低空空域无人机实时智能检测系统设计与研究
李文明 1刘小虎 2黄章进2
作者信息
- 1. 安徽警官职业学院 警察系,安徽合肥 230031
- 2. 中国科学技术大学 计算机学院,安徽合肥 230026
- 折叠
摘要
针对低空空域无人机检测的识别率低、反应速度慢等问题,本文设计并试验了一套低空空域无人机实时智能检测系统,包括监控数据采集、视频数据处理、无人机数据集、预训练模型、迁移学习训练、训练结果分析、系统集成展示的完整流程,分析并确定了无人机检测场景的关键性能指标,对基于YOLOv5的模型参数量进行了估算,并在实际场景下的3D地图中展现了无人机检测和路径跟踪的过程.实验数据表明,在指定最低召回率为90％的条件下,置信度为53％,对应的精确率为96％,同时系统在校内测试环境下集成运行正常,满足低空空域无人机实时智能检测场景的要求.
Abstract
In response to the low detection rate and slow response time in low-altitude UAV detection,a real-time intelligent detection system for low-altitude UAVs has been designed and implemented.This system includes a complete workflow consisting of monitoring data collection,video data processing,UAV dataset creation,pre-trained models,transfer learning training,analysis of training results,and system integration and demonstration.Key performance indicators for UAV detection scenes were analyzed and determined.An estimation of the model parameter quantity based on YOLOv5 was conducted.Furthermore,the process of UAV detection and path tracking was demonstrated in a real-world scenario with a 3D map.Experimental data indicated that,under the condition of a specified minimum recall rate of 90％and a confidence level of 53％,the corresponding precision rate was 96％.Moreover,the system integrated and operated normally in a campus test environment,meeting the requirements of real-time intelligent detection for low-altitude UAV scenes.
关键词
无人机检测/YOLOv5/深度学习/迁移学习/召回率Key words
UAV detection/YOLOv5/deep learning/transfer learning/recall rate引用本文复制引用
基金项目
2021年度安徽省高校自然科学研究项目(KJ2021A1470)
出版年
2024
NETL
NSTL
万方数据