Transactions on Emerging Telecommunications Technologies最新文献

This paper presents a novel basketball shooting performance optimization method that integrates deep pose estimation with a Space-Air-Ground Integrated Network (SAGIN)-enabled feedback mechanism. The core idea is to leverage advanced 3D human pose estimation techniques to capture the fine-grained body kinematics during shooting, decompose these movements into interpretable motion phases, and utilize SAGIN to provide ultra-low-latency corrective feedback. Compared to existing methods that either focus solely on biomechanical analysis or network-based performance enhancement, our framework establishes a closed-loop system capable of real-time analysis, correction, and adaptive learning. The proposed method is composed of four key components: (A) a deep pose estimation module that accurately reconstructs 3D body joints, (B) a phase-wise motion decomposition mechanism tailored to basketball shooting, (C) a SAGIN-based feedback pipeline that ensures low-latency information delivery, and (D) a unified learning objective that simultaneously optimizes pose estimation accuracy and shooting biomechanics. Experimental results demonstrate that the proposed system significantly outperforms existing methods, achieving 25.4 mm MPJPE (15%–40% reduction compared to baseline methods), 90.4% shooting accuracy (12%–18% improvement over existing systems), and 38 ms feedback latency (63% reduction compared to ground-based systems), offering a promising direction for intelligent sports training.

Fall Detection Systems (FDS) are an integral part in many Ambient Assisted Living (AAL) systems for ensuring the safety of senior citizens, especially in the underserved, isolated, and remote areas where there is unavailability of conventional communication systems. The conventional FDS systems mainly rely on cameras and wearable devices that impose significant challenges like privacy and acceptability. This paper presents a non-invasive and non-intrusive FDS leveraging ultrasonic sensors for fall detection, mitigating the challenges posed by camera systems and wearable devices, resulting into privacy preserving human fall detection. We propose a hybrid deep learning fusion approach that fuses Recurrent Neural Networks (RNN), Long Short Term Memory (LSTM), and Bi-directional LSTM (BLSTM), which achieves an accuracy of 98.14% for fall detection from time-series data. The main motivation of this study is to integrate our Fall detection system with the Space-Air-Ground-Integrated Network (SAGIN) framework to facilitate real-time alerts and emergency responses in the remote and isolated areas affected by unreliable communication systems. The integration of the FDS with the SAGIN framework presents a multi-tier processing at three levels, including Ground, Air, and Space. At the Ground level, the edge devices at the local site facilitate initial fall detection with lower latency. At the Air level, the aerial platforms like drones present an extended coverage range and facilitate data relay. And at the Space level, the satellites facilitate global connectivity, data analysis, and management for a longer course of time. Thus, the SAGIN integration with FDS systems ensures precise and real-time fall detection in remote and isolated areas, guaranteeing the availability of the communication networks. The proposed approach reduces the latency with the help of edge computing and showcases a resilient and scalable architecture for emergency response and health monitoring.

Quick Service Laboratories (QSL) provide the necessary diagnostic services that have to be performed within limited time frames and rely on coordinated solutions across its supply chain to operate successfully. The application of standard supply chain management approaches often fails to recognize the variable and unpredictable nature of QSL operations, which significantly contributes to stockouts, delays, or surplus inventory. This study looks into a different approach to the traditional methodologies of supply chain management by investigating the means when machine learning algorithms with the purpose of discovering anomalous behavior patterns are applied to QSL supply chain practices and generate value. In examining and evaluating the historical demand forecasting patterns, inventory levels, and operational performance metrics will be more easily identifiable as anomalous behaviors or dissenting levels such as demand spikes, unanticipated inventory shortfall levels, and atypical arrival patterns of inventory to generate disruption to laboratory operations. Machine learning models can be supervised or unsupervised to learn normal operation behaviors, and even detect anomalies in real time through model training. These models facilitate proactive interventions that would improve inventory management and distribution planning, as well as service delivery in general. When building on the results of our detection modeling, we found that machine learning anomaly detection could provide actionable suggestions and improved supply chain resiliency, and reduce stockouts and excess inventory, all while maintaining more controlled service levels. Our comparative evaluation of conventional monitoring and forecasting methods demonstrates superior capabilities over traditional methods in our results, by resorting to fully utilizing the complexity of simple linear and rare events found in QSL supply chains and their digital transformation story.

