Biblio

In recent years, in order to continuously promote the construction of safe cities, security monitoring equipment has been widely used all over the country. How to use computer vision technology to realize effective intelligent analysis of violence in video surveillance is very important to maintain social stability and ensure people's life and property safety. Video surveillance system has been widely used because of its intuitive and convenient advantages. However, the existing video monitoring system has relatively single function, and generally only has the functions of monitoring video viewing, query and playback. In addition, relevant researchers pay less attention to the complex abnormal behavior of violence, and relevant research often ignores the differences between violent behaviors in different scenes. At present, there are two main problems in video abnormal behavior event detection: the video data of abnormal behavior is less and the definition of abnormal behavior in different scenes cannot be clearly distinguished. The main existing methods are to model normal behavior events first, and then define videos that do not conform to the normal model as abnormal, among which the learning method of video space-time feature representation based on deep learning shows a good prospect. In the face of massive surveillance videos, it is necessary to use deep learning to identify violent behaviors, so that the machine can learn to identify human actions, instead of manually monitoring camera images to complete the alarm of violent behaviors. Network training mainly uses video data set to identify network training.

With the rapid development of multimedia and short video, there is a growing concern for video copyright protection. Some work has been proposed to add some copyright or fingerprint information to the video to trace the source of the video when it is stolen and protect video copyright. This paper proposes a video watermarking method based on a deep neural network and curriculum learning for watermarking of sliced videos. The first frame of the segmented video is perturbed by an encoder network, which is invisible and can be distinguished by the decoder network. Our model is trained and tested on an online educational video dataset consisting of 2000 different video clips. Experimental results show that our method can successfully discriminate most watermarked and non-watermarked videos with low visual disturbance, which can be achieved even under a relatively high video compression rate(H.264 video compress with CRF 32).

The requirements of much larger file sizes, different storage formats, and immersive viewing conditions pose significant challenges to the goals of compressing VR content. At the same time, the great potential of deep learning to advance progress on the video compression problem has driven a significant research effort. Because of the high bandwidth requirements of VR, there has also been significant interest in the use of space-variant, foveated compression protocols. We have integrated these techniques to create an end-to-end deep learning video compression framework. A feature of our new compression model is that it dispenses with the need for expensive search-based motion prediction computations by using displaced frame differences. We also implement foveation in our learning based approach, by introducing a Foveation Generator Unit (FGU) that generates foveation masks which direct the allocation of bits, significantly increasing compression efficiency while making it possible to retain an impression of little to no additional visual loss given an appropriate viewing geometry. Our experiment results reveal that our new compression model, which we call the Foveated MOtionless VIdeo Codec (Foveated MOVI-Codec), is able to efficiently compress videos without computing motion, while outperforming foveated version of both H.264 and H.265 on the widely used UVG dataset and on the HEVC Standard Class B Test Sequences.

Real-time data transmissions from a vehicle enhance road safety and traffic efficiency by aggregating data in a central server for data analytics. When drivers share their instantaneous vehicular information for a service provider to perform a legitimate task, a curious service provider may also infer private information it has not been authorized for. In this paper, we propose a privacy preservation framework based on the Hilbert Schmidt Independence Criterion (HSIC) to sanitize driving data to protect the vehicle’s trajectory from adversarial inference while ensuring the data is still useful for driver behavior detection. We develop a deep learning model to learn the HSIC sanitizer and demonstrate through two datasets that our approach achieves better utility-privacy trade-offs when compared to three other benchmarks.

This paper mainly explores the detection and defense of DDoS attacks in the SDN architecture of the 5G environment, and proposes a DDoS attack detection method based on the deep learning two-level model CNN-LSTM in the SDN network. Not only can it greatly improve the accuracy of attack detection, but it can also reduce the time for classifying and detecting network traffic, so that the transmission of DDoS attack traffic can be blocked in time to ensure the availability of network services.

