论文部分内容阅读
采用近红外光谱法对不同厂家的盐酸西替利嗪片进行鉴别。用固体光纤在不同时间段采集了3个生产厂家的65批盐酸西替利嗪片的近红外漫反射光谱,对光谱数据进行预处理优化,采用基线校正,9点平滑,一阶导数和向量标准化的预处理方法,采用无监督学习算法即聚类分析法进行分类,并且比较了3种不同的聚类分析方法的分类结果;用有监督学习算法即人工神经网络法,运用改进的BP算法——Levenberg-Marquardt方法对46个样本建立校正模型,并且对其余的19个未知样本进行预测。聚类分析法和人工神经网络法都能得到满意的结果,其中经过主成分分析法提取特征变量后的聚类分析结果比直接进行聚类分析和经过核主成分分析法进行特征变量提取后的聚类分析的结果差。结果表明,用主成分分析法提取了前几个主成分不一定包含绝大部分聚类特征和结构,并且运用近红外光谱法与化学计量学结合可作为一种简单、快速、无损、可靠的方法用于鉴别不同厂家的西替利嗪片。
Near-infrared spectroscopy of different manufacturers of cetirizine hydrochloride tablets were identified. Near-infrared diffuse reflectance spectra of 65 batches of cetirizine hydrochloride tablets from 3 manufacturers were collected at different time periods using solid optical fiber. The spectral data were preprocessed and optimized. Baseline calibration, 9-point smoothing, first derivative and vector The standardized preprocessing methods were classified by unsupervised learning algorithm, that is, cluster analysis, and the classification results of three different cluster analysis methods were compared. Using the supervised learning algorithm, that is, artificial neural network method, the improved BP algorithm The Levenberg-Marquardt method establishes a calibration model for 46 samples and predicts the remaining 19 unknown samples. The results of cluster analysis and artificial neural network can get satisfactory results. The result of cluster analysis after the feature variables are extracted by the principal component analysis method is more accurate than the cluster analysis and the principal component analysis after the feature variables are extracted The result of cluster analysis is poor. The results show that the principal components analysis extracted the first few principal components do not necessarily contain most of the cluster features and structures, and the use of near infrared spectroscopy combined with chemometrics can be used as a simple, fast, non-destructive and reliable The method was used to identify cetirizine tablets from different manufacturers.