Skip to main content
main-content

Über dieses Buch

This two-volume set LNICST 286-287 constitutes the post-conference proceedings of the First EAI International Conference on Artificial Intelligence for Communications and Networks, AICON 2019, held in Harbin, China, in May 2019. The 93 full papers were carefully reviewed and selected from 152 submissions. The papers are organized in topical sections on artificial intelligence, mobile network, deep learning, machine learning, wireless communication, cognitive radio, internet of things, big data, communication system, pattern recognition, channel model, beamforming, signal processing, 5G, mobile management, resource management, wireless position.

Inhaltsverzeichnis

Frontmatter

Security with Deep Learning for Communications and Networks

Frontmatter

VITEC: A Violence Detection Framework

Hundreds of millions of youths suffer from various violence each year. The negative impacts motivate much research and numerous studies on violence. However, those attempts went their own way, making the achieved results, especially from engineering, not so useful. Based on the Sensor and Social Web (SEWEB) concept, Violence Detection (VITEC) was proposed as a possible framework to facilitate multi-disciplinary researchers in their fight against violence. At its core, it consists of a primary agent, which is violence detection using physiological signals and activity recognition, and a secondary agent, which is violence detection using surveillance video. The second layer of the proposed framework contains a cloud computing service with a Personal Safety Network (PSN) database. The cloud computing service manages all data, notifications, and some more thorough processing. The upper layer is for both observed young persons and members of the PSN. The proposed framework offers business opportunities. The existing school violence/bullying intervention programs can take advantage of VITEC by providing almost instant notifications of violent events, enabling the victims to get immediate help and intensifying coordination among different sectors to fight against violence. In the long run, VITEC may provide an answer related to the vision of having a world free from violence in 2030, as addressed by the UN Special Representative of Secretary-General on violence against children.

Hany Ferdinando, Tuija Huuki, Liang Ye, Tian Han, Zhu Zhang, Guobing Sun, Tapio Seppänen, Esko Alasaarela

Software Defect Prediction Model Based on Stacked Denoising Auto-Encoder

Software defect prediction technology plays an important role in ensuring software quality. The traditional software defect prediction model can only perform “shallow learning” and cannot perform deep mining of data features. Aiming at this problem, we use the stacked denoising auto-encoder (SDAE) to superimpose into deep neural network. First, the deep network model was built through the stacked layers of denoising auto-encoder (DAE), then the unsupervised method was used to train each layer in turn with noised input for more robust expression, characteristics were learnt supervised by back propagation (BP) neural network and the whole net was optimized by using error back propagation. Simulation experiments prove that the prediction accuracy of our SDAE model is significantly improved compared with the traditional SVM and KNN prediction model.

Yu Zhu, Dongjin Yin, Yingtao Gan, Lanlan Rui, Guoxin Xia

Deep&Cross Network for Software-Intensive System Fault Prediction

With the development of information technology, the causes of software-intensive system failures become more complicated. This paper analyzes the correlation of various fault factors of software-intensive equipment and uses deep learning model to do fault prediction in complex electronic information system. Experimental results show that neural network model based on feature interaction can get better effect than some other methods.

Guoxin Xia, Dongjin Yin, Yingtao Gan, Lanlan Rui, Yu Zhu

Research on Evaluation Method of Cooperative Jamming Effect in Cognitive Confrontation

In modern warfare, due to the increasingly complex electromagnetic environment, cognitive confrontation will become the main form of war. Effective interference with enemy radars is of great importance for taking the lead in the battlefield. The evaluation of the synergistic interference effect is an important indicator to measure the performance of the interference equipment. According to the evaluation result, the interference strategy can be changed in time to achieve the best interference revenue and provide a strong guarantee for the successful penetration of the target. In this paper, the discovery probability and positioning accuracy of radar network are used as evaluation indicators to establish an evaluation model. The interference power, interference frequency, interference timing and interference pattern are used as membership functions. The distance, false alarm probability and different interference strategies are studied. The simulation shows that proper false alarm probability, closer distance and proper interference strategy can improve the interference benefit and provide a theoretical basis for obtaining the best interference effect in the actual battlefield.

Jing Ma, Bin Shi, Fei Che, Sitong Zhang

Navigation Performance Comparison of ACE-BOC Signal and TD-AltBOC Signal

With the development of intellectual property rights and navigation performance optimization, recently two new four-components signal multiplex modulation methods, ACE-BOC & TD-AltBOC have been brought forward by scholars. The navigation performances of them are aimed to be compared in this paper. Based on analyzing of the power spectrum density function of these two modulation methods, their main navigation performances of code tracking precision, anti-multipath capability, anti-jamming capability and compatibility are compared. The results show that the navigation performance of ACE-BOC modulation signal is 1.0 dB–2.1 dB prior to that of the TD-AltBOC modulation signal. This is because of the difference of their power spectrum density function figures and that of the 1.2 dB distributed power level. The study production can be referred to choose the better one between these two new modulation methods.

