文献类型：期刊文献

中文题名：基于近红外光谱特征波长筛选的湿地松木材基本密度多品系预测模型研究

英文题名：Prediction Model of Wood Basic Density in Multiple Pinus elliottii Half-sib Families Based on Near-Infrared Spectroscopy and Feature Wavelength Selection

作者：梁龙[1,2] 吴珽[1] 朱宏伟[2] 马灵玉[3] 王哲[4] 房桂干[1]

第一作者：梁龙

机构：[1]中国林业科学研究院林产化学工业研究所,江苏南京210042;[2]岳阳林纸股份有限公司,湖南岳阳414002;[3]中国林业科学研究院木材工业研究所,北京100091;[4]广东省森林培育与保护利用重点实验室,广东省林业科学研究院,广东广州510520

年份：2025

卷号：39

期号：6

起止页码：33-42

中文期刊名：木材科学与技术

外文期刊名：Chinese Journal of Wood Science and Technology

收录：;北大核心:【北大核心2023】；

基金：农业生物育种国家科技重大专项课题“纸浆材及结构材用松树木材性质与质量评价”(2023ZD0405905)。

语种：中文

中文关键词：湿地松;近红外光谱;基本密度;竞争自适应重加权采样算法;预测模型

外文关键词：Pinus elliottii;near-infrared spectroscopy;basic density;competitive adaptive reweighted sampling;prediction model

分类号：S784;O657.33

摘要：以湿地松(Pinus elliottii)三个品系EB2、A04、A05为研究对象,分析其木材基本密度轴向变异规律和制浆适应性,并探讨基于近红外光谱技术构建多品系通用基本密度预测模型的可行性。研究结果显示,三个品系基本密度均呈现从树干基部到顶部递减的趋势,其中EB2品系基本密度(513±52 kg/m^(3))适中、变异系数(10.30%)最低,表现优良的制浆潜力;不同品系间的生物学差异以及光谱自身的高冗余、高维度特性,制约了多品系全波段混合建模的稳健性及泛化能力;通过竞争自适应重加权采样(CARS)算法筛选出17个与基本密度高度相关的特征波长,据此构建模型对各品系均表现出良好的泛化能力,预测均方根误差(RMSEP)为19.6~23.23 kg/m^(3),决定系数(R2)达0.81以上,可实现湿地松木材基本密度的快速无损检测。该研究为纸浆材育种中木材性状的高通量评价提供技术支撑。
Three half-sib families(EB2,A04 and A05)of Pinus elliottii were used to analyze the longitudinal variation patterns of wood basic density and pulping potential,and to investigate the potential of developing a near-infrared(NIR)spectroscopy model to predict the basic density of multiple families.The results indicate that the basic density of three families decreased from the base to the top of the trunk.Among them,the EB2 family exhibited moderate density and the lowest coefficient of variation,demonstrating excellent pulping potential.Biological differences among families,combined with the high redundancy and high-dimensional nature of spectral data,limited the robustness and generalizability of multi-families full-spectrum models.By applying the competitive adaptive reweighted sampling(CARS)algorithm,17 feature wavelengths highly correlated with basic density were selected.The model developed with these wavelengths exhibited robust generalization across all the three families,achieving root mean square error of prediction(RMSEP)values between 19.6 to 23.23 kg/m^(3)and determination coefficients(R2)exceeding 0.81.This enables fast,non-destructive assessment of wood basic density in Pinus elliottii.This study provides technical support for high-throughput evaluation of wood properties in pulpwood breeding programs.