基于红外光谱和随机森林的蕨麻产地鉴别

Identifying the Origin of Potentilla Anserine Based on Infrared Spectroscopy and Random Forest Method

利用红外光谱与随机森林相结合的方法对不同产地蕨麻进行分类鉴别,测定了42个来自青海省不同产地的蕨麻样品的红外光谱。小波变换对红外光谱原始谱图数据进行了预处理,红外光谱数据压缩到原来的1/8,其分析精度与原始光谱数据基本相当。将42个样品划分为有30个样品的训练集和12个样品的测试集,建立随机森林预测蕨麻产地模型。使用内部交叉验证和外部数据进行验证,采用R语言实现随机森林算法,并对模型的参数进行了优化。结果表明,所建立的判别模型中训练样本和测试样本判别正确率均为100%。建立的模型能够正确地对蕨麻样品快速进行产地鉴别,红外光谱法结合随机森林可作为中药材产域分类鉴别的一种新的尝试。

The infrared spectroscopy combining with random forest method was used in the identification of Potentilla anserine from different fields of Qinghai Province. Forty-two samples of Potentilla anserine from different fields of Qinghai province were surveyed by FTIR（ Fourier transform infrared spectroscopy）. The original data matrix of FTIR was pretreated with wavelet transform. The results showed that the infrared spectroscopy data were compressed to 1/8 of its original data,but the spectral information and analytical accuracy were not deteriorated. The 42 samples of Potentilla anserine were divided into 30 training samples and 12 validation samples. Random forest model was constructed by the training samples to predict the discrimination effect of identifying the origin of Potentilla anserine with internal cross validation and external validation sample. R language was adopted to achieve algorithm of random forest. Parameters of random forest model were optimized. The prediction accuracy of the proposed model was 100% for the training samples and 100% for the test samples. It can be concluded that the method is quite suitable for the fast discrimination of producing areas of Potentilla anserine. This infrared spectral analysis technology combined the random forest was proved to be a reliable and new practical method for the identification of geographical origin of Chinese medicine. The method in the present paper is very broad prospect of application.