A Method For Predicting The Mechanical Properties Of Quercus Mongolica Based On Two-dimensional Correlation Spectroscopy And Convolutional Neural Network

Quercus mongolica is the main secondary forest tree species in northeastern China.It is hard and rot-resistant and is often used as structural wood.The flexural modulus of elasticity is one of the main mechanical properties of wood,which represents the comprehensive situation of wood mechanical properties,and is an important parameter and basis for the realization of wood application and classification.The traditional wood mechanics detection method is a destructive experiment method,which is easy to cause waste and complicated to operate,and cannot meet the actual engineering needs.The common NIR non-destructive testing method is a reliable and pollution-free analysis method,which can quickly and accurately determine the mechanical properties of wood,but the detected signal in the near-infrared spectrum is weak,and a large amount of redundant information makes the spectrogram.Unclear,etc.In response to this problem,this paper takes Quercus mongolica as the research object,and uses twodimensional correlation spectrum technology to establish a quantitative analysis model to predict the flexural elastic modulus of Quercus mongolica.First,according to the national standard GB1927～1943-2009,115 300mm×20mm×20mm Mongolian oak bending mechanical specimens were made,of which 86 were used as calibration set samples and 29 were used as prediction set samples.Infrared spectrometer,collecting data from radial and chord sections;Secondly,MSC-SG-FD preprocesses the collected original spectra to solve the problems of scattered light,baseline drift and highfrequency noise.2D synchronous correlation spectrum and 2D asynchronous correlation spectrum;Third,the linear PLS model is used to model and predict the original near-infrared spectrum,the two-dimensional synchronous correlation spectrum and the two-dimensional asynchronous correlation spectrum,respectively.Compared with the near-infrared spectrum,the two-dimensional correlation spectrum has higher spectral resolution and can improve the The interpretation ability of the spectrum,compared with the two-dimensional asynchronous correlation spectrum,the two-dimensional synchronous correlation spectrum has an autocorrelation peak,which can provide more spectral features,which is more conducive to the prediction of the flexural elastic modulus of Quercus mongolica;Then,in view of the complex nonlinear relationship between the spectrum and the true value of the flexural elastic modulus of Quercus mongolica,the RBF and F-RBF neural networks are used to establish the nonlinear model;The processed two-dimensional synchronous correlation spectrum is modeled to realize the prediction of the flexural elastic modulus of Quercus mongolica.The experimental results show that the modeling effect of 2D synchronous correlation spectrum is better than that of near-infrared spectroscopy and 2D correlation asynchronous spectrum.Taking the original 2D synchronous correlation matrix as the input,the coefficient of determination of the 2DCOSNET model calibration set is 0.9766,both of which are The root square error is 0.2878,the coefficient of determination of the prediction set of the 2DCOSNET model is 0.9502,and the root mean square error is 0.3809,which are better than the other three convolutional neural network prediction models.The 2D synchronous correlation matrix modeling preprocessed by MSC-SG-FD shows the best modeling effect of 2D synchronous correlation spectrum after MSC-SG-FD preprocessing.The coefficient of determination of the2 DCOSNET model prediction set is 0.9876,and the root mean square error is 0.2270;Taking the two-dimensional synchronization correlation matrix preprocessed by MSC-SG-FD as input,the coefficient of determination of the prediction set of the PLS model is 0.8582,the root mean square error is 0.7384,and the coefficient of determination of the prediction set of the RBF neural network model is 0.9016,and the root mean square error is 0.9016.is 0.6038,the coefficient of determination of the F-RBF model is 0.9289,and the root mean square error is0.4757;the prediction effect of the 2DCOSNET model is better than that of the PLS and FRBF models.Therefore,the convolutional neural network can use the two-dimensional synchronous correlation spectrum preprocessed by MSC-SG-FD to more accurately predict the flexural elastic modulus of Quercus mongolica.