梨可溶性固形物含量的在线近红外光谱检测

应用近红外透射检测技术在线检测梨的可溶性固形物(SSC)。在实验台上以0.5 m·s-1的速度，300 W的光照强度，采用半透射方式检测梨的光谱。实验采用的梨样品为187个，其中147个样品为校正集，40个样品为预测集，应用偏最小二乘回归(PLS)和主成分回归(PCR)建立梨可溶性固形物的在线预测模型。选取550～700 nm, 700～850 nm, 550～850 nm为建模波段范围，发现无论对于PLS还是PCR，都是550～850 nm波段的建模结果好。本实验还研究对比不同的光谱预处理方法(光谱平滑，一阶微分，二阶微分等)对预测模型性能的影响，其中5点S-G(Savitzky-Golay)光谱平滑能有效地提高光谱的信噪比，改善模型预测精度，而一阶微分、二阶微分对模型性能改善基本上没有影响；最好的预测模型相关系数r=0.948 8， 校正标准差RMSEC=0.236，预测标准差RMSEP=0.548。结果表明：PLS模型预测性能较好，梨可溶性固形物的在线检测具有可行性。

Online Detection of Soluble Solids Content of Pear by Near Infrared Transmission Spectrum

The research was to detect soluble solids content (SSC) of pear online by near infrared transmission spectrum. The movement speed of pear was 0.5 m·s-1 the power of light source was 300 W, and semi-transmission was used to collect the spectrum of pears. The total experiment samples were 187 pears, with a calibration set of 147 pears and a validation set of 40 pears. Partial least squares (PLS) and principal component regression (PCR) technique were used to develop the calibration model for online detection. Spectral ranges of 550-700 nm, 700-850 nm, 550-850 nm were used to establish the calibration models, and it was found that the model with 550-850 nm was better than others whether for PLS or for PCR. Also, the models based on different pretreatment methods such as Savitzky-Golay smooth, first derivative, second derivative and so on were compared, and the result showed that the five point Savitzky-Golay smooth could increase S/N ratio and improve performance of the model, whereas first derivative and second derivative could do little to improve performance of the model. The best model had the satisfactory calibration and prediction abilities, with the correlation coefficient (R_C)=0.948 8, root mean square error of calibration (RMSEC)=0.236 and root mean square error of validation (RMSEP)=0.548. The result in this study shows that the detection of SSC of pear online is feasible.