Network security is a prominent topic that is gaining international attention. Distributed Denial of Service (DDoS) attack is often regarded as one of the most serious threats to network security. Software Defined Network (SDN) decouples the control plane from the data plane, which can meet various network requirements. But SDN can also become the object of DDoS attacks. This paper proposes an automated DDoS attack mitigation method that is based on the programmability of the Ryu controller and the features of the OpenFlow switch flow tables. The Mininet platform is used to simulate the whole process, from SDN traffic generation to using a K-Nearest Neighbor model for traffic classification, as well as identifying and mitigating DDoS attack. The packet counts of the victim's malicious traffic input port are significantly lower after the mitigation method is implemented than before the mitigation operation. The purpose of mitigating DDoS attack is successfully achieved.

With the advent of the 5G era, high-speed and secure network access services have become a common pursuit. The QUIC (Quick UDP Internet Connection) protocol proposed by Google has been studied by many scholars due to its high speed, robustness, and low latency. However, the research on the security of the QUIC protocol by domestic and foreign scholars is insufficient. Therefore, based on the self-similarity of QUIC network traffic, combined with traffic characteristics and signal processing methods, a QUIC-based network traffic anomaly detection model is proposed in this paper. The model decomposes and reconstructs the collected QUIC network traffic data through the Empirical Mode Decomposition (EMD) method. In order to judge the occurrence of abnormality, this paper also intercepts overlapping traffic segments through sliding windows to calculate Hurst parameters and analyzes the obtained parameters to check abnormal traffic. The simulation results show that in the network environment based on the QUIC protocol, the Hurst parameter after being attacked fluctuates violently and exceeds the normal range. It also shows that the anomaly detection of QUIC network traffic can use the EMD method.

Model compression is one of the most preferred techniques for efficiently deploying deep neural networks (DNNs) on resource- constrained Internet of Things (IoT) platforms. However, the simply compressed model is often vulnerable to adversarial attacks, leading to a conflict between robustness and efficiency, especially for IoT devices exposed to complex real-world scenarios. We, for the first time, address this problem by developing a novel framework dubbed Magical-Decomposition to simultaneously enhance both robustness and efficiency for hardware. By leveraging a hardware-friendly model compression method called singular value decomposition, the defending algorithm can be supported by most of the existing DNN hardware accelerators. To step further, by using a recently developed DNN interpretation tool, the underlying scheme of how the adversarial accuracy can be increased in the compressed model is highlighted clearly. Ablation studies and extensive experiments under various attacks/models/datasets consistently validate the effectiveness and scalability of the proposed framework.

This article proposes a health monitoring system platform for cross-river bridges based on big data. The system can realize regionalized bridge operation and maintenance management. The system has functions such as registration modification and deletion of sensor equipment, user registration modification and deletion, real-time display and storage of sensor monitoring data, and evaluation and early warning of bridge structure safety. The sensor is connected to the lower computer through the serial port, analog signal, fiber grating signal, etc. The lower computer converts a variety of signals into digital signals through the single-chip A/D sampling and demodulator, etc., and transmits it to the upper computer through the serial port. The upper computer uses ARMCortex-A9 Run the main program to realize multi-threaded network communication. The system platform is to test the validity of the model, and a variety of model verification methods are used for evaluation to ensure the reliability of the big data analysis method.

Missing values are an unavoidable problem for classification tasks of machine learning in medical data. With the rapid development of the medical system, large scale medical data is increasing. Missing values increase the difficulty of mining hidden but useful information in these medical datasets. Deletion and imputation methods are the most popular methods for dealing with missing values. Existing studies ignored to compare and discuss the deletion and imputation methods of missing values under the row missing rate and the total missing rate. Meanwhile, they rarely used experiment data sets that are mixed-type and large scale. In this work, medical data sets of various sizes and mixed-type are used. At the same time, performance differences of deletion and imputation methods are compared under the MCAR (Missing Completely At Random) mechanism in the baseline task using LR (Linear Regression) and SVM (Support Vector Machine) classifiers for classification with the same row and total missing rates. Experimental results show that under the MCAR missing mechanism, the performance of two types of processing methods is related to the size of datasets and missing rates. As the increasing of missing rate, the performance of two types for processing missing values decreases, but the deletion method decreases faster, and the imputation methods based on machine learning have more stable and better classification performance on average. In addition, small data sets are easily affected by processing methods of missing values.

