| China is a large country of crop planting.It not only has a wide planting area,but also has a wide variety of planting.The problems of natural crop diseases emerge one after another.Agriculture also affects China’s national economic and social development.China exports trillions of tons of grain every year.The prediction and prevention of crop diseases are essential.Therefore,developing the means of identifying crop diseases and improving the accuracy of identification are the top priority in agricultural scientific research.In traditional manual detection,people make subjective judgment through visual observation and experience.This way not only wastes manpower and time,but also the accuracy of the experiment is low,which may delay the opportunity of disease control and cause great losses.In recent years,the theory and technology of computer vision have developed more mature.The traditional machine learning and deep learning methods are widely used in crop pest recognition.The classical methods of deep learning include deep convolution neural network,which not only has strong learning ability,but also realizes the automatic extraction and classification of image features.However,these models still have some shortcomings,such as insufficient feature information extraction and low recognition accuracy.Aiming at the above problems,this paper selects two kinds of data sets of maize diseased leaves and healthy leaves as the research object,and improves them based on VGG16 and Mobile Net V2 convolutional neural network,so as to improve the accuracy of maize leaf disease identification.The main research work is as follows:(1)Aiming at the problem of insufficient recognition accuracy caused by insufficient feature information extraction or feature disappearance in a single network,a new model recognition method based on optimized VGG16 network is proposed.In this method,the convolution layer of VGG16 model is regarded as five groups of convolution layer groups,the convolution layer groups of VGG16 model and the residual structure in Res Net model are cascaded,and the residual connection is used to reuse and transfer the feature information to be extracted,so as to improve the utilization rate of feature information;In addition,a network layer is added to the new model to eliminate the phenomenon of feature disappearance or over fitting.In order to further improve the experimental results,the experimental data are processed before the experiment,and the data set is expanded by using image enhancement techniques,such as image flipping,image rotation,random color,etc.Finally,by adjusting the parameters,the optimal result is 99.8%,which is compared with many methods.The experimental results show that this method is better than other methods in corn leaf image disease recognition.(2)Aiming at the problems of long training time,large model and many parameters of deep learning network,taking corn leaf pictures as the experimental object,a lightweight convolution neural network based on transfer learning is proposed to identify corn leaf diseases.Firstly,the migration learning method of the migration model is selected to load the pre training model of Mobile Net V2 onto the new model,and then the new model is improved and fine tuned.By setting different parameters for experiments,the optimal dropout discarding rate value and the number of dense neurons are found,and then the thawing experiments of different layers are carried out on the convolution layer in the network to obtain the most appropriate number of freezing layers;Finally,the network is built according to the three optimal solutions,and the experiment is carried out on the target data set.The results show that the new model after transfer learning is the best in the recognition of corn leaf diseases.(3)In order to provide a convenient identification method of corn leaf diseases,a system is built through Python and Qt5.Taking the optimized vgg16 model and the new model after migration learning as the identification model,the theory is put into practice through system analysis and design.Users can upload the collected corn disease images to the system,analyze and identify them in the background of the system,and then give the detection results,which effectively provides users with accurate information,so that users can make targeted protective measures. |
