蒙古栎抗弯弹性模量多模型共识的近红外检测方法

Determination of modulus of elasticity for Quercus mongolica using near-infrared spectroscopy based on consensus model

利用近红外光谱技术预测蒙古栎（Quercus mongolica）抗弯弹性模量（MOE）,提出GN（global and neigh-borhood）样本优选与CPLS多模型共识的建模方法。选取近红外光谱谱段为900-1 700 nm,径切面和弦切面采集。首先,采用一阶导数与S-G卷积平滑相结合的方法进行数据预处理;然后,利用GN算法计算光谱样本间的全局和邻域马氏距离,实现蒙古栎MOE异常样本的剔除与校正集、预测集的自动划分;最后,融合具有多个成员样本子集的PLS模型,构建CPLS共识模型,取平均值作为最终预测结果。实验采用135个300 mm×20 mm×20mm的无疵小试样为样本,剔除异常样本12个,并选取其中78个为校正集样本,45个为预测集样本。结果表明,一阶导数处理能够消除光谱背景平缓区域干扰,S-G能消除小峰值无关吸收峰的影响;GN样本优选与CPLS结合的建模方法,预测相关系数为085,相对分析误差（RPD）为193,预测效果比传统PLS建模更好,且稳定性有所提高,但该改进模型方法的RPD依然小于25,因此,可做初步分析,准确地进行定量预测依然存在局限性。

Near-infrared （NIR） spectroscopy was used to examine the modulus of elasticity （MOE） for Quercus mongolica wood and a model using global and neighborhood（GN） distance to select samples and consensus PLS（CPIS） was proposed. The NIR spectra were collected in the range of 900-1 700 nm on the radial and tangential surfaces of wood samples. Initially, first derivative combined with the S-G smooth method was used to pre-process. Then, the global and neighborhood Mahalanobis distance was calculated to eliminate the abnormal samples and divide calibration and pre- diction sets automatically. Finally, each sub-sets of PLS model was combined to establish CPLS, and the average values were reported as the examined MOE values. The samples with a size of 300 mm×20 mm×20 mm （ 135 in total） and with- out defects were used, and 12 abnormal samples were eliminated during the examination. The 78 samples were used for the calibration set, and 45 samples were for the validation set. The results showed that the first derivative can eliminate the influence of gentle area, and S-G can inhibit the irrelevant absorption peak. With the correlation of 0. 85, RPD of 1.93, and improved stability, the method combined with GN method and CPLS model was better than traditional PLS. However, RPD was lower than 2. 5, indicating that the method can perform well in the preliminary analysis rather than the accurately quantitative analysis.