Research On Rice Pest Prediction Based On Ensemble Learning

As a major food crop in China,rice provides the material basis for people’s daily life.However,rice is highly susceptible to various pests and diseases during its growth,among which insect pests are one of the important factors leading to yield reduction and quality deterioration.Pest prediction can help agricultural producers detect the trend and extent of pest occurrence in a timely manner,so that they can take appropriate agricultural measures to safeguard the yield and quality of rice.In this thesis,we conducted a study on rice insect pest prediction,starting with the infestation of rice fly and longitudinal leaf borer in some areas of Anqing City,Anhui Province,China,and the specific studies are as follows:(1)Analysis and treatment of pest influencing factors.Comprehensive literature analysis was conducted to determine the relevant factors affecting rice pests,and meteorological factors were selected as characteristics and their correlation degree was determined by gray correlation analysis.For the regional characteristics,we combined the national standard and Anhui local standard to classify the degree of pest occurrence index.In-depth analysis of the data revealed that there was a data imbalance problem,and the Adaptive Synthetic Sampling(ADASYN)algorithm at the data level was used to balance the training set while retaining the original distribution of the test set,in order to improve the focus on the prediction of pest occurrence that is biased toward the significant degree of occurrence and to improve the prediction accuracy.(2)Construction of Ada Boost-SSA-SVM pest prediction model.The Sparrow Search Algorithm(SSA)was used to optimize the hyperparameters of the SVM model,and the optimized SVM model was used as a weak learner and combined with the Ada Boost algorithm for integrated learning to construct the Ada Boost-SSA-SVM pest prediction model.In the prediction of the occurrence degree of both rice fly and rice longitudinal leaf borer,the accuracy rate exceeds more than 90% and the Macro-R exceeds 87%,which has good applicability.In comparison with models such as random forest,BP neural network and SSA-SVM,the accuracy was improved by up to 6% compared with other models,the Marco-R was improved by 24%,and Marco-F1 was improved by 14%.The data performance is more balanced,the situation is more satisfactory,and the generalization ability is stronger.(3)Development of a rice pest prediction system.In order to facilitate users’ use of the model,the Ada Boost-SSA-SVM pest prediction model is combined with Vue,Spring Boot,My SQL and other technologies to develop a rice pest prediction system,which provides functional modules for user pest data management,meteorological data management,pest prediction,etc.Finally,the prediction results are visualized and control suggestions are provided.