International Journal of Sensor Networks (24 papers in press)
Improved Fault-Tolerant SPRT Detection Method for Node Replication Attacks in Wireless Sensor Networks
by Hsin-Hsiu Chen, Cooper Cheng-Yuan Ku, David Yen
Abstract: As the Internet of Things (IoT) emerges, the application and usage of the wireless sensor network (WSN) have been increasing quickly. Meanwhile, there are many threats and risks of information security that need to be effectively dealt with. One of these network attacks is the node replication attack. The attackers may catch sensor nodes, clone them, release them into the original network, and launch various internal attacks. Many replica detection instruments/methods are proposed for this type of attack. However, most detections are characterized as high computation and communication costs. Some detection methods based on the Sequential Probability Ratio Test (SPRT) indicate much lower requirements of system overhead, but these prior works may sacrifice efficiency due to frequent retransmission of the message. In this paper, we propose a fault-tolerant method for replica detection based on the SPRT in WSNs. To improve the efficiency and reliability, we use the residual energy and slope of energy consumption of the node as appendices and then apply the SPRT to adjust the detection rate dynamically. The simulation results show that our proposed scheme achieves a better performance on the efficiency of detection and reduction of error rates.
Keywords: Clone attack; Wireless sensor network; Node replication detection; Sequential probability ratio test.
Fast Fingerprint Localization Based on Product Quantization and Convolution Neural Network in a Massive MIMO System
by Yijie Ren, Xiaojun Wang, Lin Liu, Xiaoshu Chen
Abstract: Location-based service (LBS) has recently been popular, such as auto-driving, navigation, and tracking. Fingerprint localization is one of the most effective localization schemes for both indoor and outdoor localization. In this paper, ?ngerprint localization algorithms are researched based on a massive multiple-in-multiple-out (MIMO) system. Firstly, the extraction of angle-delay channel power matrix (ADCPM) ?ngerprints and the channel model are introduced. Then, two new fast ?ngerprint localization algorithms based on product quantization (PQ) and convolution neural network (CNN) are proposed, respectively. PQ and CNN are applied to process the data in the online matching phase. Compared with other previously known positioning techniques, the test results show that the proposed algorithms achieve high accuracy, reduce delay, and greatly reduce computational complexity.
Keywords: ?ngerprint localization; massive multiple-in-multiple-out; product quantization; convolution neural network.
One Health-Inspired Early Implementation of Airborne Disease Spread Mitigation Protocols aided by IoT-based Biosensor Network
by Uche Chude-Okonkwo
Abstract: The implementation of mitigation protocols in the event of the outbreak of viral epidemics typically comes only after a significant number of infected persons have been diagnosed, within which period the infection may have spread significantly over the population. The One Health initiative provides an alternative approach to this challenge. Based on the recent advances in biosensor technology capable of accurate real-time viral diagnosis, and the advances in the internet of things technology (IoT), this paper proposes the development of a pervasive biosensor network deployed at the animal habitats in communities and strategic positions in the environments. This enables the early detection of infectious diseases and the implementation of mitigation protocols. The model simulation results show that the use of the IoT-based biosensor network to inform the implementation of the epidemic mitigation protocol ensures the reduction in infection rate and the cumulative number of infected persons in a given population.
Keywords: Biosensor network; One Health; Internet of things; epidemics.
Privacy Preservation and Security Challenges: A New Frontier Multimodal Machine Learning Research
by Santosh Kumar, Mithilesh Kumar Chaube, Srinivas Naik Nenavath, Sachin Kumar Gupta, Sumit Kumar Tetarave
Abstract: Multimodal machine learning is a vibrant multi-disciplinary field and achieved much attention due to its wide range of applications. A research problem is multimodal, for the impact of privacy preservation. It shields sensitive data in the cloud by using a single modality-based privacy system. The users biometric features are always stored in the database, primarily present in the cloud server to validate the user and his access. This facet provides a beneficial quality but at the same time has raised crucial affairs in security and privacy of biometric feature set. The main concern is to manage and stop the privacy breaches in clouds. The article discusses the detailed analysis of security schemes with a multimodal-based learning framework over sensitive data and systems at both ends. The article also accentuates frameworks and schemes that may apply in various applications to ensure privacy preservation of individuals and data security by multimodal algorithms.
