Research On The Identification Method Of Crop Diseases And Insect Pests Based On Convolutional Neural Network

My country is a big country in agricultural production,and it is very important to ensure the yield and quality of crops.However,pests and diseases have a greater impact on crop production,leading to economic losses due to crop production reduction.Therefore,research on the identification of crop diseases and insect pests is of great significance to provide reference for the next control work.For a long time,the identification and detection of plant diseases and insect pests have been mostly diagnosed by experts,which is inefficient and difficult to promote.In the current field of computer vision,convolutional neural networks are a hot topic of research,and they also have unique advantages in image recognition.It has the advantages of high accuracy and easy promotion for disease and insect identification.Aiming at the identification of crop diseases and insect pests,this paper uses the convolutional neural network algorithm as the basis to study the identification of crop diseases and insect pests.The main work is as follows:A network model of Alex Net is built,which is composed of 5 convolutional layers and 3 fully connected layers The network model was trained using the Tensor Flow framework and 6400 leaves of plant diseases and insect pests.The experimental results show that the model has a low recognition rate.In order to improve the model recognition rate,this paper changes the number of network layers of Alex Net,reduces the size of the convolution kernel,increases the number of convolution kernels,and introduces Dropout and L2 regularization optimization strategies to design 4convolutional layers and 2 fully connected The network model of re Alex Net composed of layers.Experimental results show that the correct rate of the re Alex Net model is81.27%,which is better than the Alex Net model.In order to further improve the recognition rate,data enhancement techniques such as Gaussian blurring,sharpening,image light and dark changes,and rotation are used to expand the initial pest data set,and the data set is expanded to 2,4,and 6 times.The experimental results show that the data set is expanded by 6 The accuracy of the re Alex Net model increased to 88.74%.In order to further improve the accuracy of the identification of pests and diseases,the Imag Net data set is used to select VGGl6 and Inceptionv3 networks as pre-training.After the enhancement of the pests and diseases data set,the parameter migration finetuning network is used to obtain two Convolution neural network models,tfvggl6 and tf Inceptionv3.These two single Convolution neural network models are integrated.According to the experiment,the identification accuracy rate of two migration models,tfvggl6 and tf Inceptionv3,is about 90%.From a single Convolution neural network model,the tf Inceptionv3 model fits better,and the model recognition accuracy rate is93.42%.In the multi-model ensemble,the recognition accuracy of the weighted ensemble model is 94.2%,which is higher than the single transfer learning model and is also better than the improved re Alex Net model.Aiming at the actual operation problem of crop pest identification,the weighted integrated Convolution neural network pest identification model is deployed to the server.The server side uses the Django framework,and the information exchange between the server side and the front-end page uses Json for transmission.The user uses the mobile phone to obtain the photo,clicks the upload button and then queries to obtain the recognition result.