Search

The conventional qualitative analysis of near infrared spectroscopy (NIR) commonly uses one single classification model. This paper focused on the fusion of multiple classifiers based on different single classifiers by using the fused classifier to determine different varieties of red-wines. NIR spectra of 170 red-wine samples were collected by using Fourier transform near-infrared spectrometer. Red-wine classification models were established respectively, based on PLS-DA, SVM, Fisher and Ada-Boost. Then these models were selected to obtain some different base classifiers according to Diversity Measure Feature Selective (DMFS). The highest accuracy rate of determining different varieties of red-wine test samples of four single base classifiers was up to 88.24%, and at the same time the lowest discriminant accuracy rate was 81.18%. At last, we got the fused classifier, which combined four base classifiers with weighted voting principle, and determined its test set again by using the fused classifier. The final classification accuracy rate for red-wine varieties increased to 92.94%, In contrast with one single classifier, the lowest misjudged number of fused classifiers decreased from 9 to 6.These results suggested that the performance of fused classifier is superior to one single classifier. It is feasible to use fused classifier combined with near infrared spectroscopy to determine different varieties of red-wines.

Discriminating the maturity levels of tobacco leaf with in-situ measurement can effectively reduce loss rate and quality decline due to misjudgment of the maturity levels of tobacco leaf. In the meantime, the regular way we use to determine the maturity levels of tobacco, which is depend on tobacco leaf age and judgment of tobacco grower, lacks of objectivity. So this paper proposed a method to identify maturity levels of tobacco leaf by using spectral feature parameters combined with the method of support vector machine (SVM). In this paper, a total of 351 tobacco leaf samples collected in 5 maturity levels including immature (M1), unripe (M2), mature (M3), ripe (M4), and mellow (M5) determined by experts were scanned by field spectroscope(ASD FieldSpec3) with in-situ measurement for getting their reflectance spectrum. Through spectral analysis we found that the spectrum of tobacco leaf with different levels of maturity can be distinguished in visible band but not easily be distinguished in near-infrared band, so we use the tobacco leaf spectrum in visible band as the sensitive bands to analyze and model. To find the most suitable input variables for modeling, we use continuous spectrum (350~780 nm), feature band (496~719 nm) and spectral feature parameters (the reflectance of green peak, location of green peak, first order differential value of red-edge and blue-edge, red-edge and blue-edge area, location of red-edge and blue-edge) in visible region as the input variables, and using these three kinds of input variables in the method of SVM to establish a discriminant model for identifying maturity levels of tobacco leaf. The result shows that, the model using spectral feature parameters gains the accuracy rate of 98.85%. While the accuracy rates of other two models were 90.80% and 93.10%, respectively. The conclusion was drawn that using spectral feature parameters in visible spectrum as the input variables in SVM can improve the model performance. It is feasible to use this method to identify maturity level of tobacco leaf with in-situ measurement.

In the process of practical production, it is important to accurately analyze the proportion of mixed samples with high speed, which plays a great role for quality control and formulation design in food and agricultural processing. Traditional solution is to build statistical model with a large number of representative samples, which is both labor-intensive and time-consuming. In this paper, the proportion of alcohol and acids mixed samples, and their dilute solution mixed samples(used carbon tetrachloride (CCl4) which has no near-infrared absorption characteristics as the solvent medium)，as well as sheet tobacco leaf mixed samples are respectively analyzed by using near infrared spectroscopy, SG smooth and non-negative coefficients regression, which verifies the feasibility of analyzing the proportion of the mixed samples. The results show that, the analytic proportion of transmission spectra of alcohol and acids according to non-negative coefficients regression is closer to actual molar proportion with result error less than 4%. The result of the dilute solution is much better with error less than 4%. The analytic proportion of diffuse reflectance spectra of sheet tobacco leaf according to non-negative coefficients regression is highly accurate with error less than 10%. In the meantime, it has a highly consistency between actual spectra and analytic spectra of mixed samples; and the result of F-test and T-test shows that there is no significant difference between them and the confidence level is 0.01. It has the reliability of analytical proportion in theory. With the spectral data of several pure samples, the proportion of mixed samples can be thus analyzed, which has a good application prospect for quality control and formulation design in food and agricultural processing.

Fast and non-destructive measurements of tobacco leaf pigment contents by spectroscopy in situ in the field has great significance in production guidance for nutrient diagnosis and growth monitoring of tobacco in vegetative growth stage, and it is also very important for the quality evaluation of tobacco leaves in mature stage. The purpose of this study is to estimate the chlorophyll and carotenoid contents of tobacco leaves using tobacco leaf spectrum collected in the field. Reflectance spectrum of tobacco leaves in vegetative growth stage and mature stage were collected in situ in the field and the pigment contents of tobacco leaf samples were measured in this study, taking the tobacco leaf samples collected in each and both stages as modeling sets respectively, and using the methods of support vector machine (SVM) and spectral indice to establish the pigment content estimation models, and then compare the prediction performance of the models built by different methods. The study results indicated that the difference of estimation performance by each stage or mixed stages is not significant. For chlorophyll content, SVM and spectral indice modeling methods can both have a well estimation performance, while for carotenoid content, SVM modeling method has a better estimation performance than spectral indice. The coefficient of determination and the root mean square error of SVM model for estimating tobacco leaf chlorophyll content by each stage were 0.867 6 and 0.014 7, while the coefficient of determination and the root mean square error of SVM model for estimating tobacco leaf chlorophyll content by mixed stages were 0.898 6 and 0.012 3; The coefficient of determination and the root mean square error for estimating tobacco leaf carotenoid content by each stage were 0.861 4 and 0.002 5, while the coefficient of determination and the root mean square error of SVM model for estimating tobacco leaf carotenoid content by mixed stages were 0.839 9 and 0.002 5. The innovation point of this study is that on the basis of support vector machine and spectral indice, models established by each stage and mixed stages for estimating the pigment contents of tobacco leaf samples can provide scientific basis and technical support for quality control of tobacco leaf production in field and the ensurance of tobacco leaf recovery quality.

To achieve the traceability of wine varieties, a method was proposed to fuse Near-infrared (NIR) spectra and cyclic voltammograms (CV) which contain different information using D-S evidence theory. NIR spectra and CV curves of three different varieties of wines (cabernet sauvignon, merlot, cabernet gernischt) which come from seven different geographical origins were collected separately. The discriminant models were built using PLS-DA method. Based on this, D-S evidence theory was then applied to achieve the integration of the two kinds of discrimination results. After integrated by D-S evidence theory, the accuracy rate of cross-validation is 95.69% and validation set is 94.12% for wine variety identification. When only considering the wine that come from Yantai, the accuracy rate of cross-validation is 99.46% and validation set is 100%. All the traceability models after fusion achieved better results on classification than individual method. These results suggest that the proposed method combining electrochemical information with spectral information using the D-S evidence combination formula is benefit to the improvement of model discrimination effect, and is a promising tool for discriminating different kinds of wines.