Keywords: Privacy; Preservation; Security; Biometrics; Attacks; Encryption; Cryptography; Security; Multimodal.
Fault Diagnosis and prediction with Industrial Internet of Things on Bearing and Gear Assembly
by Gagandeep Sharma, Tejbir Kaur, Sanjay Kumar Mangal
Abstract: In the era of automation, mechanical components such as bearings and gears are widely used in industrial machinery to transmit power and motion. Failure in these components directly affects the functioning of the machinery and causes the loss of money and time. Therefore, fault diagnosis and prediction of these components in advance are necessary to avoid catastrophic consequences. In this research, an experimental set-up is developed to predict the fault for various cases such as proper configuration, defective bearing, and defective gear configuration. An IIoT and conventional time and frequency domain-based techniques are used for condition-based monitoring of bearing-gear assembly. IIoT based systems can perform three major tasks; measuring and displaying the real-time vibrational responses of bearing-gear assembly, comparing it with the prescribed threshold value, and sending a warning message to the end-user using Blynk application, if the acquired acceleration values are greater than the prescribed threshold value
Keywords: Industrial Internet of Things (IIoT); Blynk application; Bearing; Gear; NodeMCU; Vibration analysis;.
An IoT based Intelligent Traffic Engagement System with Emergency Vehicles Preemption
by Bharat Pant, Harsh Sharma, Rashmi Chawla, Chirag Kumar
Abstract: The increase in vehicles beyond the handling capacity of transport infrastructure leads to congestion-like situations. As a consequence, vehicles experience extended waiting time on roads and a severe impact on emergency services. The existing works are either focused on normal traffic management or handling emergency situations. This manuscript aims to provide a complete solution to address these tasks. The proposed system updates the timer dynamically using Kullback-Leibler (KL) divergence to measure the traffic density to handle the normal traffic. In the case of emergency vehicles (EV), the system is added with EV preemption. Finally, the system has the capability to handle multiple EVs appearing simultaneously in different lanes. For the efficacy of the proposed system, multiple experiments and comparisons with state-of-the-art techniques are carried out using the Simulation of Urban MObility (SUMO) traffic simulator.
Keywords: Traffic management; emergency vehicle preemption; Internet of Things (IoT).
Detection and Defense Method of Low-Rate DDoS Attacks in Vehicle Edge Computing Network Using Information Metrics
by Xiao Bai, Shanzhi Chen, Yan Shi, Chengzhi Liang, Xiaochen Lv, F.Richard Yu
Abstract: In vehicle edge computing network, the edge nodes are usually deployed at the road-side units (RSU) and exposed to the external environment, and are vulnerable to low-rate DDoS (LR-DDoS) attacks. In this article, we raise the detection and defense issue of LR-DDoS attacks based on information metrics at the edge of vehicle network, in which multiple edge nodes cooperate to meet their defense requirements while reducing resource consumption. We formulate the problem as sampling the received traffic at the edge nodes, and automatically determine the attack traffic by comparing the real-time calculation information metrics value with the preset information metrics thresholds. Then, a defense algorithm for edge nodes to detect conflicting Mobile Subscriber International ISDN number (MSISDN) is proposed to ensure that the network continues to provide services to normal vehicles. Finally, extensive experimental evaluation and comparison are carried out to show the effectiveness of the proposed scheme.
Keywords: Attack Detection and Defense; Vehicle Edge Computing Network; Low-rate DDoS Attack; Information Metrics.