Chunxia Li, Jianjun Fan, Min Li, Yang Gao

A Video-Selection-Encryption Privacy Protection Scheme Based on Machine Learning in Smart Home Environment

The Internet of things is a new technological revolution following the computer and Internet. It aims to connect all physical objects existing in the world and forms a network with everything. In recent years, smart home gradually enters into our life. Smart home uses the Internet of things technology to connect all kinds of devices in the home, to achieve a smart home environment. Although the development of smart home has brought a qualitative leap to people’s life, there are many problems in security. Privacy security is one of the challenges to the smart home environment. Attackers can intrude various smart devices in the smart home environment, to achieve the purpose of stealing users’ personal information and privacy. Among these devices, smart cameras are the most intruded frequently. Since many cameras are installed in users’ homes to achieve real-time monitoring of the environment, but the existence of these cameras provides a channel to get information for attackers. In recent years, the leak of video privacy is emerging in an endless stream. According to the researches about privacy protection, this paper proposes a new scheme to selectively encrypt the video captured by the cameras through machine learning technology, so as to protect the personal privacy of users and improve the security of the smart home environment.

Qingshui Xue, Haozhi Zhu, Xingzhong Ju, Haojin Zhu, Fengying Li, Xiangwei Zheng, Baochuan Zuo

A Multi-agent Reinforcement Learning Based Power Control Algorithm for D2D Communication Underlaying Cellular Networks

This paper considers the power control problem in device-to-device (D2D) communication underlaying a cellular network and explores the application of the machine learning (ML) approach in power control for improving the system throughput. Two multi-agent reinforcement learning (MARL) based algorithms are proposed for performing power control of D2D users (DUs): centralized Q-learning algorithm and distributed Q-learning algorithm. In the centralized algorithm, all DU pairs sharing the same RB use a common Q table in the learning process, while in the distributed algorithm each DU pair maintains its own Q table. Simulation results show that both the centralized algorithm and the distributed algorithm can converge to the same optimum Q values, and the distributed algorithm can converge faster than the centralized algorithm. Moreover, both the proposed Q-learning algorithms outperform the random power control algorithm in terms of the system throughput and satisfaction ratio.

Wentai Chen, Jun Zheng

An Adaptive Window Time-Frequency Analysis Method Based on Short-Time Fourier Transform

Frequency hopping signal has the advantages of strong anti-jamming ability and low probability of interception. It can effectively improve communication quality and security. Therefore, frequency hopping technology is widely used in the field of information countermeasures, and has become one of the main anti-jamming technologies adopted by various countries. As a non-partner, how to quickly obtain the main parameters of frequency hopping signal in order to implement effective and timely interference is particularly important. In this paper, a blind estimation algorithm based on short-time Fourier transform (STFT) is proposed. STFT is a time-frequency analysis method with low complexity. The performance of this method depends largely on the length of the Fourier transform window. The algorithm in this paper roughly estimates the period of frequency hopping signal according to the Frequency domain characteristics of input signal, and uses this information to determine the length of local window. The obtained time-frequency distribution is purified by setting a reasonable threshold, and the purified time-frequency distribution is used to extract information and estimate the main parameters of frequency hopping signal. The results show that this method can roughly estimate the length of Fourier transform window in very low SNR environment, and fine estimation method has higher accuracy in estimating the Frequency hopping period of frequency hopping signal.

Zhiqiang Li, Xiao Wang, Ming Li, Shuai Han

Research on the Security of Personal Information in the Era of Big Data

The advent of the era of big data has made people’s lives more intelligent and convenient, and has brought enormous value to human beings. At the same time, it has brought about a lot of challenges. Personal information security is one of them. In the big data era, massive amounts of personal information are continuously input, simultaneously, the means of illegally acquiring, disseminating, and applying personal information are emerging in endlessly, Raising human thinking about information security. Therefore, the personal information security problem caused by big data should not be underestimated. Firstly, it introduces the related concepts of personal information, discusses the sources of risks faced by personal information in the process of implantation, dissemination and application. Secondly, it discusses the existing problems and solutions in China. Finally, the application difficulties and coping strategies of information security technology are analyzed, which provides theoretical support for personal information security in the era of big data.

Cheng Chi, Tengyu Liu, Xiaochen Yu, Shuo Zhang, Shuo Shi

Secure Access and Routing Scheme for Maritime Communication Network

This paper proposes a transmission scheme to ensure the maritime communication security, which includes access rules, routing selection scheme, and power allocation mechanisms. The access rules and the routing selection utilize the automatic identification system (AIS) information to choose the secure access points and routing links to prevent eavesdropping, and the power allocation limits the leaked information by means of reducing the received signal power of the eavesdroppers. The simulation results show that the intercept probability of the proposed scheme decreases by about ten to the negative two power compared with that of the contrastive scheme, and the recovering proportion for eavesdropper is less than 0.2. In addition to above, the secrecy capacity of the proposed scheme achieves about 6.8% improvement compared with the baseline scheme.

Yuzhou Fu, Chuanao Jiang, Yurong Qin, Liuguo Yin

Multipath Based Privacy Protection Method for Data Transmission in SDN

With the development of Software-Defined Networking (SDN), privacy and security issues have become an urgent problem to be solved. Although there are many ways to solve these problems, the existing technology represented by encryption cannot effectively deal with traffic analysis attacks, and there are also key management problems. For this reason, we propose a privacy protection method for SDN data transmission based on multipath, including path searching procedure for searching for all paths between the sender and the receiver, and path filtering procedure for filtering out paths to reduce path correlation, and path selection procedure for randomly selecting one path to disturbed the traffic similarity between multiple transmission. The experiment results show that our method is more effective, less similarity of traffic compared with Multipath-Floyd method and single-path method, respectively. Moreover, it is difficult for attackers to capture the traffic feature and do not need key management, which reduces the cost of the controller.

