Rapid Identification Of Tree Species By Near Infrared Spectroscopy

Woodwork have become an important part in human daily life.In addition to teas,more and more leaf products also enter into the lives of mankind.The identification of tree species is the most basic part in these industries.Near infrared spectroscopy(NIR)technology can identify tree species nondestructively,quickly and accurately.It plays an important role in tree identification and wood identification.Meanwhile,it could help the decision making in forest management.(1)Wood samples from forty valuable tree species were identified with NIRS.With 1st derivative pretreated spectrum and distance distinguish method,the recognition accuracy is 98.75%.Using this model for the prediction of unknown wood samples,the recognition accuracy is 82.21%.After pretreatment with Norris derivative filter,the recognition accuracy is increased to 95.31%.(2)Fourteen typical tree species in Northeast China were identified with NIRS.The samples were drilled with a increment borer at 1.3 meters breast-height of trees from south to north.Different wavelength and different pretreatment methods were compared on the effect of model prediction with distance distinguish method.Results showed that working with Norris derivative filter pretreatment combined with the distance distinguish method,a high recognition accurate rate of 98.21%could be reached.(3)Leaves from 9 different tree species were identified with NIRS.Two identification methods of distance method and PLS-DA were compared on the effect of model prediction.Results indicated that pretreated with the derivatives and smoothing of the spectrum,the recognition accuracy is 99.16%.When using the PLS-DA with single column identification variable matrix to identify the 1st derivative and smoothing spectrum of leaves from 4 different tree species,the recognition accuracy can reach 100%with correlation coefficient of 0.9936 and RMSEC of 0.120 and RMSEP of 0.144.When using the PLS-DA with multiple column identification variable matrix to identify the 1st derivative and smoothing spectrum of leaves from 9 different tree species,recognition accuracy can reach 100%with correlation coefficients from 0.8886 to 0.9569,RMSEC of 0.0845 to 0.15,and RMSEP of 0.0887 to 0.155.(4)Additionally,this paper studied the effects of reproducibility on the NIRS recognition model.There will be differences in the NIRS under different conditions,the longer the interval is,the more obvious the difference is,the lower the accuracy is.The S-G smoothing and Norris derivative filter optimal parameters selection method were also studied,with the change of the parameter,model recognition accuracy changes regularity.Significant differences in spectrum(1637nm and 1687nm)between softwood and hardwood were found in this paper.At 1637nm of NIRS,4 softwood samples have absorbance,and 10 hardwood samples haven’ t.At 1687nm of NIRS,10 hardwood samples have obvious absorbance,and 4 softwood samples haven’ t.This paper provides a new methods and ideas in rapid identification of common tree species and NIRS identification of softwood and hardwood.