Study On Prediction Methods Of Cotton Spider Mites In Xinjiang

Cotton is part of the most important economic crops in Xinjiang.In recent years,with the change of climate and the special tillage and management mode in Xinjiang,the ecological environment of cotton has become more conducive to the occurrence of spider mites.Because of these factors,the risk of disasters has increased.Due to the lack of systematic and effective forecasting methods,it is difficult to achieve pre-disaster prevention,and to control the post disaster.So that it became a major obstacle to cotton production.Based on this,this paper adopts remote sensing data,meteorological data and the occurrence data of historical mites to carry out the research on the prediction of damage gradually from three levels of remote sensing monitoring,regional scale forecasting,and annual forecasting of damage,and tried to be applied to actual production.Accordingly the contents of this thesis include three parts as follows:?1?Research on the remote sensing classification monitoring of damage,and provides the theoretical basis for the prediction of large area mites.We analyzed the cotton canopy hyperspectral data with different severity of damage,and looked for the optimal bands and established new spectral indices.The feature selection was performed using ReliefF algorithm and poisson correlation coefficient method,and classification algorithms of the K-Nearest Neighbor,decision tree,support vector machine,na?ve bayes,random forest,AdaBoost and artificial neural network are used to establish the model and comparative analysis.It is found that the ANN algorithm with feature selection is the optimal classification algorithm for classifying the severity of spider-mite damage in Xinjiang,and the classification accuracy is 76.62%and the F₁ is 76.68%.The classification accuracy of SVM is 75.32%and the F1 is 75.06%,and the SVM algorithm may be used as an alternate method.?2?The purpose of this study is tantamount to explore the optimal forecast method of the damage by spider mites infestation in the large-scale region of Xinjiang.In this paper,the combination of meteorological data and vegetation index,a total of 61 factors was selected as the primary factors of the predict research of cotton spider mites.After selecting features,single meteorological data,single remote sensing data and the data combination with remote sensing and meteorological data were selected for building Logistic and RVM algorithm models,and the prediction model of spider mites was generated and compared.The results demonstrate that the combination of meteorological and remote sensing data using the RVM algorithm has the best performance.The classification accuracy is 85.7%,and the F1 value is 85.7%.This method can be employed to predict cotton spider mites in certain regions.?3?A study was done on the forecast of the annual occurrence of cotton spider mites in Xinjiang.We employed the grey theory established a GM?1,1?catastrophe prediction model based on analysis of historical data from 2004 to 2013,and predicted that 2014 will be the disaster year,which has been consistent with the survey results.In addition,we established a GM?1,1?to forecast mite infestation degree based on analysis of historical data.To improve the prediction accuracy,we modified the grey model using the algorithm of the Markov chain and the BP neural network,finally we obtained the grey Markov model and the grey BP neural network model.By comparing the prediction accuracy of the three methods,it is found that the grey Markov model improves the prediction accuracy of the grey prediction from 84.31%to 94.76%,and the grey BP neural network model improves the prediction accuracy to 96.84%.Three models were used to estimate mite infestation degree in 2015 and 2016,and the grey BP neural network model was closer to the actual values,and thus exhibited the best performance.In this study,we take the cotton spider mites as the research object,based on multi-source data,combined with multiple analysis methods to study the prediction methods of damage from three aspects.The above research results can provide a theoretical basis and scientific guidance for advance prevention and comprehensive control of cotton spider mites in Xinjiang,and provide a reference for the prediction of similar diseases.