没有合适的资源?快使用搜索试试~ 我知道了~
首页深度学习认知雷达:微无人机检测与分类技术
深度学习认知雷达:微无人机检测与分类技术
需积分: 18 5 下载量 165 浏览量
更新于2024-09-07
收藏 650KB PDF 举报
本文主要探讨了"雷达深度学习"在微小型无人驾驶航空器(Micro-UASs)检测与分类中的应用。作者Gihan J. Mendis、Jin Wei和Arjuna Madanayake来自美国阿克伦大学计算机与电气工程系,他们提出了一种智能认知雷达系统,该系统利用深度学习技术,特别是低复杂度的二值化深度信念网络(Binarized Deep Belief Network,BDNN)作为分类器。 在这个系统中,关键的创新在于设计了一个基于多普勒雷达解决方案的签名模式识别器。多普勒效应被用于捕捉UAS(无人驾驶航空器)螺旋桨运动产生的频谱特性,通过使用谱相关函数(Spectral Correlation Function,SCF),使得在噪声环境中也能生成可区分的特征。SCF的使用显著提高了信号处理的鲁棒性,即使在复杂的环境条件下也能确保精确的检测和分类。 实验部分特别关注了在UAS静止而螺旋桨旋转的情况下,如何利用BDNN进行高效计算。传统的多普勒雷达可能需要大量的浮点运算,如91600次乘法,但通过深度学习方法,该工作显著降低了计算成本,实现了更快速且准确的分析。这不仅有助于减少能源消耗,还提高了系统的实时性和响应能力。 这篇论文结合了雷达技术与深度学习的优势,展示了如何利用深度学习算法在微小无人机的监测领域实现智能、高效和噪声免疫的解决方案。这对于保障公共安全,如机场监控或城市空中交通管理具有重要的实际意义。随着微UAS技术的发展,这种智能雷达系统的应用前景十分广阔。
资源详情
资源推荐
Deep learning cognitive radar for Micro UAS
detection and classification
Gihan J. Mendis
Dept. of Computer and Electrical Engineering
The University of Akron
Akron, USA
ijm11@zips.uakron.edu
Jin Wei
Dept. of Computer and Electrical Engineering
The University of Akron
Akron, USA
jwei1@uakron.edu
Arjuna Madanayake
Dept. of Computer and Electrical Engineering
The University of Akron
Akron, USA
arjuna@uakron.edu
Abstract— In this paper, we propose an intelligent cognitive
radar system for detecting and classifying the micro unmanned
aerial systems (micro UASs). In this system, we design a low-
complexity binarized deep belief network (DBN) classifier that
recognizes the signature patterns generated by using a Doppler
radar based solution. To generate the distinguishable patterns, our
work employs the spectral correlation function (SCF) that is noise
resilient. In the experiment conducted, micro UASs are clamped
to be immobile while propellers are on motion. Doppler effects
caused by propeller motions of UASs are considered. By
employing our binarized DBN, the computationally costly 91600
floating point multiplication operations required in the original
DBN are represented by using zero computational cost no
connections, simple connections, negation operations, bit-shifting
operations, and bit-shifting with negation operations. In the
simulation section, we show that the proposed system gives more
than 90% accuracy in detecting the micro UASs in the
environments with SNR ≥ -3 dB AWGN noise. Furthermore, the
classification accuracy of different micro UASs remains more than
90% for environments with SNR ≥ 0 dB AWGN noise.
Keywords—micro UAS; drones; Doppler radar; spectral
correlation function; deep belief network, low complexity; deep
learning
I. INTRODUCTION
Unmanned aerial systems (UASs) are common as
surveillance devices in the military. Micro UASs are less
expensive UASs, which are smaller in dimensions and weight.
They are also referred as “drones” and are useful for a variety of
civilian applications including agriculture, package delivery,
wildlife monitoring, leisure activities etc. However, the small
size and the slow moving speed of micro UASs make them
undetectable from regular radar systems design to detect larger
and fast moving aerial systems. Wide availability and
undetectable nature of micro UASs introduce a new security
thread [1].
Some radar based systems have been developed recently to
detect and classify micro UASs. In [2] Shin et al., a K-band radar
system with fiber-optic links for detect micro UASs is
introduced. In [3] Drozdowicz et al. presented an experimental
system for the detection and tracking of micro UASs. In [4]
Jahangir et al. used 2-D L-Band receiver arrays to detect micro
UASs and machine learning decision tree classifier was used to
reject other targets. In our previous work, we proposed a
Doppler radar based method for detecting micro UASs [5-6].
Deep learning is an emerging area of machine learning that
empower recent achievements in machine intelligent. Deep
learning methods are artificial neural network (ANN) based
machine learning techniques with multiple layers of ANNs.
Deep learning methods are capable of learning suitable features
from raw data. Therefore, they are more effective in completing
complex tasks [7]. Deep learning methods have been used for
pattern recognition applications in various areas including
speech recognition, natural language processing, audio and
music processing, image recognition, and machine vision [8-
16].
In our previous work, we used a low cost 2.4 GHz
continuous-wave Doppler radar system. This system is built
using commercially available RF components alone with a
signal processing mechanism that use spectral correlation
function (SCF) to generated noise-resilient and distinguishable
2-D signature patterns and robust deep belief network (DBN)
deep learning method as the SCF signature pattern classification
method. The radar system was set up in a laboratory
environment, data were collected for 3-micro UASs, and
collected data were used to verify the signal processing
mechanism.
978-1-5386-3988-7/17/$31.00 ©2017 IEEE
下载后可阅读完整内容,剩余4页未读,立即下载
qq_36822222
- 粉丝: 0
- 资源: 2
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
- 李兴华Java基础教程:从入门到精通
- U盘与硬盘启动安装教程:从菜鸟到专家
- C++面试宝典:动态内存管理与继承解析
- C++ STL源码深度解析:专家级剖析与关键技术
- C/C++调用DOS命令实战指南
- 神经网络补偿的多传感器航迹融合技术
- GIS中的大地坐标系与椭球体解析
- 海思Hi3515 H.264编解码处理器用户手册
- Oracle基础练习题与解答
- 谷歌地球3D建筑筛选新流程详解
- CFO与CIO携手:数据管理与企业增值的战略
- Eclipse IDE基础教程:从入门到精通
- Shell脚本专家宝典:全面学习与资源指南
- Tomcat安装指南:附带JDK配置步骤
- NA3003A电子水准仪数据格式解析与转换研究
- 自动化专业英语词汇精华:必备术语集锦
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功