The Cucumber Plant Diseases And Insect Pests Detection System Based On Android

In order to better promote the development of major projects in the Ningxia Hui Autonomous Region "13th Five-Year" research plan,realize the scale,implementation of crop production scale,industrial production,and further promote the development of the economy,it is particularly important to strengthen training and development in the middle part of the crop and management.This paper is devoted to the search for fast,convenient and accurate detection method,The depth of the neural network based on convolution,design a pest detection system,greatly facilitate the production and life of farmers,effectively promote the economic development.The convolution neural network is introduced to the pest detection field,reasonable design,the number of network nodes and the network layer will be 7264 pieces of cucumber leaves are divided into four categories:healthy leaves,mild disease leaves and serious diseases and serious disease leaf leaf.The final accuracy is up to 95.06%,and a system for detecting cucumber leaf diseases and insect pests in greenhouse has been developed.The main work is as followsFirst of all,select the more faint color as the sample background,use the camera to take samples of the cucumber,and prepare the model library for better separation of background.Secondly,the acquisition of a good photo pretreatment,pretreatment to the image processing into can training picture formats.The pretreatment mainly for image smoothing and sharpening,geometric transformation,and combined with the characteristic of solving histogram threshold,background extraction of leaf basic information and characteristics of using the technology of image segmentation,convenient the human visual recognition.Increase the number of pictures by rotating and improve the accuracy of the whole training classification.Thirdly,the convolution neural network is used to train the sample images.Segmentation and classification of a single picture sample,and extract the sample information from a single picture,divide the variables in the image,and mark and predict the five categories with the highest probability.Through it we can effectively determine the image sample information.Last,the training of the sample data is carried out and the CNN convolution neural network is used to train the classification.Using a small network structure consisting of two layers of convolution layers and a hierarchical layer of LeNet,7264 sample data are trained in 5000 classifications.By adjusting the data sample size,regularization is used to avoid over-fitting,and the training classification accuracy ratio is continuously improved.The accuracy rate increased from 86.5%to 95.1%.At the same time,the feature points in the convolution process are extracted,and 96 feature graphs are displayed in each graph.The first 32 feature graphs represent the first layer convolution feature,and the latter 64 features represent second layer convolution features.Finally,In the mobile client to complete the relevant APP interface design.