Abstract:

Objective: Using near-infrared spectrum(NIR) analysis technology, a wood elastic modulus prediction model based on support vector machine optimized by artificial bee colony algorithm(ABC-SVM) was proposed, which was expected to provide a scientific reference for non-destructive prediction of wood elastic modulus. Method: Taking 294 samples of Cunninghamia lanceolata as experimental materials, the elastic modulus of the samples was measured by conventional mechanical method, and the near-infrared diffuse reflection spectra of the samples were collected. The 350-2 500 nm spectrum was selected and the original spectrum was preprocessed by first-order derivation and 15-step exponential smoothing. Using principal component analysis to reduce the data, then the partial least squares regression(PLS) model, support vector machine regression(SVR) model and the ABC-SVM model were established to predict the elastic modulus of Cunninghamia lanceolata. The determination coefficient(R²), the root mean square error(RMSE), the mean absolute percentage error(MAPE) and the mean absolute error(MAE) were used to evaluate the advantages and disadvantages of the three models. Result: The R² of PLS model is 0.726 700, the RMSE is 6.744 9, the MAPE is 0.063 5, and the MAE is 5.065 6. As for SVR model, the R² is 0.935 305, the RMSE is 3.528 1, the MAPE is 0.023 7, the MAE is 1.840 9. Using artificial bee colony algorithm to optimize SVM parameters, the optimal parameters are c=5.670 51, g=0.031 25. The R² of ABC-SVM model is 0.935 371, the RMSE is 3.526 0, the MAPE is 0.023 7 and the MAE is 1.840 0. All three models could effectively predict the elastic modulus of Cunninghamia lanceolata. Conclusion: 1) Observing the index of determination coefficient(R²), both SVR and ABC-SVM models have better prediction performances than PLS model, and ABC-SVM has the best prediction performance. 2)According to the obtained error results(RMSE, MAPE and MAE), the MAPE of the three models is within the acceptable range, and the ABC-SVM model has the smallest indicators for the error, indicating that the model is efficient and scientific in predicting elastic modulus of Cunninghamia lanceolata.

Key words: near-infrared spectrum(NIR), elastic modulus, support vector machine(SVM), artificial bee colony(ABC) algorithm