苹果糖度的光谱模型温度补偿设计

Design of spectral model for temperature compensation in measurement of sugar content in apples

目的为减少温度对便携仪器近红外光谱模型预测的影响,尝试构建局部温度混合校正模型,结合温度信息来预测不同温度下的苹果内部品质。方法以16、24、32℃贮藏温度下红富士苹果为原料,分别用内置微型光谱仪的自制便携式水果分析仪器获得其透射光谱,结合温度传感器获取环境温度,用阿贝折射仪测定苹果糖度。建立单一温度校正模型、全局温度混合校正模型和局部温度混合校正模型对不同温度的样本进行预测。结果单一温度校正模型对不同温度下苹果糖度预测均方根误差为0.474～3.125%Brix;当采用全局温度混合校正模型时能降低温度对光谱的影响,预测均方根误差分布在0.488～0.533%Brix。根据待测样本的温度来构建多个局部温度混合校正模型,对不同温度下苹果糖度的预测均方根误差为0.462～0.500%Brix。结论局部温度混合校正模型可以结合样本温度信息预测苹果糖度,降低温度对模型的影响,同时能减少初期建模成本。

Objective To develop a local temperature-mixed calibration model for reducing the influence of temperature on the prediction of spectral model embedded in portable analyzer, in order to predicte the internal quality of apple at different temperature according to sample＇s temperature. Methods ＇Fuji＇ apples were restored at 16, 24, 32℃, respectively, and their transmittance spectra were acquired by a portable fruit analyzer with micro spectrometer. Combining the temperature sensor to obtain the ambient temperature, sugar content （SC） of apples was measured by Abbe refractometer. Simplex, global and local temperatures-mixed models were calibrated between the spectra and apples＇ SC, and the prediction performance of these temperatures-mixed models was systematically compared in predicting the SC in apple. Results Controlled at different temperature, the SC of apples were predicted by simplex temperature calibration model with root mean squared error of prediction （RMSEP） of 0.474%-3.125% Brix, while predicted by global temperature mixed model with an obviously reduced RMSEP of 0.488%-0.533% Brix, which lowering the temperature influence on the calibration model. Multi-local temperature-mixed models were developed in the multi-interval temperature ranges, and they could predict adaptively sample＇s SC according to the sample＇s temperature, with RMSEP of 0.462%-0.500% Brix. Conclusion The local temperature mixed calibration models can use the sample temperature information to predict the sugar content of apple, and reduce the effect of temperature on the model and the initial modeling costs at the same time.