基于MCCV-CARS-RF建立红提糖度和酸度的可见-近红外光谱无损检测方法

利用USB2000＋微型光谱仪采集红提400～1?000?nm透过率光谱数据，并通过理化分析测得糖度和酸度值；利用SavitZky-Golay卷积平滑法对原始光谱进行预处理，结合蒙特卡罗交叉验证法剔除奇异点，再利用竞争自适应重加权采样法降维，最终建立随机森林预测模型。糖度预测模型的校正集相关系数和均方根误差分别为0.955?8和0.315?8；验证集相关系数和均方根误差为0.956?8和0.318?5。酸度预测模型的校正集相关系数和均方根误差分别是0.945?6和0.300?1；验证集相关系数和均方根误差为0.940?5和0.311?2。结果表明，该方法适用于红提糖度和酸度的快速无损检测，且具有较高的准确度。

Nondestructive Detection of Sugar Content and Acidity in Red Globe Table Grapes Using Visible Near Infrared Spectroscopy Based on Monte-Carlo Cross Validation-Competitive Adaptive Reweighted Sampling-Random Forest (MCCV-CARS-RF)

A USB2000+ micro spectrometer was used to acquire the transmittance spectra of Red Globe table grapes in the range of 400–1 000 nm. Moreover, the sugar content and acidity value were measured chemically. The original spectra were pretreated by SavitZky-Golay smoothing (SG) and then the singularity was eliminated by Monte-Carlo cross validation method (MCCV) followed by dimension reduction by competitive adaptive reweighted sampling for development of a random forest (RF) prediction model. The correlation coefficient and root mean square error of the sugar prediction model were 0.955 8 and 0.315 8 for the calibration set, and 0.956 8 and 0.318 5 for the validation set, respectively. The correlation coefficient and root mean square error of the acidity prediction model were 0.945 6 and 0.300 1 for the calibration set, and 0.940 5 and 0.311 2 for the validation set, respectively. The results showed that this method could be suitable for rapid, nondestructive and accurate detection of sugar content and acidity in Red Globe table grapes.