中文摘要：

在水果内部品质检测分级实际生产中往往存在多通道测量,由于仪器不同或加工精度不同而存在多通道间检测模型不具通用性问题,应用多种模型传递方法研究了在线检测条件下两个不同可见/近红外光谱仪间的皇冠梨糖度预测模型传递及预测比较分析。结果表明：从仪器的光谱数据经直接校正算法（DS）和基于平均光谱差值校正的DS算法（MSSC-DS）转换后用于主仪器所建模型的预测结果相对较好,预测均方根误差小于0.5°Brix,可以满足实际生产。但通过模型转换后的预测结果均比利用从仪器数据直接建模的预测结果要差（预测均方根误差为0.381°Brix）,因而在实际生产中,需要从成本和分级精度的要求来考虑选择建模的方式。

英文摘要：

With the development of social economy and growth of people＇s living standand, the demond of fruit quality is ever increasing. Quality detection and grading of postharvest fruit is an integral part of eommoditization processing, which is also an effective way to achieve high price with good quality. Visible/NIR spectroscopy with the advantages of rapid, nondestructive and being on-line analyzing, has been widely used in agriculture. In the actual application of visible/NIR spectroscopy for on-line detection of fruit internal quality, multi-channels measurement often exists, in which the prediction model is not universal among multi channels due to different spectrometers or their different manufacture precisions. Calibration model transfer is a key problem in visible/NIR spectral quantitative analysis. Comparative analysis of some calibration model transfer methods, such as direct standardization （DS） , piecewise direct standardization （PDS）, slope/bias （S/B） between two different visible/NIR spectrometers （master and slave spectrometers, model QE65000 and QE65Pro, Ocean Optics, Inc., USA） in the sugar content on-line detection of crown pears was carried out at conveyor speed of 0.5 m/s. The results showed that the prediction values by DS algorithm and DS algorithm based on the mean spectra subtraction correction （MSSC- DS） were relatively good with low root mean square error of prediction （RMSEP） of less than 0.5° Brix, which can satisfy the industry application. And pre- processing method of MSSC can improve the prediction accuracy of calibration model transfer by eliminating and mitigating the differences between the spectra acquired on master and slave spectrometers. However, the best prediction result on salve instrument after calibration model transfer （ RMSEP was 0. 453°Brix） was still inferior to that predicted by the model developed directly using slave data （RMSEP was 0. 381°Brix）. Thus, in the actual application, appropriate modeling selection should be considered f