FENXI HUAXUE Chinese Journal of Analytical Chemistry. Savitzky-Golay. n = SG SG. Savitzky-Golay mmol /L 5700.

Transcript

1 FENXI HUAXUE Chinese Journal of Analytical Chemistry ~ 346 DOI /SP. J Savitzky-Golay 1 * PLS Savitzky-Golay SG ~ 5300 cm ~ 4160 cm - 1 SG PLS n = PLS 1 ~ 40 PLS SG SG 53 PLS mmol /L SG SG PLS Savitzky-Golay PLS 1 ~ 7 PLS Savitzky-Golay SG 8 ~ 11 0 SG SG PLS SG PLS ~ mmol /L mmol /L 5700 Necolet InGaAs No No Nos. 2007A B B No. 2007Z3-E0281 * tpan@ jnu. edu. cn

2 3 Savitzky-Golay mm ~ 4000 cm cm A v = k v C + ε A v v k v v C ε v k v k v i C i v 2 N N RMSE C i i N C i v - C i 2 RMSE v = 2 槡 N - 1 RMSE v Optimal Single wavenumber prediction bias SWPB SWPB i = C i v Optimal - C i 2 N 3 SWPB SWPB SWPB 1% 2. 3 SG SG s n 2m + 1 SG 0 ± 1 ± 2 0 SG 0 SG SG SG 5 ~ ~ 87 n = n = SG 2. 4 P = C ip - C ip 2 槡 M - 1 R p = C ip - C mp C ip - C mp C ip 槡 - C mp 2 C ip 2 - C mp C ip C ip i C mp C mp M R P R P cm cm

3 ~ 5300 cm ~ 4160 cm RMSE v Optimal 7232 cm cm - 1 SWPB SWPB 2 2 SWPB SWPB SWPB 1 Fig SWPB Near-infrared spectra of 191 serum samples Fig. 2 Distribution of single wavenumber prediction bias SWPB and chemical values 3. 2 SG PLS 1 SWPB Table 1 Mean and standard deviation of SWPB and chemical SG PLS value in calibration set and prediction set ~ 5300 cm ~ 4160 cm - 1 PLS Chemical value Mean Standard Mean deviation SWPB Calibration set mmol /L ~ Prediction set Standard deviation 5300 cm ~ 4160 cm SG PLS 1 ~ 40 PLS SG PLS PLS 3 PLS 2 SG PLS 2 SG mmol /L R P SG Fig. 3 Optimal root mean square error of predivtion corresponding to smooth points number for each PLS order derivation 0 Order 1 Order 2 Order 3 Order 4 Order 5 Order

4 3 Savitzky-Golay 345 SG PLS mmol /L SG 2 Table 2 Prediction effect of optimal model corresponding to each order derivation Polynominal degree Smoothing points PLS PLS factor No smoothing Order 1 Order 2 Order 3 Order 4 Order Order PLS SG PLS SG PLS PLS 4 PLS mmol /L SG PLS mmol /L SG PLS 4 4 PLS Fig. 4 corresponding to PLS factor before and SG SG after smoothing PLS a. PLS PLS model b. SG PLS Optimal SG SG smoothing PLS model SG PLS References 1 Burns D A Ciurczak EW. Handbook of Near-infrared Analysis 2nd ed New York Marcel dekker inc ~ CHU Xiao-Li XU Yu-Peng LU Wan-Zhen. Chinese J. Anal. Chem ~ Kasemsumran S Du Y P Maruo K Ozaki Y. Chemometrics Intell. Lab. Syst ~ CHEN Huo-Cai YANG Zhong-Guo CHEN Xing-Dan. Chinese J. Anal. Lab ~ 20 5 XU Hui-Rong WANG Hui-Sheng HUANG Kang YING Yi-Bin YANG Cheng QIAN Hao HU Jun. Spectroscopy and Spectral Analaysis ~ 2526

5 CHEN Xue-Ying LI Ye-Rui CHEN Yong WANG Long-Hu Ding Ling. Chinese J. Anal. Chem ~ YU Yan-Bo ZANG Peng FU Yuan-Hua ZHANG Lu-Da YAN Yan-Lu CHEN Bin. Spectroscopy and Spectral Analaysis ~ Savitzky A Golay M J E. Anal. Chem ~ CHU Xiao-Li YUAN Hong-Fu LU Wan-Zhen. Prog. Chem ~ CHEN Jie-Mei PAN Tao CHEN Xing-Dan. Optics Preci. Eng ~ 7 11 CAO Pu PAN Tao CHEN Xing-Dan. Optics Preci. Eng ~ 1958 Joint Optimization of Savitzky-Golay Smoothing Models and Partial Least Squares Factors for Near-infrared Spectroscopic Analysis of Serum Glucose XIE Jun 1 PAN Tao * 1 CHEN Jie-Mei 2 CHEN Hua-Zhou 1 3 Ren Xiao-Huan 1 1 Key Laboratory of Optoelectronic Information and Sensing Technologies of Guangdong Higher Educational Institutes Jinan University Guangzhou Department of Biological Engineering Jinan University Guangzhou Department of Mathematics Shanghai University Shanghai Abstract The optimal model for the near-infrared spectroscopic analysis of serum glucose was established by partial least squares PLS and Savitzky-Golay SG smoothing method. Based on the prediction effect of the optimal single wave number model a new dividing method for calibration set and prediction set was given. The calibration and prediction models were established by PLS method adopting the combination bands of cm - 1 and cm - 1 with Savitizky-Golay SG smoothing. By extending the number of smoothing points to odd and polynomial degree to fourteen smooth coefficient tables including 582 smooth modes were calculated. All PLS models corresponding to all smooth modes and all PLS factors 1-40 were constructed. The optimal model was selected by the prediction effect. And the derivation order was 1 the polynomial degree was 3 or 4 the number of smoothing points was 53 the optimal factor was 7 and the optimal reach mmol /L. The dividing method for calibration set and prediction set the extending of SG smoothing modes large-scale optimization combining SG smoothing modes and PLS factors can be effectively applied for the model optimization of near-infrared spectroscopic analysis. Keywords Serum glucose Near-infrared spectroscopy Partial least squares Savitzky-Golay smoothing Dividing calibration set and prediction set Received 19 June 2009 accepted 12 September 2009