中文摘要：

为了实现赣南脐橙按内部品质分级,对赣南脐橙可溶性固形物（SSC）进行快速在线检测研究.利用USB4000微型光纤光谱仪在0.3m·s^-1的输送速度下在线采集赣南脐橙的半透射光谱（470～1 150 nm）,并采用CARS变量选择方法对波长变量进行优选,对优选的波长变量应用偏最小二乘（PLS）回归建立脐橙SSC在线预测模型,最后利用脐橙SSC在线预测模型对完全独立的预测集样本进行预测.研究结果表明：CARS能有效筛选有用的波长变量,提高预测模型的预测精度;与全光谱PLS模型相比,CARS-PLS模型的交互验证相关系数由0.871上升为0.934,交互验证均方根误差（RMSECV）由0.560％下降为0.412％;独立预测集样本SSC的预测均方根误差（RMSEP）为0.649％,SSC预测残差落在±1.0％界限以内的样本占总预测样本数的86.3％,基本可以满足脐橙SSC在线检测分级的需要.

英文摘要：

In order to grade Gannan navel oranges by internal qualities,on-line and fast detection method of soluble solids content （SSC） was established.Semi-transmission spectra of Gannan navel oranges were acquired at moving speed of 0.3 m · s^-1 by USB4000 micro fiber spectrometer （470 ～1 150 nm）.The important wavelength variables for SSC were selected by CARS variable selection method to establish online prediction model by partial least squares （PLS） regression.The prediction model was used to predict SSC of navel oranges in fully independent prediction set.The results indicate that CARS can effectively select wavelength variables for SSC of navel oranges with improved model prediction precision.Compared to full-spectrum PLS,the model performance of CARS-PLS is improved.The correlation coefficient of cross validation is increased from 0.871 to 0.934,and the root mean square error of cross validation （RMSECV） is decreased from 0.560% to 0.412%.For fully independent prediction set,the root mean square error of prediction （RMSEP） of SSC is 0.649%.The samples that have predicted residual errors of SSC within the limits of ± 1.0% account for 86.3% of total prediction samples.The proposed method can basically satisfy the requirement of on-line detection and grading for SSC of navel oranges.