Research On Maize Leaf Disease Recognition Based On Convolutional Neural Network

As an important food crop,corn occupies a key position in my country’s agricultural field.The corn industry is not only closely related to the people’s food and clothing,but also brings high economic benefits.In the process of corn planting,various problems will be encountered,such as unscientific planting,poor disease resistance of varieties,etc.,which will have a serious impact on the yield and quality of corn.The traditional methods of identifying maize leaf diseases mainly rely on people’s past experience and consulting data to make judgments,but in the case of a wide variety of diseases and similar characteristics,these methods have certain limitations in some aspects.At present,deep learning has achieved good results in the field of agriculture.This paper explores an effective method for identifying maize leaf diseases from two aspects of convolutional neural network and transfer learning,and designs and constructs a maize leaf disease identification system.Great spot disease,rust disease,and small spot disease are accurately identified,and relevant prevention and control suggestions are also given.The main tasks of this paper are as follows:(1)Establishment of maize leaf disease database.Four kinds of images of corn leaves in natural environment were collected in the corn planting base,namely healthy,great spot,rust,and great spot,a total of 1713 images were divided into training set and validation set according to the ratio of 8:2,and two kinds of images were used.Data augmentation methods augment the dataset samples.(2)Constructing a convolutional neural network to identify maize leaf diseases.This study constructed four network models: AlexNet,Inception-v3,MobileNet-v2,and VGG16.For the first three models,each model is trained with four optimizers,and for VGG16,with different batch sizes and learning rates.The results show that MobileNet-v2 and Inceptiov3 have the highest accuracy using the Adam optimizer,while AlexNet has the highest accuracy using the Adagrad optimizer.VGG16 has the highest accuracy when the batch size and learning rate are 16 and 0.001,respectively.(3)Introduce the transfer learning method to optimize the network model.An improved method is proposed for ResNet50.Transfer learning on ResNet50 and modified ResNet50.The results show that the accuracy of the two models after combining transfer learning is improved compared to the accuracy of direct training,and the improved model proposed in this paper is also improved compared to the unimproved accuracy under the condition of direct training.On this basis,a maize leaf disease identification system was designed to achieve accurate identification of several diseases.