Near infrared analysis model transfer of holocellulose based on SWCSS-CARS algorithm
Near infrared spectroscopy is a non-destructive method that can be used to rapidly detect the content of holo-cellulose in pulpwood to improve the level of intelligent manufacturing in the paper industry.However,the existing near infrared models in practical applications are often unable to predict the spectra of wood samples measured by dif-ferent instruments,which greatly limits the wide application of this technology.Model transfer is an important solution to the problem of inter-instrument differences in near infrared spectroscopy that makes calibration models difficult to generalize across multiple instruments.Additionally,model transfer reduces the cost of re-modeling and model mainte-nance.To realize the sharing of the near infrared analysis model of the content of holocellulose in pulpwood on three Lengguang S450 grating-scanning near infrared spectrometers,the pulpwood samples collected by these three same types of near infrared spectrometers were taken as the research objects,and the SWCSS-CARS combined algorithm was proposed.The competitive adaptive reweighted sampling(CARS)wavelength optimization algorithm was used to reduce the adverse effects of invalid wavelengths in the screening wavelengths with consistent and stable signals(SWCSS)method,to improve the analysis ability of the model for two target samples.In this study,the spectral data sets of 84 pulpwood wood flour samples measured by three Lengguang S450 grating-scanning near infrared spectrome-ters with same model and their corresponding synthesized cellulose content data sets were used.The Kennard-Stone method was used to divide all the samples into a calibration set of 56 samples and a prediction set of 28 samples,and the samples in the calibration and prediction sets of the master instrument and the target instruments corresponded to each other.A partial least squares regression(PLSR)model based on SWCSS-CARS algorithm was established,and its analytical ability for target samples was compared with that of SWCSS and CARS alone.The results showed that the relative standard deviation(RPD)of the master model established by the 30 wavelengths selected by the SWCSS-CARS method based on holocellulose for the analysis of two target samples was greater than 4.6,and the value of Akaike information criterion(AIC)was 67.68,which was much smaller than 3 209.83 before the model transfer and 942.82 of the SWCSS algorithm.The dimension of the spectral matrix was reduced,and the model transfer efficiency was significantly improved.It was shown that the SWCSS-CARS algorithm can effectively remove the invalid wave-length in the SWCSS method,and successfully realize the sharing of the holocellulose content model of pulpwood among three same types of near infrared spectrometers.
holocellulose contentnear infrared spectroscopystable consistent wavelengthwavelength optimizationmodel transfer
万方数据
维普