Na Dong, Zhigeng Han, Liangmin Wang

A Biometric-Based IoT Device Identity Authentication Scheme

IoT is an important part of the new generation of information technologies and the next big thing in the IT industry after the computer and the internet. The IoT has great development potential and a wide range of possible applications, especially commercial applications. And information security of the IoT is the key to the long-term development of the whole industry. Currently, the two most significant factors in the development of the IoT are user identity authentication and privacy protection. This paper contains an analysis on the current picture of inter-device user identity authentication in the IoT and proposes an inter-device biometric authentication solution for the IoT that’s designed to work with larger devices, addressing the shortcomings of the traditional user identity authentication technologies including security and efficiency problems. A strategy for further solution optimization is also included. This paper elaborates on the specific process of user identity authentication carried out by users on devices and between devices making use of fingerprints. We’ll demonstrate the security of this solution against existing attack methods and in the last part, we enumerate various possible applications of this solution in smart homes.

Qingshui Xue, Xingzhong Ju, Haozhi Zhu, Haojin Zhu, Fengying Li, Xiangwei Zheng

A Physical and Verbal Bullying Detecting Algorithm Based on K-NN for School Bullying Prevention

School bullying is a common social problem around the world which affects teenagers, and physical violence is considered to be the most harmful while verbal bullying is the most frequent. This paper proposes an automatic physical and verbal bullying detecting method in the field of artificial intelligence. Dozens of features were extracted from acceleration and gyro data to train the physical bullying recognition while the mean value of each frame of samples was calculated for verbal bullying detection. The authors used the k-NN algorithm as the classifier. The final test accuracies of physical and verbal bullying detecting were 70.4% and 78.0%, respectively, indicating that activity recognition and speech emotion recognition can be used for detecting bullying behaviors as an artificial intelligence technique, and speech emotion recognition appeared to be better than activity recognition.

Shangbin Gao, Liang Ye

Speech Bullying Vocabulary Recognition Algorithm in Artificial Intelligent Child Protecting System

With the continuous breakthrough of various technologies, speech recognition technology has become a research hotspot. It is a way to find out the phenomenon of bullying in time by detecting whether the voice contains campus bullying vocabulary. In practical applications, an infinite network is established through sensors to transmit information, and the occurrence of campus bullying events is prevented in time. This paper studies the theory of support vector machine and its application in speech recognition. In order to identify bullying vocabulary, this paper firstly built a voice library with 250 voice audios, including 125 campus bullying word audios and 125 non-bullying audios. The first sub-frame of the speech signal was used for endpoint detection. Then mode decomposition and Fourier transform were applied. The maximum value of the primary frequency spectrum was extracted as the acoustic feature. Finally, 200 audios in the database were used for training, and 50 audios were used for speech recognition testing. The average recognition accuracy was 94%, indicating that the support vector machine theory showed a good advantage in the case of small samples for speech recognition.

Tong Liu, Liang Ye, Tian Han, Yue Li, Esko Alasaarela

Cross Station-Voltage-Level Route Planning Algorithm for System Protection Services in UHV Grid

Since the current route planning algorithms for system protection service in communication network fail to consider the situation of station voltage level crossing in ultra-high voltage grid, based on immune algorithm, a planning algorithm is proposed in this paper to lower the network operational risk by ensuring the voltage level balance degree and main-backup route similarity are both as small as possible. Firstly, the risk of node and link is analyzed to derive the network risk balance degree index, and secondly, the cross station-voltage-level planning problem model is established with the consideration of impact brought by station voltage level crossing and route similarity, and then it is solved utilizing IA. Finally, the experimental result shows that the planning algorithm proposed in this paper can lower the network risk, station voltage level balance degree and the route similarity effectively, which provides a useful reference while planning service routing.

Danyang Xu, Peng Yu, Wenjing Li, Xuesong Qiu, Luoming Meng

An Efficient Implementation of Multi-interface SSD Controller on SoC

This paper designs a high-performance multi-interface SSD controller built on Xilinx SoC. An efficient firmware is also implemented, which is elaborate to cooperate with the hardware. Parallelism techniques, such as plane-level, die-level, chip-level and channel-level, are used for improving performance. The system has hard real-time performance of sequential writing, with the minimum bad block management and wear-leveling policy to balance performance and lifetime. A transparent encryption is proposed to guarantee high security storage, that is, connecting an AES-256 core and a RAID core with DMA engine in series. Performance evaluation of physical hardware shows that writing speed can exceed 100 MiB/sec for every logical channel which combines 8 NAND Flash chips.

