Determination of cellulose,hemicellulose and lignin content using near-infrared spectroscopy in flax fiber

Abstract

The rapid and accurate determination of flax fiber composition is necessary for its application, but until now it has mainly been tested by the wet chemical method, which is time-consuming and not environmentally friendly. In this paper, near-infrared (NIR) spectroscopy was studied to determinate the main composition of flax, in which 43 flax samples were tested according to the traditional Chinese wet chemical component test standard. Five sets of spectra were generated to show the characteristic of each sample; in total 215 spectra sets were collected using a Fourier transform near-infrared spectrometer. The methods of partial least squares (PLS) and principal component regression (PCR) were used to establish the relationships between the data from the chemical and NIR methods. PLS proved to be a better quantitative method than PCR, based on the value of the coefficient of multiple determination for calibration (Rc²) and prediction (Rp²), the ratio of performance to standard deviate (RPD) and the root mean square error of prediction (RMSEP). With the best pretreatment method, the spectral range of 10,000–4000 cm^–1 yielded a better predictive result than the full range, with Rc² of 0.968, Rp² of 0.955, RMSEP of 1.060%, RPD of 4.641 for cellulose and Rc² of 0.958, Rp² of 0.906, RMSEP of 0.678%, RPD of 3.305 for hemicellulose, while the spectral range 6900–5600 cm^–1 yielded a better predictive result with Rc² of 0.936, Rp² of 0.769, RMSEP of 0.455%, and RPD of 2.366 for lignin. The study shows that NIR models can provide a simple and fast way to analyze flax fiber composition, which is also beneficial to evaluate its quality.

Keywords

flax fiber near infrared cellulose content hemicellulose content lignin content chemometric analysis

Flax, like jute, hemp, ramie and kenaf, is a bast fiber. Flax plants¹ are divided into fiber flax, oil flax and dual-use flax. Fiber flax is the main raw material used to produce middle-to-high grade linens. Linens made from flax fiber have the unique properties of being hygroscopic, cool, antiseptic and antistatic, having a natural texture and being soft in color and rough.² They have worldwide applications because of their sanitary and stiff properties. Flax is a type of cellular fiber with several nuclei, in which the morphological constituents, such as cellulose, hemicellulose, lignin and pectin, affect its physical, mechanical and chemical properties.³ Therefore, it is very important to determine the content of hemicellulose, cellulose and lignin in flax to identify its application in the textile or other industry areas. There is no standard of detecting the composition of bast fibers in the world except in China, and the existing traditional Chinese wet chemical component test standard (GB/T5889-86) has its limitations, such as being time-consuming, being a complicated operation and the potential dangers to the tester and environmental pollution.^4,5 According to the traditional Chinese wet chemical component test standard (GB/T5889-86), it takes about 6 full days to identify the content of hemicelluloses, cellulose and lignin for one flax fiber sample. One sample needs to be measured three times for data accuracy. Therefore, a simple and fast detection for flax composition is essential.

As a spectral analysis technique, the near-infrared (NIR) technique can achieve the requirement of fast and multiple composition analysis.^6,7 Recently, the NIR technique has been frequently used to quickly analyze the chemical composition of plant materials and has shown comparative advantages and value. Once reliable and robust NIR models are built, it will take only about several hours for sample preparation, and then the content of cellulose, hemicellulose and lignin can be measured through the NIR model predication without any chemical process. The NIR technique is expensive for most industries, and is mainly used by research institutes at present. The application of Fourier transform near-infrared (FT-NIR) techniques has been demonstrated in quantitatively analyzing the chemical composition of corn.⁸ NIR models were also found to be suitable for the fast and accurate analysis of the chemical composition of maizesilage,⁹ rice straw⁷ and tobacco.¹⁰ Kelley et al.⁴ obtained solid calibration models for most of the biomass components in various agricultural samples. The NIR technique proved to be useful to test hemicellulose, cellulose and lignin in ramie,¹¹ Moso Bamboo,¹² Miscanthussinensis¹³ and big bluestem.¹⁴ Through the NIR method, Miryeong et al.^15–18 determined the content of wax in flax fiber, the shive content of flax and the content of flax in flax/cotton blended products. Using the NIR method, Jiang et al.¹⁹ and Zhou et al.^20,21 obtained a series of research results about the determination of mixed hardwood lignin and carbohydrate content, cellulose content in pulp and the classification and identification of plant fibrous material.