Cotton is one of the important oil crops, and it is great significance for screening and identification of breeding materials to establish a method of the rapid, nondestructive testing of cotton seed oil content. In this study, near-infrared diffuse reflection spectroscopy of 118 high and low oil materials were adopted to establish models for fast nondestructive determining oil content of cottonseed using near infrared spectroscopy (NIR). One hundred and six cottonseed samples as calibration set that covered the range of seed oil content for upland cotton were used in this experiment. The spectral data of cottonseed were processed using the first derivative and multiplicative scatter correction (MSC). The correction NIR model of oil content was built based on partial least squares (PLS) method with the spectral regions 5 446-8 848 cm(-1) and main components (5). The determination coefficient (R2) of calibration model was 0.975, standard error of calibration (SEC) was 0.67. The authors test the model's actual ability to predict using external validation set. The correlation coefficient (r) of predicted values and the chemistry value was 0.978, the range of prediction error was 0.1% - 1.7%. The model established has good predictability. The oil content of 784 breeding stocks were predicted by NIR model, statistical analysis of predictable. results elucidated that the NIR model of oil content developed can be well applied to selective breeding and oil related study in cotton.

Wheat stripe rust caused by Puccinia striiformis f. sp. tritici, is an economically important disease in the world. It is of great significance to assess disease severity of wheat stripe rust quickly and accurately for monitoring and controlling the disease. In this study, wheat leaves infected with stripe rust pathogen under different severity levels were acquired through artificial inoculation in artificial climate chamber. Thirty wheat leaves with disease severity equal to 1%, 5%, 10%, 20%, 40%, 60%, 80% or 100% were picked out, respectively, and 30 healthy leaves were chosen as controls. A total of 270 wheat leaves were obtained and then their near infrared spectra were measured using MPA spectrometer. According to disease severity levels, 270 near infrared spectra were divided into 9 categories and each category included 30 spectra. From each category, 7 or 8 spectra were randomly chosen to make up the testing set that included 67 spectra. The remaining spectra were treated as the training set. A qualitative model for identification and classification of disease severity of wheat stripe rust was built using near infrared reflectance spectroscopy (NIRS) technology combined with discriminant partial least squares (DPLS). The effects of different preprocessing methods of obtained spectra, ratios between training sets and testing sets, and spectral ranges on qualitative recognition results of the model were investigated. The optimal model based on DPLS was built using cross verification method in the spectral region of 4000-9000 cm(-1) when "centralization" was used as the preprocessing method of spectra and the spectra were divided into the training set and the testing set with the ratio equal to 3:1. Accuracy rate of the training set was 95.57% and accuracy rate of the testing set was 97.01%. The results showed that good recognition performance could be acquired using the model based on DPLS. The results indicated that the method using near infrared reflectance spectroscopy technology proposed in this study is feasible for identification and classification of disease severity of wheat stripe rust. A new method was provided for monitoring and assessment of wheat stripe rust.

In this paper, total of 5170 flue-cured tobacco samples collected from 2003 to 2012 in the domestic and foreign origin by Shanghai Tobacco Group Technical Center were tested by near infrared spectroscopy, including the typical upper leaves 1394, central 2550, the lower part of 1226. Using projection model of based on principal component and Fisher criterion (PPF), follow the projected results to get no statistically significant differences at adjacent principal components, and the number of principal components as little as possible, in this paper, four main components to build projection analysis model, the model results show that: the near-infrared spectral characteristics of the upper and lower leaves have a significant difference that can be achieved almost entirely distinguished; while the middle leaves with upper and lower have a certain degree of overlap, which is consistent to the actual situation of the continuity of tobacco leaf. At the same time, Euclidean distance between the predicted sample projection values and the mean projection values of each class in the model, a description is given for the prediction samples to quantify the extent of the site features, and its first and second close categories. Using the dispersion of projected values in model and the given threshold value, prediction results can be refined into typically upper, upper to central, central to upper, typical central, central to the lower, the lower to central, typically the lower, or super-model range. The model was validated by 34 tobacco samples obtained from the re-drying process in 2012 with different origins and parts. This kind of analysis methods, not only can achieve discriminant analysis, and get richer feature attribute information, can provide guidance on the raw tobacco processing and formulations.

In this paper, the 7 different origin before redrying raw tobaccoafter redrying sheet tobacco's online near infrared spectroscopy were collected from sortingredrying production line specifically for "ZHONGHUA" brand. By using the projection model bulit by different origin tobacco's online spectroscopy and the method of variance and correlation analysis, we studied the uniformity and similarity quality characteristics change before and after the redrying of tobacco, which can provide support for understanding the quality of the tobacco material and cigarette product formulations. This study show that selecting about 10,000 by equally spaced sampling time from a huge number of online near infrared spectroscopy, for modeling are feasible, and representative. After manual sorting, threshing, and redrying, the uiformity of each origin tobacco near-infrared spectroscopy can be increased by 10%~35%, homogeneity of the tobacco leaf has been significantly improved. After redrying, the similar relationship embodied in the origin also have significant changes, overall it reduce significantly, that shows the quality differences embodied by origin significantly improve, which can provide greater space for formulations, it shows the need for high-quality Chinese cigarette production requires large amounts of financial and human resources to implement cured tobacco processing. The traditional means of chemical analysis, it takes a lot of time and effort, it is difficult to control the entire processing chain, Near Infrared Spectroscopy with its rapid, non-destructive advantage, not only can achieve real-time detection and quality control, but also can take full advantage of near-infrared spectroscopy information created in the production process, which is a very promising online analytical detection technology in many industries especially in the agricultural and food processing industries.

Geographical origins of wine grapes are significant factors affecting wine quality and wine prices. Tasters' evaluation is a good method but has some limitations. It is important to discriminate different wine original regions quickly and accurately. The present paper proposed a method to determine wine original regions based on Bayesian information fusion that fused near-infrared (NIR) transmission spectra information and mid-infrared (MIR) ATR spectra information of wines. This method improved the determination results by expanding the sources of analysis information. NIR spectra and MIR spectra of 153 wine samples from four different regions of grape growing were collected by near-infrared and mid-infrared Fourier transform spe trometer separately. These four different regions are Huailai, Yantai, Gansu and Changli, which areall typical geographical originals for Chinese wines. NIR and MIR discriminant models for wine regions were established using partial least squares discriminant analysis (PLS-DA) based on NIR spectra and MIR spectra separately. In PLS-DA, the regions of wine samples are presented in group of binary code. There are four wine regions in this paper, thereby using four nodes standing for categorical variables. The output nodes values for each sample in NIR and MIR models were normalized first. These values stand for the probabilities of each sample belonging to each category. They seemed as the input to the Bayesian discriminant formula as a priori probability value. The probabilities were substituteed into the Bayesian formula to get posterior probabilities, by which we can judge the new class characteristics of these samples. Considering the stability of PLS-DA models, all the wine samples were divided into calibration sets and validation sets randomly for ten times. The results of NIR and MIR discriminant models of four wine regions were as follows: the average accuracy rates of calibration sets were 78.21% (NIR) and 82.57% (MIR), and the average accuracy rates of validation sets were 82.50% (NIR) and 81.98% (MIR). After using the method proposed in this paper, the accuracy rates of calibration and validation changed to 87.11% and 90.87% separately, which all achieved better results of determination than individual spectroscopy. These results suggest that Bayesian information fusion of NIR and MIR spectra is feasible for fast identification of wine original regions.

