| Wood identification is of great significance for timber to circulate and apply in the market.The traditional wood identification methods are expensive and time-consuming,therefore,a rapid and accurate method to identify wood species is one of the important contents of wood science research.Near infrared(NIR)spectra of wood contains abundant physics,chemistry and anatomic construction.Back-propagation artificial neural network(BP-ANN)model has a strong anti-jamming,which can be effective to analysis NIR and identify wood species.This paper focused on the study about the optimization of models based on band selection and pretreatments,and comparison of models established by partial least squares discriminant analysis(PLS-DA)and soft independent modeling of class analogy(SIMCA),at the same time,the feasibility of rapid distinguish among wood species based on near infrared spectroscopy(NIRS)and BP-ANN was investigated.Finally,the paper discussed effects of aging 0h,5h,15 h,30h,60 h,90h,120 h,180h,and temperature treatment of 20?,30?,40?,50?,60?,70?,80?,100?on wood identification to determine the stability of wood recognition models.The purpose of this study is to utilize NIRS to identify wood species combined with BP-ANN.The results are as follows:(1)Based on raw NIR spectra combined with BP-ANN,the recognition accuracies of three kinds of plantations in three spectral segments including 780~2500nm,780~1100nm,and 1100~2500nm were 97.78%,95.56%,and 96.67%,respectively.Especially,the accuracy of the 780~2500nm band was up to 98.89% after the pretreatment by multiplicative scattering correction(MSC).Meanwhile,based on SIMCA method,recognition accuracies of the three spectral segments were 82.22%,81.11%,and 76.67%,respectively.Based on PLS-DA method,the recognition accuracies were 96.67%,95.56%,and 87.78%,respectively.Based on raw NIR spectra combined with BP-ANN,the recognition accuracies of six kinds of hardwood in three spectral segments were 97.62%,95.24%,and 100%,respectively.Especially,the accuracy of the 780~2500nm band was up to 100% after the pretreatment by standard normalized variate(SNV).Meanwhile,based on SIMCA method,recognition accuracies of the three spectral segments were 66.67%,50%,and 78.31%,respectively.Based on PLS-DA method,the recognition accuracies were all up to 100%.(2)Aging on wood surface had a significant effect on models of wood identification.Results showed that the recognition accuracies were higher when the aging degree of unknown samples were closed to the aging degree of modeling samples;When the models of 0 h(no aging),5 h,60 h,and 180 h for aging time were used to predict unknown samples(no aging samples and aging samples of different degrees),the recognition accuracies were 22.45%,75.77%,58.42%,and 54.34%,respectively.It showed that the recognition accuracy of model of no aging samples was the lowest,and the recognition accuracies were gradually reduced with the increase of aging degree of modeling samples.When the spectra of the samples with the same degree of aging added into the models could effectively improve the recognition accuracies.The more the spectra were added,the better the prediction effect was.All spectra collected from modeling samples of different aging degree were used to establish models for prediction of samples in unknown aging degree,and the recognition accuracy was 96.43%.(3)Different treatment temperature had a significant effect on models of wood identification.In the range of 100?,the recognition effect was well when the treatment temperature difference between predicting samples and modeling samples was not large.Especially,the recognition accuracy was up to 100% when the treatment temperature of modeling samples was consistent with the predicting samples.when the models of 20?,30?,40?,50?,60?,70?,80?,100? for treatment were used to predict unknown samples(all kinds of temperature treatment),the recognition accuracies were 69.17%,77.71%,88.96%,76.88%,64.58%,57.5%,55.42%,and 49.79%,respectively.The recognition accuracies were improved by adding spectra of which temperature was same as predicting samples.Results showed that the more these kinds of spectra were added,the predicting effects were better.All spectra collected from modeling samples of different temperature were used to establish models for prediction of unknown samples,and the recognition accuracy was up to 100%. |
PDF Full Text Request