This study explores the use of the Fledge platform, an open-source solution based on Fog Computing, to improve real-time health data management in the context of urban home care. Faced with the growing challenges of urbanization and the increase in chronic diseases, our approach aims to optimize the collection, processing, and securing of health data. The proposed architecture combines Fog and Cloud Computing, allowing for data pre-processing at the source and advanced analysis in the cloud. The results demonstrate Fledge’s exceptional performance in terms of reduced latency, data security, and integration with existing systems. The study involved setting up a test environment simulating a mobile health laboratory, using real sensors to collect physiological data. Although promising, the study has limitations in terms of sample size and experiment duration. Future research should focus on integrating advanced AI algorithms and expanding Fledge’s interoperability, paving the way for more efficient and patient-centered urban health systems.
Software-Defined Networking (SDN) represents the next generation of network infrastructure, offering flexibility and the ability to quickly integrate services tailored to specific business applications and use cases. Motivated by the rapid rise of SDN and the increasing focus on Quality of Service (QoS) within SDN networks, this project explores how SDN can replace traditional network architectures and improve QoS management. Specifically, we investigate the role of protocols such as OpenFlow, iPerf, and Open vSwitch Database Protocol, which are pivotal in configuring and managing virtual bridges, ports, tunnels, and queues in SDN environments. The project emphasizes the QoSFlow solution for effectively managing QoS in SDN networks, demonstrating its potential for enhancing flow prioritization and overall network performance.
To address the growing demand of high-speed and low-latency communications in access network segments, IEEE 802.3ca task force developed Next Generation Ethernet Passive Optical Network (NG-EPON). In this network, an Optical Line Terminal (OLT) and an Optical Network Unit (ONU) can have multiple transceivers which allow them to communicate through multiple channels simultaneously (channel bonding) to achieve aggregated high transmission rate. Due to the use of a large number of transceivers in this network, the energy consumption should be supposedly high. In this paper, we propose an Adaptive Cyclic Transmission Operation (ACTO) scheme for NG-EPON with the objective of reducing energy consumption using sleep mode without compromising performance requirements. In particular, ACTO sets transmission cycle duration of each channel taking into consideration the traffic delay performance requirements. It also presents mechanisms how energy consumption for the idle channel(s) can be minimized and how signaling associated energy consumption in OLT/ONU transceivers can be saved. Performance evaluated based on simulation shows that the proposed solution conserves 15% of energy in ONUs even during high traffic load conditions. Furthermore, the OLT side energy consumption is also reduced by approximately 13.5% in ACTO compared to a conventional solution.
Wireless Rechargeable Sensor Network (WRSN) is a new paradigm that prolongs the lifetime of Wireless Sensor Network (WSN). To improve the survival rate of sensor nodes and energy usage efficiency, this paper studied how to schedule charging paths of multiple mobile chargers and allocate charging time simultaneously. We propose an on-demand partial charging scheme for multiple mobile chargers. First, to balance the charging workload of multiple mobile chargers, we proposed a new clustering algorithm to divide the sensor nodes into several clusters. Then, we proposed an algorithm of determining the target charging sensor node. Finally, based on deep reinforcement learning technique, the charging duration calculation strategy is designed to automatically allocate charging duration for the sensors. Extensive simulations show that, compared with baseline algorithms, our scheme can increase the survival rates of sensor nodes by 3.62%~11.95% and increase the energy usage efficiency by 1.57%.~6.15% in different network scale.
This paper proposes a method for expanding the metadata of three-dimensional point cloud data using Large Language Models (LLMs). Currently, point cloud data plays a crucial role in various fields such as autonomous driving and medical image reconstruction, necessitating the expansion of metadata for efficient processing. Traditionally, metadata construction has relied on manual input, which is prone to errors. In this study, we propose a method that utilizes LLMs, particularly the Llama 3.1 model, to extract the center points of each class in the point cloud data and expand the metadata by adding these center points to the annotation files. By using center points, computational costs are reduced, and the performance of segmentation and detection models based on this data is improved.