In the present paper, six categories of standard industrial grading tobacco provided by Hongta Group are taken as experimental samples, including three different tobacco locations-upper (B), middle(C) and lower(X) parts, with each part containing two kinds of tobacco colors-orange (O) and lemon yellow (L). Two methods including projection model method based on principal component and Fisher criterion (PPF) and support vector machine (SVM) method are used to analyze color and location features of tobacco based on visible-near infrared hyperspectral data. The results of projection model method indicate that in the projection and similarity analysis of tobacco color, location and six tobacco groups classified by color and location, two kinds of color can be fully differentiated, of which the similarity value is -1.000 8. Tobacco from upper and lower parts can also be fully differentiated with similarity value 0.405 3, but they both have intersections with tobac- co from middle part. Six tobacco groups classified by color and location can be fully differentiated as well and their projection positions meet the actual external features of tobacco. The results of support vector machine method indicate that in the discriminant analysis of tobacco color, location and six tobacco groups classified by color and location, the average recognition rate of tobacco colors reaches 98%. The average recognition rate of tobacco location is 96%. The average recognition rate of six tobacco groups is 94%. Therefore, it's feasible to analyze color and location features of tobacco using visible-near infrared hyperspectral data, which can provide reference for tobacco quality evaluation, computer-aided grading and tobacco intelligent acquisition, and also offers a new approach to the analysis of exterior features of other agricultural products.

To realize the early detection of P. strii formis f. sp. tritici latent infections in wheat leaves while no disease symptoms appear, a qualitative model for identification of the wheat leaves in the incubation period of stripe rust was built using near infrared reflectance spectroscopy (NIRS) technology combined with qualitative partial least squares (DPLS). In this study, 30 leaf samples infected with P. strii formis f. sp. tritici were collected each day during the eleven-day incubation period. And 30 healthy leaf samples and 30 leaf samples showing disease symptoms infected with P. strii formis f. sp. tritici, were also collected as controls. In total, there were 390 leaf samples that were divided into thirteen categories. Near infrared spectra of 390 leaf samples were obtained using MPA spectrometer and then a model to identify the categories of wheat leaves was built using cross verification method. The effects of different spectral ranges, samples for building the model, preprocessing methods of spectra and number of principal components on NIRS prediction results for qualitative identification were investigated. The optimal identification results were obtained for the model built in the combined spectral region of 5 400-6 600 and 7 600-8 900 cm(-1) when the spectra were divided into the training set and the testing set with the ratio equal to 4 : 1, "scatter correction" was used as the preprocessing method and the number of principal components was 14. Accuracy rate, misjudgment rate and confusion rate of the training set were 95.51%, 1.28% and 3.21%, respectively. And accuracy rate, misjudgment rate and confusion rate of the testing set were 100.00%, 0.00% and 0.00%, respectively. The result showed that using near infrared reflectance spectroscopy technology, P. strii formis f. sp. tritici latent infections in wheat leaves could be detected as early as one day after inoculation (or 11 days before symptoms appearing) and the number of days when the leaf has been infected could also be identified. The results indicated that the method using near infrared reflectance spectroscopy technology proposed in this study is feasible for the identification of wheat leaves latently infected by P. strii formis f. sp. tritici. A new method based on NIRS was provided for the early detection of wheat stripe rust in this study.

To realize qualitative identification and quantitative determination of Puccinia strii formis f. sp. tritici (Pst) and P. recondita f. sp. tritici (Prt), a qualitative identification model was built using near infrared reflectance spectroscopy (NIRS) combined with distinguished partial least squares (DPLS), and a quantitative determination model was built using NIRS combined with quantitative partial least squares (QPLS). In this study, 100 pure samples including 50 samples of Pst and 50 samples of Prt were obtained, and 120 mixed samples including three replicates of mixed urediospores of the two kinds of pathogen in different proportions (the content of Pst was within the range of 2. 5% 100% with 2. 5% as the gradient) were obtained. Then the spectra of the samples were collected using MPA spectrometer, respectively. Both pure samples and mixed samples were divided into training set and testing set with the ratio equal to 2:1. Qualitative identification model and quantitative determination model were built using internal cross-validation method in the spectral region 4,000--10,000 cm(-1) based on the training sets from pure samples and mixed samples, respectively. The results showed that the identification rates of the Pst-Prt qualitative identification model for training set and testing set were both up to 100. 00% when scatter correction was used as the preprocessing method of the spectra and the number of principal components was 3. When 'range normalization + scatter correction' was used as the preprocessing method of the spectra and the number of principal components was 6, determination coefficient (RZ), standard error of calibration (SEC) and average absolute relative deviation(AARD) of the Pst-Prt quantitative determination model for training set were 99.36%, 2.31% and 8.94%, respectively, and R2, standard error of prediction (SEP) and AARD for testing set were 99.37%, 2.29% and 5. 0%, respectively. The results indicated that qualitative identification and quantitative determination of Pst and Prt using near infrared spectroscopy technology are feasible and that the Pst-Prt qualitative identification model and the Pst-Prt quantitative determination model built in this study were reliable and stable. A new method based on NIRS was provided for qualitative identification and quantitative determination of plant pathogen in this study.

The present article proposed a method of stepwise selecting characteristic wavelengths based on minimum sum of correlation coefficients (SMCC). The maximization of the ratio of inter-class Euclidean distance to the sum of inner-class Euclidean distances was used as evaluation basis in qualitative analysis of near infrared spectroscopy. Seventeen kinds of grading tobacco leaf in 2012, provided by Hongta Group, were used as experimental samples to verify the effectiveness of this new method. CO1 was selected as the reference category and ten points were selected as characteristic wavelengths. The results indicated that the average value of inner-class Euclidean distance, calculated by characteristic wavelengths, was 1.69 times as large as that calculated by all wavelengths. The average value of inter-class Euclidean distance, calculated by characteristic wavelengths, was 3.70 times as large as that calculated by all wavelengths. The average value of the ratio of inter-class Euclidean distance to the sum of inner-class Euclidean distances, calculated by characteristic wavelength, was 2.21 times as large as that calculated by all wavelengths. The ratio of characteristic wavelengths was increased. The characteristic wavelengths can express the classical differences. It was showed that SMCC was an effective way to select characteristic wavelengths in qualitative analyses of near infrared spectroscopy.