ASSESSMENT OF SPECTRUM SENSING USING SUPPORT VECTOR MACHINE COMBINED WITH PRINCIPAL COMPONENT ANALYSIS
by Manash Mahanta, Attaphongse Taparugssanagorn, Bipun Man Pati
Abstract: Cognitive Radio (CR) is an up-and-coming technology to help us rectify the issue of under-utilization of allocated spectrum as well as meet the increasing demand for free spectrum. Spectrum sensing, which is the bedrock behind this novel technological idea, empowers the CR to adjust to its surroundings by locating free spectrum or white space. Although spectrum sensing using a Support Vector Machine (SVM), which is the most effective machine learning algorithm, is already found in literature, but an SVM combined with Principal Component Analysis (PCA) and varying the Kernel scale is yet to be investigated. In this paper, we perform spectrum sensing using an SVM and evaluate the performances of various Kernel functions used in the SVM as well as how the performances of the learning algorithm change as we apply PCA and vary the Kernel scales. We found that for the purpose of maximising the Probability of Detection (PD), it is better not to use PCA while for the purpose of minimising Probability of False Alarm (PFA), it is better to use PCA. Moreover, we found that for the purpose of maximising the PD, it is most advised to keep the value of Kernel scale low, around 10 while for the purpose of decreasing the PFA, the value of Kernel scale is suggested to be high, around 40-45. We then compare the training time of the SVM Kernels. Finally, we calculate the contributions of power, variance, skewness, and kurtosis of the received signal towards the decision making process of the learning algorithm.
Keywords: Orthogonal Frequency Division Multiplexing; spectrum scarcity; cognitive radio; spectrum sensing; Machine Learning; principal component analysis; Support Vector Machine; large margin classifier; Kernel scale value; feature importance; Internet of Things;.
HAPS-SMBS 3D Installation for 5G FANET Performance Optimization Using a DNN, WCOP and NetSim Based Scheme
by Joyeeta Rani Barai, Attaphongse Taparugssanagorn
Abstract: High altitude platform stations (HAPSs) can be installed as flying super macro base stations (SMBSs) to serve the existing terrestrial infrastructure as a flying ad hoc network (FANET) keeping enhanced and continuous wireless network connections in normal and disaster scenarios. This dynamic process becomes inefficient if the mobility and locations of the user equipment (UE) are unknown and uncertain. Although increasing the number of HAPSs provides more user connectivity it is too expensive and impractical. Hence, installations of the available HAPSs in the optimum positions can solve the problem in cost effective manner. This paper proposes a novel combinational scheme consisting of the two following stages, i.e., the deep learning (DL)-based forecasting of the ground user (GU) locations and the weighted combinatorial optimisation problem (WCOP)-based HAPS-SMBS positioning to optimise the 5G FANET performances in both normal and disaster scenarios.
Keywords: super macro base station; SMBS; high altitude platform station; HAPS; DNN; weighted combinatorial optimisation problem; WCOP; NetSim; 5G.
Development and Evaluation of a Low-Cost Data Acquisition System using Heterogeneous Sensors
by Muhammad Zubair, Hafiza Nida Ishaq, Ali Raza, Hafiz Farhan Maqbool, Saqib Zafar
Abstract: Acquiring the humans musculoskeletal movements is pivotal in conducting physiotherapeutic studies, rehabilitation exercises and, designing assistive devices to cater mobility impairments. Existing studies are generally more focused on measuring a specific phenomenon and/or lack the diversity needed to cover a range of human activities. In this context, the research presented herein is aimed to develop a low-cost wearable sensory system, having a diverse set of sensors, which can acquire data of healthy and rehabilitated subjects alike during diverse locomotor activities in both indoor and outdoor environments. The system consists of twenty wearable sensors, and two off-the-shelf NIs myRIO microcontroller boards. The statistical analysis showed no significant difference between and among the subjects for various locomotor activities (P-values < 0.05), hence, showing the systems reliability and reproducibility. Gait-event identification (i.e., heel contact/toe off) has also been evaluated and showed promising results (overall time difference= ± 50 ms during level ground and ramp activities)
Keywords: Heterogeneous sensors; locomotor activities; gait events; data acquisition; public dataset; low-cost.