Yifei Niu, Hai Cheng, Songyan Liu, Huan Liu, Xiaowen Wang, Yanlin Chen

Joint BP and RNN for Resilient GPS Timing Against Spoofing Attacks

In this paper, we propose a new wide-area algorithm to secure the Global Positioning System (GPS) timing from spoofing attack. To achieve a trusted GPS timing, belief propagation (BP), recognized as one of the Artificial Intelligence (AI) approaches, and the recurrent neural network (RNN) are jointly integrated. BP is employed to authenticate each GPS receiving system in the wide-area network from malicious spoofing attacks and estimate the corresponding spoofing-induced timing error. To evaluate the spoofing status at each of the GPS receiving system, RNN is utilized to evaluate similarity in spoofing-induced errors across the antennas within the GPS receiving system. Having applied a proper training stage, simulation results show that the proposed joint BP-RNN algorithms can quickly detect the spoofed receiving system comparing with existing work.

Sriramya Bhamidipati, Kyeong Jin Kim, Hongbo Sun, Philip Orlik, Jinyun Zhang

Cloud and Big Data of AI-Enabled Networks

Frontmatter

Regional Landslide Sensitivity Analysis Based on CPSO-LSSVM

Landslide sensitivity analysis is of great significance for predicting landslide hazards. Taking the landslide in the hilly area of Sichuan Province as an example, through the interpretation of high spatial resolution remote sensing images and the analysis of the occurrence mechanism of landslides in the low hilly areas of Sichuan Province, eight landslide susceptibility evaluation factors were obtained. (elevation, slope, terrain relief, rivers, roads, geotechnical types, NDVI, fault structures). Then, using the neighborhood statistical analysis, ArcGIS technology and other methods to obtain training sample data and regional sample data. According to the characteristics of the landslide development, the Chao Particle Swarm Optimization (cpso) is used to optimize the parameters of the Least Square Support Vector Machine (lssvm), the cpso-lssvm landslide sensitivity prediction model was formed. The experimental results show that cpso-lssvm has obtained good prediction results in landslide sensitivity evaluation, and the prediction accuracy has increased to 70.5%.

Yanze Li, Zhenjian Yang, Yunjie Zhang, Zhou Jin

Multitasks Scheduling in Delay-Bounded Mobile Edge Computing

The Mobile Edge Computing (MEC) is a very novel technology in the social network. In order to satisfy the users’ delay and enhance data security and reduce energy consumption of the system. In this paper, we design an optimized strategy of edge servers chosen by reinforcement learning (AI algorithm) for resource consumption of multi-edge server. We transform the objective function into a convex optimization problem for the single edge server, and give proofs. In this scenario, it is made up of multi-users with multi-tasks and the same type of edge servers deployed on the edge of the base stations, the edge servers allocate computing resources to users. Meanwhile, the users send different tasks to the edge servers to complete computing tasks. In order to complete users’ requirements under their delay-bounded, we design an optimal path algorithm named Betweenness Centrality Algorithm (BCA) to reduce the transmission delay and we use Fuzzy Logical algorithm to classify computing tasks. We propose a resource allocation mechanism under satisfying delay-bounded. Finally, comparing to the random selected strategy and other schemes, we prove the effectiveness of the introduced algorithm, which reduces the energy consumption by 20% approximately for the single edge server, and the simulation proves that the Double Deep Q-learning (Double DQN) algorithm shows better performance than Random Selected Strategy for the multi-edge servers.

Longyu Zhou, Supeng Leng, Ning Yang, Ke Zhang

Edge Computing Based Traffic Analysis System Using Broad Learning

Current traffic analysis methods are executed on the cloud, which need to upload the traffic data and consume precious bandwidth resources. Edge computing is a more promising way to save the bandwidth resources and improve users’ privacy by offloading these tasks to the edge node. However, traffic analysis methods based on traditional machine learning need to retrain all traffic data when updating the trained model, which are not suitable for edge computing due to the poor computing power and low storage capacity of edge nodes. In this paper, we propose a novel edge computing based traffic analysis system using broad learning. For one thing, edge computing can provide a distributed architecture for saving the bandwidth resources and protecting users’ privacy. For another, we use broad learning to incrementally train the traffic data, which is more suitable for edge computing because it can support incremental updates of models on the edge nodes without retraining all data. We implement our system on the Raspberry Pi, and experimental results show that we have 98% probability to accurately identify these traffic data. Moreover, our method has the faster training speed compared with Convolutional Neural Network (CNN).

Xiting Peng, Kaoru Ota, Mianxiong Dong

The Performance of DF Relaying System Based on Energy Harvesting and Dual-Media Channels

In recent years, energy efficiency in multi-hop cooperative power-line communication (PLC) and wireless systems has recently received considerable attention. This paper considers the dual-hop PLC/wireless parallel communication system based on decode and forward (DF), where the relay harvests the high noise inherent in PLC channels to further enhance energy efficiency. In this paper, we derive the exact analytic expression of energy efficiency and the closed expression of outage probability. In order to compare and highlight the achievable gain, we also analyzed the related performances of DF relaying PLC system with energy harvesting (DF-EH). Then based on the theoretical calculation and simulation results, the influence of energy-harvesting time factor and other parameters on the system performance is analyzed. The result shows that energy-harvesting time factor and power allocation are the key factors affecting system performance. Relevant conclusions provide necessary theoretical support for the application of energy harvest technology in mixed media cooperative communication.

