基于机器学习的红外光谱丹参聚类分析

Clustering analysis based on FT-IR spectra and machine studying of Danshen

中药材的成分非常复杂,对中药材的质量分析一直是中药制药产业当中的重要一环,而且也是生产体系当中的重点、难点。以往传统的质量分析方法或是由于不可避免的主观因素占主导地位而可靠性较差,或是由于技术局限而破坏了中药材的整体性,因而对中药材进行客观的整体分析就显得极为重要。与此同时,傅里叶变换红外光谱法因其快速、准确、无损检测等特点正受到越来越多的关注,在很多行业领域中也已得到广泛的应用。本文创新地运用红外光谱技术结合模式识别技术,分别对不同产地和不同等级的丹参药材进行了建模分析,并在此基础上评价了3种常用的聚类算法——支持向量机(SVM)、自适应提升算法(AdaBoost)以及线性判别函数(LDF)——在丹参药材模型建立以及聚类分析中的应用效果。结果表明,LDF对不同产地丹参药材聚类模型的识别率和拒绝率最高;而等级分类建模方面则以SVM的识别率和拒绝率最高,且均可达到97%以上。由此可以得出结论,与传统的质量分析方法相比,红外光谱与模式识别相结合的新分析技术是解决中药整体分析的有效方法之一。

The qualitative analysis, due to the components of TCM, is playing an important role in pharmaceutical industries. Some of the traditional analytical methods, however, are short in dependability because of the unavoidable subjective complications while the others are destroying the macrocosm of TCM. On the other hand, FT-IR （Fourier Transform Infrared Spectroscopy）, known for its swiftness, veracity and scatheless scanning, has attracted great attentions and thus gained extensive applications. Together with mode-identifying, FT-IR is introduced to distinguish Danshen models of different regions or different grades; what＇s more, the impact on the model establishing of three common algorithms, SVM （Support Vector Machine）, AdaBoost （Adaptive Boosting） and LDF （Linear Discriminate Function）, is also evaluated. It is discovered LDF is the most effective to different areas as a result of its highest distinguishing rate and the rejecting rate （97%）, the same as SVM to different grades. Thus it is convinced that, comparing with traditional techniques, the IR-MI is supposed to be one of the most effective methods to the macrocosmic analysis of TMC