Prediction Of Panicle Blast Disease In Rice By Multi Model Comprehensive Analysis

Rice blast is one of the most important diseases of rice,which can be divided into seedling plague,leaf blast and ear neck blast according to the onset and site of the disease,among which the spike neck blast has the greatest impact on the yield.Accurate prediction of the extent of spike neck blast and corresponding preventive measures can improve the prediction level of disease prevention and control,and contribute to the sustainable development of agriculture.At present,the prediction models of crop diseases are mainly constructed by machine learning neural network,support vector machine and decision tree.However,these models are essentially black box models,and the ideal prediction results are too dependent on reasonable parameter input and accurate training set samples,which are difficult to be guaranteed in the complex actual situation.In view of the above situation,this dissertation takes Yingshang County of Anhui Province as the research area,extracts meteorological factors closely related to neck blast of mid season rice.Through clustering,we extract the internal information of the sample,and adopt the means of reducing the noise of the data set and mining the best partition parameters.In order to improve the accuracy of the prediction model of Rice neck blast,the following studies are carried out:(1)The related literatures were analyzed,and the relationship between the incidence of neck blast and meteorological factors was summarized,The collection of meteorological factors and relevant data were collected.Referring to the rice disease statistics,growth cycle,climate and weather data and literature in Yingshang County.We conclude the meteorological factors closely related to the occurrence and epidemic of rice neck blast.Collected the meteorological data sets for the study area from 1998 to 2015.The data are mainly derived from the "The Monitoring and warning platform for crop pests and diseases in Anhui Province",related literature and the weather data is crawled by using reptilian technology.(2)Clustering algorithm is used to mine the data clustering cluster basis,noise points and other intrinsic characteristics.Selecting K-Means combined with contour coefficient to establish cluster number in dataset.Set R and minPts in DBSCAN to identify the noise points in the dataset.These two conclusions serve as the basis for further analysis of forecasting models using SVM.(3)Modeling of each prediction model.Using DBSCAN to generate SVM prediction models for noise reduction after data sets and direct use of SVM respectively.The accuracy of the prediction model is 62.5% and 76.8% respectively after cross validation and historical data retrieval.Verified that the DBSCAN algorithm with the identification of noise points can be used as a preprocessing step of SVM to improve the robustness of disease prediction models.The prediction model of panicle neck blast in suitable area was obtained,and the suggestion to improve the robustness of prediction model was put forward.(4)A prototype system for predicting the severity of Panicle Blast Disease in mid season rice has been designed and developed,The system is intelligent,efficient,and has friendly human-machine interface.It plays a positive role in promoting scientific decisionmaking of disease prevention and control.It will play a positive role in the scientific decision-making of disease prevention and control.