Zhixiong Chen, Lijiao Wang, Cong Ye, Dongsheng Han

School Violence Detection Based on Multi-sensor Fusion and Improved Relief-F Algorithms

School bullying is a common social problem around the world, and school violence is considered to be the most harmful form of school bullying. This paper proposes a school violence detecting method based on multi-sensor fusion and improved Relief-F algorithms. Data are gathered with two movement sensors by role playing of school violence and daily-life activities. Altogether 9 kinds of activities are recorded. Time domain features and frequency domain features are extracted and filtered by an improved Relief-F algorithm. Then the authors build a two-level classifier. The first level is a Decision Tree classifier which separates the activity of jump from the others, and the second level is a Radial Basis Function neural network which classifies the remainder 8 kinds of activities. Finally a decision layer fusion algorithm combines the recognition results of the two sensors together. The average recognition accuracy of school violence reaches 84.4%, and that of daily-life reaches 97.3%.

Liang Ye, Jifu Shi, Hany Ferdinando, Tapio Seppänen, Esko Alasaarela

A UAV Based Multi-object Detection Scheme to Enhance Road Condition Monitoring and Control for Future Smart Transportation

Road condition monitoring and control is essential for smart transportation in the era of autonomous driving. In this paper, we propose to apply unmanned aerial vehicle (UAV), wireless communications and artificial intelligence (AI) to achieve multi-object detection for smart road monitoring and control. In particular, the application of UAV enables real-time image view to monitor road condition, such as traffic flow and on-road objects, in an efficient way without disturbing normal traffic. Those raw image data are first offloaded to a road side unit through wireless communications. A computing platform connected to the road side unit can execute the AI based scheme for road condition monitoring and control. The AI based scheme is developed around convolutional neural network (CNN). For demonstration, the objects of interest considered in this work include advertisement billboards, junctions, traffic signs and unsafe objects. Other objects can be extended to the developed system with more collected data. To evaluate the proposed scheme, we launched a UAV to collect real-life road images from multiple road sections of a highway. The AI based scheme is then developed using portion of the raw data. Test of the AI scheme is conducted using the rest of the dataset. The evaluation results have demonstrated that the proposed UAV based multi-object detection scheme can provide accurate results to support efficient road condition monitoring and control in future smart transportation.

Jian Yang, Jielun Zhang, Feng Ye, Xiaohui Cheng

Sentiment Analysis for Tang Poetry Based on Imagery Aided and Classifier Fusion

This paper aims to do sentiment analysis for Tang poetry from the perspective of text mining. Most previous works just focus on the literariness of Chinese poetry or establish language models statistically, which ignore the features of sentiment and specific applications. We propose a sentiment analysis system for Tang poetry based on imagery aided and classifier fusion. Especially, we extract sentimental imageries at two levels: character and word, and bring them into sentiment analysis. In addition, classifier fusion is adopted in this paper to improve classification performance. Experiments show the effectiveness of our model and our method is superior to the traditional method.

Yabo Shen, Yong Ma, Chunguo Li, Shidang Li, Mingliang Gu, Chaojin Zhang, Yun Jin, Yingli Shen

Detection of Insulator Defects Based on YOLO V3

The power system of China is still composed of power generation, transmission, substation, power distribution and other links. To ensure the safety and stability of transmission lines is an important part of large-scale transmission process, and the insulators are important in the transmission line. The existing parts, such as surface contamination, cracks, damage and other surface defects seriously threaten the operation safety of the power grid. Faults caused by insulator defects are currently the highest proportion of power system faults, so the surface defects of insulators are detected and timely completion of fault repair becomes more important. In this regard, this paper proposes a target detection algorithm based on YOLO V3 (You Only Look Once: Real-Time Object Detection), which utilizes the powerful learning ability of deep convolutional neural networks and a large number of data annotation samples. The image of the insulator photo-graphed by the machine is detected and classified, finally the intelligent detection of the intact insulator and the defective insulator is realized. The experimental results show that the YOLO V3 based insulator defect detection method can effectively identify the defective insulator strings from the aerial image of the drone. Compared with the previous insulator defect identification method, the accuracy and detection time are significantly improved, and it can realize the intelligent detection of intact insulators and defective insulators.

Fei Guo, Kun Hao, Mengqi Xia, Lu Zhao, Li Wang, Qi Liu

MCU-Based Isolated Appealing Words Detecting Method with AI Techniques

Bullying in campus has attracted more and more attention in recent years. By analyzing typical campus bullying events, it can be found that the victims often use the words “help” and some other appealing or begging words, that is to say, by using the artificial intelligence of speech recognition, we can find the occurrence of campus bullying events in time, and take measures to avoid further harm. The main purpose of this study is to help the guardians discover the occurrence of campus bullying in time by real-time monitoring of the keywords of campus bullying, and take corresponding measures in the first time to minimize the harm of campus bullying. On the basis of Sunplus MCU and speech recognition technology, by using the MFCC acoustic features and an efficient DTW classifier, we were able to realize the detection of common vocabulary of campus bullying for the specific human voice. After repeated experiments, and finally combining the voice signal processing functions of Sunplus MCU, the recognition procedure of specific isolated words was completed. On the basis of realizing the isolated word detection of specific human voice, we got an average accuracy of 99% of appealing words for the dedicated speaker and the misrecognition rate of other words and other speakers was very low.

