基于近红外光谱分析的多模型建模方法研究

以相思树的α-纤维素含量为研究对象,用一种多模型方法建立了相思树α-纤维素含量的近红外光谱分析模型。模型预测值的平均相对误差为0.97%,实验值与预测值之间的相关系数为0.963 1,模型的拟合优度为0.924 5。研究结果表明,使用的光谱数据量越大,模型的预测效果一般会越好。此外还发现了子模型中待定常数的个数与所使用光谱数据量之间的关系:建模时使用的光谱数据量越大,每个子模型中待定常数的个数一般应该越小。该结果有助于今后使用该方法建立其它近红外光谱分析模型。所建模型可用于快速测定相思树的α-纤维素含量,并有望用于其它树种某些化学成分含量的预测。     

湿地松木材近红外光谱与其结晶度的相关性

对湿地松木材近红外光谱与X射线衍射法测定的木材结晶度之间的相关性进行分析,并结合近红外光谱分析技术的基本理论,探讨降低光谱范围和选择相关光谱信息对近红外光谱预测木材结晶度的影响.结果表明:1)降低参与建模的近红外光谱范围仍然可以得到比较理想的近红外光谱模型与预测结果,当选用2 000～2 500 nm区域的光谱建立模型时,预测值与实测值的相关系数r达到0.943;2)当选择光谱范围更小但与木材纤维素吸收峰密切相关的光谱数据(1 400～1 660 nm或2 020～2 250 nm)进行建模时,模型的预测效果并未降低(r＞0.947),甚至仅采用7个光谱数据也可以得到比较理想的预测结果,预测相关系数r可达到0.930,说明采用更少的但与木材纤维素吸收峰密切相关的光谱信息,所建立的预测模型仍可得到比较理想的预测效果,这将有利于低成本、便携式近红外光谱仪的开发.     

支持向量机结合近红外光谱法测定杉木木质素的含量

采用支持向量机(SVM)结合近红外光谱(NIR)技术建立测定杉木中木质素的定量分析模型。以47个杉木样品作为实验材料,用常规方法测定了样品中木质素的含量,用近红外光谱仪采集相应的光谱,对光谱数据进行平滑、求导、小波压缩以及归一化,结合支持向量机,以径向基(RBF)作为核函数,建立了测定杉木中木质素含量的模型。校正相对误差的平方和为0.007433,预测相对误差的平方和为0.001219。结果表明,该方法测量比较准确,可以用于杉木中木质素含量的预测。     

利用近红外光谱快速分析制浆木片水分

针对实际生产中多采用混合品种木片制浆的情况,探讨了利用近红外光谱分析技术对混合木片的水分含量进行快速测定的可行性。通过国产便携式阿达玛光谱仪采集183个木片样品的光谱,经过预处理后,利用偏最小二乘法和完全交互验证方式建立4组木片的近红外光谱数据与其水分含量之间的关联模型。4个模型的相关系数均达到098以上,交叉验证均方根标准差在3%以内,相对分析误差值在68~103之间。利用建好的模型对预测集样品的水分含量进行预测,其中全局模型的预测标准差在127~240之间,结果较为精确。结果表明,尽管木材品种对木片近红外光谱水分特征吸收存在一定的影响,但在光谱预处理后,再以纯种木片与混合木片一起建立的全局模型,具有较好的适应性和较高的预测精度,可用于实际生产中快速测量混合木片的水分含量。     

近红外光谱快速预测乙酰化木材的增重率

采用近红外光谱技术对乙酰化大青杨和樟子松木材的增重率进行快速预测。在近红外波长780～2500 nm范围内,利用偏最小二乘法( PLS)建立木材横切面原始光谱及不同预处理(一阶导数、二阶导数、归一化处理和消噪)光谱乙酰化木材增重率数学模型,并进行比较分析。结果表明：乙酰化大青杨和樟子松木材分别选用归一化处理光谱和消噪光谱建立的增重率校正模型预测效果较好,预测模型相关系数( R)分别为0.90和0.70,预测标准差(RMSEP)分别为1.0072和1.3012,其中乙酰化大青杨木材增重率预测模型实测能力较佳,表明利用木材横切面近红外光谱建立的数学模型可以实现乙酰化木材增重率的快速预测。     