Environment-aware vehicle lane change prediction using a cumulative probability mapping model
by Yongxuan Sun, Bowen Zhang, Zhizhong Ding, Momiao Zhou, Mingxi Geng, Xi Wu, Jie Li, Wei Sun
Abstract: Vehicles lane change is a major cause of serious traffic accidents. Therefore, it is imperative to implement an efficient and reliable lane change prediction and warning sub-system, for example in Advanced Driving Assistance System (ADAS), to improve driving safety. Many approaches concerning the issue have been proposed so far. To have a good prediction performance, however, most of them require a large number of training data, and/or need radar or video equipment. Motivated by the demand in our development of Vehicular Ad-Hoc Network (VANET) terminals, a cumulative probability mapping (CPM) model for the prediction of lane change is proposed in this paper. The CPM model is constructed by taking into account both the vehicle motion information and the traffic context information that is acquired from VANET communications. The results based on the Next Generation Simulation (NGSIM) dataset, which is made up of real vehicle trajectories, show that the accurate rate of the proposed method is 95.8%, and the false alarm rate is 11.6%. From the view of implementation, the proposed approach has the characteristics of low cost and low computational complexity, and can be used for embedded devices.
Keywords: Lane change prediction; Probability mapping; VANET; ADAS.
Performance of Cognitive Radio Networks using Reconfigurable Intelligent Surfaces with RF Energy Harvesting for Nakagami channels
by Raed Alhamad
Abstract: In this paper, we derive the performance of Cognitive Radio Networks (CRN) with energy harvesting using Reconfigurable Intelligent Surfaces (RIS) for Nakagami fading channel with m-fading figure $M$. We derive the detection probability when the Primary Source ($PS$) harvests energy using Radio Frequency (RF) signals. A RIS is located between $PS$ and Secondary Source ($SS$) where spectrum sensing is performed. We also derive the primary and secondary throughput and optimize harvesting duration to maximize the throughput. We observed significant performance enhancement in detection probability and throughput with respect to CRN without RIS.
Keywords: CRN; RIS; Energy harvesting; throughput analysis; detection probability.
A bi-level mathematical model to protect gateways in underwater wireless sensor networks
by Zahra Maleki, Hamid Reza Maleki, Reza Akbari
Abstract: Since energy resources in underwater wireless sensor networks (UWSNs) are limited and non-renewable, energy efficiency in these networks is essential. Jamming attacks are the most common denials of service attacks in UWSNs that increase the amount of energy consumption. In this paper, a bi-level model for the protection of gateways in underwater sensor networks is introduced. In this model, the attacker intends to increase the networks energy consumption by placing a determined number of jammers on the gateways; Hence, the network manager wants to assign anti-jamming techniques to a subset of gateways to protect them against these attacks. The efficiency of this model in reducing energy consumption after a jamming attack is shown in several instances. These instances are solved with hybridization from genetic, simulated annealing, and dragon?y algorithms with CPLEX solver, and their performances are evaluated with the two-sample t-test.
Keywords: bi-level model; underwater wireless sensor networks; jamming attacks.
A Brain-inspired Multibranch Parallel Interactive Vision Mechanism for Advanced Driver Assistance Systems
by Wei Ou, Zihan Jin, Shiying Huang, Danlei Du, Jun Ye, Wenbao Han
Abstract: Brain-inspired research promotes the intersection and integration of brain science, computers, and other disciplines, driving a new round of scientific and technological revolution. This paper discusses the research content, features, and current status of researches on brain-inspired computing and proposes a brain-inspired multibranch parallel interaction model. The feature extraction of the proposed model consists of two parallel Convolutional Neural Networks (CNNs): the main feature extractor connected to the classifier and the auxiliary feature extractor. Additionally, four parallel branches are introduced in our CNN to learn the information on the importance of different feature map positions. Given the existing attacks on self-driving cars (in the case of shining attacks), we use the proposed model for training and detection. The experimental results show that the proposed model effectively reduces the redundant information on the feature maps, improves image classification accuracy to a certain extent, and has high learning efficiency and accuracy.
Keywords: Brain-inspired; CNNs; Multibranch parallel interaction; Feature extraction; Shining attack.
