Research On Identification Algorithm Of Jasmine Pests And Diseases Based On Convolution Neural Network

With the advent of the 5G era and the rapid development of artificial intelligence,deep learning technology has been widely applied to all walks of life,will bring revolutionary technological innovation to modern agriculture.As a research hotspot of deep learning technology in agricultural field,identification and detection of crop diseases and insect pests has very practical research value.Image recognition technology can identify crop diseases and insect pests accurately and quickly,it can reduce the risk of misjudgment caused by the lack of experience and theoretical knowledge.As the dominant industry in Heng County,jasmine flower industry plays an important role in the rural economy of Heng County,and in recent years,the pests and diseases of jasmine flower have been increasing.The growth status of Jasminum SAMBAC is directly related to the yield,and the invasion of plant diseases and insect pests seriously restricts the healthy and sustainable development of Jasminum SAMBAC.The research on the image recognition algorithm of jasmine diseases and insect pests will lay a foundation for the prevention and control of jasmine diseases and insect pests.The work done in this paper is as follows:(1)To collect the image data of diseases and insect pests of Jasmine suitable for this research,and make the corresponding image data set.A total of 1775 pictures of 6 kinds of jasmine diseases including anthracnose,Spodoptera Litura,leaf spot,sclerotium Rolfsii,leaf roller and leaf borer were collected.In order to solve the problem that the number of samples is too small,the image needs to be expanded.In this paper,the data set is expanded by means of space transformation,color contrast and data enhancement.After the expansion,the number of images of 6kinds of jasmine diseases and insect pests increased from 1775 to 15370,and the number of images of each kind of diseases and insect pests ranged from 2400 to2800,the results show that the accuracy of model recognition using the expanded dataset is 4.22% higher than that using the original Dataset.(2)From the two aspects of experiment and theory,the suitable network model of plant diseases and insect pests identification in Jasmine was analyzed and four common network models were selected for experimental comparison.The experimental results show that the model of light-weight MOBILENETV2 can achieve a good balance among the recognition accuracy,model size and training time.Finally Mobile Net V2 was chosen as the basic model of Jasmine disease and pest image recognition.(3)In this paper,we design an improved network model algorithm of Mobile Net V2,combine SENet module with Mobile Net V2 network,get the network model of se-Mobile Net V2.The model combines the advantages of Senet and Mobile Net V2,and enhances the sensitivity of the network.Four network models,se-Mobile Net V2,Mobile Net V2,SE-Mobile Net V1,Mobile Net V1,Mobile Net V1,are compared,the results showed that the accuracy of image recognition by SE-Mobile Net V2 was improved by 2.25% to 90.82% compared with the original model.(4)The optimization algorithm of SE-Mobile Net V2 model is analyzed.In this paper,an optimization algorithm combining Momentum and RMSPROP is designed,compared with other optimization algorithms,the algorithm proposed in this paper,which combines Momentum and RMSPROP,improves the recognition rate,converges faster and the recognition accuracy is higher,and finally reaches94.34%.