Study Of Near Infrared Spectroscopy Model For Predicting Mechanical Properties Of Larch Wood Based On Wavelet Transform

Compressive strength, modulus of elastic(MOE) and modulus of rupture(MOR) are the key performance indicators to evaluate the mechanical properties of wood. Rapid prediction of the mechanical properties makes significant contribution to ease the contradiction between supply and demand of wood, and to provide reasonable basis for the scientific cultivation and processing of plantation. In this paper, near infrared reflectance spectroscopy (NIRS) coupled with partial least squares regression(PLS) method is investigated to predict the mechanical properties of larch wood. Meanwhile, wavelet transform is introducted to denoise spectral signal. The main conclusion are as follows:(1) Compressive strength predictions for larch wood. Undertaking comprehensive analysis of the calibration and validation sets, we choose the number of best principal components when root mean square error(RMSE) reaches the minimum and the correlation coefficient(R) reaches its maximum value. The number of best principal components are both4with heartwood and sapwood models. Comparison of Sqtwolog, Rigrsure, Heursure and Minimaxi threshold, the result shows that Heursure gained the best effect of de-noising in the heartwood model, the standard error of calibration(SEC) and the root mean square error of calibration (RMSEC) were5.7921and5.7631, the R was0.7922; the standard error of prediction(SEP) and the root mean square error of prediction(RMSEP) were8.0835and8.0799, the R was0.6415. Rigrsure gained the best result in the sapwood model, the SEC and RMSEC were5.7832and5.7491, the R was0.7267; the SEP and RMSEP were6.5891and6.5362, the R was0.6594.(2) MOE predictions for larch wood. As an example of the larch heartwood sample, the best prediction accuracy is the cross-sectional model, the radial section was slightly worse, the string section was the worst. For the sapwood sample, db5-4achieved the best results in the de-noising, the SEC and RMSEC were0.5970and0.5935, the R was0.8562; the SEP and RMSEP were0.9720and0.9784, the R was0.7547.(3) MOR predictions for larch wood. We collected db5-7、bior5.5、dmey7、coif5-7and rbio5.5-7wavelet functions to pretreat spectral signals. According to results, dmey7obtained the best prediction accuracy in the heartwood model, the SEC and RMSEC were6.4152and6.3774, the R was0.7956, the SEP and RMSEP were5.8817and5.9811, the R was0.8408; db5-7acquired the best de-noising affections in the sapwood model, the SEC and RMSEC were5.7457and5.7147, the R was0.7627, the SEP and RMSEP were7.0626and7.0129, the R was0.6929.