Blockchain in Vehicular Ad hoc Networks: Applications, Challenges and Solutions
by A. F. M. Suaib Akhter, Mohiuddin Ahmed, Adnan Anwar, A. F. M. Shahen Shah, Al-Sakib Khan Pathan, Ahmet Zengin
Abstract: Blockchain has been adopted in a wide range of application domains to enhance security and privacy. VANET (Vehicular Ad hoc Network) is an important application domain in today's communication systems where incorporation of blockchain is very timely. Recent literature highlights the prospects of blockchain technology in VANET, however, it is imperative to investigate the effectiveness to ensure viability. In this paper, a thorough investigation is conducted to identify the suitability of blockchain for VANET by identifying and answering key research issues. Unlike other existing surveys, challenges related to blockchain integration, evaluation criteria, privacy preservation, cyber security, etc. are also critically analysed. Future research directions such as 6G, large scale deployment are also identified which need to be addressed by both VANET and blockchain community. Not only VANET but also Vehicular Communication Systems (VCSs) and Intelligent Transportation Systems (ITSs) have been considered in this survey.
Keywords: Blockchain; cyber; intelligent; security; system; transport; VANET; cyber security.
Time Series Representation for Information Gathering via Low Resolution Wireless Sensor Networks
by Jorge Navarro, Isaac Martin De Diego, Ana R. Redondo
Abstract: New applications of the internet of things are emerging in sectors as diverse as military, environmental, health, or food due to the improvements achieved in the development of wireless sensor networks (WSNs). In many of these applications, as the connected devices are hardly accessible and energy harvesting is not possible, it is essential to provide long-term autonomies to the end user. Hence, there is a need for developing energy-efficient strategies for data gathering that send few messages accumulating as much information as possible. This paper proposes a new strategy for sending summarised information from commercial accelerometers deployed through WSNs. An exhaustive evaluation has been performed using data from experimental devices that gathered 106 different daily time series related to animal behaviour. The obtained results show that the proposed strategy highly improves the current operation mode of the commercial devices.
Keywords: internet of things; IoT; wireless sensor networks; WSN; data science; time series; animal behaviour; information gathering; accelerometers.
Energy efficiency clustering based on fuzzy logic in heterogeneous wireless sensor networks
by Xiao Yan, Cheng Huang, Lili Wang, Xiaobei Wu
Abstract: Wireless sensor networks (WSNs) must be decentralised and independent of wireless sensor nodes that can track physical or environmental status. The operation of a node depends on the batterys life. Therefore, the batterys life is one of the key issues that limit the wide application of WSNs. In this article, a fuzzy logic system-based energy-efficient clustering (FLEEC) in heterogeneous WSN is introduced to solve the problem of maximising the network lifetime. To be more precise, three factors are considered, such as node remaining energy, distance, and proximity of the neighbour nodes and combined with fuzzy logic, to select the appropriate sensor node as cluster head (CH), thereby balancing the energy consumption of sensor nodes, and prolonging the network lifetime. Finally, the simulation results show that the FLEEC algorithm is superior to the existing algorithms in the aspects of the stable period, half node death, throughput, and network lifetime.
Keywords: wireless sensor network; WSN; fuzzy logic system; cluster head selection; energy; network lifetime.
