[Study on artificial neural network combined with near infrared spectroscopy for wood species identification].

In the present article, near infrared spectra of 89 wood samples of different geographical provenances and species were measured, and back propagation artificial neural networks(BPANN) and generalized regression neural network (GRNN) were used for modeling of wood species NIRS identifying. Parameters for two neural networks were chosen via analysis of variance, respectively; and networks were trained with optimum parameters. Considering the difference between spectra, spectra with different levels of white noise and different levels of bias were simulated and predicted by using the models built. It was found that both the two models had satisfactory prediction results, identification correct rates obtained by BPANN model applied to spectra with bias level no higher than 2% and noise level no higher than 4% were above 97%; correct rates obtained by GRNN model applied to spectra with bias level no higher than 2% and noise level no higher than 4% were above 99%.

[Determination of Pb, Cd and Cr in surface decorating materials of wood-based panels by microwave digestion-AAS].

With microwave-digestion as a sample decomposition technique and HNO3 and H2O2 as digestion reagent, the contents of Pb, Cd and Cr in surface decorating materials of wood-based panels were determined by atomic absorption spectrometry. The detection limit of Pb, Cd and Cr of FAAS is 0.12 microg x mL(-1), 0.029 microg x L(-1) and 0.146 microg x L(-1); respectively, the cadmium detection limit of GFAAS is 0.157 microg x L(-1). The precision of lead was between 0.8% and 3.0% and the recoveries ranged from 94% to 109.2%. The precision of cadmium was between 0.8% and 2.1%, and the recoveries ranged from 94.6% to 106.4%. The precision of chromium was between 1.8% and 4.9%, and the recoveries ranged from 98.8% to 107.7%. The method is accurate and reliable, and suitable and applicable to determining heavy metal of surface decorating materials of wood-based panels. The result is expected to provide scientific basis for establishing a corresponding Chinese national standard.

[Rapid identification of softwood and hardwood by near infrared spectroscopy of cross-sectional surfaces].

The feasibility of wood identification of softwood and hardwood by near infrared spectroscopy (NIR) coupled with partial least squares discriminant analysis (PLS-DA) was investigated in the present paper. The near infrared spectra (780 - 2 500 nm) were collected from wood cross-section from one softwood species (Chinese fir) and one hardwood species (eucalyptus). The results show that: (1) The identification accuracy of the calibration samples predicted by the model based on NIR coupled the PLS-DA was 100%. The correlation coefficient between the NIR predicted category variable value and the true value was 0.990, and the SEC was 0.071; (2) The identification accuracy by the model based on the spectra with 780-1 100 nm wavelengths also was 100%, and the correlation coefficient and SEC were 0. 990 and 0. 070, respectively; (3) The identification accuracy for the test samples was 100%. It was suggested that NIR can be used to rapidly and accurately identify softwood and hardwood samples. It also provides a new approach to identifying wood species.

[Study on the wood grading by near infrared spectroscopy].

The present paper discussed wood grading according to modulus of rupture (MOR) by near infrared (NIR) spectroscopy. The calibration model was built between MOR of wood and NIR data in the range of 1 000-1 400 nm with partial least square regression (PLS). The correlation coefficient (r) was 0.89 and the standard error of calibration (SEC) was 6.30 MPa. The MOR of 35 unknown samples was predicted using the model. Wood samples were graded according to their predicted values and true values. The rate of right prediction for A, B and C was 75.0%, 91.3% and 80.0% respectively, and the whole rate of right prediction was 88.6%. The result has proved that near infrared spectroscopy is a fast method for the determination of wood grade in the small clear samples.

[Application of NIR spectroscopy to estimate of MFA and fiber length of Neosinocalamus a f finis].

Near infrared spectroscopy was applied to rapidly predict microfibril angle (MFA) and fiber length of Neosinocalamus a f finis Keng by using a fiber-optic probe in diffuse reflectance mode. The MFA and fiber length were measured by X-ray diffractometry and optical microscope, respectively. Partial least squares (PLS) was used to build models based on raw and pretreated spectra, including noise spectra and noise combined with orthogonal signal correction (OSC) spectra. The results showed that the PLS models of MFA and fiber length, based on noise combined with OSC spectra, gave the strongest correlations, with correlation coefficient (R) of 0.8936 and 0.9883 and root mean standard error of prediction (RMSEP) of 0.2920 and 0.1460 in prediction set. The correlations between NIR predicted and MFA/fiber length actual values are very good. Therefore, it is concluded that MFA and fiber length of N. a f finis can be estimated by NIR spectroscopy with sufficient accuracy.

[Discrimination of bamboo fiber and ramie fiber by near infrared spectroscopy].