Liang Ye, Yue Li, Wenjing Dong, Tapio Seppänen, Esko Alasaarela

Performance Analysis Based on MGF Fading Approximation in Hybrid Cooperative Communication

Wireless communications and power line communications (PLC) are essential components for smart grid communications. In this study, the performance of dual-hop wireless/power line hybrid fading system is analyzed from two aspects of outage probability and bit error rate (BER). The system adopts a hybrid fading model based on the general Nakagami-m wireless and lognormal (LogN) power line fading based on amplify and forward (AF) relay protocol, and the Bernoulli Gaussian noise model attached to the PLC channel. Since the LogN distribution has a certain similarity with the Gamma, the key parameters of PDF with approximated LogN distribution from Gamma distribution are determined by using the moment generating function (MGF) equation and the approximation of LogN variable sum. Then the exact closed-form expression of the system outage probability and BER are obtained by integral variation. Finally, Monte Carlo simulation is used to verify the correctness of the theory and analyze the influence of different system parameters on the performance.

Cong Ye, Zhixiong Chen, Jinsha Yuan, Lijiao Wang

Age and Gender Classification for Permission Control of Mobile Devices in Tracking Systems

Not only does the human voice provide the semantics of the spoken words but also it contains the speaker-dependent characteristics, such as the gender, the age, and the emotional state of the speaker. In the last decade, speech recognition gained a great interest in identifying and tracking systems. According to the speech length of ten to thirty seconds, this paper proposes an age and gender classification method for permission control of mobile devices. Each speech signal is firstly extracted to 40 features by Mel Frequency cepstral Coefficients (MFCC). After that, the Support Vector Machine (SVM) is used to finish the age and gender classification. This paper studies six kernel models of SVM and concludes that cubic, quadratic, and medium Gaussian kernel models could improve the recognition rate up to 93.75%, 91.25% and 93.75% respectively. Therefore, it is promising for permission control of a mobile in tracking systems.

Merahi Choukri, Shaochuan Wu

Research on Fusion of Multiple Positioning Algorithms Based on Visual Indoor Positioning

Due to the limited penetration ability of GPS signal to buildings, indoor high-precision positioning combined with a variety of technologies has been paid more and more attention by researchers. Based on the traditional indoor positioning technology, a new indoor positioning method is proposed in this paper, which combines vision and inertial sensor. In this paper, we will first independently evaluate the quality of inertial positioning and visual positioning results, and then integrate them with complementary advantages to achieve the effect of high-precision positioning.

Guangchao Xu, Danyang Qin, Min Zhao, Ruolin Guo

Social Aware Edge Caching in D2D Enabled Communication

As a promising architecture, mobile edge networks can effectively mitigate the load on backhaul links and reduce the transmission delay simultaneously. With the development of artificial intelligence (AI) and machine learning, how to reasonably combine AI as well as machine learning with communication is a hot topic. In this paper, considering the content features and user preference jointly, the projective adaptive resonance theory neural network (PART NN) is used to design the community architecture. After that, we can obtain the status table of communities. In order to reduce the redundant caching, the popular contents will be cached in the user equipment (UE) of center user in advance. The cache scheme of center user is adjusted according to the status table. Two transmission links are considered, i.e., cellular link and device-to-device (D2D) link, to reduce the transmission delay. Since the content preference of UE is time-varying and the migration patterns are various, the community architecture will be updated dynamically to further improve the cache hit rate. The migration patterns of UEs are affected by social factors as well as geographical factors. The simulation results show that the community construction scheme and cache scheme effectively improve the cache hit rate and reduce the transmission delay simultaneously.

Jingyi Chen

AI-Based Network Intelligence for IoT

Frontmatter

An Emergency Resource Scheduling Model Based on Edge Computing

In recent years, more and more researches are focusing on the field of disaster emergency management, especially in emergency resource dispatching. In a disaster scenario, a control center needs to obtain a large amount of real-time information at the scene to make decisions timely and accurately. Traditional Cloud Computing has many shortcomings in terms of delay and bandwidth, and Edge Computing (EC) will be more suitable for this scenario. We apply EC to the emergency rescue to cope with the problem of latency needs. First, we propose an edge-based emergency rescue architecture that consists of three layers: Cloud Layer, Edge Layer, and Resource Layer. Based on this, we give a resource scheduling model that requires the collaboration of Cloud and Edge Layers. The Cloud Layer gives a partition for these tasks, and all sub-tasks are assigning to the edge servers to get a locally optimal solution. Finally, these solutions from different edge servers are summarized to the Cloud Layer to get a globally optimal solution. We compare our algorithm with CS-GA and VRP. The simulation results show that RSE has good performance in scheduling time and cost.