However, it was found that the research on applying NIR spectroscopy to determine the main chemical compositions of flax was limited. In this study, FT-NIR spectroscopy was used to establish the NIR models to predict the contents of hemicellulose, cellulose and lignin in flax fiber. The models derived from this work can be used to quickly analyze the main chemical compositions of flax fiber.

Methods

Sample preparation

Forty-three flax fiber specimens with different chemical components were selected in which 25 specimens were planted in France, Belgium and Netherlands. The remaining 18 fiber specimens from France were treated by enzymes aimed to get a broader variation range of chemical components. All specimens were supplied by Royal Golden Eagle Co., Ltd (Zhejiang Province, China).

The traditional Chinese wet chemical component test

Reliable data calibration is the most critical aspect because it affects the precision of the NIR measurement system for the fast chemical characterization.⁸ In this research, the cellulose, hemicellulose, lignin and other contents (including wax, water solute, pectin) in 43 samples were firstly measured following the standard procedure of the traditional Chinese wet chemical component test standard “GB/T5889-86.” It is the most common testing method for bast fiber. The detailed procedure is shown in Figure 1.

Figure 1.

Schematic diagram for the wet chemical analysis to determine bast fiber composition.

All the data (%, w/w) in Figure 1 were moisture-free. “W1, W2, W3, W4, W5” represent the weight percentage of wax, water soluble matter, pectin, hemicellulose and lignin, respectively. The remaining codes (“100- W1- W2- W3 -W4 -W5”) represent the weight percentage of cellulose. Chemical contents were measured by the Chinese wet chemical testing method and another two test rounds were added for accuracy.

NIR measurement

According to the definition of the American Society for Testing and Materials (ASTM), NIR light is a spectral region with a wavelength of 800–2500 nm (the wave number is 12,500–4000 cm^–1).²² The wave number commonly used in practical applications is the measurement unit in the NIR spectral region, and the wave number range of the NIR spectrum is from 12,500 to 4000 cm^–1 usually. The NIR spectrum is mainly caused by the non-resonant vibration of molecular vibration, which causes the molecular vibration to transition from the ground state to the high energy level. The NIR absorbance of a material is mainly associated with the overtone vibrations and combination vibrations of the chemical bonds, such as X-H (X = C,N,O). The NIR absorbance spectra offer a reliable, fast and nondestructive approach for content tests with simple sample preparation, so it is applied widely in the agriculture, pharmaceutical, textile and petroleum industries. In this study, the spectrum covered a range of 12,500–4000 cm^–1 with a spectral resolution of 8 cm^–1. Each spectrum is the average of 32 scans obtained by diffuse reflection. Considering the diversity of flax fiber composition, each sample was scanned 50 times by the NIR spectroscopy device. After achieving hygroscopic equilibrium, about 1 g of the sample (crushed into 60 items) was placed in a sample cup (18 mm in diameter) for scanning to obtain the NIR spectrum.

Chemometric analysis

Unscrambler X 10.5 software was used for NIR model construction. Principal component regression (PCR) and partial least squares (PLS) are two main statistical methods that can be used to build NIR models. These regression methods can be used to find the optimum relationship between the spectral data matrix and the properties of interest.¹² Different pretreatments can be used to correct the baseline and reduce the noise before using these methods to build NIR models, such as baseline, first derivative (FD), multiplication scattering correction (MSC) and standard normal variate (SNV). Thirty-six samples were used for calibration and seven samples were used for prediction to construct and verify the models. Those samples for calibration and prediction were randomly selected, and their data distribution should have a similar mean value and in approximately the same range. The model performance was evaluated by five important indicators,^12,13 namely the coefficient of multiple determination for calibration (Rc²), the coefficient of multiple determination for prediction (Rp²), the root mean square error of calibration (RMSEC), the root mean square error of prediction (RMSEP) and the ratio of performance to standard deviate (RPD). They were calculated as follows:

{Rc}^{2} = 1 - \frac{\sum_{i = 1}^{n_{c}} (y_{i} - {\hat{y}}_{i})^{2}}{\sum_{i = 1}^{n_{c}} (y_{i} - {\bar{y}}_{i})^{2}}

