Research On Recognition Method Of Corn Diseases Based On Deep Learning

Corn has a large planting area and high yield.For the national economy,it is an important food crop and agricultural feed.However,due to the diversity of pathogenic bacteria and imperfect plant health measures,the problem of corn leaf diseases has become more serious in recent years.Therefore,people began to pay attention to the identification of corn diseases.With the rapid rise of precision agriculture in my country,the precise identification of diseases has also attracted widespread attention.Because the traditional digital image processing technology has the disadvantages of manual design,heavy workload,high hardware cost and unsatisfactory effect when facing accurate recognition,it cannot meet the requirements of high accuracy and real-time recognition.Therefore,find a fast,Efficient and real-time disease image recognition methods are particularly important.Deep learning algorithm is a new type of machine learning algorithm capable of autonomously learning its representative features.It has now achieved significant achievements in image recognition.Based on deep learning,this paper studies the image recognition method of corn diseases,and the related work includes:(1)To study the impact of changes in the deep convolutional neural network model structure on the accuracy of corn leaf disease recognition,focus on corn disease images and health images of five categories of corn images,and use Le Net model to carry out various experiments.First,select the training set and the test set for each corn disease image according to the ratio of 8:2.Then,the influence of different convolutional neural network structure settings on the accuracy is compared through the method of experimental combination and comparative analysis,and the best parameters are selected.In addition,the Adam algorithm is selected to replace the SGD algorithm to optimize the model,and the learning rate is adjusted by the exponential decay method,the L2 regular term is added to the cross entropy function,the Re LU excitation function is selected,and the Dropout strategy is adopted.Through analysis,the final selection is a 10-layer CNN network structure.The experimental results show that the recognition rate of corn mosaic disease is 95.83%,the recognition rate of gray spot disease is90.57%,the recognition rate of leaf spot disease is 93.75%,and the recognition rate of rust and corn health is 100%.The average recognition rate can reach 96%,and the average time required to recognize a single image is 0.15 s.This model has significant advantages over traditional methods.(2)In order to better prevent and control corn diseases according to the disease degree,a method for identifying the disease degree of maize gray leaf spot based on convolutional neural network is proposed.First,classify the disease level of the leaf image of corn gray spot disease,and transform the image of the leaf disease degree of corn gray spot disease through geometric transformation methods such as rotation,scaling,mirror transformation,etc.,and the image categories and attributes are kept consistent,and the data set is expanded Samples,to solve the problem of insufficient number of training samples,and divide the generated data set into a training set and a test set according to a certain proportion.Secondly,based on the CNN model:Alex Net and VGG-16 use the above two structures to train and test the image data set of maize gray leaf spot disease degree.The parameters used by the two algorithms are kept consistent,optimized by SGD+0.9Momentum,and the weight attenuation and learning rate are set to0.0005 and 0.0001 respectively,and 64 is selected as the batch-size value.In addition to different algorithms and iteration times,the experimental process also studied the recognition results of each disease degree.Finally,according to the experimental results,it can be seen that after 200 rounds of training,the recognition accuracy of the VGG-16 model is 92.08%,and the recognition accuracy of the Alex Net model is 80.09%.The comparison and analysis of the above experiments can show that the VGG-16 model is The image recognition of gray leaf spot of maize collected in this paper has the best performance.