基于近红外光谱波段优选的针叶木材基本密度估测模型的优化研究  

Estimation of softwood basic density based on near infrared spectral bands optimal selection

作　　者：尹世逵 冯国红[1] 李春旭 赵婧含 孟永斌 王晨[1] 李耀翔[1] YIN Shikui;FENG Guohong;LI Chunxu;ZHAO Jinghan;MENG Yongbin;WANG Chen;LI Yaoxiang(College of Engineering and Technology,Northeast Forestry University,Harbin 150040,Heilongjiang,China)

机构地区：[1]东北林业大学工程技术学院

出　　处：《中南林业科技大学学报》2020年第3期85-95,共11页Journal of Central South University of Forestry & Technology

基　　金：“十三五”国家重点研发计划项目（2017YFC0504103）;中央高校基本科研业务费专项（2572015CB04）;林业工程一流学科科研创新项目

摘　　要：【目的】木材基本密度在木材质量等级评定中具有重要作用,是木材分流及精细化利用的重要依据。【方法】以东北林区典型针叶树种为研究对象,结合近红外光谱技术,构建红松、落叶松、云冷杉木材基本密度近红外估测模型,分析比较了不同波段优选算法并进行了模型优化。研究采用竞争性自适应重加权法(CARS)、无信息变量消除法(UVE)和间隔偏最小二乘法(iPLS)对木材近红外光谱波段进行优化,基于卷积平滑算法对近红外光谱数据进行预处理,结合偏最小二乘法(PLS)建立针叶木材基本密度估测模型。依据相关系数(R)、均方根误差(RMSEC)等模型参数对模型效果进行评价,对比分析确定最佳波段优选方法,得到最优针叶木材基本密度近红外估测模型。【结果】利用CARS、UVE、i PLS的波段优化方法对近红外光谱波段的筛选,可以起到优化针叶木材基本密度模型的作用,减少参与建模的近红外光谱的波段变量数,明显提升模型的运算速度,使得模型准确度更高、稳定性更好;利用间隔偏最小二乘法结合偏最小二乘法(iPLS-PLS)进行波段优选的针叶木材基本密度模型效果最好,其模型校正相关系数为0.938 0,校正均方根误差为0.021 8,验证相关系数为0.8959,验证均方根误差为0.028 0。【结论】基于波段优选及模型优化构建东北林区典型针叶树种基本密度近红外估测模型,可以有效提高运算速度及估测精度,实现针叶材基本密度的快速、准确、无损估测,为针叶木材材性研究和森林培育提供了理论依据与技术支撑,有利于进一步实现木材的高效节约与精细化利用。【Objective】The basic density of wood plays an important role in the quality assessment of wood and it is an important basis for the diversion and refinement of wood.【Method】In this study,the typical conifer species in the northeastern forest area were studied,and the near-infrared estimation model of the basic density of Korean pine,larch and spruce was constructed by using nearinfrared spectroscopy.The optimal algorithm of different bands was analyzed and optimized.The research uses competitive adaptive heavy weighting(CARS),no information variable elimination(UVE)and interval partial least squares(iPLS)to optimize the nearinfrared spectral band of wood,and based on convolution smoothing algorithm for near-infrared spectral data.Pretreatment,combined with partial least squares(PLS),was used to establish a basic density estimation model for coniferous wood.According to the model parameters such as correlation coefficient(R)and root mean square error(RMSEC),the model effect is evaluated,and the optimal band optimization method is determined by contrast analysis.The near-infrared estimation model of the optimal density of coniferous wood is obtained.【Result】The research shows that using the band optimization method of CARS,UVE and iPLS to screen the nearinfrared spectrum can optimize the basic density model of coniferous wood,and reduce the number of band variables of the near-infrared spectrum involved in modeling,and significantly improve the model.The operation speed makes the model more accurate and better.The interval-leaked least squares method combined with partial least squares(iPLS-PLS)is the best for the band-precision soft density model of coniferous wood.The coefficient is 0.9380,the corrected root mean square error is 0.0218,the verification correlation coefficient is 0.8959,and the verification root mean square error is 0.0280.【Conclusion】Based on band optimization and model optimization,a near-infrared estimation model for the basic density of typical coniferous species in northeaster

关 键 词：基本密度 近红外光谱 波段优选 偏最小二乘法 

分 类 号：S781.31[农业科学—木材科学与技术]