In the present study, near-infrared reflectance spectroscopy (NIRS) technology was applied to implement early diagnosis of two kinds of wheat rusts, i. e. wheat stripe rust and wheat leaf rust, by detecting wheat leaves as disease symptom has not appeared. The wheat leaves were divided into five categories including healthy leaves, leaves in the incubation period infected with P. strii formis f. sp. tritici, leaves showing symptom infected with P. strii formis f. sp. tritici, leaves in the incubation period infected with P. recondita f. sp. tritici and leaves showing symptom infected with P. recondita f. sp. tritici. Near infrared spectra of 150 wheat leaves were obtained using MPA spectrometer and then a model to identify the categories of wheat leaves was built using distinguished partial least squares (DPLS). For building the model, second-order derivative method was regarded as the best preprocessing method of the spectra and the spectral region 4000 - 8000 cm(-1) was regarded as the optimal spectral region. Using the model with different training sets and testing sets, the average identification rate of the training sets was 96.56% and the average identification rate of the testing sets was 91.85%. The results proved the model's stability. The optimal identification rates were obtained while the ratio of training set to testing set was 2 : 1 and the number of principal components was 10. The identification rate of the training set was 97.00% and the identification rate of the testing set was 96.00%. The results indicated that the identification method based on the NIRS technology developed in this study is feasible for early diagnosis of wheat stripe rust and wheat leaf rust.

In the present study, tobacco quality analysis of different producing areas was carried out applying spectrum projection and correlation methods. The group of industrial classification data was near-infrared (NIR) spectrum in 2010 year of middle parts of tobacco plant from Hongta Tobacco (Group) Co., Ltd. Twelve hundred seventy six superior tobacco leaf samples were collected from four producing areas, in which three areas from Yuxi, Chuxiong and Zhaotong, in Yunnan province all belong to tobacco varieties of K326 and one area from Dali belongs to tobacco varieties of Hongda. The conclusion showed that when the samples were divided into two parts by the ratio of 2 : 1 randomly as analysis and verification sets, the verification set corresponded with the analysis set applying spectrum projection because their correlation coefficients by the first and second dimensional projection were all above 0.99. At the same time, The study discussed a method to get the quantitative similarity values of different producing areas samples. The similarity values were instructive in tobacco plant planning, quality management, acquisition of raw materials of tobacco and tobacco leaf blending.

In this study, tobacco quality analysis of main Industrial classification of different years was carried out applying spectrum projection and correlation methods. The group of data was near-infrared (NIR) spectrum from Hongta Tobacco (Group) Co., Ltd. 5730 tobacco leaf Industrial classification samples from Yuxi in Yunnan province from 2007 to 2010 year were collected using near infrared spectroscopy, which from different parts and colors and all belong to tobacco varieties of HONGDA. The conclusion showed that, when the samples were divided to two part by the ratio of 2:1 randomly as analysis and verification sets in the same year, the verification set corresponded with the analysis set applying spectrum projection because their correlation coefficients were above 0.98. The correlation coefficients between two different years applying spectrum projection were above 0.97. The highest correlation coefficient was the one between 2008 and 2009 year and the lowest correlation coefficient was the one between 2007 and 2010 year. At the same time, The study discussed a method to get the quantitative similarity values of different industrial classification samples. The similarity and consistency values were instructive in combination and replacement of tobacco leaf blending.

In this study, tobacco quality analysis of industrial classification of different producing area was carried out applying spectrum projection and correlation methods. The group of industrial classification data was near-infrared (NIR) spectrum in 2010 year from different tobacco plant parts and colors of Hongta Tobacco (Group) Co., Ltd. 6 064 tobacco leaf samples of 17 classes from Yuxi, Chuxiong and Zhaotong, in Yunnan province and 6 industrial classifications were collected using near infrared spectroscopy, which from different parts and colors and all belong to tobacco varieties of K326. The conclusion showed that, the probability of the grading belonging by the first dimension was 84%, the probability of the producing area belonging by the second dimension was 71%. The study can explain the difference of tobacco quality of industrial classification and producing area by a projection method to get the quantitative similarity values. The quantitative similarity values were instructive in combination of tobacco leaf blending.

A total of 96 red wines aged with 3 kinds of methods were included in this study, including 44 wines aged in oak barrel, 26 wines aged in stainless steel tank added with oak chips and 26 wines aged in stainless steel tanks. The infrared spectra of the wines were scanned by Fourier transform infrared spectrometer with attenuated total reflection (ATR) accessories. To classify the 96 different aged wines, discriminant partial least squares (DPLS) method and support vector machine (SVM) method were used to establish models respectively. In order to examine the stability of the discriminant model, modeling was repeated 10 times with two-thirds of samples randomly selected as cross-validation. All the models had high discriminating power with the classification accuracy of the cross-validation and the validation all higher than 90%. These results suggest that the infrared ATR spectroscopy combined with pattern recognition method is a promising tool for discriminating different aging wines.

Four hundred ninety five samples from 3 parts of tobacco were collected from 11 different ecological environments in 2010 to analyze the characteristics of tobaccos in different ecological environments with near infrared spectroscopy. The results indicate that the differences are remarkable both between lower and upper parts and between lower and middle parts of tobacco, while negligible between upper and middle parts of tobacco under the same ecological environment. Good stability and consistency were showed by the similarity of characteristics of tobaccos under different ecological environments, and the projection correlation coefficient of analysis set and validation set is over 98%. This article also proposes a method of measuring the similarity between characteristics of tobaccos from different ecological environments. The results can provide reference data to quantify tobacco's planting programming, production processing and quality management.

A quantitative identification model for testing the purity of hybrid maize seeds was built by near infrared reflectance spectroscopy with quantitative partial least squares (QPLS). The NIR spectra of 123 seeds powder samples (Nongda108 and mother178) with the purity of 600-100% were collected using MPA spectrometer. All samples were divided into two groups: calibration set (82 samples) and validation set (41 samples). Synergy interval partial least squares (SiPLSu) was used for selecting effective spectral regions and building models. The influences of different spectral regions and different calibration samples on the prediction results and different main components were compared. The result showed that the spectral regions 6 000 8 000, 6 000-9 000 and 6 000-10 000 cm(-1) all had better prediction results (R2 over 95%). Spectral region 6 000-10 000 cm(-1) was regarded the optimum spectral region for building the model with less main components(8), and the determination coefficient (R2) of calibration and validation sets were 96.61% and 97.67% respectively, SEC (standard error of calibration) and SEP (standard error of prediction) were 2.15% and 1.78% respectively, RSDs (relative standard deviation) were 2.04% and 1.94% respectively. Even with different calibration samples, the average determination coefficients (R2) of calibration and validation sets were 96.21% and 95.75%, SEC (standard error of calibration) and SEP (standard error of prediction) were 2.29% and 2.23% respectively, RSDs (relative standard deviation) were 2.81% and 2.73% respectively, which further proved the model's stability. With the increase in the number of main components, the identification rates in calibration set and validation set gradually increased, when the number of main components reached 8, the model determination coefficients reached the best (96.61% and 97.67%), and related coefficients of true value and predicted value were 98.29% and 98.87% respectively. The results have important value for rapid and accurate testing of hybrid maize seed purity.

