Research On Wood Physical And Mechanical Properties Estimation And Model Optimization Based On NIR

In view of the contradiction between the imbalance of supply and demand of wood in China,it is necessary to adopt a suitable method to quickly estimate the basic properties of wood,improve the efficiency of wood utilization,and provide theoretical basis and technical support for the rational cultivation and processing of plantations.The compressive strength,flexural modulus and flexural strength of wood are the main performance parameters for evaluating the mechanical properties of wood.The basic density of wood is an important parameter for the quality assessment of wood.It has important applications in the study of basic properties of wood,forest cultivation,and forestry breeding.Therefore,how to quickly and accurately obtain the timber properties of the northeast forest area is crucial.This study is based on near-infrared spectroscopy combined with partial least squares and BP neural network algorithm modeling methods,from the near-infrared spectrum acquisition method of wood samples,the pretreatment method of near-infrared spectral data,the band method of spectral data,etc.The wood basic density,flexural strength,flexural modulus,and compressive strength estimation model are optimized to select the best modeling method to obtain the optimal wood material estimation model.(1)Using near-infrared spectroscopy combined with artificial neural network algorithm modeling method,the optimization model of three mechanical properties of larch bending strength,grain bending elastic modulus and grain bending strength is studied.In this experiment,near-infrared spectra were collected from three sections of cross-section,chord section and radial section of deciduous wood standard samples,and the optimal model for predicting the mechanical properties of larch wood was determined.The results show that the mechanical properties of the larch constructed by the spectral data collected by the cross-section are the best,and the model of the mechanical properties of the larch is slightly less effective.The mechanical properties of the larch in the chord The estimated model is the worst.(2)Taking the eucalyptus sample basic density true value and near-infrared spectroscopy data as input,respectively,through convolution smoothing,first-order derivative and second-order derivative preprocessing method are used to preprocess near-infrared spectroscopy data.A near-infrared estimation model based on partial least squares(PLS)for the basic density of eucalyptus wood was established.The results show that the basic density model of Eucalyptus wood pretreated by first derivative is optimal in the range of 350?2500nm.After denoising optimization of near-infrared spectral data and constructing the basic density model of Eucalyptus wood,the basic density model of Eucalyptus wood pretreated by first derivative is optimal in the range of 500?2300nm,and the correlation coefficient of calibration set is 0.9871.The corrected root mean square error is 0.0016,the correlation coefficient of the verification set is 0.9486,and the predicted root mean square error is 0.0021.(3)Taking the Korean pine,larch and spruce species as the research object in the northeast forest area,using competitive adaptive weighting method(CARS),no information variable elimination method(UVE)and interval partial least squares(iPLS)The near-infrared spectral band is optimized.The convolution smoothing algorithm is used to preprocess the near-infrared spectral data.The partial density estimation model of coniferous wood is established by partial least squares(PLS)method.The optimal band optimization method is determined by contrast analysis.Excellent coniferous wood basic density model The research shows that the screening of near-infrared spectral bands by the band optimization method of CARS,UVE and iPLS can play a role in optimizing the basic density model of coniferous wood.The least-squares method of interval partial least squares(iPLS-PLS)is the best for the band-precision basic density model of coniferous wood.The model correlation coefficient is 0.9380,the corrected root mean square error is 0.0218,and the correlation coefficient is verified.0.8959,verify the root mean square error is 0.0280.