Functional dependencies (FDs) are widely applied in data management tasks. Since FDs on data are usually unknown, FD discovery techniques are studied for automatically finding hidden FDs from data. In this paper, we develop techniques to dynamically discover FDs in response to changes on data. Formally, given the complete set Σ of minimal and valid FDs on a relational instance r, we aim to find the complete set Σ$^\textrm\textbackslashprime$ of minimal and valid FDs on røplus\textbackslashDelta r, where \textbackslashDelta r is a set of tuple insertions and deletions. Different from the batch approaches that compute Σ$^\textrm\textbackslashprime$ on røplus\textbackslashDelta r from scratch, our dynamic method computes Σ$^\textrm\textbackslashprime$ in response to \textbackslashtriangle\textbackslashuparrow. by leveraging the known Σ on r, and avoids processing the whole of r for each update from \textbackslashDelta r. We tackle dynamic FD discovery on røplus\textbackslashDelta r by dynamic hitting set enumeration on the difference-set of røplus\textbackslashDelta r. Specifically, (1) leveraging auxiliary structures built on r, we first present an efficient algorithm to update the difference-set of r to that of røplus\textbackslashDelta r. (2) We then compute Σ$^\textrm\textbackslashprime$, by recasting dynamic FD discovery as dynamic hitting set enumeration on the difference-set of røplus\textbackslashDelta r and developing novel techniques for dynamic hitting set enumeration. (3) We finally experimentally verify the effectiveness and efficiency of our approaches, using real-life and synthetic data. The results show that our dynamic FD discovery method outperforms the batch counterparts on most tested data, even when \textbackslashDelta r is up to 30 % of r.

At present, cloud service providers control the direct management rights of cloud data, and cloud data cannot be effectively and assured deleted, which may easily lead to security problems such as data residue and user privacy leakage. This paper analyzes the related research work of cloud data assured deletion in recent years from three aspects: encryption key deletion, multi-replica association deletion, and verifiable deletion. The advantages and disadvantages of various deletion schemes are analysed in detail, and finally the prospect of future research on assured deletion of cloud data is given.

Data secure deletion operation in storage media is an important function of data security management. The internal physical properties of SSDs are different from hard disks, and data secure deletion of disks can not apply to SSDs directly. Copyback operation is used to improve the data migration performance of SSDs but is rarely used due to error accumulation issue. We propose a data securely deletion algorithm based on copyback operation, which improves the efficiency of data secure deletion without affecting the reliability of data. First, this paper proves that the data secure delete operation takes a long time on the channel bus, increasing the I/O overhead, and reducing the performance of the SSDs. Secondly, this paper designs an efficient data deletion algorithm, which can process read requests quickly. The experimental results show that the proposed algorithm can reduce the response time of read requests by 21% and the response time of delete requests by 18.7% over the existing algorithm.

Steady advancement in Artificial Intelligence (AI) development over recent years has caused AI systems to become more readily adopted across industry and military use-cases globally. As powerful as these algorithms are, there are still gaping questions regarding their security and reliability. Beyond adversarial machine learning, software supply chain vulnerabilities and model backdoor injection exploits are emerging as potential threats to the physical safety of AI reliant CPS such as autonomous vehicles. In this work in progress paper, we introduce the concept of AI supply chain vulnerabilities with a provided proof of concept autonomous exploitation framework. We investigate the viability of algorithm backdoors and software third party library dependencies for applicability into modern AI attack kill chains. We leverage an autonomous vehicle case study for demonstrating the applicability of our offensive methodologies within a realistic AI CPS operating environment.

