智慧林业背景下翅荚木速生材综纤维素含量机器学习预测

Machine Learning Prediction of Holocellulose Content in Fast-growing Wood of Zenia insignis Chun in the Background of Smart Forestry

林产工业自动化和智慧林业的发展对无损检测手段提出了新的要求,本研究以应用潜力巨大的速生材翅荚木为原材料提出了一种综纤维素快速检测方法。该方法利用傅里叶红外光谱技术,并采用不同的光谱数据预处理方法和机器学习模型来测定翅荚木的综纤维素含量。本研究具体采用了傅里叶红外光谱作为试验数据基础,并测量了翅荚木样本的综纤维素含量及光谱数据。经过MCCV剔除异常值后,选择四种光谱预处理技术(FD/SD/SNV/MSC)对光谱数据进行预处理,并采用PCA算法对其进行特征优化。接着采用BPNN和SVR两种机器学习模型建立起综纤维素含量预测模型,并对不同的处理方法和建模方式进行评估。结果表明:采用SNV和MSC预处理的光谱数据具有较高的综纤维素含量预测能力。此外,采用MSC-PCA-SVR预测模型的预测效果最佳,决定系数为0.97,均方根误差为0.004 596,能够很好地实现对翅荚木综纤维素含量的无损检测,总体而言,基于SVR建立的翅荚木综纤维素含量预测模型预测效果最佳。

The development of automation in forest industry and intelligent forestry has put forward new requirements for non-destructive testing methods,and in this study,a rapid testing method for holocellulose using fast-growing wood of Zenia insignis Chun was proposed,which has great potential for application,as the raw material.Fourier transform infrared spectroscopy and different spectral data preprocessing methods and machine learning models were used to determine the holocellulose content of Zenia insignis Chun in this method.In this study,Fourier transform infrared spectroscopy was specifically used as the basis for the experimental data,and the holocellulose content and spectral data of the Zenia insignis Chun samples were measured.After the elimination of outliers by MCCV,four spectral preprocessing techniques(FD/SD/SNV/MSC)were selected to preprocess the spectral data and the features were optimized using the PCA algorithm.Then two machine learning models,BPNN and SVR,were used to build a prediction model for holocellulose content,and the different processing methods and modeling approaches were evaluated.The results showed that the spectral data preprocessed with SNV and MSC had a high predictive ability for holocellulose content.In addition,the best prediction results were obtained using the MSC-PCA-SVR prediction model,with a coefficient of determination of 0.97 and a root mean square error of 0.004596,which can well realize the demand for nondestructive testing of the holocellulose content of Zenia insignis Chun,and the prediction effect of the prediction model of the holocellulose content of Zenia insignis Chun,which was established on the basis of SVR,was the best prediction result.