Grading Disease Severity And Developing Diagnosis System Of Wheat Fusarium Head Blight Based On Computer Vision Technology

Fusarium head blight(FHB)is one of the most important diseases in wheat production,mainly occurring in the ear.Wheat ear infected FHB always vomits the toxin DON,which seriously threatens human and animal health,and national food security.Due to the FHB severity cannot be accurately identified,the cost of pesticide application is increasing every year,and the agricultural ecological environment is also seriously polluted.In order to effectively protect food safety,reduce pesticide application costs,and protect the agricultural ecological environment,it is particularly important to quickly and non-destructively detect wheat FHB with different levels of disease.Currently,with the rapid development of information technology,a large number of mobile devices such as mobile phones and tablet computers have certain computing capabilities,so that image or video data can play an important role in the diagnosis of crop diseases.To provide technical and methodological support for wheat disease prevention and control,this paper chosen wheat FHB as the research object,explored a fast and accurate method for identifying the severity of wheat FHB based on computer vision technology,and designed a Android smartphone-based diagnosis system of wheat FHB.Mainly results from the following four aspects were obtained:(1)The segmentation method of wheat ear based on U-Net network was proposed.Wheat ear in field environment was chosen as research target,to explore the method of identifying disease severity of wheat FHB.Firstly,the dataset of wheat ears was constructed to provide a data benchmark for wheat ear segmentation.Secondly,the segmentation model of wheat ear based on U-Net network was established to effectively implement wheat ear segmentation in the field environment.The results showed that the accuracy of the wheat ear segmentation model constructed in this paper is 0.981,and the segmentation time is less than 1s.This indicates that the proposed method can quickly and accurately segment wheat ear in the field environment,and provide significant technical support for disease identification of wheat FHB.(2)The disease spot segmentation of wheat FHB based on Pulse Coupled Neural Network(PCNN)was proposed.Based on artificial bee colony(ABC)and PCNN,the PCNN with K-means clustering of the improved ABC(IABC-K-PCNN)was used to segment disease spot of wheat FHB.The results showed that the proposed method in this paper is better than the traditional those methods under different segmentation evaluation indexes,which have an accuracy of 0.925 and a run time of 5.11 s.It is concluded that the proposed method can effectively segment the disease spot of wheat FHB,and provide method support for rapid and accurate identification of disease levels.(3)The severity classification method of wheat FHB based on fusion characteristics was proposed.The Relief-F algorithm was used to excavate the shallow and deep features of wheat FHB with different levels of disease,and to accurately identify wheat FHB.Firstly,according to the differences in the color and texture images of the disease with different severity,the two types of features of the disease were extracted as shallow features;then the parameters of the AlexNet model trained on the ImageNet 2012 dataset were migrated to the disease dataset of wheat FHB,the deep convolution features of wheat FHB were extracted as deep features.Secondly,the weights of shallow and deep features were calculated using the Relief-F algorithm,and the shallow and deep features were fused into the new feature based on the weights.Finally,Random Forest(RF)ws used for model training to achieve effective classification and identification of different disease levels.The results showed that when the sample size is relatively small,the recognition accuracy of the fusion feature model is 0.942,and the diagnosis time is 6.21 s.Compared with the use of shallow and deep features alone,it is improved by 2% to 5%,and the fusion feature model is more robust.(4)Based on the above research methods,a wheat FHB diagnosis system based on Android smartphone was developed.It consists of three parts: client,server and database.The client was designed by Android Studio,and its functions mainly include image collection,image storage,GPS positioning,data upload,diagnosis result query,historical disease query and voice interaction.The server was completed through the mixed programming of Java,Matlab,and Python.Tomcat as the server mainly implements data reception,segmentation of wheat ear,disease classification,and data storage.The database was implemented by MySQL and mainly includes: "disease database" and "disease diagnosis knowledge base".The "disease database" mainly stores uploaded disease data,and the "disease diagnosis knowledge base" mainly stores knowledge bases that implement different treatment measures for different disease levels.Finally,through testing and verification,a mobile terminal based on Android smartphone can real time collect images of wheat FHB and upload them to the server.On the server,after processing the images,the results are stored in the "disease database",and the diagnosis knowledge suitable for the current condition is selected from the "disease diagnosis knowledge base" and feedback to the user.