(1)

{Rp}^{2} = 1 - \frac{\sum_{i = 1}^{n_{p}} (y_{i} - {\hat{y}}_{i})^{2}}{\sum_{i = 1}^{n_{p}} (y_{i} - {\bar{y}}_{i})^{2}}

(2)

RMSEC = \sqrt{\frac{\sum_{i = 1}^{n_{c}} ({\hat{y}}_{i} - y_{i})^{2}}{n_{c}}}

(3)

RMSEP = \sqrt{\frac{\sum_{i = 1}^{n_{p}} ({\hat{y}}_{i} - y_{i})^{2}}{n_{p}}}

(4)

where n_c and n_p are the number of samples for calibration and prediction, respectively, y_i is the measured value for the ith sample,

{\hat{y}}_{i}

is the predicted value for the i^th sample and

{\bar{y}}_{i}

is the mean of y_i for all the samples. Rc², Rp² measure how successful the fit is in explaining the variation of the data. A value close to 1 indicates a good fit.

RPD was calculated to assess the predictive ability of the NIR model.¹³ The higher the value of RPD, the more powerful the predictive ability the model obtains. In specific agricultural applications, RPD larger than 1.5 is regarded as good for preliminary screenings and initial predictions; RPD between 2.0 and 2.5 is considered satisfactory for prediction; RPD greater than 3.0 indicates that the model could predict efficiently.²³ RPD was calculated as follows

RPD = \frac{\sqrt{\sum_{i = 1}^{n_{p}} (y_{i, p} - {\bar{y}}_{i, p})^{2} / \sum_{i = 1}^{n_{p}} (y_{i, p} - {\bar{y}}_{i, p})^{2} (n_{p} - 1) (n_{p} - 1)}}{RMSEP}

(5)

where n_p is the number of samples for prediction,

y_{i, p}

is the true value of the ith sample for prediction and

{\bar{y}}_{i, p}

is the mean of

y_{i, p}

of all the samples for prediction.

Results and discussion

Flax samples

Table 1 summarizes the cellulose, hemicellulose and lignin contents of 43 flax samples. For calibration, cellulose ranged from 64.08% to 82.58% with a SD of 4.23, hemicellulose ranged from 9.38% to 19.48% with a SD of 2.46 and lignin ranged from 1.27% to 7.06% with a SD of 1.37. For prediction, cellulose ranged from 65.28% to 80.88% with a SD of 5.01, hemicellulose ranged from 11.05% to 18.20% with a SD of 2.21 and lignin ranged from 2.18% to 4.73% with a SD of 0.95. For robust model construction, NIR modeling required similar average and SDs for calibration and prediction.^11,14,24 As we saw in the table, the SDs of cellulose for calibration and prediction were 4.23 and 5.01, respectively, which are similar. Also, we can come to the same conclusion through the SD of hemicellulose and lignin. The broader the composition of flax samples for calibration was, the wider the application range of the calibration model. For more reasonable prediction results, the composition for prediction should be limited in the range for calibration.

Table 1.

Composition (%, w/w) of flax samples by the chemical method for calibration and prediction

		Cellulose	Hemicellulose	Lignin
Calibration	Number of samples	36	36	36
	Mean	74.58	14.32	3.35
	Range	64.06–82.58	9.38–19.48	1.27–7.06
	SD	4.23	2.46	1.37
Prediction	Number of samples	7	7	7
	Mean	75.88	14.40	3.21
	Range	65.28–80.88	11.05–18.20	2.18–4.73
	SD	5.01	2.21	0.95

Sample spectra

Considering the complex spectral fluctuate characteristic of flax samples, each sample was scanned 50 times to form an initial spectra database. The average of 10 single random spectra in the database was used as a representative curve set, and finally five sets of spectra were generated to show the characteristic of each sample. In total 215 spectra sets can be generated, in which 180 spectra sets are used to develop and 35 spectra sets are used to verify the NIR models, which is presented in Figure 2.

Figure 2.

Two-hundred-and-fifteen spectra sets with 32 scans at a resolution of 8 cm^–1.

