Mutual relative localization and identification is an important feature for the stabilization and navigation of multi-Unmanned Aerial Vehicle (UAV) systems. Camera-based communications technology, also referred to as Optical Camera Communications (OCC) in the literature, is a novel approach that could bring a valuable solution to such a complex task. In such system, the UAVs are equipped with LEDs that act as beacons and with cameras allowing them to locate the LEDs of other UAVs. Specific blinking sequences are assigned to the LEDs of each of the UAVs in order to uniquely identify them. This camera-based relative localization and identification system is immune to Radio Frequency (RF) electromagnetic interference and operates in Global Navigation satellite (GNSS) denied environments. In addition, since many UAVs are already equipped with cameras, the implementation of this system is inexpensive. In this article, we study in detail the capacity of this system and its limitations. Furthermore, we show how to construct blinking sequences for UAV LEDs in order to improve system performance. Finally, experimental results are presented to corroborate the analytical derivations.
Sensing-Assisted LoS/NLoS Identification in Dynamic UAV Positioning Systems
Shandong University（山东大学）；Joint SDU-NTU Centre for Artificial Intelligence Research (C-FAIR)（山东大学-南洋理工大学联合人工智能研究中心）；School of Software, Shandong University（山东大学软件学院）；Peking University（北京大学）；State Key Laboratory of Photonics and Communications, School of Electronics, Peking University（北京大学电子工程学院光通信与光子学国家重点实验室）；Donghua University（东华大学）；School of Information and Intelligence Science, Donghua University（东华大学信息智能科学学部）
Wavelet Transform Analytics for RF-Based UAV Detection and Identification System Using Machine Learning
在这项工作中，我们通过使用机器学习算法和预先训练的基于卷积神经网络的算法SqueezeNet作为分类器，对无线干扰（如WiFi和蓝牙）下基于射频（RF）的无人机检测和识别系统（DDI）进行了全面的比较分析。在射频信号指纹识别研究中，信号的瞬态和稳态可以用来从射频信号中提取唯一的签名。通过利用无人机（UAV）的射频控制信号进行DDI，我们分别考虑了信号的每个状态进行特征提取，并比较了无人机检测和识别的利弊。使用各种小波变换（离散小波变换、连续小波变换和小波散射变换）从信号中提取特征，我们使用这些特征建立不同的模型。我们研究了这些模型在不同信噪比（SNR）水平下的性能。通过使用小波散射变换从30 dB SNR下的RF信号的稳态中提取特征（散点图），并使用这些散点图来训练SqueezeNet，我们在10 dB SNR下实现了98.9%的准确度。
