The Study On Deep Learning Of Maize Disease Recognition Based On Edge Computing

Agriculture is the primary industry in China,and the safety of agricultural products is the key guarantee for national stability and people's happiness.China's population ranks the first in the world,but the per capita arable land area is relatively small,so the issue of food security is particularly important.The quality of crops directly affects the yield and quality of grain.As an important product of crops,maize leaves can not be accurately distinguished from other defects such as disease spot and damage in appearance,which requires a lot of artificial means to be classified and prevented.However,the disease of maize leaves can be detected quickly and accurately by computer vision technology.Based on deep neural network,transfer learning and edge computing,a classification system of maize disease was analyzed and studied in this paper.On the one hand,with the spread of mobile computing and the Internet of things(Io T),billions of mobile devices and Io T devices are connected to the Internet,generating huge bytes of data at the edge of the network.Driven by this trend,it is urgent to push the frontier of artificial intelligence to the edge of the network and fully release the potential of edge big data.As the most popular deep learning technology supporting modern technology,how to reasonably integrate with the Internet of things and apply it in the field of agriculture has become a very important research direction at present.On the other hand,deep learning is now widely used in a variety of agricultural scenarios,including the crop data being captured by end-users such as sensors and cameras in the Internet of things.This data needs to be analyzed in real time using deep learning or used to train deep learning models.However,deep learning models require a lot of computational resources.At this point,the combination of edge computing and deep learning becomes a feasible method by taking advantage of the fact that the edge.This paper introduces a cloud-edge transfer-learning model based on Edge computing and deep learning,which mainly includes the following tasks:(1)Study the accuracy of corn leaf disease recognition in different depth convolutional neural network models.The accuracy and loss values were obtained by comparing the training of Alex Net,Goog Le Net and Res Net18 models,and the generalization ability and suitability of the three models for corn leaf disease recognition were observed.Finally,the Res Net model was selected,and the original data were amplified by a factor of 2,4 and 8 using the data amplification technology.It was found that the number of training images was different,and the generalization ability of the model was significantly different.As the amount of data increases,the model chosen in this paper has better generalization capabilities.(2)A cloud-edge transfer-learning model based on Edge computing and deep learning is proposed.This model will accomplish three tasks: the first task is to Transfer the training data we collect at the terminal to the Cloud computing device,and the neural network model will be trained by the Cloud device;The second task is to transfer the trained model to the edge equipment.The third part will test or continue to train the image to the edge end,edge end through transfer learning to identify or train the image data.(3)Simulate the two states of the cloud server,High load server and Low load server,and then conduct the experiment in two stages.The first stage is the data transmission stage: sending data to the cloud server and edge devices.In the first stage,the maize disease image data set in the experiment was 100 M.It was found that when the terminal data and the edge equipment transmitted data to each other,the edge segment was closer to the terminal,so the transmission speed could greatly meet the timeliness of image recognition.The second stage is network training stage: the convolutional neural network is trained by using the data sent by the terminal.Under the state of Low load server,the cloud server takes a short time to train the model and makes full and effective use of computing resources.If we combine the cloud server with the edge device,we can train the network model when the cloud server is in the Low load server state.The cloud server then sends the trained model to the edge.Finally,the edge device can perform tasks such as retraining and recognition of the new data by using the model of migration learning and training,without considering the load state of the cloud server.Finally,the experimental results show that the cloud-edge transfer-learning model structure proposed in this paper can reduce the Cloud pressure,improve the running speed and reduce the total running time.