In general, all NIR spectra exhibited three main peaks. The first peak was relatively wide, between 6550 and 7100 cm^–1, which was in relation to cellulose (6711 cm^–1) and the O-H stretch in the first overtone of H₂O (6897 cm^–1). The second peak at 5180 cm^–1 was associated with the O-H stretch/O-H deformation of H₂O from a combination of cellulose.^25,26 The third peak potentially corresponded to the asymmetrical C=O stretch in the third overtone of cellulose (4745 cm^–1).^24,25

NIR model development

Forty-three flax samples with 215 spectra sets were analyzed using the PLS and PCR methods. The PLS and PCR methods for models built based on Rc², Rp², RMSEC, RMSEP and RPD are comparative studied in Tables 2–4, respectively. PLS demonstrated better model performance than PCR, as proven by the larger Rc², Rp² and RPD. Therefore, in this study, PLS was applied as a more powerful tool to quantitatively analyze the contents of cellulose, hemicellulose and lignin in the flax fiber.

Table 2.

Results of the PLS and PCR models for cellulose with different pretreatments based on the full and reduced spectral range

Method	Indicator	Full range (12,500–4000 cm^–1)		Reduced range (10,000–4000 cm^–1)
Method	Indicator	Original spectra	Optimum pretreatment (baseline and FD)	Original spectra	Optimum pretreatment (FD and SNV)
PLS	Rc ²	0.935	0.998	0.931	0.968
	RMSEC	1.097	0.177	1.134	0.777
	Rp ²	0.921	0.899	0.934	0.955
	RMSEP	1.402	1.594	1.287	1.060
	RPD	3.471	3.059	3.860	4.641
PCR	Rc ²	0.777	0.849	0.772	0.872
	RMSEC	2.037	1.676	2.062	1.545
	Rp ²	0.871	0.814	0.813	0.891
	RMSEP	1.797	2.163	2.167	1.649
	RPD	2.525	2.173	2.079	3.189

PLS: partial least squares; PCR: principal component regression; FD: first derivative; SNV: standard normal variate; Rc²: coefficient of multiple determination for calibration; RMSEC: root mean square error of calibration; Rp²: coefficient of multiple determination for prediction; RMSEP: root mean square error of prediction; RPD: ratio of performance to standard deviate.

To avoid the background noise, light scattering and the inhomogeneity of the sample, different mathematical pretreatments were used to process the spectra. In this study, the optimum pretreatment results were FD and SNV for cellulose and baseline and FD for hemicellulose and lignin. After optimum pretreatment, the performance of cellulose, hemicellulose and lignin models in the calibration and prediction sub-database significantly increased when they were compared to those established by the original spectra information. A reduced spectral range was also applied for further optimizing the models. Tables 2 and 3 show that the reduced range (10,000–4000 cm^–1) had significantly better model fitness than the full range (12,500–4000 cm^–1) model, with Rc² of 0.968, Rp² of 0.955, RMSEC of 0.777%, RMSEP of 1.060% and RPD of 4.641 for cellulose and Rc² of 0.958, Rp2 of 0.906, RMSEC of 0.511%, RMSEP of 0.678% and RPD of 3.305 for the hemicellulose. Table 4 shows that for the lignin model, the reduced range (6900–5600 cm^–1) had significantly better model fitness than the full range (12,500–4000 cm^–1) model, with Rc² of 0.936, Rp² of 0.769, RMSEC of 0.351%, RMSEP of 0.455% and RPD of 2.366. This demonstrated that cellulose and hemicellulose models provided more accurate prediction; the lignin model provided satisfactory prediction. All of the models tested could be used for most applications.²⁷

Table 3.

Results of the PLS and PCR models for hemicellulose with different pretreatments based on the full and reduced spectral range

Method	Indicator	Full range (12,500–4000 cm^–1)		Reduced range (10,000–4000 cm^–1)
Method	Indicator	Original spectra	Optimum pretreatment (FD)	Original spectra	Optimum pretreatment (baseline and FD)
PLS	Rc ²	0.774	0.984	0.795	0.958
	RMSEC	1.182	0.315	1.126	0.511
	Rp ²	0.759	0.826	0.825	0.906
	RMSEP	1.083	0.922	0.924	0.678
	RPD	2.239	2.693	2.571	3.305
PCR	Rc ²	0.717	0.760	0.745	0.711
	RMSEC	1.324	1.219	1.258	1.338
	Rp ²	0.774	0.755	0.720	0.734
	RMSEP	1.049	1.093	1.169	1.139
	RPD	2.055	2.058	1.967	2.108

