Prediction Method Of Birch Mechanical Strength Parallel To Grain Based On Near Infrared Spectroscopy

Compressive strength parallel to grain and tensile strength parallel to grain of wood is the important physical indicators,using traditional methods to detect,there is a long time,the disadvantage of the high cost,but also because of the anisotropy and inhomogeneity of lumber,the measuring accuracy of complex operation and low,so accurate,non-destructive to testing is especially important in practical engineering.Near infrared spectroscopy（NIR）analysis technology is a simple,convenient,fast and effective nondestructive testing technology.It can predict wood mechanical properties through indirect analysis and modeling of near infrared spectroscopy data,which can be used as an effective method to detect wood mechanical strength.In this paper,birch wood was taken as an example.According to the national standard,100wood mechanical specimens with compressive strength parallel to grain and grain and tensile strength parallel were prepared respectively.Spectral data were collected in the near-infrared spectral band of 900～1700nm and the true value of mechanical strength was measured.According to the collected spectral data,first of all,the samples were divided into 75 calibration sets and 25 prediction sets in a ratio of 3:1 by using Sample set partitioning based on joint x-y distances（SPXY）.Then,in order to eliminate the effects of scattered light,baseline drift and noise interference,the SG convolution smoothing with compressive strength spectral data and MSC-1^ST-SG combined method with tensile strength spectral data were preprocessed respectively.Secondly,based Model Population Analysis（MPA）framework development model algorithm to remove the data redundancy and information between variables,to adopt to the different proportion appearance characteristics of the data set number of wavelengths will reduce the accuracy prediction model,the introduction of Sampling Error Profile Analysis（SEPA）as Variable Iterative Space Shrinkage Approach（VISSA）improvement strategy,SEPA-VISSA method was used to optimize the characteristic wavelength of birch compressive strength spectrum;in order to improve the Monte Carlo Uninformative Variables Elimination（MC-UVE）,which is often unstable in wavelength selection due to the small difference between spectral data and random sampling,the integration strategy is adopted.And combined with Iteratively Variable Subset Optimization（IVSO）to optimize the IVSO in the case of multiple characteristic wavelengths caused by the complex calculation process,using CMC-UVE-IVSO hybrid algorithm to extract the tensile strength spectral characteristic wavelength.Finally,in view of the linear prediction model,due to the anisotropy and inhomogeneity of lumber,the spectral characteristics of the wavelength and the nonlinear relationship between the real value of mechanical strength,can reduce the accuracy of the model,so based on Particle Swarm Optimization（PSO）to optimize the kernel function parameters and establish the Relevance Vector Machine（RVM）nonlinear mechanical strength prediction model.The experimental results show that:The nonlinear prediction model of RVM based on the kernel function of PSO optimization,in which the prediction determination coefficient of SEPA-VISSA-RVM birch compressive strength prediction model is 0.9449,the root mean square error of prediction is 2.0432,and the relative analysis error is 4.2936.The prediction coefficient of CMC-UVE-IVSO-RVM birch tensile strength prediction model is 0.9684,the root mean square error of prediction is 11.7840,and the relative analysis error is 3.3198.These two models have higher prediction performance compared with the model built by Partial Least Squares（PLS）and Extreme Learning Machine（ELM）.It can realize more accurate and stable nondestructive testing of birch’s compressive and tensile strength.