A CCTV-based intrusion detection system is a security technology that detects intrusions by analyzing real-time video feeds, commonly used in public spaces, businesses, and other critical infrastructures to safeguard assets and individuals. These systems are essential for ensuring security, as they provide continuous monitoring and alert capabilities. However, conventional systems often rely on simple rule-based models that lack adaptability to complex and dynamic environments, leading to frequent false positives and missed detections. For example, environmental factors such as lighting changes, shadows, or the movement of animals can trigger unnecessary alarms, creating inefficiencies and diminishing trust in the system's reliability. Additionally, these systems usually require intricate manual configurations, where users need to define specific detection rules for each camera or surveillance area. This can become particularly challenging in large-scale environments with multiple cameras, as it increases the risk of human error, slows down deployment, and limits the system’s overall management efficiency.To address these limitations, the proposed model acquires real-time streaming data from CCTV cameras using the RTSP protocol and utilizes the YOLO deep learning object detection algorithm to accurately detect objects such as people, vehicles, and other relevant entities within the video. YOLO’s high-speed and precise object detection capabilities allow for more reliable identification of potential intrusions, even in real-time scenarios. Once the objects are detected, their positional data is fed into a spatiotemporal grid frequency analysis model, which examines the frequency of object occurrences over time and across different areas within the surveillance zone. Areas with low occurrence frequencies are automatically designated as restricted zones using a threshold mechanism, reducing the need for manual input and constant monitoring. This automated approach not only lightens the load on users but also significantly enhances the system’s ability to detect actual threats while minimizing false alerts. By integrating advanced deep learning techniques and automation, the model offers a more efficient, accurate, and user-friendly solution for modern security needs
Recently, there has been a lot of research on the Text2Image generative model as social and technological interest in generative models has increased. In addition, the interest in Korean contents (K-Contnets) has increased. In this paper, we propose "K-Contents Specialized Text2Image Pipeline". "K-Contents Specialized Text2Image Pipelilne" is specialized for understanding Korean prompts and aims to generate images specialized for various K-Contents. To build the pipeline, we collect and preprocess text and image data. Then, we train the Korean Text Encoder on the text data and train three specialized K-Contents generation pipelines on the image data. We also propose to use the SDXL model’s Img2Img Refiner with the SD model for time-efficient image generation. After that, we use the learned pipelines to generate real images and analyze their performance. Finally, we discuss the utilization of the above pipelines, limitations, and future research.
In the field of ecological monitoring, a significant aspect is bird sound classification. As different movements are observed by birds, classification of its sounds is quite important. The target information can be fully explained with the help of suitable feature extraction and selection schemes. In this paper, the first approach utilizes a parameter pliable Variational Mode Decomposition (PP-VMD) technique for extracting the features and then for feature selection Colliding Bodies Optimization (CBO) algorithm is utilized. The chosen features are then classified with the conventional machine learning classifiers. The second approach uses the concept of spectral clustering (SC) and K-means algorithm with clustering the initial centers (KIC) for feature extraction and Gravitational Search Algorithm (GSA) is used for feature selection before classifying it with machine learning classifiers. The proposed two strategies are implemented on publicly available bird sound classification dataset and the best results are obtained when PP-VMD with CBO and SVM classifier is used reporting an accuracy of 94.45%.
In modern battlefields, the stability of wireless communications is crucial for intelligence transmission, command coordination, and maintaining strategic advantage. However, with the rapid advancement of communication technologies, the electromagnetic environment has become increasingly complex, making intelligent decision-making in communication countermeasures a prominent research focus. However, most existing studies adopt deep learning and reinforcement learning methods, which require extensive data samples and computational resources. To address these limitations, this paper proposes a game-theoretic model for intelligent radio communication countermeasures under incomplete information. It quantitatively evaluates common attack and defense tactcis used in communication countermeasures and explores the mixed-strategy equilibrium of the game when both the attackers and defenders consist of multiple types. Furthermore, simulation-based comparative analyses are conducted to assess the factors influencing the utility of both the attacker and defender. The findings offer theoretical insights and decision-making support for the deployment of communication countermeasure strategies in practical military operations, contributing to significant strategic value.
This paper proposes an effective testing method that combines blockchain and metaverse technologies. It analyzes the interaction between blockchain networks and virtual reality worlds, leading to the development of a secure and efficient testing process. The paper evaluates this interaction in various scenarios, offering insights to enhance both reliability and security. By integrating metaverse and blockchain technologies, this study makes a significant contribution to improving safety in the digital environment. Additionally, it proposes key evaluation factors and formulas to objectively assess the integration of blockchain and metaverse services. These evaluation metrics can be valuable in developing assessment tools and simulation instruments.
With the worldwide increase in disaster occurrences and their scale of damage, a resilient Information and Communication Technology (ICT) system that remains available in disaster-affected areas despite Internet, telecom, or mobile service disruptions is strongly desired. In particular, disaster first responders who rush to disaster-stricken areas to carry out investigations, search and rescue operations for people, require tools to collect, share, report, and record disaster information. This paper proposes a portable and elastic edge computing network to support the activities of disaster first responders. The portability and elasticity features enable us to adaptively apply the proposed edge computing network in various disaster situations. In the edge computing node, Artificial Intelligence (AI) functions, such as automatic speed recognition, are equipped to offer hands-free operation and on-site intelligent image processing for first responders. We developed a pilot model of an edge computing node and confirmed the requirements to include AI functionality through experimental evaluation.