The research on discrimination of natural bamboo fiber, bamboo pulp fibers and ramie fiber used in textile was demonstrated by near infrared (NIR) spectroscopy. First, the spectra of three kinds of fiber were scanned by NIR spectrometer. Then, the spectral data were pretreated by first derivatives. In the end, the databases and discriminating model of natural bamboo fiber, bamboo pulp fibers and ramie fiber were built. The kind of unknown fiber can be discriminated by the NIR model. The results indicated that the natural bamboo fiber, bamboo pulp fibers and ramie fiber can been discriminated quickly by NIR spectroscopy without destroying samples.

[FTIR studies of masson pine wood decayed by brown-rot fungi].

Fourier transform infrared (FTIR) spectroscopy was used to study the chemical changes of masson pine (pinus massoniana lamb.) decayed by the brown-rot fungus Wolfiporia cocos (Schwein.) Ryvarden & Gilbn. for different durations up to 23 weeks. The ratios of height of the lignin/holocellulose and holocellulose/lignin IR peaks were measured, and the klason lignin content and holocellulose content of the sound wood and brown-rotted wood with different level of decay were analyzed by wet chemical methods. The relationship between the two chemical components and the ratios of IR peak height was also established. FTIR spectra showed that, during the first 15 weeks of decay, the intensity of absorption bands at 1 736 and 1 372 cm(-1) ascribed to holocellulose decreased gradually, accompanied by a successive increase in the intensity of band at 1 510 and 1 225 cm(-1) ascribed to lignin. However, the intensities of holocellulose bands at 1736 and 1372 cm(-1) had a little increase, and the intensities of lignin bands at 1 510 and 1 225 cm(-1) had a very slight decrease after 15 weeks of decay. There was a very good correlation between the ratios of height of the lignin/holocellulose (I1510/1736, I1510/I1372, I1225/I736 and I1225/I1372) and the klason lignin content or holocellulose content. The coefficients of determination for the klason lignin content and the holocellulose content were 0.97-0.99 and 0.96-0.97, respectively. High coeffieients of determination were also obtained between the holocellulose/ lignin peak height ratios and the holocellulose content (R2 = 0.96). The above results suggest that, in the system studied, the klason lignin content and holocellulose content of wood decayed to differnent levels could be determined with reasonable accuracy by the FTIR technology.

[Study on denoising near infrared spectra of wood based on wavelet transform].

Near infrared (NIR) spectra of wood samples are often confused by a series of noise, which greatly influences accurate analytical result. In order to improve analytical precision, the authors need to pretreat the spectrum data. Derivative can correct baseline and background effects, increasing the resolution ratio of the spectra. However, it also increases the noise at the same time. The present paper aims at using wavelet transform to eliminate the noise of the near infrared first derivative spectrum of wood with the methods of 9 point smoothing spectrum, 25 point smoothing spectrum, the nonlinear wavelet hard-threshold spectrum, the nonlinear wavelet soft-threshold spectrum, 9 point smoothing+wavelet transform and 25 point smoothing spectrum+ wavelet transform. The results show that the wavelet transform has particular advantage on noise elimination of the near infrared spectra while reserving the useful information of spectrum. It can also improve the signal to noise ratio of spectrum, promising the prospect of a wide application in the wood near infrared spectroscopic analysis.

[Quantitative prediction of holocellulose, lignin, and microfibril angle of Chinese fir by BP-ANN and NIR spectrometry].

The amount of holocellulose, lignin, and microfibril angle of Chinese fir was predicted by using back-propagation neural network (BP-ANN) combined with near infrared (NIR) spectrometry. First, the data of original spectra were pretreated by Savitzky-Golay smoothing algorithm and the second derivative, then the data of near infrared spectrometry with 171 points were compressed to 86 points by using wavelet transform, and finally, the models were established by using BP-ANN. The models were validated using leave-n-out cross-validation approach, and the influences of the number of hidden neurons, learning rate, momentum, and epochs were discussed in the present paper. The prediction samples, which were not used in the model generation, were predicted by using the obtained models, the correlation coefficients (R2) of holocellulose, lignin and microfibril angle were 0.91, 0.90 and 0.87, respectively. The root mean square errors of prediction (RMSEP) of the established models were 0.86%, 0.33%, and 4.9%, respectively. The obtained results showed that the method is fast and nondestructive and can basically satisfy the requirement of quantitative analysis.

[Rapid modeling method for spectroscopic analysis of chemical components of bamboo].

