成熟度是水果评价的重要标准，直接影响水果的品质和经济价值。针对红提采摘成熟度评判困难，果肉营养价值参差不齐、产品竞争力低等问题，建立基于可见/近红外光谱技术的红提成熟度判别模型。该研究选取红提生长过程的4个阶段（分别为：未成熟、半成熟、成熟、过熟）的样本并进行光谱信息采集。选择550 nm~1 000 nm的光谱波段建模，分别将经过预处理的光谱用竞争性自适应加权算法（Competitive Adaptive Reweighted Sampling，CARS）、无信息变量消除算法（Uniformative Variable Elimination，UVE）和连续投影算法（Successive Projection Algorithm，SPA）进行特征波长提取，建立支持向量机（Support Vector Machines，SVM）、极限学习机（Extreme Learning Machine，ELM）和偏最小二乘判别分析（Partial Least Squares Discriminant Analysis，PLS-DA）的判别模型，最终建立可见/近红外光谱技术的红提成熟度的最佳判别分类模型。研究结果表明，在Savitzky-Golay（SG）卷积平滑处理算法光谱预处理后运用SPA算法进行特征波段提取建立的ELM模型成熟度判别分类效果最佳，SVM模型次之，PLS-DA模型最差。因此，红提成熟度的最佳判别分类模型为SG-SPA-ELM，该模型的训练集和测试集的准确率分别为97.50%和96.67%。利用可见/近红外光谱技术对红提成熟度进行判别是可行的，该研究为红提成熟度的判别找到了一种新的无损检测方法。

Maturity is an important criterion for evaluating fruit, as it directly affects the quality and economic value. A discriminant model based on visible/near-infrared spectroscopy was established to determine the maturity of red globe grapes to simplify the process of assessing the maturity, uneven nutritional value, and low competitiveness of red globe grapes. Spectral information on the samples was collected from four stages of the red globe grapes growth period (immature, semi-mature, mature, and super-mature). The spectral band of 550~1 000 nm was selected for modeling. Pre-processed spectra were extracted by competitive adaptive reweighted sampling, uninformative variable elimination, and successive projection algorithm (SPA) to establish the discriminant models of support vector machine, extreme learning machine (ELM), and partial least squares discriminant analysis (PLS-DA), respectively. The best discriminant classification model for the maturity of red globe grapes based on visible/near-infrared spectroscopy was established. The results showed that the ELM model for discrimination and classification of maturity, established by applying the SPA for feature wavelength extraction after spectral pre-processing using the Savitzky–Golay (SG) algorithm showed the best results, followed by the support vector machine model and then the PLS-DA model. Therefore, the best discriminant classification model for red globe grape maturity was SG-SPA-ELM. The accuracy of this model was 97.50% and 96.67% for the training sets and test sets. Therefore, visible/near-infrared spectroscopy can be applied to determine the maturity of red globe grape using a non-destructive method.

国家自然科学基金项目（31871863；32072302）；湖北省自然科学基金项目（2012FKB02910）；湖北省研究与开发计划项目（2011BHB016）