| China is a big agricultural country.With the rapid development of digital agriculture,its existence not only has a great impact on crop yield,but also has a positive significance for the prevention and control of pests and diseases,which will reduce economic losses to a certain extent.For a long time,the identification and detection of pests and diseases are mostly diagnosed by experts,which is inefficient and difficult to promote.Scholars continue to explore how to use machine learning to identify diseases more accurately.This paper took soybean leaves as the research object,mainly to explore the common gray spot and frost spot in soybean(also known as downy mildew)and how to accurately identify these two diseases.Through in-depth understanding of domestic and foreign related research and analysis of the relevant technology and theoretical knowledge of their own learning,provide a feasible scheme to combine the optimization model and system.The research contents are as follows:(1)Processing of data sets.On the one hand,the images are obtained from the project data sets in kaggle and aistudio,and on the other hand,they are obtained from the Internet to supplement the sample database.Due to the different sizes of the obtained images,it is necessary to standardize their sizes and then increase the obtained image data sets.The number of samples for the two diseases is 1000 each.,a total of 2000 sample data strengthen the main method is adopted by the rotating mirror and operation,the increase of the amount of data to a certain extent will enhance the accuracy in network model identification,then respectively through the median filtering,gaussian filtering and mean filtering process for processing,to explore what kind of filter way get better recognition rate.(2)Model construction and optimization.In this paper,a 17-layer network model architecture including 10 convolutional layers,4 pooling layers and 3 fully connected layers is constructed.Before the model training,the data is first processed,and it is divided into training set and test set according to 7:3.After that,the test set processed by three filtering methods uses the convolutional neural network to determine the recognition accuracy.The experimental results It shows that the recognition accuracy of the test set processed by Gaussian filtering is the highest,which is 83.57%.After that,the main parameters in the network model are analyzed and optimized,and the problem of model overfitting is solved.The optimal parameter collocation is obtained by studying the number of batch training samples and the learning rate.The experimental results show that when the number of batch training samples is 32 and the learning rate is 0.001,the accuracy rate reaches93.89%.Finally,the combination of activation function and optimization algorithm is explored to further improve the recognition accuracy.Experiments show that when the activation function is selected as ELU and the optimization algorithm is selected as Adam,the accuracy of the model reaches the highest 98.97%.(3)Establishment of the system.in order to more conveniently identify soybean diseases,this paper builds a disease identification system using the Django framework.The main functions designed in this system include uploading images,calling models and returning identification results.Users can easily use this system.After uploading disease pictures,they will get the system’s judgment results in time,which reduces users’ dependence on agricultural technical experts,facilitates them to treat diseases faster in the next step,and also provides information for the identification of soybean diseases.strong technical support. |
