Hyperspectral Estimation Model For Physiological Parameters Of Pine Tree Under Bursaphelenchus Xylophilus Stress

Bursaphelenchus xylophilus is a kind of devastating diseases on pine trees,which creates serious crisis on the conifer forest resources and ecological environment.The infection of Bursaphelenchus xylophilus could significantly influence the physiological and biochemical characteristics of host plants.In this paper,we obtained the hyperspectral data of the healthy and infected black pine trees and masson pine trees and measured the corresponding physiological and biochemical parameters by real-time sampling and experimental analysis.Then,the relationships between hyperspectral data and physiological parameters were investigated by a variety of methodsto establish various simulation models based on hyperspectral specific characteristics..The simulation accuracy of these models were analyzed and compared each other,and the optimal estimation models were selected.The models could be used to estimate the physiological parameters of the pine trees quickly and accurately.So,the better methods may provide references to improve the accuracy of monitoring models on Bursaphelenchus xylophilus diseases.Furthermore,it can also supply a good basis for setting up the extensive,rapid and accurate early warning system on forest pests and diseases..The results were as follows:1.The accuracy of chlorophyll and carotenoid content estimation model of infected masson pine trees and black pine trees based on red edge parameters were higher than the model based on vegetation index,and the red edge parameters were more sensitive to the pigment content information.The carotenoid content of infected masson pine trees and black pine trees were sensitive to the same spectral indices mND705 and d?min,the accuracy of the models were not high.The estimation of the black pine trees were better than masson pine trees.The overall accuracy of water content estimation model of infected masson pine trees and black pine trees based on moisture indices were not high,and the accuary of black pine trees model were slightly higher.2.The chlorophyll and carotenoid content of infected masson pine trees and black pine trees were estimated based on the BP neural network,the accuracy and prediction of the BP neural network model based on red edge parameters were better compared to the principal component by PCA.In the same way,the BP neural network models based on moisture indices were better than PCA-BP neural network models.The BP neural network achieved satisfactory results,and it is a preferred method to simulate the dynamic changes of vegetation physiological parameters.3.The accuracy of the stimation models based on PSO(Particle Swarm Optimization)-SVR(Support Vector Regression)were high,it was used to simulate the physiological parameters of infected pine trees feasible.The accuracy of chlorophyll,carotenoid and water content estimation model of infected masson pine trees and black pine trees based on RF regression algorithm were not higher.As a result,using RF regression model to simulate the dynamic change of vegetation physiological parameter could not achieve the desired effect.4.It's clear that the optimal decomposition level of the infected masson pine trees and black pine trees' spectral curve is 7 layers.The optimal decomposition number of the infected masson pine trees and black pine trees spectral curve was 7.According to the wavelet decomposition coefficient,the wavelet energy coefficient were calculated and the stepwise regression estimation model of physiological parameters were established.More model parameters were selected based on wavelet coefficients,the satisfactory simulate accuracy also been acquired.It is a feasible method to simulate the vegetation pigment content.The simulation accuracy of the water content estimation model based on the wavelet analysis to infected masson pine trees and black pine trees were relatively low.The optimal wavelet basis functions were needed to improve the simulate accuracy,the wavelet decomposition characteristics could be further applied to simulate the dynamic changes of vegetation physiological parameters accurately.5.Through comparative analysis variety estimation models of vegetation physiological parameters of infected masson pine trees and black pine trees,the results show that the BP neural network could improve the accuracy of the estimation model significantly.BP neural network model based on red edge parameter and moisture index has the highest precision,and BP neural network model based on principal component analysis were the second.Compared to the stepwise regression model,the accuracy of the PSO-SVR model based on the red edge parameter and the water index has increased,RF regression algorithm were used to simulate the physiological parameters of infected masson pine trees and black pine trees which achieved unsatisfactory effect.