Internet of Things (IoT) has become a prevalent topic in the world of technology. It helps billion of devices to connect to the internet so that they can exchange data with each other. Nowadays, the IoT can be applied...Internet of Things (IoT) has become a prevalent topic in the world of technology. It helps billion of devices to connect to the internet so that they can exchange data with each other. Nowadays, the IoT can be applied in anything, from cellphones, coffee makers, cars, body sensors to smart surveillance, water distribution, energy management system, and environmental monitoring. However, the rapid growth of IoT has brought new and critical threats to the security and privacy of the users. Due to the millions of insecure IoT devices, an adversary can easily break into an application to make it unstable and steal sensitive user information and data. This paper provides an overview of different kinds of cybersecurity attacks against IoT devices as well as an analysis of IoT architecture. It then discusses the security solutions we can take to protect IoT devices against different kinds of security attacks. The main goal of this research is to enhance the development of IoT research by highlighting the different kinds of security challenges that IoT is facing nowadays, and the existing security solutions we can implement to make IoT devices more secure. In this study, we analyze the security solutions of IoT in three forms: secure authentication, secure communications, and application security to find suitable security solutions for protecting IoT devices.展开更多
Internet of Things (IoT) networks present unique cybersecurity challenges due to their distributed and heterogeneous nature. Our study explores the effectiveness of two types of deep learning models, long-term memory ...Internet of Things (IoT) networks present unique cybersecurity challenges due to their distributed and heterogeneous nature. Our study explores the effectiveness of two types of deep learning models, long-term memory neural networks (LSTMs) and deep neural networks (DNNs), for detecting attacks in IoT networks. We evaluated the performance of six hybrid models combining LSTM or DNN feature extractors with classifiers such as Random Forest, k-Nearest Neighbors and XGBoost. The LSTM-RF and LSTM-XGBoost models showed lower accuracy variability in the face of different types of attack, indicating greater robustness. The LSTM-RF and LSTM-XGBoost models show variability in results, with accuracies between 58% and 99% for attack types, while LSTM-KNN has higher but more variable accuracies, between 72% and 99%. The DNN-RF and DNN-XGBoost models show lower variability in their results, with accuracies between 59% and 99%, while DNN-KNN has higher but more variable accuracies, between 71% and 99%. LSTM-based models are proving to be more effective for detecting attacks in IoT networks, particularly for sophisticated attacks. However, the final choice of model depends on the constraints of the application, taking into account a trade-off between accuracy and complexity.展开更多
针对车联网中拒绝服务(denial of service,DoS)攻击难以防范且现有监督学习方法无法有效检测零日攻击的问题,提出了一种混合DoS攻击入侵检测系统.首先,对数据集进行预处理,提高数据的质量;其次,利用特征选择滤除冗余特征,旨在获得代表...针对车联网中拒绝服务(denial of service,DoS)攻击难以防范且现有监督学习方法无法有效检测零日攻击的问题,提出了一种混合DoS攻击入侵检测系统.首先,对数据集进行预处理,提高数据的质量;其次,利用特征选择滤除冗余特征,旨在获得代表性更强的特征;再次,采用集成学习方法将5种基于树结构的监督分类器堆叠集成用于检测已知DoS攻击;最后,提出了一种无监督异常检测方法,将卷积去噪自动编码器与注意力机制相结合来建立正常行为模型,用于检测堆叠集成模型漏报的未知DoS攻击.实验结果表明,对于已知DoS攻击检测,所提系统在Car-Hacking数据集和CICIDS2017数据集上的检测准确率分别为100%和99.967%;对于未知DoS攻击检测,所提系统在上述两个数据集上的检测准确率分别为100%和83.953%,并且在两个数据集上的平均测试时间分别为0.072 ms和0.157 ms,验证了所提系统的有效性和可行性.展开更多
在分布式物联网的大规模应用背景下,各实体设备中密码技术作为信息安全的底层支撑架构,正面临着侧信道攻击(SCA)这一物理层安全威胁的严峻挑战. SM4分组密码算法作为我国自主研制的商用密码算法标准,已深度集成于分布式物联网安全协议中...在分布式物联网的大规模应用背景下,各实体设备中密码技术作为信息安全的底层支撑架构,正面临着侧信道攻击(SCA)这一物理层安全威胁的严峻挑战. SM4分组密码算法作为我国自主研制的商用密码算法标准,已深度集成于分布式物联网安全协议中,但其实现层面的侧信道脆弱性问题亟待解决.针对SM4密钥扩展算法的侧信道攻击研究存在空白,现有攻击方法多依赖多能迹统计特性,而单能迹攻击研究匮乏.研究提出一种基于贝叶斯网络结合建模侧信道攻击的单能迹侧信道攻击方法,针对单条能量轨迹,通过构建概率图模型,结合置信传播算法,实现对轮子密钥的高效推测,进而恢复主密钥.仿真实验与实测实验表明该攻击方法有效,在理想实测环境下主密钥恢复成功率达85.74%,即使在实测能迹中添加大量高斯白噪声,使得信噪比仅为10 d B的条件下,成功率仍可达70%.与传统方法相比,所提方法在成功率、所需能量轨迹数量和攻击时间等方面优势显著,为分布式物联网系统含密设备的侧信道攻击研究提供了新的思路与技术手段,也为相关防护设计提供了理论依据和参考.展开更多
在异构接入与无线协议并存的物联网(Internet of Things,IoT)环境中,传统静态认证已难以构建有效防线。对此,分析ZigBee协议下中间人重放攻击的流程和漏洞,研究基于物理不可克隆函数熵源的动态密钥生成策略、多因子行为-环境融合认证模...在异构接入与无线协议并存的物联网(Internet of Things,IoT)环境中,传统静态认证已难以构建有效防线。对此,分析ZigBee协议下中间人重放攻击的流程和漏洞,研究基于物理不可克隆函数熵源的动态密钥生成策略、多因子行为-环境融合认证模型以及轻量化可信执行环境(Trusted Execution Environment,TEE)构建策略,提出零信任防御体系构建路径,实现认证过程动态化与计算区域隔离化。展开更多
文摘Internet of Things (IoT) has become a prevalent topic in the world of technology. It helps billion of devices to connect to the internet so that they can exchange data with each other. Nowadays, the IoT can be applied in anything, from cellphones, coffee makers, cars, body sensors to smart surveillance, water distribution, energy management system, and environmental monitoring. However, the rapid growth of IoT has brought new and critical threats to the security and privacy of the users. Due to the millions of insecure IoT devices, an adversary can easily break into an application to make it unstable and steal sensitive user information and data. This paper provides an overview of different kinds of cybersecurity attacks against IoT devices as well as an analysis of IoT architecture. It then discusses the security solutions we can take to protect IoT devices against different kinds of security attacks. The main goal of this research is to enhance the development of IoT research by highlighting the different kinds of security challenges that IoT is facing nowadays, and the existing security solutions we can implement to make IoT devices more secure. In this study, we analyze the security solutions of IoT in three forms: secure authentication, secure communications, and application security to find suitable security solutions for protecting IoT devices.
