近红外光谱结合连续投影算法检测综纤维素含量

Determination of Holocellulose Content Using Near Infrared Spectroscopy and Successive Projections Algorithm

中文关键词:  近红外光谱  连续投影算法  制浆材  综纤维素含量

Key Words:near infrared spectroscopy  SPA  pulping wood materials  holocellulose content

中文摘要:

      为了简化模型并实现制浆材综纤维素含量近红外光谱法的快速准确检测，以连续投影算法（SPA）筛选出有效波长组合进行了建模实验研究和分析。选择5种制浆材原料共82个样品，测量其综纤维素含量及光谱数据，经蒙特卡罗交叉验证法剔除异常样品后，剩余样品按2∶1划分为校正集和预测集。校正集先以多元散射校正（MSC）方法预处理，再利用SPA选择波长结合偏最小二乘（PLS）回归建立了综纤维素含量的近红外分析模型，并与相关系数法、竞争性重加权自适应选择（CARS）算法所选波长的建模及预测效果进行了比较。结果表明，SPA算法选择出25个波长能充分表征全谱图中的综纤维素含量信息，预测精度最高，预测均方根误差（RMSEP）和预测决定系数（R2p）分别为0.8306和0.9801，满足工业应用精度需求。

Abstract:

      In order to simplify the model and realize the rapid and accurate determination of the pulping wood’s holocellulose content by near infrared spectroscopy， the effective wavelength combination was selected by successive projections algorithm （SPA） to conduct research and analysis of modeling experiment. A total of 82 samples of 5 kinds of pulpwood were prepared to measure the holocellulose content and sample spectral data. After removing the outliers by Monte Carlo cross validation method， all remaining samples were split into calibration and prediction sets by the ratio of 2∶1. The calibration sets was pretreated by MSC method， and using SPA to select wavelengths， a near infrared analysis model of the holocellulose content was established by partial least square （PLS） regression. The model performance was compared with that of the correlation coefficient method as well as competitive adaptive reweighed sampling （CARS） algorithm. The results showed that the 25 wavelengths selected by the SPA could fully characterize the holocellulose content information hidden in the whole spectrum. The PLS regression model based on SPA had the highest prediction accuracy with RMSEP=0.8306 and R2p=0.9801 respectively，which met the precision requirements of industrial applications. The investigation also provided a more effective method for rapid determination of the holocellulose content of pulp woods．