Study On Detection Of Basic Density Of Wood Based On Near Infrared Spectroscopy

The basic density of wood is an important index to characterize the physical properties of wood,not only can evaluate the actual weight of wood,but also related to the physical and mechanical properties of wood shrinkage,hardness,strength and so on.The conventional density detection method is tedious,time-consuming,and has a certain damage to the wood,which affects the subsequent processing of wood.So It is very important to detect the basic density of wood.Near infrared spectroscopy is a fast and efficient indirect analysis technology,compared to other wood detection technologies,it has the advantages of simple operation,fast detection,non-destructive and easy to realize online detection,and it provides an effective approach for nondestructive testing and information acquisition of wood.In this paper,Xylosma racemosum is taken as the research object.According to the national standard,120 samples are made and their basic density are determined.The samples are divided into calibration set and prediction set with the ratio of 2:1,80 for calibration set and 40 for prediction set.Due to the growth characteristics of wood caused by different facets of the differences,considering the nature of each section and the actual production conditions,this paper obtains the near infrared spectral data of the radial section of samples by near infrared spectrometer ranging from 900 nm to 1700 nm.And it analysis the relationship between spectral data and basic density of the samples,builds the NIR fast prediction model of Xylosma racemosum basic density.Firstly,Monte Carlo sampling method is utilized to eliminate the singular sample.Comparing to leverage value method and half resampling method,Monte Carlo sampling method is the singular value recognition method based on statistics,and considers the influence of each sample on the whole data.Secondly,the multi spectral correction +SG smoothing is used to preprocess the spectral data to eliminate the influence of spectral drift,surface scattering and noise.Then,the BiPLS-SPA algorithm is applied to extract the characteristic wavelength,which can reduce the influence of the uncorrelated variables on the prediction effect,and reduce the calculation amount of the minimum redundant wavelength of SPA in the whole band.Finally,it uses the feature spectra to build the wavelet neural network model to predict the wood basic density,wavelet neural network has strong learning ability,and has the advantages of simple structure and fast convergence speed.Six singular samples are eliminated by Monte Carlo sampling method,their serial numbers are 6,23,8,37,52,64 and 72.Compared with the full sample modeling,correlation coefficient of prediction model is increased from 0.846 to 0.904.The experiment uses derivative spectra,smoothing algorithm,standard normal transformation and multiple scattering correction to preprocess the spectral data,since the fewer number of wavelength in this paper,derivative will produce a certain error,the application of multiple scatter correction combined with SG smoothing is applied to eliminate the influence on scattering effect and noise.The BiPLS combined with SPA algorithm is utilized to extract the feature wavelength.BiPLS divides the spectrum into 5~15 sub interval,compared with the results of different sub interval,when spectrum divided into 10 sub interval,the RMSECV is minimum,and the optimal interval combination is [3 5 6 7 9],the number of variables is decreased from 117 to 59.The SPA algorithm is used for further feature extraction,and the number of variables is decreased from 59 to 6,those are the most relevant feature wavelength with the basic density.The wavelet neural network prediction samples show better results than the commonly used linear modeling method of partial least squares(PLS)and BP neural network.The correlation coefficient is 0.968,the root mean square error of prediction is 0.0144.The results show the approach proposed by this paper is more suitable for building the relationship between wood basic density and near infrared spectrum.