基于两种模式识别技术的盐酸左氧氟沙星注射液近红外光谱定量分析

测定同一厂家生产的53个不同批次盐酸左氧氟沙星注射液的近红外光谱. 先利用小波变换技术对光谱变量进行去噪, 并对其有效的压缩, 以提高建模效率, 再分别利用神经网络及支持向量机技术建立盐酸左氧氟沙星注射液样品的定量分析模型, 并讨论了建模过程中相关参数的优化选择. 仿真实验表明, 建立的SVM定量分析模型的相关性要优于BP网, 同时SVM定量分析模型的RMSECV及RME两个指标值也显示其预测效果良好, 泛化能力强.

Application of Two Pattern Recognition Techniques to Near-infrared Spectroscopy Quantitative Analysis of Levofloxacin Hydrochloride for Injection

The 53 samples of Levofloxacin Hydrochloride for injection from different batches of a factory were surveyed by near-infrared (NIR) spectroscopy. The spectrum variables of all the samples had been efficiently compressed and de-noised through the wavelet transformation ( WT) technology before the models were established by pattern recognition techniques. The two quantitative analysis models of Levofloxacin Hydrochloride for injection established via support vector machine ( SVM ) and artificial neural network (ANN) were studied separately in this experiment using radial basis function ( RBF) SVM and back propagation (BP) network, and the related parameters were also discussed in detail. The simulation results show that the correlation of predicted values and chemical determination values of SVM model is better than that of ANN model, and SVM model owns excellent generalization for quantitative analysis results and high prediction accuracy.