采用可见/近红外光谱技术在线检测水果糖度, 每个水果品种要单独建模, 模型升级维护耗时费力。 探讨建立苹果、 梨等薄皮水果可溶性固形物(SSC)在线检测通用数学模型的可行性。 利用自行设计的可见/近红外漫透射光谱在线检测系统, 在积分时间80 ms、 单线速度5个/s的条件下, 采集新梨7号、 砀山酥梨、 玉露香梨和富士苹果四种水果的可见/近红外漫透射光谱。 分析了四种水果的可见/近红外漫透射光谱响应特性, 采用变异系数法和连续投影算法, 筛选通用数学模型建模用光谱变量, 并建立了偏最小二乘和最小二乘支持向量机梨与苹果梨通用数学模型。 采用新样品评价模型的预测能力, 变异系数法筛选光谱波段建立的偏最小二乘通用数学模型预测精度最高, 通用模型预测梨和苹果梨模型预测均方根误差分别为049%和055%, 通用模型预测相关系数分别为088和093； 独立模型预测新梨7号、 玉露香梨、 砀山酥梨和富士苹果的预测相关系数分别为093, 091, 088和095, 预测均方根误差分别为040%, 042%, 041%和046%。 通用数学模型的预测精度略低于每个品种的独立数学模型, 但是通用模型的通用性高于单一模型。 实验结果说明采用变异系数法结合偏最小二乘法建立薄皮水果在线检测通用数学模型, 实现四种水果糖度在线检测是可行的。Mathematical Model

It takes a plenty of time for model updating and maintenance when use visible near infrared spectroscopy to measure the soluble solids contents(SSC)of fruits online and each of the fruit varieties need to be modeled separately. This paper aimed to explore the feasibility of establishing the online detection universal mathematical models of thin skinned fruits such as the apples and pears . The online visible near infrared spectroscopy diffuse transmission spectra detection system which was designed by ourselves was applied. Under the condition of integral time 80 ms, single speed 5 s-1 collecting visible near infrared spectroscopy diffuse transmission spectra of Xinli No. 7, Dangshan pear, Yulu pear and Fuji apple. The spectra response characteristics of near infrared diffuse transmission of four kinds of fruit were analyzed by using the variation coefficient method and continuous projection algorithm screened the modeling spectral variables of the universal mathematical mode and establish partial least squares and least squares support vector machine universal mathematical models of apple and pears finally. New samples were used to evaluation the predictive ability of the universal model. The coefficient variation method by screening similar band to established the universal mathematical model of partial least squares spectral had the highest prediction accuracy. Pear and apple pear universal models correlation coefficient (rp) of prediction are 088 and 093 and the root mean square error of prediction (RMSEP) are 049% and 055% respectively; the correlation coefficient of independent model for predict Xinli No7, Yulu pear, Dangshan pear and Fuji apple were 093, 091, 088 and 095, and the root mean square error of prediction are 040%, 042%, 041% and 046% respectively. The prediction accuracy of the universal mathematical model is slightly lower than the independent mathematical model prediction accuracy of each variety, but the generality of universal model is higher than the single model. The experiment results shows that using coefficient variation method combined with partial least squares method to establish the online detection general mathematical model of thin skinned fruit is feasible in achieving four kinds of fruit sugar online detection .