Trust Assessment based Stable and Attack Resistant Grouping Strategy for Data Dissemination in IoV
by Richa Sharma, Teek Parval Sharma, Ajay Kumar Sharma
Abstract: The coalescence of Vehicular Ad-hoc Network (VANET) with the Internet of Things (IoT) results in the concept of the Internet of Vehicles (IoV). IoV structures a strong
backbone for Intelligent Transport System (ITS), however due to highly dynamic vehicular networks the trustworthiness of disseminated messages is significant since a fake message might cause fatal mishaps on the road. To address these issues, in
this article trust assisted grouping approach has been proposed to limit the malicious communication among vehicles in both dense and sparse traffic scenarios and enhance security. The key highlights consolidated in our proposed approach incorporate trust-based GH selection based on direct and indirect trust assessment parameters. Trust assessment is performed by Road Side Unit (RSU) to maintain the degree of trust among vehicles of the group by evaluating direct trust assessment based parameters like realization, reputation, experience and indirect trust assessment based parameters like suggestion factor. Vehicle attaining supreme trust value is elected as Group Head, which enhances trust among Group Member. The performance and accuracy of our approach is rigorously evaluated under various
trust assessment parameters (including Group head lifetime, average group lifetime, fake message generation attack detection, Sybil attack detection, data overhead and detection performance) in both sparse and dense scenarios. Furthermore, the efficiency of our approach is evaluated under malicious vehicles presence and performance is benchmarked compared to two earlier trust assessment models which concludes that the malicious vehicles are unable to gain higher trust assessment values. Extensive simulations performed in Network Simulator (NS3) and mobility
simulator Simulation Mobility (SUMO) indicate that our approach achieve promising results compared to two earlier trust assessment models.
Keywords: IoV; Direct trust assessment; Attack Model; NS3; SUMO.
Spatial properties verification approach of wireless sensor networks using model checking
by Jia Wang, Jun Niu, Zheke Yuan, Fan Fei
Abstract: Reasonable spatial deployment of wireless sensor networks (WSNs) can extend their life cycles and enhance the reliability of the systems relying on them. The spatial layouts of WSNs need to be strictly verified to ensure they satisfy desired requirements. Model checking has made some progress in verifying spatial properties of WSNs, and the Spatial Logic for Closure Space (SLCS) can characterize some spatial specifications of WSNs. However, SLCS is imperfect in characterizing and reasoning about some quantitative properties of WSNs. In this paper, we firstly propose a novel WSN spatial model, namely Quantitative Closure Space Model (QCSM), to model WSNs with quantitative descriptions. Then, we use the improved SLCS logic, which is extended with a novel degree-correlation operator to enrich the expressiveness of SLCS, to characterize more kinds of spatial specifications of WSNs. Experiments indicate that our approach can automatically and effectively verify some complex spatial specifications of WSNs.
Keywords: Wireless sensor networks; spatial deployment; spatial specification; modal logic; spatial logic for closure space; quantitative closure space model; degree-correlation operator;.
Preserving privacy in Mobile Crowdsensing
by Bayan H. Alamri, M.M. Monowar, Suhair Alshehri, Haseeb Zafar, Iftikhar Ahmad
Abstract: Mobile crowdsensing (MCS) is a technique where individuals voluntarily utilise their devices to collect data to measure phenomena. In this article, a review of privacy-preserving in MCS is presented. First, it highlights MCS definitions, architecture, and unique characteristics. Then, it provides background knowledge about MCS. Afterward, a privacy-oriented MCS taxonomy in terms of privacy-oriented; data reliability, incentive, and task allocation user recruitment mechanisms, is devised. This work explores contemporary state-of-the-art issues related to privacy and security. It reviews 35 recent research published by high-quality sources and provides a topic-oriented survey for these efforts. It shows that only 16% of the papers evaluate their schemes through experiments on real smartphones, and Huawei is the most widely used mobile (45%). It shows an increasing trend in publications from 2017 till now. It highlights recent challenges faced the privacy in MCS and potential research directions for developing more advanced methods to optimise MCS
Keywords: mobile crowdsensing; MCS; privacy preservation; data reliability; untrustworthy; incentive; user recruitment.
Early Detection of a Fall Using Wi-Fi and Deep Learning
by Warayut Surasakhon, Attaphongse Taparugssanagorn, Sarawut Lerspalungsanti, Khirakorn Thipprachak
Abstract: There have been millions of dramatic falls in the elderly leading cause of traumatic injury and death. While the problem is more serious, the number of elders is continuously increasing. To relieve this issue, early detection of a fall is very helpful for elders who are concerned about falling or a health problem when alone and can automatically alert medical responders. In this paper, we present a fall detection system using Wi-Fi channel state informations (CSIs). Wi-Fi technology is selected since it is ubiquitous allowing our system to work everywhere even in the private area like in a bathroom and to be easily implemented with low cost. The CSI provides information not only about the environment, but also human movements. To interpret fall or non-fall event, we employ several classification algorithms as well as deep learning models, which can provide very accurate results up to 97.7%.