The primary objective of this study is to develop a robust model that assists financial institutions in identify potential customers with a higher likelihood of adopting car loans. We use a unique dataset of remittance transactions and vehicle financing data provided by a commercial bank in Uzbekistan. To balance data, we apply data sampling techniques. We then compare the performance of Logistic Regression (LR), Decision Tree (DT), and Random Forest (RF) models in these two datasets. Our analysis reveals than all models perform better on the Synthetic Minority Over-sampling Technique Edited Nearest Neighbours (SMOTE-ENN) dataset. DT outperforms the other models. Based on these insights, we recommend using DT and SMOTE-ENN techniques on imbalanced datasets. The results of this study offer practical implications for data scientists and financial institutions in remittance-receiving countries aiming to leverage remittance flows, boost cross-selling, and increase revenue.
The Fourth Industrial Revolution (IR4.0) has transformed various sectors, including cultural heritage preservation, education, and tourism, through advanced technologies like virtual reality (VR). This study explores the intersection of technology and cultural heritage learning, highlighting the crucial role of user experience (UX) in VR applications. While VR holds significant potential for cultural heritage learning, existing research often emphasizes utility and usability over UX. To address this gap, the study evaluates the UX of VR applications designed for cultural heritage, specifically focusing on the Istana Jahar historical site. The objectives include assessing dimensions such as effectiveness, efficiency, attractiveness, satisfaction, emotion, engagement, attention, and perception. The research employs quantitative methods using the Design Science Research Methodology (DSRM) to guide the development and evaluation of the VR application. A cluster random sampling technique was used to recruit 113 students from Universiti Malaysia Kelantan (UMK), ensuring a representative sample with relevant educational backgrounds. Participants interacted with the VR application and completed a survey based on the developed conceptual framework. Findings indicate high mean scores across UX dimensions: Effectiveness (4.487), Efficiency (4.363), Attractiveness (4.528), Satisfaction (4.363), Emotion (4.401), Engagement (4.385), Attention (4.366), and Perception (4.416). These results demonstrate that the elements within the conceptual framework effectively enhance the UX in cultural heritage learning applications.
With higher access capacity, non-orthogonal multiple access (NOMA) is a promising wireless access technology to meet the demands of massive machine-type communications (mMTC) and internet of things (IoT) applications. Focused on NOMA-based uplink multiple access transmission, this paper proposes data detection schemes that integrate sparse code multiple access (SCMA) and successive interference cancellation (SIC) techniques, termed the SCMA-SIC process. Simulation comparisons of various NOMA receiver designs, including hybrid-domain NOMA and iterative detection and decoding (IDD) for coded SCMA, show that the SCMA-SIC process achieves the best system performance. In addition, with power control strategies for uplink access, the SCMA-SIC process still performs the best.
To protect the services and data of the main servers, Rapidly Deployed Cyber Attack Bait is proposed for the unpredictable cybersecurity attack in this research. Based on Software Defined Network, the data packet and operation of cybersecurity attack could be redirected to the on demand deployed bait container. By suitable artificial intelligence strategy, the idle bait containers corresponding to the type of cybersecurity attack can be on demand deployed automatically and rapidly. Verification shows the feasibility of the virtualized bait container could be about 1.16 times to 2 times enough for the future coming attacker and also record the all operations of the cybersecurity attack individually.
In 1995, a research project started at the NTT Laboratories with the goal of fostering innovation (creating new value through innovative ideas) and promoting DX (digital transformation of society, business, and organizations) through the “loosely connected and autonomous networks ${ }^{\wedge \wedge}$ of computers. This became known as “the Brokerless Theory, ${ }^{9,}$ the world's first P2P (Peer-to-Peer) theory, which attracted global attention. From this new concept of “loosely connected and autonomous networks/ ${ }^{9}$ numerous innovative social models, business models, and internet services emerged, including Skype (Sky P2P), social networking services, blockchain. Recent research has revealed that a new concept called “the External Vector” is effective for fostering innovation and $\mathbf{D X}$. The External Vector refers to building new, weak, loose, and autonomous connections between different fields, from which new value can be created. Along with technological innovation, which creates new value frameworks as technology advances, the External Vector serves as one of the two essential driving forces behind the emergence of innovation and DX. This paper proposes the following three points: 1)The External Vector is effective for fostering regional innovation and regional DX. 2)Building connections using the External Vector requires verbalizing (theorizing) the essence of connections and training in the practical use of the verbalized concept. 3)The combination of “P2P Concepts and Principles ${ }^{\wedge \wedge}$ and “Go Concepts and Principles ${ }^{\wedge} \wedge$ is effective for verbalizing (theorizing) and training. Furthermore, the effectiveness of combining P2P and Go is quantitatively demonstrated through a questionnaire survey conducted with 486 participants.
