the benchtop near infrared and standard normal variate for the handheld near infrared, yielding low N F , high R 2 and r 2 , low RMSEC, RMSECV, RMSEP, and high RPD. The LOD and LOQ for both spectrometers were 0.33 and 1.10%, respectively, and no significant difference was found between the predicted MY values by the benchtop and handheld near infrared spectrometers. The models were, in general, not sensitive to sample source and size of the validation set. When spectra from the benchtop near infrared were standardized using a reverse standardization strategy, calibrated against MY , and transferred to the handheld near infrared, prediction performance dropped, from r 2 of 0.99 to 0.98, RMSEP increased from 0.96% to 1.53%, and RPD decreased from 10.1 to 6.3. Despite the reduced prediction performance, the handheld near infrared with a transferred calibration model from the benchtop near infrared was still useful for screening, quality control, and process control applications.

中文翻译：

---

使用手持式近红外光谱仪和偏最小二乘回归来量化姜黄粉中的间苯胺黄掺假

使用手持式近红外光谱仪预测干姜黄粉中的间苯二酚黄 (MY) 掺假水平 (0-30% w/w) 的功效与使用偏最小二乘回归模型的台式近红外光谱仪进行了测试。近红外仪器之间的差异是分辨率（即 1 nm（手持式）与 0.5 nm（台式））和扫描期间的样品容器（塑料袋（手持式）与石英玻璃杯（台式））。使用模型因子数 (NF)、校准和验证的确定系数（分别为 R 2 和 r 2 ）、校准、交叉验证和验证的均方根误差来评估开发的校准模型的预测性能（RMSEC、RMSECV 和 RMSEP），预测误差与标准偏差的比率 (RPD)，以及检测限 (LOD) 和定量 (LOQ)。最佳台式校准模型基于使用 Savitzky-Golay 一阶导数算法预处理的光谱数据，用于台式近红外和用于手持式近红外的标准正态变量，产生低 NF 、高 R 2 和 r 2 、低 RMSEC、RMSECV、RMSEP和高 RPD。两种光谱仪的 LOD 和 LOQ 分别为 0.33% 和 1.10%，台式和手持式近红外光谱仪的预测 MY 值之间没有显着差异。通常，这些模型对样本来源和验证集的大小不敏感。当台式近红外光谱使用反向标准化策略进行标准化，针对 MY 校准，并传输到手持式近红外光谱时，预测性能下降，从 r 2 的 0.99 到 0.98，RMSEP 从 0.96% 增加到 1.53%，RPD 从 10.1 减少到 6.3。尽管预测性能有所下降，但带有从台式近红外传输的校准模型的手持式近红外仍可用于筛选、质量控制和过程控制应用。