Premium
Method for extracting Raman spectra characteristic variables of biological sample based on Hilbert–Huang transform
Author(s) -
Zhao Xiaoyu,
He Yan,
Liu Zihao,
Zhang Wei,
Tong Liang
Publication year - 2020
Publication title -
journal of raman spectroscopy
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.748
H-Index - 110
eISSN - 1097-4555
pISSN - 0377-0486
DOI - 10.1002/jrs.5866
Subject(s) - hilbert–huang transform , raman spectroscopy , hilbert transform , mathematics , hilbert spectral analysis , principal component analysis , instantaneous phase , pattern recognition (psychology) , analytical chemistry (journal) , chemistry , artificial intelligence , statistics , spectral density , physics , computer science , optics , filter (signal processing) , energy (signal processing) , chromatography , computer vision
Abstract Because Raman peaks of the biological sample are superimposed on each other, the use of characteristic peak attribution is limited to some extent. In this study, we show that Hilbert–Huang transform (HHT) provides a Raman spectral feature extracting method, especially for biological samples. First, the empirical mode decomposition algorithm was used to decompose Raman spectra into intrinsic mode functions (IMFs). It is worth noticing that the IMF frequency is single or nearly single, so its further transformation (instantaneous amplitude, instantaneous angle, instantaneous angular frequency, Hilbert spectrum, and Hilbert marginal spectrum) from Hilbert processing is monotonous instead of the raw overlapping. Then, the Hilbert marginal spectrum was selected by one‐way analysis of variance and related with the rice type to establish a partial least squares regression (PLS) model with a 95.00% accuracy. This result is better than those based on characteristic variables screened by PLS, interval PLS, principal component analysis, independent component analysis, successive projections algorithm, haar, db, and coif (85.00%, 90.00%, 82.50%, 77.50%, 90.00%, 92.50%, 80.00%, and 85.00%). These results illustrate that HHT can accurately extract the characteristic variables from biological Raman spectra. The classification accuracy based on HHT is slightly lower than those based on bior 2.4, three‐layer decomposition (97.50%) and sym 5, five‐layer decomposition (97.50%). Significantly, no parameters need to be set such as the wavelet mother function and the decomposition layer in the HHT feature extraction process. This paper provides a HHT method for Raman spectral feature extraction, which is simple and effective.