非线性算法在近红外预测木材密度中的应用研究

研究基于近红外光谱技术的木材密度预测。运用基于高斯核变换的非线性偏最小二乘法建立密度预测模型,并且对所建模型的评价参数进行了对比分析。结果表明该方法建立的预测模型能对样品的密度进行有效预测。研究表明样品近红外光谱信息与样品的实际密度值之间不是单纯的线性关系,非线性模型可以更好地表征二者之间的关系。     

基于近红外光谱与BP神经网络预测落叶松木屑的含水率

利用近红外光谱(NIR)技术结合BP神经网络定量预测了落叶松木屑的含水率。首先对采集的落叶松木屑原始近红外光谱进行9点平滑及多元散射校正预处理,然后利用主成分分析法提取光谱数据主成分作为BP神经网络的输入,最后建立BP神经网络预测模型并采用交叉验证法对模型进行验证。所建模型校正集的相关系数R为0.98,校正集的均方根误差RMSEC为0.001 7;预测集的相关系数R为0.99,预测集的均方根误差RMSEP为0.001 5。研究表明,此方法可以实现对落叶松木屑含水率的快速预测。     

速生材制浆性能与其材性的多元非线性回归分析及预测方法

采用多元非线性回归方法分析速生阔叶木材的制浆得率与其化学组分、纸浆的抗张指数、撕裂指数分别与纤维形态的相关性,建立速生材制浆性能的预测模型,选取代表性的速生木材进行材性分析和制浆性能试验研究,用实测值与预测模型算出预测值进行对比.结果证明:预测值与实际测定值拟合得很好,预测模型的精度高.     

近红外光谱结合支持向量机快速识别树种

利用阿达玛变换近红外光谱结合支持向量机,对制浆造纸常用木材树种的快速识别进行研究。将各树种近红外光谱先进行多点平滑和标准正态变换预处理以消除噪音干扰和光散射导致的测量偏差,然后基于不同建模策略建立一对多和一对一两种支持向量机模型,考察这两种模型对多树种属间分类和种间分类的预测能力,并与传统的偏最小二乘判别分析分类法进行对比。结果表明,支持向量机预测模型对桉木、相思木、杨木、水杉等树种的属间分类正确率达到98%以上,种间分类正确率均达到95%以上,在处理复杂分类问题时模型稳健性明显优于传统分类方法,从方法上证明了近红外技术工业化应用的可能性,为进一步建立近红外在线检测木片材性分析系统奠定了基础。     

近红外光谱法测定毛竹综纤维素的含量研究

研究了用近红外光谱(NIR)结合多变量统计分析技术对毛竹综纤维素含量的快速测定。用常规实验室方法测定了54个竹材样品的综纤维素含量,用近红外光谱仪采集相应样品的光谱,对原始光谱进行二阶导数和25点平滑预处理后,从54个竹材样品中挑选41个代表性的样品建模,选择1011~1675nm和1930~2488nm波段区间,用偏最小二乘法(PLS1)和完全交互验证方式建立毛竹综纤维素含量的预测模型。结果表明,毛竹综纤维素含量和近红外光谱之间存在非常好的相关性,预测模型的相关系数(RP)为0.95,预测模型的标准偏差(SEP)为0.76%。     

The use of near-infrared scanning for the prediction of pulp yield and chemical properties of Pinus patula in the Mpumalanga escarpment area of South Africa

Near-infrared (NIR) scanning technology is regarded as a potential tool for rapid determination of wood properties, which can substitute time-consuming and costly traditional methods. Pinus patula is the most important softwood species in South Africa, and this study is aimed at developing NIR calibration models for quick prediction of its pulp yield and chemical composition. A total of 85 trees from 17 plots, covering the range of site conditions in the Mpumalanga escarpment area, were sampled. Two samples were taken from each tree: a 1 m billet above breast height and a 20 mm disc at breast height. The billet was pulped using the kraft pulping process to determine pulp yield. The disc was ground into sawdust and the chemical composition was determined using conventional wet chemistry. Sawdust was scanned on a NIR spectrophotometer to produce NIR spectra. Calibration models to predict pulp yield, cellulose and lignin content were developed by applying chemometrics and partial least squares regression. Validation and determination of prediction accuracy of the models were performed using independent data. The prediction of cellulose and lignin were acceptable with correlations of determinations (r ²) of 0.71 and 0.70 respectively. Standard errors of prediction were generally low (less that 0.86) for all the models. The prediction r ² for both total and screened pulp yield were only 0.62. Although the cellulose and lignin models can be used with confidence, the expansion of the sample size for follow-up research must be considered in order to increase the variability of tested wood properties and improve the prediction strength of the models. The NIR calibration provided in this study can contribute to the efficient examination of forest site-to-wood quality relationships that would enhance precision forest management and wood processing efficiency.     

