NIR-based Prediction Model Of Larch Wood Density And Its Influencing Factors

Near infrared(NIR)technology,as one of the rapidly developing modern analytical techniques,can achieve effective prediction of the basic density of wood in the production and processing of wood.In the actual industrial process,the NIR technique is susceptible to the interference of several factors,which cause the variation of NIR spectra and thus affect the prediction accuracy of NIR prediction models.Among them,changes in wood surface roughness,temperature and moisture content have a greater impact on the accuracy of the NIR analysis technique.Therefore,this paper takes larch as the research object,analyzes the larch wood spectra under the influence of different factors,studies the influence of different surface roughness,different moisture content and different temperature conditions on the NIR prediction model of the basic density of larch wood,and establishes the relevant calibration model to provide the theoretical basis for improving the accuracy and universality of the NIR prediction wood density model.The main research contents are as follows:(1)Effect of different surface roughness on near infrared prediction model of basic density of Larch wood.Three kinds of surface roughness were set to simulate the surface roughness differences caused by different machining accuracy,namely,unpolished(M0),150-mesh(M1)and 320-mesh(M2).Artificial selection,backward interval partial least squares(BiPLS)and synergy interval partial least squares(SiPLS)were used to complete the optimization of band.Partial least squares(PLS)method was used to construct a single prediction model of different surface roughness and a near-infrared global model containing three kinds of surface roughness samples.The results showed that the SiPLS method performed the best among the band selection methods with correlation coefficients of 0.865 9,0.766 0,and 0.725 6 for the validation set,respectively.The prediction ability of the global model built by mixing three different surface roughness is better than that of the single model,and the correlation coefficients of the verification set are 0.843 9,0.876 9 and 0.754 2,respectively.The results show that SiPLS method can extract effective information from near infrared spectrum of larch and reduce the difficulty of modeling.The influence of surface roughness variation can be reduced by constructing near-infrared model that includes all surface roughness.(2)Effect of different water content on near infrared prediction model of basic density of Larch wood.Six moisture content conditions were set: absolute dry(H0),10%(H1),30%(H2),50%(H3),70%(H4)and saturated(H5).Based on the improved BP neural network(BPNN),a single model and a global model for near infrared prediction of basic density of larch wood were established,and cross-verified.The results show that the prediction model established by multi-universe algorithm optimized BPNN(MVO-BPNN)is better than other modeling methods.It is also better to build a global model of water content than a single model.The results show that MVO-BPNN method can be applied to predict the basic density of larch wood under different water content conditions.A single model has poor performance in predicting different water content samples,and only performs well when the water content is consistent with the water content used in modeling.(3)Effect of different temperature on the spectrum of near infrared prediction model of basic density of Larch wood.Set five temperature conditions:-30 ℃,-20 ℃,0 ℃,20 ℃and 30 ℃.The pretreatment effect of generalized least square weighting(GLSW)was analyzed,and the model was established by PLS method.Using 20 ℃ as the reference,piecewise direct standardization(PDS)and double window piecewise direct standardization(DWPDS)methods were used to calibrate and transfer different temperatures,and compared with the global temperature model.The results showed that temperature had an obvious linear effect on near infrared spectrum of larch in the range of 1 800 nm ～ 2 000 nm.The pretreatment effect of GLSW was better than other methods,the correlation coefficient of correction set was 0.964 5,and the correlation coefficient of verification set was 0.801 4.After DWPDS correction,the correlation coefficients of the verification set are all over 0.7.The results show that DWPDS calibration transfer method can effectively reduce the influence of temperature on near infrared spectrum and improve the accuracy of prediction model.