Development Of Remote Intelligent Diagnosis System For Typical Crop Leaf Diseases

At present,the economic development of China has been showing a stable growth trend.Among them,the development of agriculture is of great significance to the national economy.However,the losses caused by agricultural diseases and insect pests are incalculable every year in China.Therefore,it is very important to identify and diagnose pests and diseases as soon as possible,so as to minimize the economic losses caused by pests and diseases.At present,the diagnosis of pests and diseases mainly adopts the method of artificial identification.When farmers in some economically underdeveloped areas encounter diseases,because they lack relevant knowledge,they even invite agricultural experts from the city to diagnose them in the fields.This is not only very inconvenient,but also easy to delay the time of diagnosis and miss the golden node for the treatment of pests and diseases.In this study,an intelligent diagnosis system for typical crop leaf diseases was developed based on Android mobile phone.Users can use this system to identify common leaf diseases of several typical crops.Tests show that the recognition accuracy of leaf diseases reaches more than90%.In this way,the time and cost needed by farmers to diagnose crop diseases are greatly saved,and measures can be taken to treat the diseases early to avoid greater losses.The main research contents and results are as follows:(1)Collect the disease pictures of five typical crops of rice,wheat,maize,cotton and soybean to make data sets.Five kinds of most common leaf diseases were selected for each crop.The pictures were downloaded from the Plantvillage database and supplemented with the pictures of diseases on the Internet such as the China Agricultural Pest and Disease Network.(2)Then do the pre-processing work for the collected pictures.Because the pictures are different in size,illumination,background,noise and so on,these will have a greater impact on the diagnosis and recognition of crop diseases.Therefore,to minimize the irrelevant factors and simplify the operation,we need to preprocess the pictures,such as normalization,artificial clipping,random flipping,random rotation,filtering and noise reduction,etc.(3)Improve the network structure of VGG16,reduce three full connection layers to two,and add a Dropout layer to prevent network over-fitting.Considering the problem of insufficientsample size of data set,the identification algorithm of crop leaf diseases is realized by migration learning.The weight of the remaining layers remains unchanged,and the fully connected layer is fine-tuned.After training,the network model of disease identification is obtained.The average recognition accuracy of 25 kinds of diseases can reach 96.18%.(4)Train the network model on PC,then transplant the model to Android mobile phone by Tensorflow Lite.Import the trained network model file and dynamic link file,then design the interface,define the variables,and finally generate the APK file to complete the development and design of the APP program.