Non-destructive testing, as a new, rapid, non-destructive technology, is the direction of agricultural produce testing in the future. In this study, the nitrogen content of cucumber leaves was predetermined using near infrared spectroscopy technology. The main results were as follows: The authors measured the nitrogen content in cucumber leaves with Kjeldahl method and near infrared spectroscopy, then established a model between them, and processed a external verification next. The verification results showed that the determination coefficient of the model was 0.4066, relative standard deviation is 0.1559, and calibration standard deviation is 0.72; Then the authors predicted the cucumber leaves nitrogen content with this model, and the results showed that the mean absolute percent error was 0.59, average relative error was 13.88, and correlation coefficient of the chemical values and predicted values was 0.6377. So it was proved that this model had a certain feasibility in vegetable leaves nitrogen testing.

The present paper builds a model based on Monte Carlo method in the projection of the blending tobacco. This model is made up of two parts: the projecting points of tobacco materials, whose coordinates are calculated by means of the PPF (projection based on principal component and Fisher criterion) projection method for the tobacco near-infrared spectrum; and the point of tobacco blend, which is produced by linear additive to the projecting point coordinates of tobacco materials. In order to analyze the projection points deviation from initial state levels, Monte Carlo method is introduced to simulate the differences and changes of raw material projection. The results indicate that there are two major factors affecting the relative deviation: the highest proportion of tobacco materials in the blend, which is too high to make the deviation under control; and the quantity of materials, which is so small to control the deviation. The conclusion is close to the principle of actual formulating designing, particularly, the more in the quantity while the lower in proportion of each. Finally the paper figures out the upper limit of the proportions in the different quantity of materials by theory. It also has important reference value for other agricultural products blend.

The appearance features of tobacco reflect its inner quality. Many factors, such as different plant parts, variety and maturity, provide standard and foundation for tobacco production processing. According to the different position of tobacco plant parts, tobacco plants leaves can be divided into five parts as tip, upper-middle, middle, lower-middle and priming leaf respectively. Five hundred tobacco leaf samples (100 each for one of five tobacco plant parts) from Yunnan province in 2008 were collected using near infrared spectroscopy, which all belong to tobacco varieties of K326. The similarity analysis of tobacco plant parts was carried out using mathematical model of SIMCA similarity analysis. The conclusion showed that the tobacco plant parts similarity results based on near-infrared spectroscopy corresponded to the relative tobacco plant parts in Yunnan province. The farther two tobacco plant parts were away from each other, the lower the similarity of corresponding parts was. And the similarity results of adjacent tobacco plant parts were different. The study discussed a method of confirming PC numbers and realized the quantitative similarity analysis between classes. It is instructive in replacement or adjustment of tobacco leaf blending and evaluation of tobacco industrial grading.

Near infrared spectroscopy technology was applied to study rapid and nondestructive discrimination method of hybrid maize seed purity. With NongDa108 hybrid seeds and mother 178 seeds, a discrimination model for the purity of maize single seed was built by near infrared reflectance spectroscopy with distinguished partial least squares (DPLS). A total of 200 seeds including 100 hybrid seeds and 100 mother seeds were divided into two groups: calibration set (150 samples) and validation set (50 samples), and each group had same number of hybrid and mother seeds. To eliminate human errors as much as possible we used two sample cups with transmission hole diameter of 3.0 and 4.5 mm, respectively, at the bottom for spectrum acquisition. The location of sample cups and seeds were fixed during spectrum acquisition process. The result showed that the average identification rate with 3 mm transmission hole diameter was 99.82%, significantly higher than that of 4.5 mm whose average identification rate was just 90.96%. There was no significant difference among the identification rates of one replicate and two replicates spectrum on endosperm face, two replicates spectrum on embryo face and four replicates. The rates of validation set reached about 99%, slightly more than that of one replicate on embryo face. The identification rates of one spectrum and two replicates spectrum on endosperm face in calibration and validation set were 100%, with the spectral region between 4000 and 8000 cm(-1). With 3.0 mm transmission hole diameter and 4000-8000 cm(-1) spectral region, the seed purity identification rates in calibration and validation sets built up by one spectrum on endosperm face were 100%. With the increase in principal components, the identification rates in calibration set and validation set gradually increased, and when principal components reached 9, the rate in both of sets were 100%. The results have important value for rapid and nondestructive testing of hybrid maize seed purity.

In the process of analyzing and designing tobacco redrying formula using near infrared spectroscopy, a great deal of spectra for different ratio mixed samples are badly needed. However these sample spectra are very hard to obtain in the actual production process. Furthermore, for the samples of different grades it is difficult to achieve the goal of even mixing, in consequence of introducing inevitable errors. In order to solve the above problems, the present paper proposes to use "theory of even mixed spectrum" produced by linear additive near infrared spectroscopy in place of the spectra of actual mixed samples. This way can not only eliminate the errors caused by uneven mixture, but also leave out the course of mixing samples and measuring spectra and save time, effort, and material simultaneously. This article analyzes the comparison between linear additive spectra and the spectra of actual mixed samples from the following four aspects: original spectra, derivative spectra, principal components, and the data of PPF projection, accordingly verifing the feasibility and superiority of the linear additive spectra.

The purpose of the present paper is to establish a method of the fingerprint of LiuWei DiHuang pills by the near infrared spectroscopy. First, the authors established the fingerprint of LiuWei DiHuang pills by high performance capillary electrophoresis (HPCE), and calculated its similarity. At the same time, the authors scanned its near infrared spectrogram. Then the authors established the mathematical model between the similarity of fingerprint and the near infrared spectrogram. Through the optimization of the model, the correlation(r), calibration standard deviation and the average relative error of the modeling set were 0.9046, 0.058 and 6.12%, respectively. It proved that the linearity between calculated and forecast of fingerprint was clear. The results showed that the method of the fingerprint of LiuWei DiHuang pills by the near infrared spectroscopy was feasible.