As shown in Table 5, the minimum central approximate position of the combination bands and frequency doubling absorption band of the C-H and O-H groups, which were relevant to cellulose, hemicellulose and lignin, was around 4347 cm^–1 and the maximum was 10,500 cm^–1. It was found from Figure 2 that the spectral range of 12,500–10,000 cm^–1 contained the excess noise. As a result (see Tables 2 and 3), the reduced spectral range from 10,000 to 4000 cm^–1 was chosen to build the NIR model of cellulose and hemicellulose, which was proved to be better than the full range. Also, the reduced spectral range from 6900 to 5600 cm^–1 of model development for lignin content was basically in agreement with the NIR region from 6900 to 5400 cm^–1 for lignin characterization.²⁸

Table 4.

Results of the PLS and PCR models for lignin with different pretreatments based on the full and reduced spectral range

Method	Indicator	Full range (12,500–4000 cm^–1)		Reduced range (6900–5600 cm^–1)
Method	Indicator	Original spectra	Optimum pretreatment (Smoothing)	Original spectra	Optimum pretreatment (baseline and FD)
PLS	Rc ²	0.755	0.744	0.738	0.936
	RMSEC	0.687	0.702	0.710	0.351
	Rp ²	0.714	0.735	0.491	0.769
	RMSEP	0.506	0.488	0.676	0.455
	RPD	1.991	2.046	1.780	2.366
PCR	Rc ²	0.567	0.688	0.614	0.747
	RMSEC	0.913	0.775	0.862	0.698
	Rp ²	0.551	0.587	0.596	0.595
	RMSEP	0.635	0.609	0.602	0.603
	RPD	1.710	1.708	1.773	1.886

Table 5.

The central approximate position of the combination bands and frequency doubling absorption band of the C-H and O-H groups

Group	Wavenumbers (cm^–1)
Group	C-H	O-H
Combination bands	4347	5000
The primary frequency doubling	5700	7000
The secondary frequency doubling	8700	10,500

Figure 3 shows plots of predicted versus measured contents of cellulose (a), hemicellulose (b) and lignin (c) in the calibration sub-database. The x-axis represents measured contents of cellulose, hemicellulose and lignin from the traditional wet chemistry component test method according to GB/T5889-86, and the y-axis represents their predicted contents according to optimal NIR models of the calibration sub-database. In Figure 3, the prediction versus reference plots showed that the slope of the linear equation was close to 1 (0.968, 0.958 and 0.936, respectively), in which X was the measured content and Y was the predicted content. This indicated the high predicting quality of the three PLS models, which was proven by the other statistics listed in Tables 2–4.

Figure 3.

Partial least squares regression plots of predicted versus measured contents of cellulose (a), hemicellulose (b) and lignin (c) in 180 spectra sets in the calibration sub-database of flax fiber.

Tables 6–8 provide the actual data of the five measurements of each prediction sample. The samples for prediction represented different contents from those for calibration, but demonstrated consistent results with the samples for calibration. It was also noticed that the SD of samples for prediction was very low, and the average CV (%) values of cellulose, hemicellulose and lignin were 0.39%, 1.87% and 9.52%, while through the traditional test method, the CV (%) values were 1.6%, 3.17% and 17.12%, respectively, in this paper. Generally, the CV (%) value was lower than that of traditional methods, especially for hemicellulose and lignin, for which it was less than half.

Table 6.

Predicted versus measured content of cellulose in the prediction sub-database

Measured content (%)	Predicted content (%)
Measured content (%)	1	2	3	4	5	Mean	SD	CV (%)
65.28	64.31	65.75	65.87	65.92	65.49	65.47	0.67	1.02
72.34	72.42	72.88	72.73	72.72	72.93	72.74	0.20	0.27
76.61	75.25	75.44	76.16	76.01	76.01	75.77	0.40	0.53
77.84	78.96	79.10	78.67	78.86	78.90	78.90	0.16	0.20
78.67	78.58	78.66	78.71	79.00	78.86	78.76	0.17	0.21
79.57	79.43	79.82	80.13	79.71	79.45	79.71	0.29	0.36
80.88	78.53	78.77	78.42	78.67	78.62	78.60	0.13	0.17

Table 7.