A rapid modeling method for predicting the chemical components contents of bamboo was presented. The holocellulose contents and lignin contents of 54 samples from three growth years, two longitudinal positions and three radial positions were analyzed according to traditional chemical methods. Eleven samples were selected based on their holocellulose content and lignin content from these 54 samples to cover the range of holocellulose content and lignin content. Eleven samples were mixed at preset ratio with each other to give 21 mixed samples, the holocellulose content and lignin contents of which were computed. Another 22 samples with different chemical component contents were selected from the same 54 samples. The relationship between the chemical component contents and the diffuse reflectance NIR spectra of these samples was established using partial least squares regression. The correlation coefficient of prediction model for holocellulose content and lignin content was 0.92 and 0.93, respectively. The standard error of prediction for holocellulose content and lignin content was 1.04% and 0.913, respectively. The prediction results were similar to those from the prediction models developed by traditional methods. The results presented in this study demonstrate that samples can be prepared rapidly by the mixture of samples with each other and their chemical component contents can be computed. The technique will significantly reduce sampling time and analyzing time without adversely affecting the quality of the model.

[Study of the Eucalyptus and Poplar by generalized two-dimensional infrared correlation spectroscopy].

Poplar and Eucalyptus were identified fast by Fourier transform infrared spectroscopy (FTIR) combined with two-dimensional correlation spectroscopy (2D) in the present paper. The two kinds of wood were similar to each other in one-dimensional IR spectra but quite different in 2D FTIR spectra. In the range between 800 and 1500 cm(-1), they are similar and three strong auto-peaks were aroused around 1221, 954 and 879 cm(-1) in synchronous spectrum and four weak auto-peaks were aroused around 1470, 1150, 1105 and 1008 cm(-1), respectively. In the range between 1500 and 1800 cm(-1), one strong auto-peak appeared with Poplar around 1665 cm(-1) and one weak auto-peak appeared around 1600 cm(-1) in synchronous spectrum. They formed one pair of cross-peaks, whereas three auto-peaks not only at 1650 cm(-1) but also at 1725 and 1600 cm(-1) appeared with Eucalyptus and they became one 3 X 3 peak cluster. In addition, Eucalyptus has two weak auto-peaks at 1580, 1510 cm(-1) and four negative cross-peaks at (1725 and 1580 cm(-1), (1650 and 1580 cm(-1)), (1600 and 1580 cm(-1)) and (1510 and 1580 cm(-1)) and three positive cross-peaks at (1725 and 1510 cm(-1)), (1650 and 1510 cm(-1)) and (1600 and 1510 cm(-1)), respectively, which suggests that corresponding absorbance bands of Eucalyptus are more susceptive to the thermal perturbation than that of Poplar. The difference of 2D correlation between Eucalyptus and Poplar was obvious. The results proved that 2D correlation spectra could enhance the resolution of infrared spectra and increase the capacity of identification, which make it a powerful, rapid and new approach to identifying Eucalyptus and Poplar.

[Study of thermal perturbation of natural bamboo fiber by two dimensional correlation analysis and Fourier transform infrared spectroscopy].

The Fourier transform infrared spectroscopy (FTIR) combined with generalized two-dimensional correlation analysis was applied to study the mini-heating process of natural bamboo fiber. The absorption peaks of natural bamboo fiber and bamboo in the FTIR spectra were different, which showed the contents of lignin and hemicelluloses of natural bamboo fiber was lower than those of bamboo. The changes in absorption peaks of natural bamboo fiber in the FTIR spectra at different temperatures were inconspicuous during heating up from 50 to 120 degrees C, which showed that there was not oxidation reaction in natural bamboo fiber during the process. With the help of 2D correlation analysis, the changes of different groups of natural bamboo fiber and bamboo during heating process were reflected. The strongest autopeak of them was all aroused at 1 665 cm1 in synchronous spectrum. The difference was that there were several weak auto-peaks and cross peaks in the natural bamboo fiber, but in the bamboo, one stronger 5 x 5 group was aroused in the 833-1230 cm(-1). Region the reason was the difference in chemistry composition and the change degree during heating process. In conclusion, the 2D correlation analysis of FTIR can be a new method to analyze the microcosmic dynamic change in the structure of natural bamboo fiber and bamboo during the mini-heating process and also offers an important theory gist for the study of oxidation mechanism of them.

[Influence of sample surface roughness on mathematical model of NIR quantitative analysis of wood density].

Near infrared spectroscopy is widely used as a quantitative method, and the main multivariate techniques consist of regression methods used to build prediction models, however, the accuracy of analysis results will be affected by many factors. In the present paper, the influence of different sample roughness on the mathematical model of NIR quantitative analysis of wood density was studied. The result of experiments showed that if the roughness of predicted samples was consistent with that of calibrated samples, the result was good, otherwise the error would be much higher. The roughness-mixed model was more flexible and adaptable to different sample roughness. The prediction ability of the roughness-mixed model was much better than that of the single-roughness model.

[Determination of holocellulose and lignin content in Chinese fir by near infrared spectroscopy].

