Research On Classification Of Pinus Massoniana Pests Level Based On Hyperspectral Data

In order to improve the role of hyperspectral remote sensing in forest health monitoring,and in the application of pest infestation levels.With the aggravation of pine forest epidemic in China and hyperspectral remote sensing has a significant advantage,this paper mainly uses hyperspectral remote sensing to study and explore the classification of Pinus massoniana pest infestation levels.The hyperspectral data obtained by PSR-3500 ground object spectrometer and the hyperspectral image data obtained by HJ-1A domestic environmental small satellite are taken as the original data.Taking Masson pine with different levels of diseases and insect pests in Cuiping District of Yibin City as the research object,we study the inversion model of the degree of disease and insect damage.According to the correlation coefficient(r)value,combined with the root mean square error(RMSE),the inversion results are comprehensively analyzed.Using fuzzy classification and cluster analysis to classify the severity of the disease,then this paper establishes a spectral library of different levels of P.massoniana pests infestation.Finally,taking gray segmentation and forest health analysis as reference data,combined with blind source extraction algorithm(MSCPE_BSE)based on mean square prediction error,the grade judgments of various grades of insect pests Masson pine.The main contents of this paper are as follows:(1)First of all,the spectral reflectance obtained from the field survey was used for correlation analysis between the disease index and the original spectrum,and the correlation analysis between the disease index and the first order differential spectrum.The first order differential spectrum of diseases and Pinus massoniana pest infestation is more obvious than the original spectrum,in the"Green Peak"(550-600nm),"Red Edge"(711.5-724.9nm),and"Yellow Valley"(550-582nm)bands,the reflectance of different grades of pests and diseases is quite different.As the severity of pests and diseases increases,the peak value decreases and the valley value increases.Red light bands and near-infrared bands can reflect the severity of different pests and diseases of Masson pine.(2)Secondly,the sensitive parameters are extracted,and combined with principal component analysis and partial least squares regression analysis to verify and confirm the selected value.Through the correlation analysis of 26 selected red edge characteristic parameters,vegetation index and disease index,the results show that in addition to the red edge amplitude Dmin,red valley value Ro,mNDVI705,SDr/SDy,PSRI,VOG3 are positively correlated,the other spectral parameters are negatively correlated.Univariate linear model,quadratic function model,cubic function model,and multiple stepwise regression model(including 6-element linear,12-element linear,multiple quadratic)were used for regression analysis.The improved multiple regression model had the best effect,and the prediction set RMSE=0.038,R~2=0.99.Then the classification is based on the disease index of the inversion model.The membership function model in the fuzzy classification is the best,with an overall accuracy of 0.9673 and a Kappa coefficient of0.9123.In the cluster analysis,fuzzy C-means clustering is better than K-means clustering results,and the correlation coefficients of the results are all above 0.88.Finally,a four-level Pinus massoniana pest and disease level spectrum database was established based on the disease classification results.(3)Finally,the detection effect is compared with 8 classic target detection algorithms.In order to verify the feasibility of MSCPE_BSE for the detection of P.massoniana pests and diseases,this paper uses the spectrum of P.massoniana pests and diseases as the source signal.MSCPE_BSE detected the masson pine level targets and evaluated the ROC curve.When the false alarm rate(pfa)is consistent,the algorithm performs better than the other 8 target detection algorithms.