Improving Raman spectroscopic identification of rice varieties by feature extraction
Min Sha, Ding Zhang, Zhengyong Zhang, Wei Jinhong, Yuan Chen, Mengtian Wang, Jun Liu
Abstract
Abstract Raman spectroscopy combined with pattern recognition can identify rice varieties excellently. In this study, a method was established to select the key feature for identification model. Seventy‐two Raman spectra of three varieties of rice were analyzed. Seventy‐one independent variables and characteristic bands (420–560, 820–980, 1,000–1,200, and 1,300–1,500 cm −1 ) were obtained by principal component analysis (PCA). Window analysis further narrowed the range of characteristic bands (451–550, 951–1,000, and 1,351–1,450 cm −1 ). Hierarchical cluster analysis (HCA) obtained 30 wavenumbers with small correlation. The prediction accuracy was 91.71%, whereas the time was reduced by 10 times when these 30 wavenumbers were used to establish the identification model. The method combined PCA, window analysis, and HCA with support vector machine can be used as an effective feature extraction method to improve the efficiency for identification of rice varieties. Under the circumstances of large sample size or relatively complex data, the screening of Raman spectrum information is an important means to simplify the model and improve the prediction efficiency.