Songyan Zhong, Tao He, Mengyu Li, Lanlan Rui, Guoxin Xia, Yu Zhu

Mobility-Aware Task Parallel Offloading for Vehicle Fog Computing

When applying fog computing paradigm into Internet of vehicle, vehicles are regarded as intelligent devices with computation and communication capability. These moving intelligent devices are often employed to assist various computation-intensive task offloading in vehicle fog computing, which brings real-time responses. However, vehicles mobility and network dynamics make it challenging to offload tasks to ideal target nodes for user-vehicle. In this paper, leveraging the result of vehicle mobility-awareness, we investigate the task offloading problem in vehicle fog computing aiming to minimize service time. Specifically, we consider that a task can be decomposed into subtasks in any proportion and offloaded from user-vehicle to multi service vehicles in parallel via vehicle-to-vehicle (V2V) links. Mobility information of vehicles collected by RSU is modeled to predicted the states of V2V links based on hidden Markov model (HMM). Then, we refine a rule to select target service-vehicles and the size of each subtask according to predicted results. Comparing with random and single-point task offloading, the proposed approach indicates a better performance on amount of finished task and service time in vehicle dense area.

Jindou Xie, Yunjian Jia, Zhengchuan Chen, Liang Liang

A Floor Distinction Method Based on Recurrent Neural Network in Cellular Network

Indoor localization is nowadays becoming a hot topic and research trend for future large-scale location-aware services, particularly in high-rise buildings with complex structures. However, the indoor positioning methods existing are just with high interests of two-dimensional planar information, and the crucial height information for accurate position result is awfully neglected. Furthermore, without considering the shadow effect caused by indoor constant changing impact on the terminal to be located, positioning methods cannot achieve a desirable localization accuracy for building environment. In this paper, we proposed a fast and reliable method using deep neural network for floor-level distinction and position estimation based on ubiquitous radio waves in mobile communication system. The framework composed of autoencoder to extract the effective feature vectors and recurrent neural network classifier to solve the misclassification caused by timing-discontinuity of received signal. It is shown that the accuracy of floor distinction is over 90.2% in different structural construction environments, which can provide comparable to current top-performing floor localization methods.

Yongliang Zhang, Lin Ma, Danyang Qin, Miao Yu

A Multi-classifier Approach for Fuzzy KNN Based WIFI Indoor Localization

WIFI fingerprint positioning technology has been widely studied and developed for a long time and lots of experiment systems have been established. However, the time-varying and nonlinear features of the WIFI signal impede the development in the application level. The performance of existing positioning systems could be very unstable due to the signal varying. Our object is to combine the fuzzy technology and multi-classifier approach to improve the system accuracy and robustness. The proposed method adopts the fuzzy integral to fusion the results obtained from different fuzzy K-nearest neighbor (KNN) classifiers generated by adaBoost. Experiment results demonstrate that our approach improves the average positioning errors and their standard deviations by 21% and 26% separately compared to the traditional KNN algorithm.

Yuanfeng Du, Dongkai Yang

Research on Application and Development of Key Technologies of Dynamic Wireless Charging System in New Intelligent Transportation System

With the development of electronic science and technology, Intelligent Transportation has entered a new stage. In recent years, IOT (Internet of Things) technology has brought many impacts on people’s daily life and social development. IOV (Internet of Vehicle) technology is an important application of IOT technology in Intelligent Transportation System. The rapid development of IOV technology provides a guarantee for Intelligent Transportation. Therefore, accelerating the landing of IOV has a profound impact on the development of new Intelligent Transportation. Dynamic wireless charging technology has become one of the five cutting-edge technologies to accelerate the landing of IOV. Firstly, the article introduces the new Intelligent Transportation System and its key technology. Then the article analyzes the advantages of wireless charging and the composition of wireless energy transmission system. Finally, the article introduces the composition of the dynamic wireless charging system on electric vehicles, the application and development of dynamic wireless charging key technology in new Intelligent Transportation.

Lina Ma, Shi An, Wanlong Zhao

Binocular Vision-Based Human Ranging Algorithm Based on Human Faces Recognition

In the field of security, timely and effective identification is very important for safeguarding public safety, national security and information security. Face recognition is an important technology in these areas. The calculation of range plays an important role in protecting safety and tracking suspects. Binocular stereo vision ranging has wide application in non-contact precise measurement and dangerous scenes. In this paper, a binocular range measurement system based on face recognition is proposed. The system can detect and recognize faces and calculate its real time range. It could realize tracking real time faces and calculate its distance from the cameras and locate them. And it suits the feature of special places of high security and preventing the suspicious people from entering and out.

Xiaolin He, Lin Ma, Weixiao Meng

Deep Reinforcement Learning Based Task Offloading in SDN-Enabled Industrial Internet of Things

Recent advances in communication and sensor network technologies make Industrial Internet of Things (IIoT) a major driving force for future industry. Various devices in wide industry fields generate diverse computation tasks with their distinct service requirements. Note that the distribution of such tasks has essential intrinsic patterns and varies according to factors like region, season and time. Different from previous efforts to develop algorithms in specific scenarios for reducing task execution latency without considering the task generation patterns of IIoT, we propose a DRL-based Task Offloading algorithm (DRLTO) to learn such generation patterns and maximize the task completion rate. A SDN-enabled multi-layer heterogeneous computing framework is also introduced to efficiently assign tasks according to the obtained knowledges towards their features. Extensive experiments validate that our algorithm can not only significantly improve the average task completion rate, but also achieve near-optimal results in lots of IIoT scenarios.