In the present paper, taking 66 wheat samples for testing materials, ridge regression technology in near-infrared (NIR) spectroscopy quantitative analysis was researched. The NIR-ridge regression model for determination of protein content was established by NIR spectral data of 44 wheat samples to predict the protein content of the other 22 samples. The average relative error was 0.015 18 between the predictive results and Kjeldahl's values (chemical analysis values). And the predictive results were compared with those values derived through partial least squares (PLS) method, showing that ridge regression method was deserved to be chosen for NIR spectroscopy quantitative analysis. Furthermore, in order to reduce the disturbance to predictive capacity of the quantitative analysis model resulting from irrelevant information, one effective way is to screen the wavelength information. In order to select the spectral information with more content information and stronger relativity with the composition or the nature of the samples to improve the model's predictive accuracy, ridge regression was used to select wavelength information in this paper. The NIR-ridge regression model was established with the spectral information at 4 wavelength points, which were selected from 1 297 wavelength points, to predict the protein content of the 22 samples. The average relative error was 0.013 7 and the correlation coefficient reached 0.981 7 between the predictive results and Kjeldahl's values. The results showed that ridge regression was able to screen the essential wavelength information from a large amount of spectral information. It not only can simplify the model and effectively reduce the disturbance resulting from collinearity information, but also has practical significance for designing special NIR analysis instrument for analyzing specific component in some special samples.

With 112 licorice seed samples with different hard rates ranging from 0.3% to 99.3%, harvested in different years from 2002 to 2007 and from different locations of China including Xinjiang municipality, Ningxia province, Inner-Mongolia municipality, Gansu province, Shanxi province and Heilongjiang province, a model for determining hard rate of licorice seeds was tried to be built by near infrared reflectance spectroscopy with quantitative partial least squares (QPLS). All the seeds samples were divided into two groups: calibration set (including 84 samples) and validation set (including 28 samples). The influences of different spectral regions, different main components and different calibration samples on the prediction results were compared. The result indicated that the spectral regions of 4000-8000, 5000-9000, 5000-8000, 5000-7000 and 5000-6000 cm(-1) all had satisfied and similar prediction results, then 5000-6000 cm(-1) was regarded as the optimum spectral region for building the model because of its faster operation speed. The model with 6 main components had better relative high determination coefficient (R2) and low standard errors and absolute errors. With the spectral range from 5000 to 6000 cm(-1) and 6 main components, there was a better fitting between the predictive value and true value. Determination coefficients (R2) of calibration and validation sets are 90.23% and 91.24%, the coefficients of correlation are 0.9532 and 0.9579, the standard errors are 10.31 and 9.72, and the average absolute errors are 8.01% and 7.45% respectively. Even with different calibration samples, the models have high determination coefficient (R2 over 90%), low standard errors (about 10.00) and low absolute errors (about 7.90%). The building of NIR model for determining hard rate of licorice seeds could promote the application of hard seeds in cultivation.

To break the dilemma on judging hard seeds and soft seeds of licorice and other legume families nondestructively, a distinguishing model for the hardness of licorice single seed was tried to be built by near infrared reflectance spectroscopy with distinguished partial least squares(DPLS). A total of 244 licorice seeds were divided into three groups: calibration set (120 samples), validation set (60 samples) and prediction set (64 samples), and each group has the same number of hard seeds and soft seeds. To eliminate the human error as far as possible, a specially made sample cup was designed for spectrum acquisition. Then the locations of the seed and the fiber-optic probe were fixed during each spectrum acquisition process. The influences of different replicate time, different spectral region and different calibration samples on the identification rate were compared. The result indicated that four replicates could increase the identification rate of the model significantly, the identification rates of the model of four replicates in calibration, validation and prediction set samples were 95.83%, 95.00% and 96.88% respectively, while that of one replicate were 93.33%, 91.67% and 82.81% respectively. The model of the spectral region between 4,000 and 80,000 cm(-1) was better than that of other regions, and the identification rate in calibration, validation and prediction set samples were 95.53%, 95.94% and 94.53% respectively. Even with different samples, the predication rates were all more than 90%. The identification rates of hard seed and soft seed in prediction set samples were 92.50% and 96.56% respectively. The prediction for seeds with different size and different color showed that this model was not suitable for bigger and smaller seeds, especially not for black seeds. NIR offered a new way to distinguish the hardness of licorice singe seed quickly, precisely and nondestructively, which will advance the study on the mechanism of hardness of crop seeds.

In the present study, different drought tolerance rice from different countries and areas were selected and grown in water field and drought field respectively, including 4 traditional varieties of drought rice, 18 varieties of modified drought rice, 2 varieties of drought traits rice, 2 varieties of drought tolerance rice, and a total of 30 different varieties of drought tolerance rice were involved. Using near infrared diffuse reflection spectra of leaves from water field and drought field, we studied the rice drought tolerance identification analysis. Results showed that: using the average spectra of several leaves' spectra, selecting 4,500-7,500 cm(-1) as effective analysis spectra zone, choosing the first derivative and multiple scattering correction (MSC) as spectra preprocessing method, we can set up the calibration models between the spectra of leaves from drought field and the yield of rice. Simultaneously, we concluded that the performance of calibration model for rice yield and drought tolerance identification indexes in the upper booting stage was better than in the previous booting stage whose correlation coefficient of cross validation could reach 0.8. But there was no obvious relation between the spectra from water field and the yield, the drought tolerance identification indexes. We explained the difference in these two series models' performance from the relationship between some parameter of the leaves' biochemistry (chlorophyll, moisture, etc) and yield, the drought tolerance identification indexes.

In the present research, a set of maize powder samples was used to study the calibration transfer between two fourier transform near-infrared (FTNIR) spectrophotometers, and a method of moving window support vector regression machines (SVR) was used to correct the differences between the two instruments. Bruker Vector 22/N was referred to as "master" on which the maize protein calibration model was built. Bruker MPA was referred to as "slave" instrument. A transformation matrix was constructed based on the spectra of a sample set (for calibration transfer) measured on both instruments. After transfer, NIR spectra acquired on "slave" will appear as if they were measured on master instrument. The calibration model available for the master can then be used to predict the transformed spectra measured on the slave. The transfer parameters were computed as follows. For wavelength i, the absorbance vector obtained on the master instrument was regressed against the corresponding absorbance matrix of a spectral window obtained on the slave instrument. Method of SVR was used for regression Moving the wavelength i and corresponding window, the transfer parameter for each wavelength can be obtained. For the two FTNIR spectrophotometers, a window size of 31 wavelengths and a subset of 15 transfer samples were chosen to establish the SVR regression model between "master" and "slave". Applying the calibration model to the prediction samples after being corrected by the transfer parameters, a good transfer performance can be achieved. The correlation coefficient (r) is 0.9434, while the relative standard deviation (RSD) is 4.23%. These results suggest that the SVR method can be used to successfully transfer the calibration model for protein of maize developed on a FTNIR spectrophotometer to another.

