基于可见/近红外光谱技术的板栗产地识别

采用可见/近红外光谱分析技术对河北和安徽两个产地的板栗进行检测分级，获得板栗样品在600~1 100 nm波长区间的可见/近红外光谱，采用偏最小二乘判别分析（PLSDA）进行建模，比较不同预处理方法和波长范围对PLSDA模型精度的影响。结果显示，不同预处理方法对模型性能影响较大，一阶导数预处理在全波长范围所建PLSDA模型性能最优，校正集和验证集的决定系数分别为0.884和0.863，均方根误差分别为0.170和0.191。不同波长范围也会影响模型的预测和识别性能，在波长区间为750~1 000 nm的光谱所建立的PLSDA模型性能要高于全波长光谱所建立的模型性能，其中经过Savitzky Golay平滑预处理后，模型性能的提高最为明显，且其与原始光谱在校正集和验证集的敏感性和特异性均达到最优，即识别率均可达到100%。因此，可见/近红外光谱分析技术能够对板栗产地进行鉴别。

Detection of chestnut production place based on visible and near infrared spectroscopy

This paper focused on identifying the geographic origin of chestnut by using visible and near infrared (VIS/NIR) spectroscopy. The VIS/NIR spectra were obtained at the spectral range of 600-1 100 nm for chestnut samples. Partial least square discriminant analysis models were developed, and different spectral pre processing methods were used and compared to evaluate the effect on mathematical models. Besides, different spectral ranges were also compared to determine the influence on the accuracy of the PLSDA model. The results showed that different spectral pre processing methods could influence the PLSDA models. PLSDA model for the spectra based on 1st derivative test could provide optimal performance with the determination coefficients for calibration and validation sets of 0.884 and 0.863, RMSEC and RMSEP of 0.170 and 0.191, respectively. The spectral range also affected mathematical models. The performance for the PLSDA models over the spectral range of 750-1 000 nm overall was better, especially for the spectra based on Savitzky Golay smoothing pre processing method, which could provide noticeable improvement for model performance. The models for the original spectra and the spectra based on Savitzky Golay smoothing pre processing method in calibration and validation sets had optimal sensitivity and specificity, which suggested that the recognition rate for calibration and validation sets could reach 100%. Thus, visible and near infrared spectroscopy can recognize the geographic origin of chestnut.