Early Hyperspectral Identification Of Erannis Jacobsoni Djak Disaster

Early monitoring and early warning of forest pest disaster has always been the focus of forest ecosystem security protection.Forest pests have the characteristics of fast propagation speed,large occurrence area and long damage duration,which seriously affect the normal growth and development of the forest and even cause a large area of forest to die.Timely and reliable prediction of the early occurrence and development of forest pest disasters and effective prevention and control measures can minimize the damage of forest trees.Based on this,this study took ajirgen mountain forest area in binder,Kent province,Mongolia,which was affected by Erannis jacobsoni Djak pests,as the test area,used ground non-imaging hyperspectral data and ground survey data,and carried out early hyperspectral identification of pests by hyperspectral features and machine learning algorithms.Ground non-imaging hyperspectral features include smooth spectrum?SSR?,differential spectrum?DSR?and spectral continuous wavelet coefficient?CWC?,etc;Ground survey data includes canopy color and leaf loss rate data?sample indicator?,as well as the forest tree chlorophyll content?CHLC?,fresh weight moisture content?LWCF?,dry weight moisture content?LWCD?,photosynthetic rate?A100?and light intensity when the light intensity is100?mol m^-2 s^-1 At 200?mol m^-2 s^-1 corresponding photosynthetic rate?A200?and other early indicators of pests.First,the Pearson?Person?correlation analysis method is used to analyze the sensitivity between hyperspectral features and early indicator indicators of pests.Second,the indicator indicators are quickly extracted through the Findpeaks function?Fp?and the continuous projection algorithm combined mode?Fp-SPA?Sensitive spectral features,and finally through the random forest classification?RF?,support vector machine classification?SVC?and Fischer discriminant?FD?to build an early hyperspectral identification model of the Larix gmelinii based on a single indicator of insect pests,and The classification accuracy of the model was evaluated.This study made the following conclusions:?1?Sensitivity analysis of hyperspectral features to early indicators of pests.The early indicators of pests such as CHLC,LWCD,LWCF,A100 and A200 have different degrees of sensitivity to hyperspectral features such as SSR,DSR and CWC.The most significant sensitivity is CWC,followed by DSR,and SSR sensitivity is the worst.?2?Sensitive hyperspectral feature extraction of early indicators of pests.Using the Fp-SPA function model,the sensitive spectral characteristic bands were effectively extracted from the spectral features such as SSR,DSR and CWC.?3?The early hyperspectral recognition results of insect pests show that CWC has a good ability to recognize insects early,and the bior3.7-RF model based on CHLC is the best,with an overall accuracy of 0.73 and a Kappa coefficient of 0.64.In terms of model selection,the RF model identification effect is better than SVC and FD.It can be seen that in the early identification of the Erannis jacobsoni Djak insect pests,the most potential hyperspectral feature is CWC,the most perceptible indicator is CHLC,and the optimal early recognition model is RF.This not only has an important reference value for the early prediction and prevention of forest pests,but also provides a feasible way for the early remote sensing monitoring of forest pests.