快速检测木材微纤丝角的近红外光谱分析技术

介绍了一种新的测量木材微纤丝角的无损检测技术--近红外光谱分析,并详细阐述了测量时木材样品的选择及制备,测量工作包括X射线衍射和近红外光谱采集、多变量数据分析与模型建立的方法及步骤.证明了近红外光谱分析技术可以用于快速准确地预测木材的微纤丝角.     

粗皮桉木材力学性质的近红外光谱方法预测

以人工林粗皮桉木材为研究对象,采用常规力学测试方法和近红外光谱方法对其无疵小试样力学性质进行研究。用近红外光谱仪采集试样表面的近红外光谱,对采集的近红外漫反射光谱进行导数预处理并对不同波段光谱建立校正模型,以1/3试样作为预测集对校正模型进行验证。结果表明:二阶导数预处理、350～25000nm全光谱波段、径切面和弦切面平均光谱值对粗皮桉木材力学性质模型预测效果最好。抗弯弹性模量和抗弯强度、顺纹抗压强度的实测值与近红外光谱方法的预测值存在较好的相关性,相关系数均大于0.88,相对分析误差大于2.0,表明利用近红外光谱方法预测人工林粗皮桉木材力学性质效果较好。     

Rapid assessment of wood chemical properties and pulp yield of Eucalyptus camaldulensis in Thailand tree plantations by near infrared spectroscopy for improving wood selection for high quality pulp

Near-infrared (NIR) spectroscopy has been demonstrated as a means for rapid nondestructive determination of the chemical composition and final pulp yield of Eucalyptus camaldulensis in Thailand tree plantations. Multiple linear regression (MLR) analysis and partial least squares (PLS) analysis were introduced to develop statistical models in terms of calibration equations for total pulp yield, screened pulp yield, and contents of -cellulose, pentosans, and lignin in wood. In MLR analysis, a reasonably good calibration equation was found only for pentosans (standard error of prediction (SEP): 0.98%). The PLS analysis improved the accuracy of prediction for every criterion variable, especially for pentosans (SEP: 0.91%) and lignin (SEP: 0.52%). Also, in the case of screened pulp yield, we were able to use such a statistical result as an indicator of the characteristics of the pulp and paper. Thus, NIR spectroscopy could be satisfactorily used as an effective assessment technique for Eucalyptus camaldulensis plantation trees.     

桦木顺纹抗压强度的SEPA-VISSA-RVM近红外光谱预测

木材顺纹抗压强度是评价木材力学性能的重要指标,而传统测量方法操作复杂、精确度低。以桦木为例,提出基于近红外光谱技术(NIR)的SEPA-VISSA-RVM木材顺纹抗压强度模型,实现对其更加精确的预测。试验选取100个木材试件,在900~1700 nm近红外光谱波段上采集数据并测量抗压强度真值;然后采用卷积平滑(SG)方法进行光谱预处理;使用采样误差分布分析(SEPA)作为变量空间迭代收缩算法(VISSA)的改进策略进行特征波长优选;最后通过粒子群优化算法(PSO)优化核函数参数并建立相关向量机(RVM)的预测模型。试验表明:在特征波长优选方面,以偏最小二乘法(PLS)建模为基础的SEPA-VISSA方法,其预测决定系数为0.9593,预测均方根误差为2.8995,相对分析误差为3.0256,光谱变量数由512减小到111个,占总波长的22%,均优于VCPA、CARS和VISSA算法;在建模预测方面,以SEPA-VISSA所选波长为基础的RVM模型,PSO优化的拉普拉斯(Laplacian)核函数的核宽度为10.4043,决定系数为0.9449,预测均方根误差为2.0432,相对分析误差为4.2936,预测效果优于PLS和SVR。因此,基于近红外光谱的SEPA-VISSA-RVM建模能够实现对桦木顺纹抗压强度更准确和稳定的无损检测。     

Nondestructive sampling of Eucalyptus globulus and E. nitens for wood properties. III. Predicted pulp yield using Near Infrared Reflectance Analysis