Near infrared (NIR) diffuse reflection spectra can be used to obtain samples' chemical component content and physical parameters with few pretreatments. As a fast and nondestructive technique, NIR has been widely accepted. The diffuse reflection light is detected after interaction with sample. Generally, it is thought that the light carries the interior information of sample, but no one can definitely depict how much information of inner sample can be collected. In the present study, three kinds of flue-cured tobacco were used to design our tests. Each kind of tobacco was prepared in two forms, slice and powder. Then we got three slice tobacco samples and three powder tobacco samples. Every experimental sample consisted of one slice tobacco and one powder tobacco. The slice tobacco was put in the bottom of the sample-cup and the powder tobacco was placed on top of the slice tobacco. Combining different slice tobaccos with different powder tobaccos, the authors can get nine experimental samples. Every experimental sample was scanned from the bottom of the sample-cup for each sample using Bruker MPA FT-NIR instrument four times repeatedly, then we got 36 pieces of NIR spectra totally. In order to find the relationship between the deep light penetration and variance contribution rate of different principal components (PC) of NIR spectra, cluster analysis was carried out using different combination of PC. When the first and second PC were chosen, the samples were clustered according to the exterior slice tobacco; Using the third and fourth PC, the samples were clustered according to the interior powder tobacco. Combining the variance contribution rates of different PC, we could elementarily describe NIR spectra information composition according to the path length of the light penetration quantitatively. Results showed that the first and the second PC contained about 98% spectra information which represented the exterior information, and the third and the fourth PC contained about 1.5% which represented the interior information of the sample. These results can help us profoundly understand the importance of PC selection in NIR qualitative and quantitative analysis.

The diffused-reflectance near-infrared (NIR) spectrum of medicinal rhubarbs was collected by Fourier transform spectroscopy instrument Principal components (PC) and wavelet packet entropy (WPE) were then calculated from the spectrum. Based on these two kinds of features, the models of identification of medicinal rhubarbs were developed using Fisher classifier. The results show that the error rates of cross-validation and prediction using WPE are all lower than those using PC. The model was built by WPE feature extraction method combined with Fisher classifier, the error rate of cross-validation is 6.52%, while that for prediction is 2.04%. The research result provides a method for identifying medicinal rhubarbs quickly.

Sample's moisture content and granularity would have an important influence on NIR veracity. Preprocesssing methods have been used for NIR analysis to achieve better NIR veracity. With Yunnan flue-cured tobacco as study materials, the authors innovatively applied microwave fast drying technology and home pulverizer to tobacco preprocessing, which could finish the preprocessing in 5 min, faster than the traditional methods such as oven drying and sifting. Results showed that: microwave fast drying technology could improve NIR analysis veracity when samples' moisture content had a wide range; influence of samples' granularity on NIR veracity could be eliminated after 25 s home pulverizer shatter. This study can offer a new method for NIR analysis workers to simplify NIR preprocessing.

The influence of sample test conditions on the NIR veracity was studied with homemade grating diffuse NIR instrument using Yunnan flue-cured tobacco. Deducing analysis error was achieved by model self-emendation when a global NIR model was set up. Without regarding the influence of loading samples and test conditions, the test repetition error, re-loading error and samples tightness error, which were brought by instrument S/N, accounted for 50%, 30% and 20% of the total error, respectively. Depressing sample could reduce errors brought by sample tightness. Changes in test conditions could bring more analysis error, which was larger than the total of repetition error. These results theoretically explain the influence of sample test conditions on the NIR analysis veracity, which can provide basic theory data for farther improvement of homemade instrument and offer a new idea for resolving this problem.

The influence of sample annum and the distribution of sample component on NIR veracity was studied with homemade grating diffuse NIR instrument using Yunnan flue-cured tobacco. Results showed that sample annum had an obvious influence on the total sugar and nicotine models, but had an unconspicuous influence on the total-nitrogen model. Models set up by samples, whose component content distribution was normal school, was better than those set up by even distribution. The conclusion in this study has a significant referenced value for the method and principle to select representative samples to modeling from a large amount of specimens.

Alfalfa hay has high nutritive value, and it is one of the most important protein feed for domestic animals. The quality parameters of alfalfa hay, including CP, Ash, NDF, ADF, ADL and IVDMD, were predicted using Fourier transform near infrared reflectance spectroscopy with PLS regression in this test. Then the 6 models were validated by cross-validation and external-validation. The results indicated that FT-NIR models of alfalfa hay quality have considerable accuracy and precision: the correlation coefficient of cross-validation is 0.953 88 to 0. 990 19, and the RMSECV is 1.980-0.345; The correlation coefficient of external-validation is 0.963-0. 990. By using FT-NIR, analysis can rapidly and accurately determine the quality of alfalfa without any chemical reagent. This method is of great significance for analysing the trait of alfalfa production, the quality determination, the estimation of germ plasm resource, and the identifying and selecting of hybridized generations in alfalfa research of China.

The technology of near infrared reflectance spectroscopy (NIRS) have been widely used in many research areas, owing to its rapidness, high efficiency, low cost and no pollution. The present paper mainly illustrates the significance of the applications of NIRS to grassland ecology research, and explains some innovative implications of near infrared reflectance spectroscopy in the determination of a variety of forage nutrients, minerals, and the components of soil, prediction of the composition of for-age mixtures, animal performance, grass resistance of diseases and insect pests and other complex ecological characters, and doing research on biochemical markers, isotope discrimination and so on. By synthesizing these applications properly, it is concluded that NIRS could be used as a holistic tool in grassland ecology research to determine the chemical components, and analyze the complex dynamic character of grassland ecosystem and the total specialty of the system running. According to this paper, the authors also hope to promote the application conditions of NIRS in the grassland ecology research in China, and accelerate the modernization of research measures in this area.

The influence of instrument parameters, such as scan internal, length of instrument run, frequency of scan background etc, on NIR veracity was studied with a homemade grating diffuse NIR instrument using Yunnan flue-cured tobacco. Results showed that scan interval, such as 8 nm or 16 nm, had no evident influence on NIR quantitative analysis. To improve scan speed, the scan interval the authors decided to use was 16 nm. NIR model was set up which could revise the influence of the length of instrument run. This instrument can clue on baseline shift to decide the frequency of background scanning, which can deduce NIR analysis error and improve NIR veracity.

The theory of local partial least square (LPLS) algorithm was described based on locally weighted regression algorithm (LWR). The influence of data processing parameters, such as principal component numbers and local set-up sample number in LPLS mode, on the NIR veracity was studied with homemade grating diffuse NIR instrument using Yunnan flue-cured tobacco. Results showed that for nicotine model, the principal component number decided by cross validation was not the best choice, and better results could be achieved by reducing the principal component number; using 30-50 samples to set up NIR model, the veracity of total sugar, total nitrogen, and nicotine could be improved by 7%, 14% and 10%, respectively. So, LPLS algorithm can effectively improve NIR model's veracity, and is a good method to set up robust NIR models.

The present study piled a cube array with the winter wheat's cap reflection hyperspectra sampled from 35 experimental districts during 8 growth periods, and then used a multiway partial least squares (NPLS) algorithm to establish a model to assess the per-mu-yield of winter wheat. The model was used to predict the per-mu-yields of other 13 experimental districts. The correlation coefficient of predicted and actual values was 0.9366, and the mean relative error was 4.44%. These results showed a good prediction of the model. The study has guiding meaning to actual yield assessment.

