基于仿真数据迁移学习的固定翼无人机检测  

Fixed-Wing UAV Detection Based on Simulated Data Transfer Learning

作　　者：付玉[1,2] 张垚 赵萌[1,2] 王绵沼 郑江鹏[1,2] 贾晨 陈胜勇 Fu Yu;Zhang Yao;Zhao Meng;Wang Mianzhao;Zheng Jiangpeng;Jia Chen;Chen Shengyong(Engineering Research Center of Learning-Based Intelligent System,Ministry of Education,Tianjin 300384,China;School of Computer Science and Engineering,Tianjin University of Technology,Tianjin 300384,China)

机构地区：[1]学习型智能系统教育部工程研究中心,天津300384 [2]天津理工大学计算机科学与工程学院,天津300384

出　　处：《系统仿真学报》2023年第5期998-1007,共10页Journal of System Simulation

基　　金：国家自然科学基金(61906133,62020106004,61903275,61902078)。

摘　　要：数据在视觉检测任务中发挥重要作用,针对足够数量的真实固定翼无人机数据难以获取的问题,构建了一个包含大量仿真和少量真实的固定翼无人机数据集,采用权重迁移的思想,通过对仿真固定翼无人机数据的训练达到对真实固定翼无人机数据的检测。在此基础上又提出一个两阶段学习策略,利用多尺度特征融合进一步降低无人机的漏检率。仿真实验结果表明,利用仿真数据检测真实固定翼无人机在未来目标检测研究中有潜在应用前景。Data play an important role in visual inspection tasks,but it is difficult to obtain a sufficient amount of real fixed-wing UAV data.Therefore,a data set containing a large number of simulated fixedwing UAV data and a small number of real fixed-wing UAV data is constructed,and the real fixed-wing UAV data are detected by training simulated fixed-wing UAV data based on the idea of weight transfer.On this basis,a two-stage learning strategy is proposed to further reduce the missed detection rate of UAVs by using multi-scale feature fusion.The simulation results show that simulated data can be used to detect real fixed-wing UAVs,which has potential application prospects in future target detection research.

关 键 词：视觉检测 固定翼无人机 权重迁移 多尺度特征 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]