With the development of the Internet, the number of connected devices and the amount of transferred data increase annually leading to more complex network applications and limitations in data processing speed. The Data Plane Development Kit (DPDK) addresses these challenges by optimizing packet processing for high-speed network applications, including offloading processing tasks to Network Interface Cards (NICs) hardware using the rte_flow interface, which allows configuration of NICs offloads with specific flow rules. However, with the increasing complexity of network applications, the number of rules used is also growing rapidly, and more complexity. This presents a challenge for building and developing network systems capable of efficiently handling a large number of rules. This paper presents an architecture for processing the rte_flow interface using FPGA-based Smart NICs to leverage the reprogrammable capabilities of hardware accelerators. The architecture uses Ternary Content-Addressable Memory (TCAM) to reduce the key size in the rule tables, significantly increasing the number of rules and rule tables supported by the FPGA. Additionally, a custom DPDK driver (named VTL driver) is developed to manage and implement algorithms aimed at optimizing the number of rules in the hardware. To evaluate our proposed architecture, the Alveo U200 FPGA-based accelerator card is used to implement these features including 8 receive rule tables and 1 transmit rule table per ethernet port. Each table supports up to 2048 wildcard matches and 16384 exact match entries. The system achieves a network throughput of 100 Gbps per port and a rule insert rate of 204 million rules per second. These results significantly enhance hardware acceleration efficiency in networking.
Fusing multimodal sensors for 3D object detection has been extensively researched in the field of autonomous driving. However, existing multimodal sensor fusion methods still struggle to provide reliable detection across different modalities under diverse environmental conditions. Specifically, straightforward methods like summation or concatenation in radar-camera fusion may lead to spatial misalignment and fail to localize objects in complex scenes. To address this, we propose Adaptive Cross-Attention Gated Network (ACAGN) to enhance radar-camera fusion capabilities in Bird's-Eye View (BEV) space. Our approach integrates a deformable cross-attention and an adaptive gated network mechanism. The deformable cross-attention aligns radar and camera features from BEV with greater spatial precision, handling variations between those features effectively. Meanwhile, the adaptive gated network dynamically filters and prioritizes the most relevant information from each sensor. This dual approach improves stability and robustness of detection, as demonstrated through extensive evaluations on the nuScenes dataset.
Digital image forgery is the process of manipulating an image to deceive or mislead observers with real or manipulated content. Median filtering is widely used to smooth images and obscure traces of tampering, making its detection critical for image forensics. However, identifying median filtering becomes more complex when additional operations, such as compression, resampling, or noise addition, are applied. To address this issue, we propose a lightweight convolutional neural network (CNN) model named SobelMNet, specifically designed for detecting median filtering in compressed images. The proposed model utilises a Sobel filter-based preprocessing step to enhance the residual differences between the original and manipulated images. These residuals, which capture subtle features indicative of median filtering, are analysed by CNN for classification. Further, the proposed model is evaluated on grayscale low-resolution images generated from the Dresden dataset for both binary and multiclass classification tasks. The model achieved a remarkable detection accuracy of 99.43% in median filter detection and outperformed state-of-the-art methods in various scenarios, including combinations of median filtering with Gaussian blur, resampling, and additive white Gaussian noise (AWGN) with an average accuracy of 98.32%. Finally, its lightweight architecture ensures computational efficiency, making it practical for real-world forensic applications.
This paper presents an Augmented Reality (AR)-enhanced approach to improving the accuracy and efficiency of signal-based indoor positioning systems, particularly in controlled environments such as laboratories, cleanrooms, and data centers. Traditional positioning systems face challenges related to manual fingerprint database creation, accuracy, and environmental variability. The proposed system automates the database creation process using AR for real-time distance measurement and visual reference capture, significantly reducing manual effort and errors. In the positioning phase, the system combines signal strength data with AR-based distance validation and visual reference comparisons to improve location accuracy. The integration of AR enhances the system's ability to provide more reliable positioning results in environments where conditions, such as lighting and surface textures, are stable. This paper outlines the architecture and key components of the system while future work will focus on empirical validation and optimization of the system for real-world applications.
