Brand Identification of Soybean Milk Powder based on Raman Spectroscopy Combined with Random Forest Algorithm
Zhengyong Zhang, Xiaojing Shi, Yaju Zhao, Yinsheng Zhang, Haiyan Wang
Abstract
Abstract Raman spectroscopy can characterize the rich molecular vibration information of soybean milk powder samples, but difficulties arise in its direct use for sample classification and identification. Therefore, it is urgent to develop an intelligent identification technology based on Raman spectroscopy. For brand identification of soybean milk powder, this work investigates and discusses a variety of spectral processing technologies including wavelet denoising, normalization, principal component analysis, and the results show that appropriate spectral processing can improve the recognition accuracy of the random forest algorithm. Under the optimal conditions (db2 wavelet, normalization, principal component analysis, 30 decision trees), the best recognition effect of soybean milk brand identification can be achieved. The Raman spectral signal acquisition time of each sample is 40 s, and the spectra pretreatment and identification operation time only takes a few minutes. The analytical approach established in this paper has the advantages of convenient and fast Raman spectra acquisition, fast and accurate model identification.