优化的近红外光谱LS-SVM模型测定小麦蛋白质

Determination of Wheat Protein by Optimized Near Infrared Spectrum LS-SVM Model

小麦蛋白质测定的常规方法为湿化学分析法,具有操作复杂、污染环境和耗时较长等缺点,为此建立了一个近红外光谱LS-SVM模型,以实现小麦蛋白质含量的简便、快速测定。首先对样品光谱进行"均值中心化+去趋势"预处理,并用SPXY法划分校正集和测试集样本;然后采用改进的二进制蝙蝠算法(IBBA)进行建模参数和特征波长的联合优化,根据优化结果对校正集数据建立LS-SVM模型,并用测试集数据验证其性能;最后通过与常用的PLS、CARS-PLS及未优化的SVM、LS-SVM建模结果进行比较,确认该模型的有效性。结果表明,该模型的各项性能指标优异,能够满足实际检测工作的要求。

The conventional method of wheat protein determination is wet chemical analysis, which has the disadvantages of complex operation, polluting environment and long time-consuming. A LS-SVM model of near infrared spectroscopy was established to realize the simple and rapid determination of protein content in wheat. Firstly, the sample of spectra was pretreated by "mean centralization+de-trending", and the samples of calibration set and test set were divided by SPXY method. Then, an improved binary bat algorithm(IBBA) is used to optimize the modeling parameters and characteristic wavelengths. LS-SVM model is built based on the calibration set data according to the optimization results, and its performance is verified by the test set data. Finally, the validity of the model is confirmed by comparing the modeling results with PLS, CARS-PLS and non-optimized SVM and LS-SVM. The experimental results show that the performance of the model is excellent and can meet the requirements of practical detection.