Study On Characteristic Band Selection For The Quality Of Some Natural Products By Near Infrared Spectroscopy

Near infrared reflectance spectroscopy is a rapid, convenient and accurate analysis technique used for the online detection, the simultaneous determination of multi-component and quality control. With the development of computer technology and chemo-metrics, near infrared spectroscopy is applied widely to animal husbandry, agriculture, food, tobacco, chemical industry and medicine and so on. Partial least squares is a typical and far-reaching chemo-metrics with strong prediction and simple model. Although the partial least squares regression based on the full spectrum contains whole information of target, the calibration model also contains chemical information of non-target and weak absorbance of target, the quality and accuracy of the model will be influenced. So it is important to select feature bands to improve and promote advantage of near infrared spectroscopy.Some active components of local natural resources in Chong Qing were determined by near infrared spectroscopy, especially the accuracy of NIR was improved with characteristic band selection. Interval partial least squares was studied to select characteristic band:1. during analyzing protein value of bamboo shoots, interval partial least squares (iPLS) and backward interval partial least squares (BiPLS) were applied to select characteristic spectral bands of protein. The results show that the performance of iPLS and BiPLS were better than PLS based on the whole spectrum and BiPLS gave the best result. The variables in these models were decreased effectively and the prediction accuracy was improved.2. Near infrared spectroscopy combined with three different iPLS algorithms determine content of raw fat in bamboo shots, and select the feature bands. The predictive result shows that the method is fast and accuracy, and is possible for quality control during buying bamboo shoots.3. Near infrared spectroscopy was employed to simultaneous determine content of costunolide and dehydro-a-curcumene. And interval partial least squares was applied to search for an optimized combination of information spectral intervals about two active constitutents. iPLS model could diminish runtime and select the optimal intervals.4. This experiment was attempted to determine the content of Chinese goldthread by near infrared spectroscopy coupled with interval partial least squares and analysis optimal bands. The variables in these models were decreased effectively and the prediction accuracy was improved.1. Wavelength Selection of FT-NIR Spectroscopy for Determination of Protein of Chimonobambusa quadrangularis Shoot based on iPLS and BiPLS ModelsDuring analyzing protein content of Chimonobambusa quadrangularis shoot by near-infrared spectroscopy, interval partial least squares (iPLS) and backward interval partial least squares (BiPLS) were applied to select characteristic spectral bands of Chimonobambusa quadrangularis shoot. Methods:The whole spectrum was divided into12and17intervals, respectively. Then, PLS regression models were established by using whole spectrum and smaller bands, and then BiPLS was established with combination of each interval. These models were evaluated by the mean square error of cross validation (RMSECV) and the root mean square error of prediction (RMSEP). Results:The results indicated that after selecting characteristic bands the iPLS models and BiPLS model were better than PLS model based on the whole spectrum, the optimal combinations of7([5,3,6,12,4,2,11]) spectral intervals were chosen by BiPLS to obtain a satisfactory result from12intervals, theRMSECV and RMSEP were0.321%and0.218%respectively. Conclusion:The variables in these models were decreased effectively and the prediction accuracy was improved.2. Wavelength Selection of FT-NIR Spectroscopy for Determination of Fat of Chimonobambusa quadrangularis Shoot based on iPLS and BiPLS ModelsNear infrared spectrometry combined with chemometric methods was used for establishing a new method to determine and select characteristic spectral bands of raw fat in Chimonobambusa quadrangularis shoot. The number of interval was determined. Interval partial least squares (iPLS).backward interval partial least squares (BiPLS) and forward interval partial least squares (FiPLS) were employed to select effective spectral regions and establish the quantitative calibration methods of raw fat. The results indicated that after selecting characteristic bands FiPLS model was the best. Correlation coefficient obtained is93.50, RMSECV is0.586%, RMSEP is0.500%. The predictive results show that proposed method is rapid, non-destructive and credible, which can be applied to control the quality of Chimonobambusa quadrangularis shoot.3. Simultaneous determination of costunolide and dehydro-a-curcumene in Aucklandia lappa Decne by near infrareds spectroscopyAbstract:Near-infrared spectroscopy was applied to determine the content of costunolide and dehydro-a-curcumene in Aucklandia lappa Decne combined with chemometrics. Interval partial least squares (iPLS) and forward interval partial least squares (FiPLS) were applied to select characteristic spectral bands of costunolide and dehydro-a-curcumene in Aucklandia lappa Decne. The best number of intervals was settled, and then interval partial least squares (iPLS) and forward interval partial least squares (FiPLS) were employed to build the models. Pretreatments of spectra were discussed in detail. The predictive results show that FiPLS has the satisfactory performance. The root mean square error of prediction (RMSEP) of costunolide and dehydro-a-curcumene is0.478%and0.346%.The predictive results show that the proposed method is rapid, non-destructive and credible, which can be applied t o control the quality of Aucklandia lappa Decne.4. Characteristic Band Selection of Near-infrared Spectrum for Determining Berberine Value of Chinese Goldthread Based on IplsAbstract:During analyzing berberine value of Chinese goldthread by near-infrared spectroscopy, iPLS (interval partial least squares) was applied to select characteristic spectral bands of berberine. The whole spectrum was divided into different intervals (15,16,17......23,24). Then, PLS regression models were established by using whole spectrum and smaller bands. These models were evaluated by root mean square error of prediction(RMSEP). The results indicated that after selecting characteristic bands, the R, RMSEP were94.99,0.096%, respectively. The variables in these models were decreased effectively and the prediction accuracy was improved.