Research On Rice Disease Image Recognition Method Based On SAE Feature Fusio

Rice is the second largest grain crop in China,with an output of more than 200 million tons in 2019.Affected by weather,pesticides and other factors,the occurrence of rice diseases is increasing day by day.The traditional manual identification methods can not adapt to the occurrence of a large number of diseases,and the results of manual naked eye identification are uncertain,unable to take treatment measures in time,and even have counterproductive results.The field investigation in Heilongjiang reclamation area found that the damage coefficient of rice by rice blast,sheath blight and white leaf blight was grade I,and bacterial leaf streak and flax leaf spot were grade II.They were the five most representative diseases with great harm.Therefore,this paper mainly studies the identification of these five kinds of diseases.The continuous maturity of machine learning and artificial intelligence technology,as well as the rapidity of using smart phones to obtain high-resolution images,have laid a theoretical and application foundation for this research.The main research contents are as follows.(1)Sparse auto encoder(SAE)is used to extract the features of rice disease images.This paper studies the network structure and algorithm principle of SAE,analyzes the feature extraction process of SAE and classical automatic encoder with sparsity constraint,reduces the feature dimension,and then proposes a model structure of multi feature fusion using sparse automatic coding network to improve the classification speed and accuracy of classifier.(2)Research on rice disease identification model based on SAE.The dimension of the collected feature data is reduced by using the automatic encoder model with sparsity constraint,and its parameters are fine tuned.The accuracy of the model is compared by using 10 fold cross validation method and confusion matrix.In the process of network training,the optimal super parameters are obtained through many experiments to improve the performance of the model.In the SAE network model structure,the support vector machine(SVM)optimized by switch particle swarm optimization(SPSO)is used as the classifier.By selecting different penalty parameter C and kernel function parameter g,the recognition results are compared and analyzed to determine the optimal parameters of SVM model.Compared with the traditional machine learning recognition method,the convergence speed is faster,and the average recognition rate is 95.3%.(3)Design and implement rice disease identification system.The system includes two modules: feature extraction and classification and recognition.The system integrates SAE algorithm to realize multi feature fusion and extraction,and applies the optimized SVM model for recognition,which can realize the intelligent recognition of five kinds of common rice disease images.