哈姆林甜橙果实内在品质的可见-近红外光谱无损检测法

采用可见- 近红外漫反射光谱技术，结合偏最小二乘法，以不同时间采摘的哈姆林甜橙果实为样品建立其可溶性固形物、含酸量和VC 的无损检测数学模型，同时对不同光谱预处理方法和不同建模波段范围对模型的预测性能进行对比分析。结果表明：原始光谱在400～1000nm 波段的模型预测精度较高。经多元散射校正和5 点移动平均平滑预处理后，果实可溶性固形物含量的PLS 模型最好，校正集样品的相关系数为0.995RMSEC和RMSEP分别为0.026%、0.028%；预测集样品的相关系数为0.992。经多元散射校正和9 点移动平均平滑预处理后，果实含酸量的PLS 模型最好，校正集样品的相关系数为0.997，RMSEC 和RMSEP 分别为0.012%、0.013%；预测集样品的相关系数为0.997。经多元散射校正和9 点移动平均平滑预处理后，果实VC 含量的PLS 模型最好，校正集样品的相关系数为0.998，RMSEC 和RMSEP 分别为0.009%、0.009%；预测集样品的相关系数为0.999。可见由不同时间采摘的果实组成的样品集所建立的数学模型可以提高模型的预测精度，从而提高模型的适用范围。应用可见-近红外漫反射光谱检测哈姆林甜橙果实的内在品质可行。

Non-destructive Measurement of Soluble Solids,Vitamin C and Titratable Acidity of Hamlin Sweet Orange using Vis/NIR Spectrometry

The potential of reflectance visible/near infrared spectroscopy (VNIRS) was investigated for measuring total soluble solids (TSS), vitamin C (VC) and titratable acidity (TA) in Hamlin orange fruit (Citrus sinensis L.). VNIR spectra of Hamlin orange fruits harvested at different times were measured and related with the contents of TSS, VC and TA by partial least squares (PLS) method to establish nondestructive models for predicting the TSS, VC and TA in the fruit. Meanwhile, the effects of different spectral pretreatment methods and spectral waveband range on the performance of the established models were also investigated. The results showed that the PLS models of original spectra within the waveband range from 400 to 1000 nm gave optimal predictions for TSS, VC and TA. Through multiple scatter calibration and 5-point moving-average smoothing pretreatment, an optimal TSS prediction model was obtained, with a correlation coefficient of 0.995 and a root mean square error of calibration (RMSEC) of 0.026% for the calibration sample set and a correlation coefficient of 0.992 and a root mean square error of prediction (RMSEP) of 0.028% for the validation sample set. Multiple scatter calibration and 9-point moving-average smoothing pretreatment gave an optimal TA prediction model, with a correlation coefficient of 0.997 and a RMSEC of 0.012% for the calibration sample set and a correlation coefficient of 0.997 and a RMSEP of 0.013% for the validation sample set. An optimal VC prediction model was also obtained through multiple scatter calibration and 9-point moving-average smoothing pretreatment, with a correlation coefficient of 0.998 and a RMSEC of 0.009% for the calibration sample set and a correlation coefficient of 0.999 and a RMSEP of 0.009 for the validation sample set. These results suggest that the use of a sample set comprising Hamlin organe fruits collected at different harvesting times can improve the accuracy of a PLS prediction model.