| As one of the most important industries in China,agriculture has become the foundation of the national economy.The variety and quality of agricultural products directly affect people’s daily life.Since agricultural production plays a pivotal role in China’s modern economic development,the monitoring and prevention of pests and diseases are extremely important.The primary and most important point in the prevention and control of crop pests and diseases is how to accurately and quickly identify the diseases and pests that harm crops,so as to ensure the normal growth of crops.This paper proposes an identification technique based on deep convolutional neural network,which improves the identification rate and the identification efficiency.This paper analyzes the current research status of convolut ional neural network technology and identification of crop pests and diseases,and the following researches are mainly carried out based on the relevant theories of convolutional neural networks:(1)As for the problems caused by traditional manual identification methods and fuzzy recognition,such as cumbersome work,artificial error,difficult statistics of effective disease area,escape of pests,complex preprocessing and low identification accuracy,this paper proposes a convolutional neural network based on AlexNet and GoogleNet to identify pests and diseases.This method which is based on the convolutional neural network has the advantages of simple preprocessing,fast identification rate and simple operation.Through comparison experiments,t he identification accuracy based on AlexNet or GoogleNet is maintained at about 80%.Comparing with the traditional methods,this network not only improves a lot on the identification accuracy,but also has the advantages of fast speed of training and short time of identification.(2)Based on the algorithms of AlexNet and GoogleNet,this paper proposes an improved convolutional neural network based on migration learning and data expansion.By migrating the knowledge gained from the AlexNet network,combined with the Inception module of the GoogleNet network,the identification accuracy is increased to 93%.Through data expansion,the identification accuracy is increased to 97%.This improved convolutional neural network can fully meet the application requirements of the actual situation.(3)Cobined with smartphone and WebGIS server,this paper designs and develops an operational data platform for multi-user and multiple pest and disease data that are cumbersome and complex,which is not conducive for related research and other issues.The operation data platform not only can help you to identify pests and diseases and manage data through smartphone,but also can help you to receive information and manage GPS data through WebGIS server,which is of great significance for helping relevant technicians to carry out complex research work. |