Predicted versus measured content of hemicellulose in the prediction sub-database

Measured content (%)	Predicted content (%)
Measured content (%)	1	2	3	4	5	Mean	SD	CV (%)
11.05	11.61	10.99	11.08	11.37	11.49	11.31	0.27	2.31
12.82	11.89	12.25	12.26	11.59	11.86	11.97	0.29	2.40
13.36	13.59	13.38	13.26	13.20	13.06	13.30	0.20	1.53
13.68	15.11	15.03	14.55	14.86	15.34	14.98	0.30	1.92
15.55	15.17	15.20	14.54	15.05	14.96	14.98	0.27	1.78
16.13	15.62	16.12	16.14	16.14	15.86	15.98	0.23	1.46
18.20	18.82	18.11	18.17	18.11	18.11	18.26	0.31	1.72

Table 8.

Predicted versus measured content of lignin in the prediction sub-database

Measured content (%)	Predicted content (%)
Measured content (%)	1	2	3	4	5	Mean	SD	CV (%)
2.18	2.14	2.02	1.88	2.30	2.26	2.12	0.17	8.16
2.24	1.79	2.11	2.00	1.66	1.42	1.80	0.28	15.32
2.71	3.31	3.34	2.22	2.77	2.78	2.88	0.46	16.08
2.85	2.82	2.29	2.39	2.80	2.49	2.56	0.24	9.32
3.32	2.95	3.02	3.04	2.72	3.31	3.01	0.21	6.99
4.46	3.64	3.81	3.98	4.54	4.44	4.08	0.39	9.63
4.73	5.41	5.37	5.30	5.47	5.40	5.39	0.06	1.15

Conclusions

FT-NIR spectroscopy together with chemometric analysis was applied to the quantitative prediction of cellulose, hemicellulose and lignin contents of flax. PLS was proved to be a better quantitative method than PCR, based on larger Rp² and RPD and smaller RMSEP when establishing the NIR prediction model. The spectral range from 10,000 to 4000 cm^–1 with the FD and SNV treatment yielded a better prediction model, with Rc² of 0.968, Rp² of 0.955, RMSEP of 1.060% and RPD of 4.641 for cellulose; the spectral range from 10,000 to 4000 cm^–1 with the baseline and FD treatment yielded a better prediction model with Rc² of 0.959, Rp² of 0.906, RMSEP of 0.678% and RPD of 3.305 for hemicellulose; and the spectral range from 6900 to 5600 cm^–1 with the baseline and FD treatment yielded a better prediction model with Rc² of 0.936, Rp² of 0.769, RMSEP of 0.455% and RPD of 2.366 for lignin. NIR models were established to show good potential as a highly efficient tool for rapid and accurate measurement of cellulose and hemicellulose content in flax fiber. Moreover, the model was also good for the prediction of lignin content to some extent. In future work, the accuracy of lignin prediction needs to be improved and the sample size for calibration needs to be increased reasonably for a wider range to improve the model robustness.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This worked was supported by the earmarked fund for China Agriculture Research System for Bast and Leaf Fiber Crops: CARS-19; China Academy of Agricultural Science and Technology Innovation Project: ASTIP-IBFC07.

References

Faruk

Bledzki

Fink

, et al. Biocomposites reinforced with natural fibers: 2000-2010. Progr Polym Sci 2012; 37: 1552–1596.

. Textile materials, 1st ed. Beijing: China Textile & Apparel Press, 2006, pp. 17–19.

Flynn

Amiri

Ulven

. Hybridized carbon and flax fiber composites for tailored performance. Mater Des 2016; 102: 21–29.

Kelley

Rowell

Davis

, et al. Rapid analysis of the chemical composition of agricultural fibers using near infrared spectroscopy and pyrolysis molecular beam mass spectrometry. Biomass Bioenergy 2004; 27: 77–88.

Zheng

Zhang

. Degumming of ramie fibers by alkalophilic bacteria and their polysaccharide-degrading enzymes. Bioresource Technol 2001; 78: 89–94.

