Research On Crop Leaf Disease Recognition Based On Convolutional Neural Network And Ensemble Learning

Crop diseases are an important factor leading to poor growth and development of crops.Therefore,it is very necessary to accurately identify the types of crop diseases and take targeted measures.In traditional disease detection and identification,farmers mainly rely on human observation and are accustomed to diagnosis based on experience.However,this method takes a long time,is subjective,and has a high rate of misjudgment.Therefore,the use of intelligent means to identify crop diseases is of positive significance.Therefore,this paper mainly uses convolutional neural network for migration learning to classify crop leaf diseases,and introduces integrated learning strategies to improve the accuracy and recognition accuracy of crop diseases.The main research contents and conclusions of this paper are as follows:1.The research objects of this experiment mainly choose 11 types of disease data sets of 4 crops,with a total of 20,637 pictures.First,by using the six network models of Le Net,Alex Net,Res Net-18,Res Net-50,VGG16 and Xception under the SGD optimizer(stochastic gradient descent optimizer),the effect of identifying the leaf disease data set of 4 crops and 11 crops is carried out.Comparison and analysis show that the four network models Res Net-18,Res Net-50,VGG16 and Xception have higher accuracy in identifying crop leaf diseases.2.In the above experiments,it was found that the accuracy rate,recall rate and F1 value of the six network models were low.By studying the influence of the hyperparameter batchsize on the model performance,compare the four network models at batchsize=8 and batchsize=16.Accuracy,recall and F1 value,it can be concluded that when batchsize is set to 16,the accuracy of the model has been slightly improved,and the accuracy,recall and F1 of the model have been greatly improved by 15%-29%.Makes the overall performance of the network model better,so using a larger batchsize can effectively improve the overall performance of the model.3.Explore the impact of different optimizers on crop leaf disease identification.By doing four network models of Res Net-18,Res Net-50,VGG16 and Xception,the crop leaf disease identification is performed under the SGD optimizer(stochastic gradient descent optimizer),RMSprop optimizer,and Adam optimizer(adaptive time estimation method).By comparing and analyzing the experimental results,it can be concluded that when the Adam optimization algorithm is used for model optimization,its advantages are the most obvious,achieving the highest classification accuracy,as well as higher accuracy,recall and F1 values.4.Through the above experiments,it can be concluded that in the identification and classification of crop leaf diseases,the Xception network model is used in the case that the optimizer is Adam and batchsize is set to 16,and the accuracy of leaf disease recognition in the test set reaches 99.74%.,The accuracy rate reached 80.05%,the recall rate reached 80.70%,and the F1 value reached 80.25%.5.Introduce an integrated learning strategy to optimize the identification and classification of crop leaf diseases.By integrating the four network models of Res Net-18,Res Net-50,VGG16 and Xception in different combinations,the identification and classification experiments of 11 leaf diseases of 4 crops were carried out.The experimental results prove that the overall performance of the classifier can be improved by the ensemble learning method.And increasing the number of sub-classifiers or enhancing the performance of sub-classifiers can improve the overall performance of the integrated classifier.6.By comparing the experimental results of all integrated classifiers,it is finally possible to integrate the four sub-classifiers based on Res Net-18,Res Net-50,VGG16 and Xception.The integrated classifier constituted has reached the highest recognition accuracy rate of 99.93%,the highest The accuracy rate of83.98%,the highest recall rate of 84.55% and the highest F1 value of 84.26%.It is verified that the crop leaf disease identification method based on ensemble learning proposed in this paper is better than a single convolutional neural network.