Research On Application Of Maize Leaf Disease Recognition Based On Convolutional Neural Network

As one of the main food crops in our country,the output of corn is related to the food and clothing of the people and also affects the development of the national economy.Due to the weak disease resistance of varieties and irregular cultivation and cultivation,corn diseases have increased year by year in recent years,and the types have increased year by year.This has led to a large reduction in corn production and brought huge economic losses to the country and farmers.There are many types of corn diseases,some of which have similar phenotypes and symptoms,which are difficult to identify by visual observation.Methods that rely on human subjective judgment require a strong background of professional knowledge.In large-scale recognition scenarios,the recognition speed is slow and the efficiency is low.Deep learning is widely used in crop disease identification,but there are fewer applications for corn disease identification.Based on the Convolutional Neural Network(CNN)algorithm,this paper designs and implements a corn leaf disease identification APP,which can accurately identify common corn leaf spot,corn leaf spot,corn rust and healthy leaves,and provide disease details And prevention methods to help farmers master the prevention and control measures of related diseases.The main tasks completed in this paper are as follows:(1)In the field of corn disease recognition,there is currently no publicly available,high-quality image data set.In this paper,a dataset of corn diseases was constructed based on the four common corn leaf images taken in the field.The training set,validation set,and test set are divided according to the ratio of 6:2:2 for model training and testing.In the research process,data enhancement technology is used to optimize the performance of convolutional neural network.(2)Study the effects of different convolutional neural network models on corn leaf disease recognition.Six CNN models are trained on the corn disease data set.Combining the training process and visual analysis of the model,it is found that compared with other networks,the training process of the DenseNet169 network is more stable and the model converges faster.The accuracy of the final test set is DenseNet169(99.9%),ResNet50(99.7%),Xception(99.4%),InceptionV3(99.2%),NASNetMobile(98.4%),MobileNetV2(98.19%),and the results show that DenseNet169 has better recognition effect and generalization ability.(3)Calculate the accuracy rate,recall rate,and the harmonic mean of the accuracy rate and the recall rate of the test set,and combine the confusion matrix performance result as the evaluation index of the model performance to provide a reference for the selection of the model.The results show that the DenseNet169 network’s recognition accuracy of the four sample labels are 100%,100%,100%,100%,and the recall rates are 99%,100%,100%,and 100%,respectively.The harmonic mean of the accuracy and the recall rate Respectively 0.99,1.00,1.00,and 1.00 have the most significant effects among all models and the confusion matrix performance is also the best.(4)At present,there are few researches on corn disease recognition based on mobile applications.This paper proposes a corn disease recognition model based on convolutional neural network.Adopting the C/S architecture and the development model of separation of front and back ends,the corn disease identification APP is designed and implemented.Users can quickly obtain corn disease identification results and corresponding disease prevention methods by taking photos.By testing 40 real images of corn diseases,the accuracy of APP recognition was 92.5%.It has high guiding value in practice,and provides important support for the intelligent research and application of maize disease identification in my country.