To solve the problem of an excessive number of component vulnerabilities and limited defense resources in industrial cyber physical systems, a method for analyzing security critical components of system is proposed. Firstly, the components and vulnerability information in the system are modeled based on SysML block definition diagram. Secondly, as SysML block definition diagram is challenging to support direct analysis, a block security dependency graph model is proposed. On this basis, the transformation rules from SysML block definition graph to block security dependency graph are established according to the structure of block definition graph and its vulnerability information. Then, the calculation method of component security importance is proposed, and a security critical component analysis tool is designed and implemented. Finally, an example of a Drone system is given to illustrate the effectiveness of the proposed method. The application of this method can provide theoretical and technical support for selecting key defense components in the industrial cyber physical system.

Unsupervised cross-domain NER task aims to solve the issues when data in a new domain are fully-unlabeled. It leverages labeled data from source domain to predict entities in unlabeled target domain. Since training models on large domain corpus is time-consuming, in this paper, we consider an alternative way by introducing syntactic dependency structure. Such information is more accessible and can be shared between sentences from different domains. We propose a novel framework with dependency-aware GNN (DGNN) to learn these common structures from source domain and adapt them to target domain, alleviating the data scarcity issue and bridging the domain gap. Experimental results show that our method outperforms state-of-the-art methods.

In recent years, with the occurrence of climate change and various extreme events, the research on the resistance of physical information systems to large-scale complex faults is of great significance. Propose a power information system to deal with complex faults in extreme weather, establish an anti-interference framework, construct a regional anti-interference strategy based on regional load output matching and topological connectivity, and propose branch active power adjustment methods to reduce disasters. In order to resist the risk of system instability caused by overrun of branch power and phase disconnection, the improved IEEE33 node test system simulation shows that this strategy can effectively reduce the harm of large-scale and complex faults.

Cyber threats can cause severe damage to computing infrastructure and systems as well as data breaches that make sensitive data vulnerable to attackers and adversaries. It is therefore imperative to discover those threats and stop them before bad actors penetrating into the information systems.Threats hunting algorithms based on machine learning have shown great advantage over classical methods. Reinforcement learning models are getting more accurate for identifying not only signature-based but also behavior-based threats. Quantum mechanics brings a new dimension in improving classification speed with exponential advantage. The accuracy of the AI/ML algorithms could be affected by many factors, from algorithm, data, to prejudicial, or even intentional. As a result, AI/ML applications need to be non-biased and trustworthy.In this research, we developed a machine learning-based cyber threat detection and assessment tool. It uses two-stage (both unsupervised and supervised learning) analyzing method on 822,226 log data recorded from a web server on AWS cloud. The results show the algorithm has the ability to identify the threats with high confidence.

Networked cyber-physical systems must balance the utility of communication for monitoring and control with the risks of revealing private information. Many of these networks, such as wireless communication, are vulnerable to eavesdrop-ping by illegitimate recipients. Obfuscation can hide information from eaves-droppers by ensuring their observations are ambiguous or misleading. At the same time, coordination with recipients can enable them to interpret obfuscated data. In this way, we propose an obfuscation framework for dynamic systems that ensures privacy against eavesdroppers while maintaining utility for legitimate recipients. We consider eavesdroppers unaware of obfuscation by requiring that their observations are consistent with the original system, as well as eaves-droppers aware of the goals of obfuscation by assuming they learn of the specific obfuscation implementation used. We present a method for bounded synthesis of solutions based upon distributed reactive synthesis and the synthesis of publicly-known obfuscators.