In the present paper, 66 wheat samples were used as experimental materials, 33 of them were used for building the quantitative analysis model of protein content, and the rest composed the prediction set. Using Moore-Penrose matrix, we estimated directly the regression coefficients of the regression analysis model with Fourier transform near-infrared (FTNIR) whole spectroscopy. The samples of prediction set were analyzed, and the correlation coefficient is 0.979 9 between the prediction values of the near-infrared model and the standard chemical ones by Kjeldahl's method, and the average relative error is 1.76%. Using Moore-Penrose matrix, we can not only get the near-infrared spectroscopy analysis model's regression coefficients, but also know their contribution at every wavelength point. Consequently we can understand and explain the physical and chemical significance of the FTNIR whole spectroscopy regression model.

Intact-kernel samples of normal maize inbred lines and hybrids were collected from field experiments of three locations. Calibration equations were developed by partial least square regression (PLS) of chemical values of near infrared reflectance spectroscopy (NIRS) data and tested through both cross and external validation. In addition, 40 progenies of F1 and F2 generation not included in calibration and validation sets were verified to further evaluate the reliability of three calibration equations. The authors found the coefficients of correlation (r) of 0.98, 0.93 and 0.97 between NIRS predicted and actual protein, starch and oil content in these materials, respectively. However, the greatest relative errors were 2.7% (protein), 2.46% (starch) and 7% (oil). Thus, the accuracy of prediction could be comparable to chemical methods. The feasibility of developing NIRS equations with samples of inbred lines to determine grain quality of hybrids was also examined. The analysis of principal components of spectrum of the inbred lines and hybrids supported a new theory that plant spectrum properties could be heritable.

This paper introduced the application of support vector machines(SVM) regression method based on statistics studytheory to the quantitative analysis with near-infrared (NIR) spectroscopy. Sixty-six wheat samples were used as experimental materials, and thirty-three of them were used as calibration samples. The protein contents and NIR spectra of the calibration samples were used to build SVM regression models by four different kernel functions. The protein content of the predicting samples are estimated by four different SVM regression models. All of the correlation coefficients between the estimated values by different SVM regression models and the standard chemical values of protein content by Kjeldahl's method are more than 0.97. The average absolute error is less than 0.32. To investigate the predicting effect, it is compared with PLS regression models. The result suggested that the SVM regression, which was built to estimate the protein content of wheat samples, can also be used in the quantitative analysis of real samples by NIR.

This paper introduces the principle and method with which the model about the quantitative analysis of Fourier transformation near infrared (NIR) spectroscopy by MAXR regression procedure can be established. In this way, the authors have selected the wave length information by Matlab language design programming in order to establish the quantitative analysis models with near infrared spectroscopy. Taking sixty-six wheat samples as experiment materials, quantitative analysis models to determine protein content are established with thirty-three samples. The relative coefficient are 0.977 1 and 0.976 5 respectively and the standard error are 0.335 and 0.340 between the predication result of the two models which include respectively two or three wave length information and Kjeldahl's value for the protein content of the another thirty-three wheat samples. When selecting the wave length information, the MAXR regression procedure can establish the optimum regression models which contain 1 or 2...or k wavelength information respectively. MAXR regression procedure is a useful method when selecting the optimum wavelength information because of its shorter computation time, and the method not only can carefully select the essential wavelength information to establish NIR spectroscopy quantitative analysis models of resisting multicollinearity information disturbance, but also to establish the work for selecting optimum wavelength information which can direct to design the special NIR analysis instrument for analyzing specific component in the special samples.

Support Vector Machine (SVM) is a method for the research on identifying two types of problem. It is the latest branch in the statistics study theories, and the identification model has a strict mathematics foundation. In this paper, the basic principle and method of SVM are not only introduced, but also applied to chemometrics. One hundred and three rhubarb samples were used as experimental materials. The identification models were established with near-infrared spectroscopy and SVM training method with the intention of identifying whether the rhubarb samples are true or false. The thirty-three samples in training set were identified by the identifying models with the accurate rate of 100%, while seventy estimate samples had an accurate rate of 96.77%. The research result provided the method of identifying the traditional Chinese medicine rhubarb quickly. So, it shows the feasibility of establishing the models with near-infrared spectroscopy and SVM method to identify biological samples. This paper introduced the theme of SVM training method in order to beget the attention of the research members who deal with chemometrics.

This study is based on the agriculture product near infrared spectra database, which is a foundation database. The database has very important effects on agriculture products quality analysis and agriculture breeding. What the NIR researchers and NIR users care about is how to utilize information of the foundation database fully. To share the NIR resource, unifying the scanning term to get high quality spectra is the first step. This article uses wheat powder as sample to study the influence of different resolution, different He-Ne frequency and sample granularity on the wheat powder protein model. The results show that scanning sample by 4, 8 or 16 cm(-1) resolution has little influence on the wheat powder protein math model. The change in He-Ne frequency has influence to wavenumber accuracy, but when the change is within 1 cm(-1), the influence is indistinctive. For FT-NIR instruments with He-Ne to have better stability, we needn't often adjust the He-Ne wavenumber. Sample granularity has more distinctive influence on the NIR math models.

A method to share the information resource in the database of Fourier transform near-infrared(FTNIR) spectrum information of agricultural products and utilize the spectrum information sufficiently is explored in this paper. Mapping spectrum information from one instrument to another is studied to express the spectrum information accurately between the instruments. Then mapping spectrum information is used to establish a mathematical model of quantitative analysis without including standard samples. The analysis result is that the relative coefficient r is 0.941 and the relative error is 3.28% between the model estimate values and the Kjeldahl's value for the protein content of twenty-two wheat samples, while the relative coefficient r is 0.963 and the relative error is 2.4% for the other model, which is established by using standard samples. It is shown that the spectrum information can be shared by using the mapping spectrum information. So it can be concluded that the spectrum information in one FTNIR spectrum information database can be transformed to another instrument's mapping spectrum information, which makes full use of the information resource in the database of FTNIR spectrum information to realize the resource sharing between different instruments.

The latent root regression model with near-infrared spectra of 40 soybean samples was founded for analyzing the content of soybean protein in this study. The contents of protein in another 32 soybean samples were predicted by this model. The predicting results were compared with PLS, which shows that the latent root regression model can practically be used for the quantitative analysis of the biological samples with near-infrared spectra. This method is a new kind of chemometrics calibration method, which is modified from PCR. Because the method takes the role of sample composition into account when extracting the principal component from the NIR spectra of samples, the model has a good result in analyzing samples. Further more, the results showed that it is necessary to take account of the role of sample composition when building quantitative analysis model using NIR spectra.