引用本文:石鹏程,赵振根,李庆龙.深度置信网络在四旋翼无人机传感器攻击检测中的应用[J].控制理论与应用,2026,43(4):774~782.[点击复制]
SHI Peng-cheng,ZHAO Zhen-gen,LI Qing-long.Application of deep belief network in sensor attack detection for quadrotor UAVs[J].Control Theory & Applications,2026,43(4):774~782.[点击复制]
深度置信网络在四旋翼无人机传感器攻击检测中的应用
Application of deep belief network in sensor attack detection for quadrotor UAVs
摘要点击 183  全文点击 26  投稿时间：2024-04-03  修订日期：2025-08-02
查看全文  查看/发表评论  下载PDF阅读器   HTML
DOI编号  10.7641/CTA.2024.40199
  2026,43(4):774-782
中文关键词  扩展卡尔曼滤波器  深度置信网络  攻击检测  四旋翼无人机  自适应滤波
英文关键词  quadrotor UAV  extended Kalman filter  deep belief network  attack detection  adaptive filtering
基金项目  国家自然科学基金基础科学中心项目(62388101), 国家自然科学基金面上项目(62473195), 国家自然科学基金重点项目(62233009), 中国博士后 科学基金面上项目(2021M701701), 中央高校基本科研业务费专项资金项目(NS2024017)资助.
作者单位E-mail
石鹏程 南京航空航天大学 自动化学院 spc0506@nuaa.edu.cn 
赵振根* 南京航空航天大学自动化学院 zhaozhengen@nuaa.edu.cn 
李庆龙 南京航空航天大学 自动化学院  
中文摘要
      为了实现四旋翼无人机的传感器攻击快速准确检测, 本文提出了一种基于状态估计和深度学习的攻击检 测算法. 首先, 算法利用扩展卡尔曼滤波器(EKF)估计无人机状态, 并从传感器测量中提取特征信息. 接着, 采用滑 动时序窗口构建检测信息, 并通过深度置信网络(DBN)建立检测信息与传感器状态(是否受攻击)之间的非线性映射 关系. EKF简化了传感器状态检测信息的获取过程, 而DBN准确拟合了复杂的非线性关系, 从而显著提高了检测精 度. 为增强状态估计的可靠性, 本文还设计了一种自适应EKF算法, 能够在检测到传感器攻击时动态调整测量噪声 的协方差矩阵. 仿真结果表明, 所提出的EKF-DBN检测算法在准确率和检测效率上优于传统方法.
英文摘要
      To achieve fast and accurate detection of sensor attacks on quadrotor UAVs, this paper proposes an attack detection algorithm based on state estimation and deep learning. Firstly, the algorithm uses the extended Kalman filter (EKF) to estimate the UAV’s state and extract feature information from sensor measurements. Then, a sliding temporal window is applied to construct detection information, and a deep belief network (DBN) is used to establish a nonlinear mapping between the detection information and the sensor state (whether under attack). EKF simplifies the acquisition of sensor state detection information, while DBN accurately fits the complex nonlinear relationship, significantly improving detection accuracy. Furthermore, an adaptive EKF algorithm is designed to dynamically adjust the measurement noise covariance matrix upon detecting a sensor attack, enhancing the reliability of state estimation. Simulation results show that the proposed EKF-DBN detection algorithm outperforms traditional methods in terms of accuracy and detection efficiency.