  Within-tree variation in kraft pulp yield, predicted using near infrared reflectance analysis, was studied in thirty trees of E. globulus and fifty trees of E. nitens to develop a non-destructive sampling strategy. Trees, aged 5 to 9 years, were sampled across a range of sites in southern Australia. Simulated core samples were removed at six fixed heights easily accessible from the ground (0.5, 0.7, ... 1.5 m) and at seven percentage heights (0, 20, 30, ... 70%). Whole-tree values, calculated from percentage height data, were correlated with the core data to determine the optimal sampling height. Core samples were found to be good predictors of whole-tree pulp yield for E. globulus, with simulated cores taken from the recommended sampling height (1.1 m) explaining more than 50% of variation in whole-tree pulp yield. Results for E. nitens were variable with large site differences apparent. On high quality sites, core samples from the recommended sampling height (0.9 m) were good predictors of whole-tree pulp yield, explaining around 60% of the variation. On poor quality sites, cores were poor predictors of whole-tree pulp yield. Radial orientation of cores was not important and predicted pulp yield was not related to tree size, basic density or fibre length. To estimate stand mean pulp yield to an accuracy of ±1% would require sampling 6 trees of E. globulus and 4 trees for E. nitens using either multiple discs or core samples. A single sampling height (1.1 m) is recommended for sampling for basic density, fibre length, fibre coarseness and predicted pulp yield in E. globulus. For E. nitens the recommended sampling height for basic density and fibre length is 0.7 m and 0.9 m is recommended for predicted pulp yield on good quality sites. Received 17 September 1998     

Nondestructive determination of paper properties of Populus × euramericana with near infrared spectroscopy

Five Populus x euramericana wood samples representing three different sites were selected and nearinfrared （NIR） spectra were obtained. For these sections, basis weight, brightness and three mechanical properties （tensile index, tearing index and bursting index） were determined by standard analytical methods. Calibrations were developed for each paper property using the NIR spectra, data on paper properties, using partial least squares （PLS） regression. The results show that the coefficients of correlation of calibration and validation for basis weight were 0.8824 and 0.8299, respectively; the standard error of calibration （SEC） and prediction （SEP） were 1.150 and 1.170, respectively. In testing for brightness, the correlation coefficient of calibration was 0.9621 and for validation 0.9612, while the SEC and SEP were 0.997 and 1.300, respectively; paper brightness and NIR spectroscopy were highly correlated. NIR spectroscopy can be used to predict tensile, tearing and bursting indices of paper samples rapidly. We found that the paper properties fitted by NIR and GB methods were highly correlated. The coefficients of correlation of calibration and validation for basis weight exceeded 0.8000, while the SEC and SEP were very small. These results reveal that the five paper properties of Populus x euramericana and those predicted by the NIR model were highly correlated. We conclude that the NIR models can be used for the prediction of paper properties.     

四种算法用于近红外测定制浆材材性的对比研究

采集了常见制浆材(桉木、相思木及杨木)样品的近红外光谱,测定了样品的基本密度、综纤维素、木质素和苯醇抽出物含量,用人为控制水分的方法测定了样品的水分含量。对原始光谱进行预处理后,分别运用偏最小二乘法(PLS)、LASSO算法、支持向量机法(SVR)和人工神经网络法(BP-ANN)建立基本密度、水分含量、综纤维素、木质素和苯醇抽出物含量的预测模型。对预测模型进行独立验证,结果显示:LASSO算法建立的基本密度和综纤维素模型性能最优,其预测均方根误差(RMSEP)分别为0.006 3 g/cm~3和0.49%,绝对偏差(AD)范围分别为-0.008 8~0.009 6 g/cm~3和-0.85%~0.87%;PLS建立的水分含量模型及苯醇抽出物模型最优,RMSEP值分别为1.21%和0.24%,AD范围分别为-1.99%~2.03%和-0.35%~0.38%;SVR建立的木质素模型最优,RMSEP值为0.43%,AD范围为-0.76%~0.74%,均满足制浆造纸工业中对误差的要求。     

一种快速测定棕榈藤材材质特性的近红外光谱分析法

分析了棕榈藤生物质材料特点与近红外光谱关系；阐述了近红外光谱分析技术可以快速准确地预测棕榈藤材的材质特性，并详细介绍了其中包括藤材的选择及加工、藤材近红外光谱采集、藤材材性真值测量、多变量数据分析与模型建立的实验方法与操作步骤。