论文部分内容阅读
利用近红外光谱技术结合支持向量机对植物油脂酸值含量进行回归预测。收集大豆油、花生油等油样共37份,应用激光近红外光谱仪对油样进行光谱采集,采用标准正态变量变化、多元散射校正和正交信号校正3种不同方法进行预处理。运用网格搜索法进行参数寻优,寻找最佳参数组合(C,g),建立支持向量机回归模型进行定量预测。研究表明,经过SNV、MSC和OSC预处理数据建立的模型的惩罚因子C均只有1,大大降低了模型出现过拟合现象的概率,提高了模型的泛化能力、稳健性和预测能力;预处理方法MSC和SNV建立的SVR模型校正集相关系数R较高,均达到99%;OSC建立的SVR模型具有最佳的预测性能,预测相关系数R达到93%以上;采用激光近红外光谱技术预测植物油脂酸值含量的方法是可靠的,为实现植物油脂酸值的快速检测提供了重要的依据。
Using NIRS and Support Vector Machines to Predict the Oleic Acid Content in Plant. A total of 37 oil samples such as soybean oil and peanut oil were collected. The oil samples were collected by laser near infrared spectroscopy (NIRS), and the samples were pretreated by three different methods: standard normal variation, multivariate scatter calibration and orthogonal signal correction. The method of grid search is used to find out the best parameter combination (C, g), and the support vector machine regression model is established for quantitative prediction. The results show that the penalty factor C of the model established by SNV, MSC and OSC pretreatment data is only 1, which greatly reduces the probability of model overfitting and improves the generalization ability, robustness and predictive ability of the model. The correlation coefficient R of the calibration set of SVR model established by MSC and SNV is high, both of which reach 99%. The SVR model established by OSC has the best prediction performance and the prediction correlation coefficient R reaches above 93%. The prediction by laser near-infrared spectroscopy The method of vegetable oil acid value is reliable, which provides an important basis for the rapid detection of vegetable oil acid value.