测试部位、温度对苹果品质近红外分析准确度的影响

本实验以苹果为试材建立常温近红外光谱无损检测模型，研究测试部位和样品温度对分析模型精度的影响，并探讨温度补偿模型最少用果量。结果表明，阳面2 点和阴面2 点光谱混合后平均所得光谱建立模型的预测精度最高，水分和颜色模型交互验证相关系数分别达0.8274 和0.8005；当不同温度样品添加量达到校正集样品数的20% 时基本可以消除温度波动对定量分析模型的影响；温度补偿验证结果表明，可溶性固形物、水分和颜色温度混合校正模型预测标准误差分别比常温模型低15.52%～29.77%、13.68%～31.68% 和45.9%～88.46%。

Effects of Testing Position and Temperature on Accuracy of Near-Infrared Prediction Model for Apple Quality

A nondestructive measurement model predicting quality of apple fruit was developed at room temperature by near infrared (NIR) spectroscopy. The effects of testing position and temperature on NIR model accuracy of evaluating apple qualities was studied, and the minimal quantity of samples used to compensate temperature fluctuations was discussed. Results showed that the accuracy of model constructed by averaging over spectra gained from two positions of each side (sun-expose side and shaded side) is the highest. One minus variance ratios of moisture and color model reach to 0.8274 and 0.8005, respectively. Samples of different temperatures (twenty percent of calibration sets) are needed to add in room temperature model to compensate temperature fluctuations. Hybrid calibration model accuracies of soluble solid content, moisture and color (°Hue) after temperature compensation are much higher than those of model at room temperature. Standard errors of the prediction of hybrid calibration model of soluble solid content, moisture and color are lower than those of model at room temperature by 15.52% to 29.77%, 13.68% to 31.68% and 45.9% to 88.46%, respectively.