Cornish

Myers

Kelley

. Latex quantification in homogenate and purified latex samples from various plant species using near infrared reflectance spectroscopy. Ind Crop Prod 2004; 19: 283–296.

Jin

Chen

. Near-infrared analysis of the chemical composition of rice straw. Ind Crop Prod 2007; 26: 207–211.

Liu

Hayes

, et al. Fast classification and compositional analysis of cornstover fractions using Fourier transform near-infrared techniques. Bioresource Technol 2008; 99: 7323–7332.

Cozzolino

Fassio

Fernández

, et al. Measurement of chemical composition in wet whole maize silage by visible and near infrared reflectance spectroscopy. Animal Feed Sci Technol 2006; 129: 329–336.

10.

Zhang

Cong

Xie

, et al. Quantitative analysis of routine chemical constituents in tobacco by near-infrared spectroscopy and support vector machine. Spectrochimica Acta A 2008; 71: 1408–1413.

11.

Jiang

Han

Zhang

, et al. Fast compositional analysis of ramie using near-infrared spectroscopy. Carbohydr Polym 2010; 81: 937–941.

12.

Sun

Zhou

, et al. Determination of hemicellulose, cellulose and lignin in moso bamboo by near infrared spectroscopy. Sci Rep 2016; 5: 1–11.

13.

Jin

Chen

Shi

, et al. Determination of hemicellulose, cellulose and lignin content using visible and near infrared spectroscopy in Miscanthussinensis. Bioresource Technol 2017; 241: 603–609.

14.

Zhang

Loretta

, et al. Development of near-infrared spectroscopy models for quantitative determination of cellulose and hemicellulose contents of big bluestem. Renew Energ 2017; 109: 101–109.

15.

Miryeong

David

Danny

, et al. Fourier transform near-infrared spectroscopy for determining linen content in linen/cotton blend products. Text Res J 2005; 75: 583–590.

16.

Miryeong

Franklin

David

. Transfer of near-infrared calibration model for determining fiber content in flax: effects of transfer samples and standardization procedure. Appl Spectrosc 2007; 61: 414–418.

17.

Miryeong

David

Herbert

, et al. Partial least squares regression calibration for determining wax content in processed flax fiber by near-infrared spectroscopy. Appl Spectrosc 2006; 60: 437–440.

18.

Miryeong

Franklin

Wiley

, et al. Application of partial least-squares regression to near-infrared reflectance spectroscopic determination of shive content in flax. Appl Spectrosc 2003; 57: 551–556.

19.

Jiang

Zhou

Han

, et al. Classification and identification of plant fibrous material with different species using near infrared technique—a new way to approach determining biomass properties accurately within different species. Frontier Plant Sci 2017; 7: 2000–2000.

20.

Zhou

Jiang

Brian

, et al. Prediction of mixed hardwood lignin and carbohydrate content using ATR-FTIR and FT-NIR. Carbohydr Polym 2015; 121: 336–341.

21.

Zhou

Han

Gao

, et al. Rapid determination of cellulose content in pulp using near infrared modeling technique. BioResources 2018; 13: 6122–6132.

22.

ASTM E 1944:1998. Standard practice for describing and measuring performance of laboratory Fourier transform near-infrared (FT-NIR) spectrometers: level zero and level one tests. Reapproved 2007:1.

23.

Arana

Jaren

Arazuri

. Maturity, variety and origin determination in white grapes (Vitisvinifera L.) using near infrared reflectance technology. Near Infrared Spectrosc 2005; 13: 349–357.

24.

Zou

Zhao

Malcolm

, et al. Variables selection methods in near-infrared spectroscopy. Analytica Chimica Acta 2010; 667: 14–32.

25.

Burns

Ciurczak

. Handbook of near-infrared analysis, New York: CRC Press, 2007.

26.

Manfred

Jose

Karin

. A review of band assignments in near infrared spectra of wood and wood components. Near Infrared Spectroscopy 2011; 19: 287–308.

27.

Williams

. Implementation of near-infrared technology. Near Infrared Technol Agric Food Ind 2001; 2: 143–143.

28.

Jiang

. Quantative determination of fibrous materials chemical composition and near infrared model construction, Doctoral Dissertation, Donghua University, 2013, pp. 75–76.