Embedded devices are becoming increasingly pervasive in safety-critical systems of the emerging cyber-physical world. While trusted execution environments (TEEs), such as ARM TrustZone, have been widely deployed in mobile platforms, little attention has been given to deployment on real-time cyber-physical systems, which present a different set of challenges compared to mobile applications. For safety-critical cyber-physical systems, such as autonomous drones or automobiles, the current TEE deployment paradigm, which focuses only on confidentiality and integrity, is insufficient. Computation in these systems also needs to be completed in a timely manner (e.g., before the car hits a pedestrian), putting a much stronger emphasis on availability.To bridge this gap, we present RT-TEE, a real-time trusted execution environment. There are three key research challenges. First, RT-TEE bootstraps the ability to ensure availability using a minimal set of hardware primitives on commodity embedded platforms. Second, to balance real-time performance and scheduler complexity, we designed a policy-based event-driven hierarchical scheduler. Third, to mitigate the risks of having device drivers in the secure environment, we designed an I/O reference monitor that leverages software sandboxing and driver debloating to provide fine-grained access control on peripherals while minimizing the trusted computing base (TCB).We implemented prototypes on both ARMv8-A and ARMv8-M platforms. The system is tested on both synthetic tasks and real-life CPS applications. We evaluated rover and plane in simulation and quadcopter both in simulation and with a real drone.

A new privacy Laplace common recognition algorithm is designed to protect users’ privacy data in this paper. This algorithm disturbs state transitions and information generation functions using exponentially decaying Laplace noise to avoid attacks. The mean square consistency and privacy protection performance are further studied. Finally, the theoretical results obtained are verified by performing numerical simulations.

The value and size of information exchanged through dark-web pages are remarkable. Recently Many researches showed values and interests in using machine-learning methods to extract security-related useful knowledge from those dark-web pages. In this scope, our goals in this research focus on evaluating best prediction models while analyzing traffic level data coming from the dark web. Results and analysis showed that feature selection played an important role when trying to identify the best models. Sometimes the right combination of features would increase the model’s accuracy. For some feature set and classifier combinations, the Src Port and Dst Port both proved to be important features. When available, they were always selected over most other features. When absent, it resulted in many other features being selected to compensate for the information they provided. The Protocol feature was never selected as a feature, regardless of whether Src Port and Dst Port were available.

Physical-layer key (PLK) generation scheme is a new key generation scheme based on wireless channel reciprocity. However, the security of physical layer keys still lacks sufficient theoretical support in the presence of eavesdropping attacks until now, which affects the promotion in practical applications. By analyzing the propagation mode of multipath signals under non-line-of-sight (nLoS), an improved spatial cross-correlation model is constructed, where the spatial cross-correlation is between eavesdropping channel and legitimate channel. Results show that compared with the multipath and obstacle distribution of the channel, the azimuth and distance between the eavesdropper and the eavesdropped user have a greater impact on the cross-correlation.

With the intelligent development of power system, due to the double-layer structure of smart grid and the characteristics of failure propagation across layers, the attack path also changes significantly: from single-layer to multi-layer and from static to dynamic. In response to the shortcomings of the single-layer attack path of traditional attack path identification methods, this paper proposes the idea of cross-layer attack, which integrates the threat propagation mechanism of the information layer and the failure propagation mechanism of the physical layer to establish a forward-backward bi-directional detection model. The model is mainly used to predict possible cross-layer attack paths and evaluate their path generation probabilities to provide theoretical guidance and technical support for defenders. The experimental results show that the method proposed in this paper can well identify the dynamic cross-layer attacks in the smart grid.

Since the cyber and physical layers in the distribution system are deeply integrated, the traditional distribution system has gradually developed into the cyber-physical distribution system (CPDS), and the failures of the cyber layer will affect the reliable and safe operation of the whole distribution system. Therefore, this paper proposes an CPDS planning method considering the reliability of the cyber-physical system. First, the reliability evaluation model of CPDS is proposed. Specifically, the functional reliability model of the cyber layer is introduced, based on which the physical equipment reliability model is further investigated. Second, an optimal planning model of CPDS considering cyber-physical random failures is developed, which is solved using the Monte Carlo Simulation technique. The proposed model is tested on the modified IEEE 33-node distribution system, and the results demonstrate the effectiveness of the proposed method.