Research On Crop Disease Identification Based On Super Parameter Optimization Algorithm And Improved Co-location Network

The research of crop disease identification is to distinguish different kinds of diseases according to the pictures of crop leaves,improve the accuracy of disease classification,reduce the abuse of pesticides after the correct diagnosis of crop diseases,which can significantly reduce the planting cost of farmers,reduce the economic losses caused by the wrong diagnosis of diseases,and avoid unnecessary losses caused by the use of pesticides in the planting process We should pay more attention to environmental pollution.How to predict and diagnose crop diseases efficiently,quickly and accurately in the first time,so as to take effective prevention measures is more practical.In recent years,deep learning method has become the mainstream method of crop disease recognition.However,the existing crop disease classification based on deep learning still has the following problems:(1)the parameters of deep learning network are complex,and it can’t meet the existing parameter adjustment requirements by relying on the experience judgment of researchers and manually adjusting the super parameters of the model.We can explore the use of intelligent algorithm instead of gradient descent algorithm to reduce the trial and error cost and improve the practicability.(2)The traditional research on plant disease recognition based on image mainly focuses on the recognition of plant leaf images sampled from the laboratory with pure color background,but the classification model trained by such data has poor effect on plant image recognition under complex background in natural environment.Aiming at the above problems,based on the deep learning image classification theory,this paper aims at the problem of super parameter optimization in crop disease recognition and the problem of poor robustness in real scene application.The main work of this paper is as follows:(1)The data set of crop diseases was constructed.Through the extraction and data enhancement of corn disease data set in plantvillage data set,the cornland data set is constructed,which contains three kinds of corn disease pictures and one kind of healthy corn leaves.Then,the plantvillage data set is supplemented and enhanced by web crawler,which integrates a crop leaf disease data set,multipleplant data set,collected in a mixed laboratory and real scene.(2)A super parameter optimization method based on orthogonal particle swarm optimization is proposed.In this paper,the problem of super parameter optimization in crop disease classification is studied,and a super parameter optimization method based on improved particle swarm algorithm is proposed for small sample classification of corn disease.In this paper,a new orthogonal particle swarm optimization(OPSO)algorithm is proposed by improving the traditional particle swarm optimization algorithm.This method can replace the traditional gradient descent method.The experimental results show that the model using OPSO algorithm performs better in classification accuracy and cross loss than the model using traditional BP algorithm.This method is also applied to VGg pre training model and integrated model,and the integrated model achieves high accuracy in corn disease classification model.(3)A fine-grained classification method of crop diseases based on CO location is proposed.The training results of classification model still have strong generalization and robustness in real scene applications.This method can effectively improve the classification accuracy.In the case of less image data,the open plantvillage data set combined with self collected image data is used for model training,and the classification accuracy is still high in the multiplant test set simulating the real scene.It proves that this method fully retains the feature information of the key disease location,and has a certain degree of accuracy It can resist the interference of noise.