文摘Internet of Things (IoT) networks present unique cybersecurity challenges due to their distributed and heterogeneous nature. Our study explores the effectiveness of two types of deep learning models, long-term memory neural networks (LSTMs) and deep neural networks (DNNs), for detecting attacks in IoT networks. We evaluated the performance of six hybrid models combining LSTM or DNN feature extractors with classifiers such as Random Forest, k-Nearest Neighbors and XGBoost. The LSTM-RF and LSTM-XGBoost models showed lower accuracy variability in the face of different types of attack, indicating greater robustness. The LSTM-RF and LSTM-XGBoost models show variability in results, with accuracies between 58% and 99% for attack types, while LSTM-KNN has higher but more variable accuracies, between 72% and 99%. The DNN-RF and DNN-XGBoost models show lower variability in their results, with accuracies between 59% and 99%, while DNN-KNN has higher but more variable accuracies, between 71% and 99%. LSTM-based models are proving to be more effective for detecting attacks in IoT networks, particularly for sophisticated attacks. However, the final choice of model depends on the constraints of the application, taking into account a trade-off between accuracy and complexity.
文摘针对车联网中拒绝服务(denial of service,DoS)攻击难以防范且现有监督学习方法无法有效检测零日攻击的问题,提出了一种混合DoS攻击入侵检测系统.首先,对数据集进行预处理,提高数据的质量;其次,利用特征选择滤除冗余特征,旨在获得代表性更强的特征;再次,采用集成学习方法将5种基于树结构的监督分类器堆叠集成用于检测已知DoS攻击;最后,提出了一种无监督异常检测方法,将卷积去噪自动编码器与注意力机制相结合来建立正常行为模型,用于检测堆叠集成模型漏报的未知DoS攻击.实验结果表明,对于已知DoS攻击检测,所提系统在Car-Hacking数据集和CICIDS2017数据集上的检测准确率分别为100%和99.967%;对于未知DoS攻击检测,所提系统在上述两个数据集上的检测准确率分别为100%和83.953%,并且在两个数据集上的平均测试时间分别为0.072 ms和0.157 ms,验证了所提系统的有效性和可行性.
文摘在分布式物联网的大规模应用背景下,各实体设备中密码技术作为信息安全的底层支撑架构,正面临着侧信道攻击(SCA)这一物理层安全威胁的严峻挑战. SM4分组密码算法作为我国自主研制的商用密码算法标准,已深度集成于分布式物联网安全协议中,但其实现层面的侧信道脆弱性问题亟待解决.针对SM4密钥扩展算法的侧信道攻击研究存在空白,现有攻击方法多依赖多能迹统计特性,而单能迹攻击研究匮乏.研究提出一种基于贝叶斯网络结合建模侧信道攻击的单能迹侧信道攻击方法,针对单条能量轨迹,通过构建概率图模型,结合置信传播算法,实现对轮子密钥的高效推测,进而恢复主密钥.仿真实验与实测实验表明该攻击方法有效,在理想实测环境下主密钥恢复成功率达85.74%,即使在实测能迹中添加大量高斯白噪声,使得信噪比仅为10 d B的条件下,成功率仍可达70%.与传统方法相比,所提方法在成功率、所需能量轨迹数量和攻击时间等方面优势显著,为分布式物联网系统含密设备的侧信道攻击研究提供了新的思路与技术手段,也为相关防护设计提供了理论依据和参考.
文摘在异构接入与无线协议并存的物联网(Internet of Things,IoT)环境中,传统静态认证已难以构建有效防线。对此,分析ZigBee协议下中间人重放攻击的流程和漏洞,研究基于物理不可克隆函数熵源的动态密钥生成策略、多因子行为-环境融合认证模型以及轻量化可信执行环境(Trusted Execution Environment,TEE)构建策略,提出零信任防御体系构建路径,实现认证过程动态化与计算区域隔离化。
