Image Recogniton Of Tea Leafhopper (Empoasca Pirisuga Matumura) And Field Population Monitoring Based On Deep Learning Algorithm

Tea leafhopper is one of the main pests of tea gardens in my country,which has a significant impact on the yield and quality of tea.Mastering the density of insect populations in tea gardens is the key to rationally adopting prevention and control measures to control tea leafhoppers.This research combines the color plate trapping technology and the Internet of Things technology to develop a remote camera system that can continuously collect the population of the tea leafhoppers in the field.Based on deep learning,the tea leafhoppers in the photos are identified and counted.The dynamic monitoring of the population of the tea leafhoppers will help guide tea farmers to take reasonable control measures in time.The main research contents of the subject are as follows:Based on the Internet of Things technology,a remote camera system was built for the tea leafhopper in the field,including the tea leafhopper trapping module,image acquisition module,microcontroller,4G wireless transmission module,solar power module and remote monitoring cloud platform.The tea leafhopper trapping module is responsible for trapping the tea leafhopper;the function of the image acquisition module is to collect pictures of the tea leafhopper;the solar power module provides stable power supply for each module in the entire system;the microcontroller is responsible for controlling the others in the system The module realizes the functions of image acquisition and data transmission control of the entire system;the 4G wireless transmission module is responsible for maintaining stable data transmission with the server.The remote monitoring cloud platform is responsible for receiving the pictures collected by the equipment,processing them and establishing a database for storage.Based on image deep learning,the recognition technology of the tea leafhopper is studied,including(1)the study of image preprocessing methods.Specifically,the histogram changes are used to process images with overexposed or dark images,and the Copy-Paste strategy is used for data enhancement to improve the generalization of the model.(2)Selection and fine-tuning of target detection algorithm: Choose the model YOLOv3 that has a better detection effect for small targets,and improve its detection effect on small targets by improving NMS and using K-means mean clustering.(3)Target recognition research: According to the selected deep learning model,the average accuracy AP(average precision)is used as an indicator to compare the effects of conventional target detection algorithms and improved target detection algorithms.The improved YOLOv3 algorithm has improved recognition accuracy.At the same time,the accuracy of the algorithm is compared with different fine-tuning strategies.Taking the original YOLOv3 AP as the baseline,the Copy-Paste strategy improves the model AP by 3.6%,and K-Means clustering makes the model AP increase by 1.8%,and the improved NMS obtains an AP increase of1.6%,indicating that the model fine-tuning is effective.Research on the application of field population monitoring of the green tea leafhopper: Combining the remote camera system of the tea leafhopper and the research on the identification technology of the tea leafhopper,a complete field population monitoring system of the tea leafhopper has been formed.The system studied in this paper was applied to the population monitoring of green tea leafhoppers in actual tea gardens,and the feasibility of the application of this system was studied.Taking the result of manual monitoring as the real value,comparing the result of the system monitoring with it,the research shows that the accuracy rate of the monitoring of tea leafhoppers using this system is 78.4%,and the trend of the number of leafhoppers monitored by the system for 15 consecutive days is the same as the trend of the true value,using the data analysis software to analyze the difference between the system monitoring results and the real value,and the P-value of the two sets of data is 0.07,indicating that there is no difference between the two sets of data,which proves that the system is effective for field population monitoring of the tea leafhopper.