"基于深度学习的云端车辆网络物理入侵检测系统"

需积分: 0 200 浏览量更新于2024-01-05 收藏 11.86MB PDF 举报

"基于深度学习的基于云的车辆网络物理入侵检测"是一项研究领域，旨在利用深度学习和云计算技术来识别和防止车辆网络系统中的物理入侵行为。本文使用了一种名为Cloud-Based Cyber-Physical Intrusion Detection的方法，该方法利用云计算来分析收集到的数据，并运用深度学习算法来检测潜在的入侵行为。该研究是由来自格林威治大学的George Loukas、Tuan Vuong、Ryan Heartfield、Georgia Sakellari、Yongpil Yoon和Diane Gan等人共同完成的。他们旨在解决车辆网络系统中可能面临的安全风险和威胁问题，例如黑客攻击和恶意软件的入侵。通过采用深度学习算法，这项研究提供了一种有效的方法来监测和检测车辆网络系统中的潜在入侵行为。深度学习是一种机器学习技术，它通过模拟人脑神经网络的运行方式来进行数据分析和模式识别。借助这种技术，研究人员能够对大量的数据进行自动学习，并能够识别出异常行为和潜在的入侵活动。另外，云计算技术在这项研究中也扮演了重要的角色。利用云计算平台，研究人员能够更加高效地处理和分析大量的数据，并能够实时监测和防御潜在的入侵行为。云计算技术还能够帮助车辆网络系统灵活扩展，适应不断增长的数据量和计算需求。研究结果表明，该方法在车辆网络系统中的物理入侵检测方面取得了良好的效果。通过分析大量的车辆网络数据，深度学习算法能够准确地识别出异常行为和入侵活动，并能够及时采取措施来阻止和防御这些威胁。这项研究对于车辆网络系统的安全具有重要的意义。通过使用深度学习和云计算技术，车辆网络系统能够更好地抵御黑客攻击和恶意软件的入侵，并能够保护车辆和驾乘者的安全和隐私。总之，基于深度学习的基于云的车辆网络物理入侵检测是一项具有重要意义的研究领域。通过利用深度学习和云计算技术，研究人员能够有效地监测和防御车辆网络系统中的物理入侵行为。这项研究对于车辆网络系统的安全和可靠性具有重要的意义，能够提高车辆和驾乘者的安全和隐私保护水平。

G. Loukas et al.: Cloud-Based Cyber-Physical Intrusion Detection for Vehicles Using Deep Learning

ad hoc networks for the communication between driverless

automobiles, raising the question of what happens when one

of the vehicles misbehaves and launches a cyber attack on

other vehicles on the same network. In this initial work, they

have used NS-2 and mobility simulation tools to evaluate

the performance of an intrusion detection system based on

artiﬁcial neural networks. Its input features were some of the

ones typically used in standard network intrusion detection

systems, such as payload sizes, hop counts etc. The same

authors have extended their work to take into account magne-

tometer sensors [32] and gyroscopes [33], and address grey

hole attacks [34].

A common characteristic of all detection mechanisms pro-

vided above, whether local or distributed across multiple

vehicles, is a focus on minimising the processing load, either

by applying lightweight techniques from statistics or by pre-

deﬁning simple behavioural rules that are easy to monitor.

This is because they are all limited by the onboard capabil-

ities of the vehicle at hand. As a result, they usually can-

not leverage modern classiﬁcation techniques, such as those

currently developed in the ﬁeld of deep learning. In effect,

the stronger the detection algorithm, the greater the energy

consumption, and in turn the less attractive a solution is for

a resource-constrained vehicle. To overcome this limitation,

we turn to the emerging ﬁeld of cloud robotics [35]. Our

proposal is to ofﬂoad the bulk of the processing required to

beneﬁt from deep learning to a more powerful infrastructure

(whether a single server, cloudlet or cloud). By computa-

tion ofﬂoading, we refer to the process of executing certain

computational tasks on more resourceful computers which

are not in the user’s immediate computing environment. The

concept has similarities with the online forensics techniques

used for cloud-based detection of malware and tainted data

on Android smartphones [36], [37]. However, instead of

crowd-sourcing detection, we focus on utilising computa-

tional ofﬂoading to allow access to deep learning based tech-

niques without the processing and energy cost which would

otherwise be prohibitive for a vehicle. Figure 1 illustrates

the conceptual difference between onboard and ofﬂoaded

intrusion detection for a vehicle. In both cases, data collec-

tion and aggregation occurs on the vehicle. In the onboard

case, the reasoning (the analysis of the data to determine

whether there is an attack or not) is also on the vehicle. In the

ofﬂoaded case, the data aggregated onboard are sent via a

network to a cloud infrastructure or equivalent to perform the

reasoning.

Over the past couple of years, the growing maturity in

deep learning algorithms has led to wider use outside of

its traditional applications in image and natural language

processing, e.g. in detecting malware [38] or rogue certiﬁ-

cates from trusted certiﬁcate authorities [39]. It has also been

used to improve intrusion detection accuracy in traditional

computer networks [40], [41], but not for cyber-physical

systems, such as vehicles. An exception is the recent work by

Kang et al. [42], which is geared towards the automotive

industry and detection of attacks on CAN bus. While a

FIGURE 1. Conceptual comparison between onboard and offloaded

intrusion detection for a vehicle.

promising start, the particular work is limited to a generic

command injection attack, which is detected by monitoring

a single type of data source and using a simple and generalist

deep neural network architecture, which does not account for

the overall state of the vehicle through multiple features or for

temporal information (the fact that the impact of the attack

changes over time during the attack). Also, it has been evalu-

ated only in simulation.

III. CONTRIBUTIONS

Here, we progress considerably beyond the state of the art

with the following key contributions:

• We design and evaluate two neural network architectures

for the real-time analysis of multiple sources of data col-

lected periodically onboard the vehicle and representing

both its cyber and physical processes.

• We produce a prototype implementation of deep learn-

ing based intrusion detection for cyber-physical attacks

on a real robotic vehicle, tested for three different types

of attacks and compared against the best-performing

generalist machine learning techniques typically used in

intrusion detection.

• We evaluate both experimentally and using a mathemat-

ical model the practicality of a computational ofﬂoad-

ing conﬁguration for providing resource-constrained

cyber-physical systems with access to high-end intru-

sion detection.

Note that an early version of this work has been included in

the second author’s PhD dissertation [43]. Here, we expand

through evaluation of two deep learning approaches, a more

practical setup with a single deep learning model for all

attacks rather than individual for each attack, as well as eval-

uation on an unseen attack and an updated literature review.

3494 VOLUME 6, 2018

剩余17页未读，继续阅读

俞林鑫

粉丝: 19
资源: 288

"基于深度学习的云端车辆网络物理入侵检测系统"

使用深度学习的基于云的实时网络入侵检测-王雪妍1

基于深度神经网络的车载网络安全入侵检测新方法-王雪妍1

使用区块链保护个人数据-王雪妍1

Saduj-OS:2021操作系统课程作业

风华产业论文最终版.doc

lab7-os

传媒行业：2月中国手游市场规模增幅明显，IP价值有望继续提升.pdf

2021六（3）班语文学习总结.docx

精通J2EE—Eclipse Struts Hibernate Spring 整合应用案例源代码1

医院后台管理系统-医院后台管理系统-医院后台管理系统-医院后台管理系统-医院后台管理系统-医院后台管理系统-医院后台管理系统-医

最新资源