化学计量学和近红外光谱法用于木材与竹材品质研究

【作者】 焦淑菲；

【导师】 张卓勇；

【作者基本信息】 首都师范大学 ， 分析化学， 2009， 硕士

【摘要】 综纤维素（包括纤维素和半纤维素）和木质素是组成木材和竹材的主要成分，它们与木材和竹材的加工利用密切相关。在造纸工业中，纤维素含量多少与纸浆得率和纸浆质量密切相关；木质素的含量是制定蒸煮和漂白工艺条件的重要依据。测量植物纤维原料中综纤维素含量的方法主要是依据国标GB/T 2677.10-1995，测量木质素含量的方法主要是依据国标GB/T 2677.8-1994。综纤维素和木质素含量的传统分析方法操作手续繁琐，成本高，无法实现大批量样品的快速测定。因此需要发展一种快速、高效、准确的新型分析检测技术。近红外光谱是一种绿色分析技术，具有快速、无损、简单、准确等特点。由于近红外光谱重叠严重，因此在进行定量分析中必须使用化学计量学技术与计算机数据处理来提取相关信息。本文将化学计量学方法与近红外光谱技术相结合研究了木材与竹材的品质参量。具体研究内容包括以下四个方面：1.近红外光谱法结合反向传播的人工神经网络测定桉树和毛竹中木质素与综纤维素的含量。用常规湿化学方法测定了72株桉树样品的综纤维素含量，54株毛竹样品的木质素含量以及53株毛竹样品的综纤维素含量。为了提高信噪比和计算速度，对原始近红外光谱进行平滑、压缩、归一化预处理。采用Leave-n-out交叉验证法优化了隐含层神经元的个数、学习率、动量因子和学习次数。三组数据的预测均方根误差分别为0.57%、0.88%和1.40%，实验结果比较令人满意。2.近红外光谱法结合径向基函数网络测定毛竹中综纤维素的含量。以53个毛竹样品作为实验材料，对光谱数据进行平滑、求导、压缩以及归一化,用毛竹的近红外光谱数据建立径向基函数网络模型。预测模型的预测均方根误差为0.0323。结果表明，该方法测量比较准确，可以用于毛竹中综纤维素含量的预测。3.近红外光谱法结合支持向量机测定桉树和杉木中木质素与综纤维素的含量。分别以72个桉树样品、58个杉木样品以及47个杉木样品作为实验材料，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，以径向基函数(RBF)作为核函数，建立了定量分析模型。三组数据校正相对误差的平方和分别为0.01047、0.003626和0.007433，预测相对误差的平方和为0.005057、0.009576和0.001219。结果表明，该方法测量比较准确。4.近红外光谱法结合广义回归神经网络测定桉树中综纤维素的含量。以72个桉树样品作为实验材料，对光谱数据进行平滑、求导、压缩以及归一化,用桉树的近红外光谱数据建立广义回归神经网络模型。预测模型的预测均方根误差为0.0198。结果表明，该方法测量比较准确，可以用于桉树中综纤维素含量的预测。更多还原

【Abstract】 Holo cellulose (including cellulose and hemicellulose) and lignin are the chief constituent oftimber and bamboo wood. They correla te closely to the processing and utilization of timber andbamboo wood. In the paper industry, the content of cellulose correla te closely to the get rate ofpaper pulp and the quality of paper pulp, the content of lignin is basic to a making of steamingand bleaching technology conditions. The measuring of the content of holo cellulose in fibremateria ls follows the nationa l standard 2677.10-1995, the measuring of the content of ligninfollows the nationa l standard 2677.8-1994. The traditiona l analysis methods of the content ofholo cellulose and lignin are complicated and expensive. Therefore, a new fast, efficient, accurateanalysis technology must be developed.Near infrared spectrum is a green analytica l technology. It is fast, non-destructive, easy andaccurate. But near infrared spectrum overlaps seriously and it needs analysis by chemometricsmethods and computer data processing. The character parameter of timber and bamboo wood isresearched by near-infrared reflecta nce spectroscopy and chemometrics methods. The results asfollows:1. Holocellulose content of eucalyptus and bamboo, lignin content of bamboo were predicted bynear-infrared reflecta nce spectroscopy and back propagation artificia l neura l network (BP-ANN).Holocellulose content of 72 eucalyptus samples, lignin content of 54 bamboo samples andholocellulose content of 53 bamboo samples was measured according to wet-chemica l method.In order to improve the signa l-noise ratio and accelera te computation speed of neura l network,the raw spectral data were pretreated by smoothing, compress and scaling. The number of hiddenneurons, learning rate, momentum, and epochs were optimized by using lea ve-n-out crossvalidation approach. The root mea n square error of prediction are 0.57%, 0.88% and 1.40%.These results are satisfactory.2. Via near-infrared reflecta nce spectroscopy combined with radia l basis function (RBF) neura lnetwork, a model for determining holocellulose content of bamboo was established . 53 bamboosamples were used as experimental materia l. The spectral data were pretreated by smoothing,derivative, compress and scaling. A real data set from near-infrared reflecta nce spectroscopy ofbamboo were used to build up models with RBF. The root mea n square error of predicted model is 0.0323. These results demonstrate that the method is precise. It can be used to determinateholocellulose content of bamboo.3. Via near infrared spectroscopy combined with support vector machine (SVM), models fordetermining holocellulose content of eucalyptus and Chinese fir, lignin content of Chinese firwere established . 72 eucalyptus samples, 58 Chinese fir samples and 47 Chinese fir samples wereused as experimental materia l. The spectra of samples were recorded by near infraredspectrometer. The spectral data were pretreated by smoothing, derivative, compress and scaling.The radia l basis function (RBF) was used as the kernel function. A model with support vectormachine was established. The sum of the square of the relative calibration error are 0.01047,0.003626 and 0.007433, the sum of the square of the relative prediction error are 0.005057,0.009576 and 0.001219. These results demonstrate that the method is precise.4. Via near-infrared reflecta nce spectroscopy combined with genera lized regression neura lnetwork (GRNN), a model for determining holocellulose content of eucalyptus was established .72 eucalyptus sample s were used as experimental materia l. The spectral data were pretreated bysmoothing, derivative, compress and scaling. A real data set from near-infrared reflecta ncespectroscopy of eucalyptus were used to build up models with GRNN. The root mea n squareerror of predicted model is 0.0198. These results demonstrate that the method is precise. It can beused to determinate holocellulose content of eucalyptus.更多还原