Research On The Real-time UAV Target Recognition System Based On Embedded Technology And Improved YOLOv4 Algorithm

In recent years,with the rapid development of the neural networks,target recognition has been rapidly improved in both accuracy and speed,and widely used in disaster response,military strikes,video surveillance,and other fields.Traditional offline target recognition systems are usually processed through multiple phases.Firstly,the acquired image is copied to a local computer,then the target is identified,and sent to the user.This method falls behind not the current technological development,especially for the applications like military attack and suspect-tracking,which need to be processed in real time with high accuracy.With the increase of the load of unmanned aerial vehicle(UAV)and the development of embedded equipment,it is urgent to meet the needs of real-time target recognition with new technologies.In this study,Dajiang M600 Pro UAV is used as the airborne platform of the target recognition system.Image acquisition and target recognition are carried out on this platform.Then,a cross-platform protocol for data transmission is developed based on the Netty framework.Finally,all the above software,hardware,and subsystems are integrated to develop a programmed and user-friendly graphical UAV real-time target recognition system.Given the above objectives,the main research contents of this study are as follows:(1)To get the results with high accuracy and fast speed,this study takes YOLOv4 target recognition algorithm as the primary research object.The YOLOV4 algorithm performs well on high-performance equipment,but the recognition accuracy decreases and the performance drops on the embedded equipment.To improve the performance and accuracy of the YOLOv4 algorithm in the embedded environment this work improved and optimized the YOLOv4 algorithm by integrating it with the K-means algorithm and the NVIDIA Tensor RT reasoner.(2)Because the software platforms of embedded devices and ground devices are usually different,it is necessary to develop a cross-platform transmission protocol for the devices.In order to ensure the reliability and programmability of the transmission,the protocol is based on the Netty framework and constructed with the TCP protocol to support features like breakpoint resume and delimitation.(3)Based on(1),(2)and related hardware platforms,a real-time target recognition system for UAV is designed and implemented with Java programming language,Spring framework,and Angular framework.The system supports features like image acquisition,target recognition,result transmission,result display,and data archiving.At the same time,the system can directly login through the acquired IP address,and simultaneously display the identified targets from multiple users and addresses.The hardware development of UAV and embedded system is completed,and the capability of the process-based target recognition system is verified through experiments.Also,this research focuses on the systematic research of real-time target recognition based on UAV,with the optimized YOLOv4 algorithm,the system can detect targets that are missed by the original algorithm.Additionally,the techniques of the systematic processing procedure developed in this study can be used in other UAV-based applications.Therefore,this study has a certain reference and application value in UAV-based real-time target recognitions.