应用近红外光谱和小波网络构建的木材基本密度预测模型  

Prediction Model of Wood Basic Density by Near Infrared Spectroscopy and Wavelet Networks

作　　者：潘屾 王克奇[1] 梁玉亮 张怡卓[1] 

机构地区：[1]东北林业大学,哈尔滨150040

出　　处：《东北林业大学学报》2018年第2期59-62,共4页Journal of Northeast Forestry University

基　　金：国家林业局"948"项目(2015-4-25);中央高校基本科研业务费专项资金项目(2572017DB05);黑龙江省自然科学基金项目(C2017005)

摘　　要：以柞木为研究对象,将120个样本以2∶1的比例分为校正集和预测集,80个校正集,40个预测集;使用900~1 700 nm的近红外光谱仪,获取样本径切面的近红外光谱数据;采用蒙特卡洛采样法剔除奇异样本,采用多元散射校正和S-G平滑对光谱数据进行预处理,消除光谱漂移、表面散射和噪声的影响;通过Bi PLS-SPA算法对特征波长进行提取,构建小波神经网络模型,预测柞木基本密度;将建模方法与常用的偏最小二乘(PLS)和BP神经网络进行了对比,验证小波网络的有效性。结果表明:小波神经网络对预测集样本验证结果更好,相关系数为0.968,预测均方根误差为0.014 4。With oak, 120 samples were divided into calibration set and prediction set, 80 calibration sets and 40 prediction sets in the ratio of 2 ： 1. The near-infrared spectrometer in 900-1 700 nm was used to obtain the near infrared spectroscopy da- ta. The singular samples were removed by Monte-Carlo sampling method. The spectral data were preprocessed by multivari- ate scatter correction and SG smoothing to eliminate the effects of spectral drift, surface scattering and noise. The charac- teristic wavelength was extracted by BiPLS-SPA algorithm. The wavelet neural network model was constructed to predict the basic density of oak wood. The modeling method was compared with the traditional partial least squares （PLS） and BP neural network to verify the effectiveness of the wavelet neural network. The wavelet neural network was more effective in predicting sample sets with a correlation coefficient of 0.968 and a prediction root mean square error of 0.014 4.

关 键 词：木材基本密度 近红外光谱 小波神经网络 

分 类 号：S781.31[农业科学—木材科学与技术] O433.4[农业科学—林学]