Classification And System Design Of Rape Root Swelling Disease Based On Improved YOLOv3

As the position of food safety continues to rise,the process of growing food and oil production and produce becomes even more important.As the country’s food and oil production comes mainly from oilseed rape,the quality and yield of oilseed rape requires extra attention.Under normal circumstances,the amount of oil produced and the quality of the oil produced by oilseed rape is very good,but as oilseed rape is grown in soil,it is likely to be infected with diseases during the growing process,which can lead not only to a significant reduction in the end-of-year oil production,but also to the death of large quantities of oilseed rape.In order to solve the problem of oilseed rape diseases,in-depth research can be carried out in the direction of studying the grading of oilseed rape diseases,the defence against oilseed rape diseases and the selection of oilseed rape seed quality,thus reducing the probability of a large number of oilseed rape infections with viruses.In order to address the problem of reduced oil and grain yields due to oilseed rape diseases,this paper examines the grading of oilseed rape rhizome disease grades.At present,in order to study the grading of rape root disease grades,agricultural staff need to manually grade each rape root disease grade obtained according to the grading criteria of rape root disease grades,but because the data set of rape root images is large,and at the same time manually in the process of grading rape root disease grades will consume a lot of time and manpower,which is not only inefficient but also This is not only inefficient but also has low grading accuracy.With the continuous development of artificial intelligence and computer vision technology,it is possible to design a system that can automatically classify the grade of rape rootworm by combining convolutional neural network models and software design and development ideas,thus significantly improving the efficiency of manual classification of rape rootworm.This paper focuses on pre-processing the original rape root images,selecting the optimal predictive network model and optimising it by building different convolutional neural network models to obtain the final predictive network model.The selected optimal prediction model is finally integrated into the Android system for development,and the design and implementation of an automatic classification system for rape root disease grade is completed.The main research work completed in this paper is as follows.(1)Pre-processing of oilseed rape root images.As the original rape root images contain a large amount of noise and contain multiple overlapping root swellings,pre-processing operations need to be performed on the original rape root images to lay the foundation for the later dataset building.The method used in this paper is to first automatically segment the original oilseed rape root images to obtain a batch of single-root oilseed rape data images,and then to perform OTSU threshold segmentation and morphological opening operations on the single-root oilseed rape images,which reduces the noise in the images and also highlights the characteristics of the oilseed rape roots more.(2)Establishment and comparison of a grade grading model for oilseed rape root diseases.Based on the oilseed rape data images and grading criteria provided by the Institute of Oilseed Crops,Chinese Academy of Agricultural Sciences,a dataset containing4000 oilseed rape root images can be built by pre-processing the original oilseed rape images and using the Label Img tool.In order to select the optimal model for predicting the grade of oilseed rape root disease,four different convolutional neural network models were used to iteratively train the oilseed rape dataset,using GPU to accelerate the speed of the network model training,finally,after analysing and comparing the model training data,it can be found that the YOLOv3 network model has the best prediction effect.(3)Optimization of a graded prediction model for oilseed rape rhizome disease.In order to optimise the prediction accuracy of the original prediction network model and the speed of model training,the loss function and activation function in the original YOLOv3 network model were optimised in this paper.After comparing the data of the optimised YOLOv3 network model and the original YOLOv3 model,it was found that the optimised YOLOv3 network model was more accurate.At the same time,the test accuracy of the test set in the optimised prediction model was 92%,which basically met the experimental requirements.(4)Implementation of an oilseed rape rhizome disease grade grading system.In order to improve the efficiency of manual grading of rape root diseases,this paper packages the rape root disease grading prediction model and integrates the packaged prediction model into the Android system for integrated development.Finally,after testing the system,the whole system was implemented.