Research And Application Of Pear Disease And Pest Identification Algorithm Based On CNN

With the advancement of the rural revitalization strategy,government investment and social capital will be more invested in agriculture and rural areas,and the convenience brought by scientific and technological innovation will also be applied to agricultural planting.However,from the perspective of fruit tree planting,the increasing cost of agricultural input and labor input has seriously affected farmers ’ enthusiasm for planting.The continuous development of urbanization has gradually shrunk the rural labor force.In addition,the inheritance of agricultural planting technology has a fault in the new generation of agricultural personnel,and a large number of farmland is barren.Entrusted by a farm,this paper proposes to establish an intelligent pesticide application system to reduce the manpower and material input of agriculture and solve the above problems in view of the difficulty of manual application and indiscriminate excessive application in the prevention and control of pear diseases and insect pests.The main contributions of this paper are as follows :Aiming at the problem that there is no public data set of pear tree diseases and insect pests,through the Internet,field shooting to obtain training materials,data cleaning and data marking one by one,the data set of pear tree rust,scab,Fusarium wilt and healthy leaves was preliminarily established.The combination of classification model and target detection model is used to detect field pests and diseases,and a variety of classification models and target extraction methods are compared.Considering the detection speed,detection accuracy,training model size,system application environment and other factors,the combination of YOLOv5 s target detection and Mobile Net V2 for pest and disease leaf classification is finally determined.Based on the Jetson Nano edge computing development board,a real-time detection and application control system is designed.The system can be flexibly mounted on drones and agricultural robots.Under the condition of limited computing power and energy consumption,it can accurately identify the camera input stream,realize real-time detection of pear pests and diseases and output application control signals offline.