Research On Recognition System For Garden Plant Pest Based On Convolutional Neural Network

Garden plant pest species,large harm,seriously affecting the practical and ornamental garden plants,so that the significance of garden plants to beautify the city and give people a sense of beauty no longer exists.How to reduce the occurrence of garden plant pests and provide people with corresponding pest identification and prevention methods,which is of great significance for the scientific management of gardens.The correct identification of pest species is the first and key step in pest control.On the basis of previous researches,this study established a set of garden plant pest identification system which can be used on mobile phone based on convolutional neural network.The preliminary research conclusions are as follows:(1)An image batch processing method based on visual saliency is proposed.How to realize the collection,processing and related production of pictures in the data set is the basis of this research.This study mainly identified 41 types of pests including underground pests,leaf-eating pests,dry borer pests and stabbing pests.The images of these identified pest types are obtained from the original image data of the team,related databases and websites..In image processing by ITTI,GBVS,LC,AC,FT and HC six significant contrast method,selection of automatic cutting GBVS applied to the image of pests,and combine with Grab Cut pests automatic segmentation,at last the presence of segmentation of the two images in accordance with the proportion of 3:1:1 was divided into training set,validation set and test set,for later research provides data base,also for the future research provides image capture and processing idea for other objects.(2)The transfer learning of the convolutional neural network is realized,and the specific model of garden plant pest identification is obtained and the verification result of0.93 is obtained.The thesis first conducts simple training and comparison on the three models of Inception-V3,Res Net-50 and Mobile Net-V2 to determine and select the Inception-V3 model to extract the features of the convolutional layer,and reduces the 2048-dimensional features of the last layer through t-SNE.Then,through multi-parameter setting comparison,the optimal parameters of the SVM and Soft Max classifiers are obtained,and finally the optimal recognition results of the two classifiers are obtained.The overall use of transfer learning ideas to obtain the specific model of garden plant pest recognition and recognition results.(3)The garden plant pest identification model was further optimized to improve the accuracy of identification and proved that removing the background would reduce the identification result.Under the same experimental method,the recognition accuracy rate after background removal was reduced by 8%.Exploring the cause found that some different types of pests are slightly similar in shape,color,and texture,but their living environment and hazardous parts are slightly different,so The background can also be regarded as a feature of different types of pests,then removing the background will reduce the accuracy.(4)Based on the convolutional neural network,a set of garden plant pest identification system that can be used on mobile phones was created,and good test results were obtained on the real machine.This article studies the development of mobile phone applications related to garden plant pests.The user realizes the identification of the garden plant pests by logging in the software,and then through the album images and photographs,and finally provides the user with the recognition results and the images of such pests.It allows users to make further judgments based on the pictures themselves,and at the same time gives the habits,hazard characteristics and control methods of such pests,providing convenience for users,allowing users to identify pests anytime,anywhere.The software also provides a problem and feedback path,which makes it easier to receive the problems encountered by users in the application and deal with them in time,and also provides a source of ideas for the next step of software optimization.