Sensor Data-Driven UAV Anomaly Detection using Deep Learning Approach
Sensor Data-Driven UAV Anomaly Detection using Deep Learning Approach
复制标题
DOI：
10.1109/milcom52596.2021.9653036
复制
发表时间：
2021-11
期刊：
影响因子：
--
通讯作者：
Julio Galvan;A. Raja;Yanyan Li;Jiawei Yuan
中科院分区：
文献类型：
--
作者：
Julio Galvan;A. Raja;Yanyan Li;Jiawei Yuan
由于无人机（UAV）的高机动性和丰富的传感能力，它们越来越多地用于执行一系列军事和民用任务。与此同时，无人机也面临着各种安全问题，包括外部攻击和内部硬件/软件故障。因此，检测无人机的异常状态是保护其免受恶意对手攻击并防止潜在坠毁的关键任务。在本文中，我们提出了一种通过使用深度学习方法实时监测和分析传感器数据的无人机异常检测系统。该系统利用卷积神经网络（CNN）从原始传感器数据中自动提取和学习特征，然后对其进行处理以支持异常检测。我们使用我们的无人机网络安全仿真平台构建无人机IMU传感器数据的数据集，以支持我们的CNN模型的训练。本文还对不同的深度学习模型进行了评估和比较。我们验证了所提出的检测系统的性能，使用广泛的实验评估，这表明，我们的系统在不同的条件下实现了高的检测精度。
Thanks to the high mobility and rich sensing capabilities of unmanned aerial vehicles (UAVs), or drones, they are increasingly leveraged to perform a series of military and civilian tasks today. Meanwhile, UAVs are also facing various security and safety concerns raised by both external attacks and internal hardware/software failures. Therefore, detecting the abnormal status of a UAV is a critical task to protect it against malicious adversaries and prevent potential crashes. In this paper, we propose an anomaly detection system for UAVs by monitoring and analyzing their sensor data in real-time using deep learning approaches. The proposed system leverages the convolutional neural network (CNN) to extract and learn features automatically from raw sensor data and then process them to support anomaly detection. We construct a data set of UAV IMU sensor data using our UAV cybersecurity simulation platform to support the training of our CNN model. Different deep learning models are also evaluated and compared in this paper. We validate the performance of the proposed detection system using extensive experimental evaluation, which demonstrates that our system achieves high detection accuracy under different conditions.