The contents of holocellulose and lignin of wood are important determinants of the pulping quality of wood. The determination of holocellulose and lignin contents using traditional chemical methods is a costly and time-consuming process. Near infrared reflectance (NIR) analysis offers a fast, nondestructive testing and low cost alternative for prediction of wood quality. In the present article, the total amounts of holocellulose and lignin contents of 48 samples were analyzed according to standard wet-chemical method. All samples were milled using a Standard Wiley knife mill with a 2 mm screen. The 2 mm material was sieved with a 40-60 mesh sieve. Then, near infrared (NIR) spectra were collected in diffuse reflectance from samples of meal contained in a spinning cup by an analytical spectral devices (ASD) Lab Spec at wavelengths between 350 nm and 2 500 nm. The raw spectra were pretreated by the second derivative and smoothing, then the NIR model was built using partial least-squares statistical analysis and full cross validation. The coefficients of correlation (r) of calibration and validation for holocellulose were 0. 96 and 0.93, respectively; the standard error of calibration (SEC) and the standard error of prediction (SEP) were 0.39 and 0.50, respectively. For lignin, the values for r of calibration and validation were 0.99 and 0.90, while the SEC and SEP were 0.10 and 0.28, respectively. It was concluded that NIR analysis is a reliable, fast and nondestructive testing predictor of holocellulose and lignin content of wood in Chinese fir.

[Rapid determination of Klason lignin content in bamboo by NIR].

Lignin is one of the main components of lignocellulosic materials. The main purpose of wood cooking and bleaching is to remove lignin by chemical agent in paper industry. Whereas the lignin content shows wide variations depending on its tree specie, site condition, part and so on, it is essential to analyze the lignin content of different raw material. The aim of this paper is to develop a rapid near infrared (NIR) reflectance spectroscopic method to characterize the Klason lignin content of bamboo. Fifty four samples from three growth years, two positions along the longitudinal directions and three positions along the radial directions within a bamboo pole were prepared. The Klason lignin contents of 54 samples were analyzed according to traditional chemical method, the spectra of these samples were collected by NIR in the range of 350 to 2500 nm, and the relationship between the lignin content and the spectra of these samples was established by multivariate statistical technique. After second derivative pretreatment of raw spectra, the Klason lignin contents of the bamboo samples were quantified using partial least-squares statistical analysis (PLS1) and full cross validation in the range of 1011-1675 nm and 1930-2488 nm. High coefficients of correlation (r) were obtained between the predicted NIR results and those obtained from traditional chemical method. The correlation coefficient of calibration model and prediction model was 0.99 and 0.97, respectively. The standard error of calibration (SEC) and standard error of prediction (SEP) was 0.36% and 0.59%, respectively. It was found that the lignin content in bamboo could be determined rapidly with reasonable accuracy by the NIR method.

[The analysis of wood microfibril angle by near infrared spectroscopy and X-ray diffractometry].

In the present paper, near infrared (NIR) spectroscopy and X-ray diffractometry were used to rapidly predict the microfibril angle (MFA). MFA is one of the most important factors affecting wood properties. Wood resource utilization and tree improvement programs require cost-effective methods for the rapid analysis of thousands of samples. In the experiment, the average MFA of each sample were rapidly measured by X-ray scanning diffractometry. Then, the PLS model was built between NIR data and MFA by the X-ray measured. As a result, a very strong linear relationship has been found between NIR spectra fitted and X-ray measured, and the coefficients (r2) of calibration and prediction models were 0.867 and 0.816 respectively.

[Water in wood and its near infrared spectroscopic analysis].

Water is one of the most important characters of wood, so the effect of moisture content on the results must be considered in wood property analysis by NIR spectroscopy. In the present study, the absorbance of wood with various moisture content in NIR spectra was analyzed. Then the moisture contents in wood were predicted by NIR spectroscopy. The coefficient of determination is 0. 99; and SEC and SEP are 0.041 and 0.043 respectively. In addition, a strong liner relationship between wood density with a 12% moisture content and wood NIR spectra with different moisture contents (7%, 12%, 20%, 30% and 60%) was found. The results indicate that the model built with different moisture contents can predict wood density of 12% moisture content.

[Near infrared spectroscopy of wood sections and rapid density prediction].

In the present report, the use of calibrated near infrared (NIR) spectroscopy for rapid prediction of solid wood density is described. NIR spectra were obtained from the three sections (cross, radial and tangential section) of each Chinese fir sample. The authors found that the spectra of the three sections are different and the correlation coefficients between the laboratory-determined density and the NIR-fitted data are different, too. The prediction results showed that the model based on NIR spectra taken from the cross section is the best, and the correlation coefficients are 0.94 (cross), 0.85 (radial) and 0.81 (tangential), respectively. Using the cross model to predict the density of unknown wood samples, we can see: r2 is 0.977 and the standard deviation (STDEV) is only 0.006.