Data fusion of laser induced breakdown spectroscopy (LIBS) and infrared spectroscopy (IR) coupled with random forest (RF) for the classification and discrimination of compound salvia miltiorrhiza

Abstract

As an important traditional Chinese medicine, compound salvia miltiorrhiza has many functions, such as reducing blood pressure, promoting blood circulation and relieving pain, nourishing heart and tranquilizing mind, etc. Therefore, it is significant important to monitor the quality of compound salvia miltiorrhiza. In this work, a novel strategy based on data fusion of LIBS and IR coupled with RF was constructed for the classification and discrimination of compound salvia miltiorrhiza. The LIBS and IR spectra collection of compound salvia miltiorrhiza samples was carried out. In data fusion section, normalization was used for eliminating orders of magnitude difference between the LIBS and IR spectral data intensity, variable importance (VI) and variable importance in the projection (VIP) were applied to select the feature variables in the LIBS and IR spectral data, and variable importance threshold was optimized based on out of bag (OOB) error estimation. Then, the optimized feature variables was fused by a feature layer data fusion. Finally, the RF discrimination model was constructed based on LIBS-IR fusion data, and the obtained result was compared with the RF discrimination models based on LIBS and IR spectra. The results show that the RF discrimination model based on LIBS-IR fusion data selected by VI can achieve the best predictive performance for the quality analysis of compound salvia miltiorrhiza with the sensitivity ​= ​0.9333, the specificity ​= ​0.9667 and the accuracy ​= ​0.9619 for testing set. The result fully certificates the feasibility of the quality analysis of compound salvia miltiorrhiza via data fusion based on LIBS-IR spectra proposed in the work, and it will provide a new technology for the identification and analysis of Chinese medicinal materials.