千禧番茄营养丰富且酸甜可口, 不同千禧番茄品种的风味和营养价值均有明显差异, 尤其是番茄红素、 柠檬酸、 维生素C和氨基酸含量的差异较大。 传统人工分类方式效率低、 主观性强、 误检率高等问题亟待解决。 为筛选综合营养价值高且风味佳的千禧番茄品种, 实现千禧番茄快速准确分类, 提出了基于千禧番茄光谱特征的分类模型构建及GWO优化SVM算法研究, 以期解决千禧番茄自动化分类问题。 以四个品种千禧番茄作为研究对象, 试验样本240个, 将其按2:1比例划分为训练集160个和测试集80个样本, 利用可见-近红外光谱采集系统获取350~1 000 nm范围内的千禧番茄反射强度, 经光谱校正得样本反射率; 为增强信噪比, 截取481.15~800.03 nm范围内的光谱波段作为有效波段。 由于数据采集过程受无关信息干扰影响建模效果, 故将平滑点数设置为3进行Savitzky-Golay(SG)平滑预处理。 预处理后采用连续投影算法(SPA)提取特征波长变量, 优选得到11个特征波长反射率作为输入矩阵X, 预设样本变量1, 2, 3和4作为输出矩阵Y, 利用支持向量机(SVM)建立SPA-SVM千禧番茄定性分类模型, 训练集和测试集平均分类准确率分别为59.38%和48.75%; 在此基础上, 引入灰狼优化算法(GWO)对训练集160个样本训练, 寻求SVM最优惩罚系数(c)和核函数参数(g), 根据模型训练结果对测试集80个样本预测, 建立SPA-GWO-SVM千禧番茄分类模型, 训练集和测试集平均分类准确率分别为100%和81.25%。 研究结果表明: 经灰狼算法优化后的支持向量机模型性能明显提高, 其中训练集和测试集平均分类准确率分别提高了40.62%和32.50%, 灰狼优化算法可用于提高支持向量机的分类性能, 实现对千禧番茄品种的分类。 本研究为千禧番茄及其他果蔬快速准确分类提供了新的思路和方法。

Qianxi tomatoes are rich in nutrition, tasting sweet, sour and delicious, different varieties of qianxi tomato's flavor and nutritional value is obviously different, especially lycopene, citric acid, vitamin C and amino acid content varies greatly and the traditional artificial classification method of low efficiency, strong subjectivity, high rate of error detection and other issues are pressing to be solved. Therefore, in order to screen the high comprehensive nutritional value and good flavor of the qianxi tomatoes to achieve the rapid and accurate classification of the qianxi tomatoes, a classification model based on qianxi tomatoes spectral features and a GWO optimized SVM algorithm was proposed to solve the problem of automated qianxi tomatoes classification. In this study, a total of 240 qianxi tomatoes of four varieties were taken as the research objects, divided into 160 training sets and 80 test sets according to the ratio of 2:1. The qianxi tomatoes fruit reflective intensity in the range of 350 to 1 000 nm was obtained by using a visible/near-infrared spectral acquisition system, and the sample reflectance by spectrally corrected was obtained and analyzed. The effective information of the qianxi tomatoes spectrum in the range of 481.15 to 800.03 nm was intercepted to enhance the signal-to-noise ratio. Since the modeling effect is affected by the interference of irrelevant information in the data acquisition process, Savitzky-Golay (SG) smoothing pretreatment was performed with the smoothing point to 3. After SG smoothing pretreatment, the characteristic wavelength variables are extracted by successive projections algorithm (SPA), the reflectance of the optimal selected 11 characteristic wavelength variables as the input matrix X, preset sample variables 1, 2, 3, and 4 as output matrix Y, the SPA-SVM qualitative classification model of qianxi tomatoes was established. The average classification accuracy of the training set is 59.38%, the test set is 48.75%. On this basis, the gray wolf optimization (GWO) algorithm was introduced to train 160 samples training set, seeking the optimal penalty coefficient (c) and the nuclear function parameter (g) of the SVM. Based on the training results of the model, the classification results of 80 test set samples were predicted to establish the SPA-GWO-SVM qualitative classification model of qianxi tomatoes and the average classification accuracy of the training set is 100%, the test set is 81.25%. The research results show that the performance of the support vector machine model optimized by the grey wolf algorithm has been improved significantly. The average classification accuracy of the training set is improved by 40.62%, and the average classification accuracy of the test set is improved by 32.50%, which shows that the gray wolf optimization algorithm can be used to improve the performance of the support vector machine classification model and realize the classification of qianxi tomatoes. This study provides a new idea and method for the rapid and accurate classification of qianxi tomatoes and other fruits and vegetables.