Jiadai Wang, Yurui Cao, Jiajia Liu, Yanning Zhang

A Reinforcement Learning Based Task Offloading Scheme for Vehicular Edge Computing Network

Recently, the trends of automation and intelligence in vehicular networks have led to the emergence of intelligent connected vehicles (ICVs), and various intelligent applications like autonomous driving have also rapidly developed. Usually, these applications are compute-intensive, and require large amounts of computation resources, which conflicts with resource-limited vehicles. This contradiction becomes a bottleneck in the development of vehicular networks. To address this challenge, the researchers combined mobile edge computing (MEC) with vehicular networks, and proposed vehicular edge computing networks (VECNs). The deploying of MEC servers near the vehicles allows compute-intensive applications to be offloaded to MEC servers for execution, so as to alleviate vehicles’ computational pressure. However, the high dynamic feature which makes traditional optimization algorithms like convex/non-convex optimization less suitable for vehicular networks, often lacks adequate consideration in the existing task offloading schemes. Toward this end, we propose a reinforcement learning based task offloading scheme, i.e., a deep Q learning algorithm, to solve the delay minimization problem in VECNs. Extensive numerical results corroborate the superior performance of our proposed scheme on reducing the processing delay of vehicles’ computation tasks.

Jie Zhang, Hongzhi Guo, Jiajia Liu

Sensor-Based Human Activity Recognition for Smart Healthcare: A Semi-supervised Machine Learning

Human action recognition is an integral part of smart health monitoring, where intelligence behind the services is obtained and improves through sensor information. It poses tremendous challenges due to huge diversities of human actions and also a large variation in how a particular action can be performed. This problem has been intensified more with the emergence of Internet of Things (IoT), which has resulted in larger datasets acquired by a massive number of sensors. The big data based machine learning is the best candidate to deal with this grand challenge. However, one of the biggest challenges in using large datasets in machine learning is to label sufficient data to train a model accurately. Instead of using expensive supervised learning, we propose a semi-supervised classifier for time-series data. The proposed framework is the joint design of variational auto-encoder (VAE) and convolutional neural network (CNN). In particular, the VAE intends to extract the salient characteristics of human activity data and to provide the useful criteria for the compressed sensing reconstruction, while the CNN aims for extracting the discriminative features and for producing the low-dimension latent codes. Given a combination of labeled and raw time-series data, our architecture utilizes compressed samples from the latent vector in a deconvolutional decoder to reconstruct the input time-series. We intend to train the classifier to detect human actions for smart health systems.

Abrar Zahin, Le Thanh Tan, Rose Qingyang Hu

CoLoRSim: A Highly Scalable ICN Simulation Platform

In the past few years, Internet usage is shifting from host-to-host communication to content distribution. As a result, information-centric networking (ICN) is emerging as a promising candidate paradigm for the future Internet. Meanwhile, many researchers have developed specified simulation platforms for the ICN architecture they proposed, to evaluate the performances in the research stage. In this paper, we present a simulation platform for a recent proposed ICN architecture named CoLoR. With the help of this simulation platform, performance for CoLoR can be evaluated through large-scale simulations. New protocols designed for CoLoR can also be studied.

Hongyi Li, Hongbin Luo

Manifold Learning Based Super Resolution for Mixed-Resolution Multi-view Video in Visual Internet of Things

In a Visual Internet of Things (VIoT), the video sequences of different viewpoints are captured by different visual sensors and transmitted simultaneously, which puts a huge burden on storage and bandwidth resources. Mixed-resolution multi-view video format can alleviate the burden on the limited storage and bandwidth resources. However, the low resolution view need to be up-sampled to provide high quality visual experiences to the users. Therefore, a super resolution (SR) algorithm to reconstruct the low resolution view is highly desirable. In this paper, we propose a new two-stage super resolution method. In the first depth-assisted high frequency synthesis stage, depth image based rendering (DIBR) is used to project a high resolution view to a low resolution view to estimate the super resolution result. Then in the second high frequency compensation stage, the local block matching model based on manifold learning is used to enhance the super resolution result. The experimental results demonstrate that our method is capable to achieving a PSNR gain up to 4.76 dB over bicubic baseline and recover details in edge regions, without sacrificing the quality of smooth areas.

Yuan Zhou, Ying Wang, Yeda Zhang, Xiaoting Du, Hui Liu, Chuo Li

Posture Recognition and Heading Estimation Based on Machine Learning Using MEMS Sensors

With the popularity of smartphones and the performance improvement of embedded sensor, the smartphone has become the most important terminal device in motion recognition and indoor positioning. In this paper, the methods of the smartphone posture recognition and the pedestrian heading estimation are proposed. We analyze the signal characteristic of the accelerometer and the gyroscope, the representative feature information is extracted and a classifier based on DT model is proposed. Besides, considering the different postures of the smartphone, we propose an improved heading estimation method, which utilizes a weighted-average operation and combines the principal component analysis-based (PCA-based) method and the angle deviation method innovatively. The results of the experiments show that the average accuracy of posture recognition is nearly 97.1%, which can satisfy the pattern recognition in the process of pedestrian navigation. The average error of the proposed heading estimation is 6.2° and the performance is improved than the single PCA-based and angle deviation method.

Boyuan Wang, Xuelin Liu, Baoguo Yu, Ruicai Jia, Lu Huang

Backmatter

Weitere Informationen

Premium Partner

    Bildnachweise