Research on the Quantitative Analysis of Near Infrared Spectroscopy of Asiatic Moonseed Based on Support Vector Machine and Wavelet Transform

Abstract

With rapidly analysis, no pollution, no damage, simple operation, low analysis cost, environmental protection and many other advantages, the Near Infrared(NIR) Spectroscopy analysis has made breakthrough progress in the Chinese medicine field. Under the circumstance of small sample size, the 80 Asiatic Moonseed extract samples (the different ratio of material to liquid) from different cultivation area were surveyed by the fourier transform near infrared diffuse reflectance spectrometer in this study. For reducing the complexity of model and raising the accuracy rate of prediction, the processing of dimension reduction was carried on using the wavelet transform(WT) method for the compression of the spectral variables, the compression ratio can reach 98.75%. And the models of two kinds extract of Asiatic Moonseed are carried on using Support Vector Machine technology in order to analyze the content alkaloid of extract. The simulation results show that, the prediction decision coefficient (R2) is 0.9836, the average relative error(ARE) is 0.0137, the root mean square error of Cross-Validation(RMSECV) is 0.0201 in the Asiatic Moonseed extract samples (the ratio of material to liquid 1:2), and the predictive decision coefficient is 0.9908, the average relative error is 0.0175, and the root mean square error of Cross-Validation is 0.0198 in the Asiatic Moonseed extract samples (the ratio of material to liquid 1:5). The prediction decision coefficient was achieved above 95% for the two kinds of samples, the process of modeling is simple, the model is also stable, and the prediction accuracy of model can satisfy certain practical application. Therefore, the research owns certain application value, and can also provide certain technical reference for other infrared spectrum's qualitative analysis.