基于改进YOLO的飞机起降阶段跟踪方法  

Tracking method of aircraft in take-off and landing phase based on improved YOLO

作　　者：郭晓静 李欣 隋昊达 GUO Xiao-jing;LI Xin;SUI Hao-da(College of Electronic Information and Automation,Civil Aviation University of China,Tianjin 300300,China;College of Computer Science and Technology,Civil Aviation University of China,Tianjin 300300,China)

机构地区：[1]中国民航大学电子信息与自动化学院,天津300300 [2]中国民航大学计算机科学与技术学院,天津300300

出　　处：《计算机工程与设计》2022年第11期3250-3256,共7页Computer Engineering and Design

基　　金：中央高校基本科研基金项目(3122020026)。

摘　　要：针对起降阶段飞机目标识别跟踪中人工与雷达配合检测效率低的问题,提出一种轻量化单目标识别跟踪模型。引入Mobile Ntv3e改进YOLOv4的特征提取网络,利用基于马尔科夫链蒙特卡罗采样的K-means++算法对自建数据集标注的边界框尺寸信息进行聚类分析。将数据集按7:3进行样本训练与测试,实现根据飞机垂尾处的公司标记自动分类识别、跟踪。实验结果表明,改进后算法的准确率达到94.29%,每秒处理的帧数达到67f/s,速度相比原来的YOLOv4算法提高了35f/s,提升了检测效率,具有较好的理论价值与应用前景。Aiming at the low detection efficiency of manual and radar cooperation in aircraft target recognition and tracking during take-off and landing stage,a lightweight single target recognition and tracking model was proposed.MobileNetv3 was introduced to improve the structure of YOLOv4.K-means++algorithm,which was based on Markov Chain Monte Carlo sampling,the bounding box size information marked by the self-built data set was clustered.The data set was 73 for sample training and testing to realize automatic classification,identification and tracking according to the company mark at the vertical tail of aircraft.Experimental results show that the accuracy of the improved algorithm reaches 94.29%,and the number of frames per second processed reaches 67 f/s.It increases by 35 f/s in comparison with original YOLOv4 algorithm.The detection efficiency is improved,which has better theoretical value and application prospects.

关 键 词：飞机目标识别 目标跟踪 YOLO算法 深度可分离卷积 聚类分析 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]