Keywords: fall detection; activity recognition; channel state information; CSI; machine learning; deep learning.
Security Demonstration for the Quantum Noise-based Physical Layer Using Variable Keys
by Shuai Shi, Ning Xiao
Abstract: With the continued advancement of science and technology, a large amount of important information is carried by optical fibre networks. Therefore, it is imperative to use secure transmission strategies to protect important information. The Y-00 cipher that employs multi-order modulation to prevent eavesdropping on ciphertext is a practical candidate for providing data protection at the physical layer. The Y-00 cipher combines the mathematical encryption of multilevel signalling and quantum noise to provide high security to fibre communications. This paper proposes a quantum noise-based physical layer secure transmission scheme, combining the Y-00 cipher with time-domain spectral phase encoding (TDSPE). The operation methods of the Y-00 cipher in the data encryption and TDSPE in the key distribution are introduced. Then, the system performance is investigated by transmission experiments. The noise-masking phenomenon is demonstrated and quantified. The probability of the eavesdropper guessing cipher text correctly is evaluated. Last, the proposed secure transmission is achieved at 1 Gbps over a 100.2 km optical fibre link, with an intensity level of 1024 and a noise masking number of 71. The experimental results prove the effective feasibility and high security.
Keywords: quantum noise-based; variable keys; physical layer; optical fibre link; Y-00 cipher; TDSPE.
Multi-applicable text classification based on deep neural network
by Jingjing Yang, Feng Deng, Suhuan Lv, Rui Wang, Qi Guo, Zongchun Kou, Shiqiang Chen
Abstract: Most long text classification methods based on deep learning have problems such as semantics sparsity and long-distance dependence. To tackle these problems, a novel multi-applicable text classification based on deep neural network (MTDNN) is proposed, which contains a bidirectional encoder representation from transformer (BERT), a dimension reduction layer, and the bidirectional long short-term memory (Bi-LSTM) combining attention mechanism. BERT is used to pre-train the words into the word embedding vectors. The dimension reduction layer extracts the feature phrase representations with higher weight from the word embedding vectors. The Bi-LSTM captures both the forward and backward context representations. An attention mechanism is employed to focus on the information outputted from the Bi-LSTM. The experimental results illustrate that the accuracy of the MTDNN for long text, short text classification, and sentiment analysis reaches 94.95%, 93.53% and 92.32%, respectively. The results show that our method outperforms the other state-of-the-art text classification methods.
Keywords: text classification; deep neural network; BERT; long short-term memory; LSTM; attention mechanism; multi-applicable.
An adaptive multi-group slime mould algorithm for node localization in wireless sensor networks
by Xiankang He, Lijun Yan, Shi-Jian Liu, Jixiang Lv, Jeng-Shyang Pan
Abstract: Node localization is a common and significant practical application question in wireless sensor network (WSN). The goal of this problem is to use anchor nodes in the network to estimate the geographical location of the unknown node. A novel algorithm, named adaptive multi-group slime mould algorithm (AMSMA), is proposed in this study. The improved slime mould algorithm uses the multi-group strategy and adaptive communication mechanism to alleviate the lack of population diversity, development and exploration imbalance of the slime mould algorithm. The proposed AMSMA was tested under CEC2013 test suite. Compared with SMA and corresponding optimization algorithms, the AMSMA is more effective and efficient. In addition, a novel localization algorithm based on AMSMA is proposed. The AMSMA-Distance Vector-Hop (AMSMA-DV-Hop) is applied to the localization of WSN. Compared with some other existing localization algorithms, the proposed AMSMA-DV-Hop is an effective algorithm for the localization of WSN.
Keywords: WSN; multi-group; adaptive; slime mould algorithm