Unmanned Aerial Vehicles (UAVs) are ubiquitous in diverse applications, underscoring the need for efficient path-planning, particularly in complex three-dimensional (3D) environments. Current heuristic path-planning algorithms are largely designed for two-dimensional (2D) contexts, with a select few adapted for 3D open spaces. The application of these algorithms in 3D indoor or maze environments, however, remains largely unexplored. To address this gap, this study implements Dijkstra's, Greedy BFS, A*, Beam Search A*, Iterative Deepening A* (IDA*), Theta*, Weighted A* (WA*), Dynamic Weighted A* (DWA*), D* in 3D-environment and presents a comparative analysis within 3D indoor and maze scenarios. We assess their performance based on parameters such as path length, computational time, the number of points needed to reach the goal, and notably, memory consumption-a key consideration in UAVs due to their limited onboard memory. Through this analysis, we provide crucial insights into the behavior of these algorithms in complex 3D environments, thus informing the selection and development of optimal path-planning strategies for future UAV applications.

With the advancement of low-carbon economy and the emergence of Space-Air-Ground Integrated Network (SAGIN), carbon tax policies have become essential to driving sustainable optimization in freight industries operating within heterogeneous and dynamic network environments. This paper proposes a dynamic freight order allocation optimization model suitable for SAGIN scenarios under carbon tax constraints, simultaneously addressing multiple objectives including customer satisfaction, operational costs, resource utilization, and carbon emissions control. First, the model incorporates carbon tax policies to dynamically adjust emission costs and transportation efficiency in response to real-time data transmitted via SAGIN. Second, it captures the demand fluctuations characteristic of actual freight operations through dynamic and random settings enabled by SAGIN's ubiquitous connectivity. To comprehensively evaluate freight platforms' sustainability in SAGIN contexts, multiple sustainability assessment indicators are developed, including carbon emissions, empty running distance, overall revenue, and transportation efficiency. Experimental results confirm that the model significantly improves operational efficiency, reduces costs, and minimizes emissions under carbon tax constraints, providing robust decision-making support for achieving low-carbon targets while maintaining economic development and operational effectiveness.

Rise of online social networks has transformed how people interact, exchange information and connect with one another. But this digital evolution has also brought forth a significant challenge: the proliferation of abusive content. Detecting as well as mitigating abusive content is important for fostering a safe and inclusive online environment. This survey provides a comprehensive overview of the state-of-the-art methods for abusive content detection in online social networks. The paper begins by defining abusive content and its various manifestations in the digital realm and then delves into the evolving landscape of online social networks, highlighting the unique challenges posed by their dynamic and user-generated nature. Firstly, a scientometric analysis of the literature pertaining to the last 30 years (1993–2023) has been performed through which a deep analysis of prominent keywords, documents, institutions, and countries have been conducted. Further, this survey explores the key approaches to abusive content detection, including machine learning methods, natural language processing techniques, and deep learning models. The importance of dataset curation and annotation, which play a pivotal role in training robust and effective models, has been discussed. The survey also highlights various challenges, ethical implications, and future research directions that can guide the development of more effective and responsible abusive content detection systems in social networks.

Segmentation is a crucial step to divide an image into background and foreground sections and uses different colors to identify the target portion. Different thresholding-based segmentations have been developed in the current research analysis; however, the problems are difficult to solve and take a long time. To overcome these existing limitations, a novel Masi entropy-based multi-level thresholding with an effective optimization algorithm is introduced for image segmentation. The input images are collected from the open-source dataset, namely the Berkeley Segmentation Dataset 500 (BSDS500) and the Cityscapes dataset. To remove noise and enhance image quality, use the Quantized Haar Wavelet Assisted Histogram Equalization (QuaWHe) technique in the pre-processing stage. After noise removal, image segmentation was performed by the 2D practical Masi entropy histogram function (2D-MentH) with the Reinforcement Learning-assisted fire-fly-oriented multiverse optimizer (RL-FF-MVO) algorithm. The RL-FF-MVO mechanism helps to select the optimal set of threshold values from the values obtained using the 2D-MentH mechanism. By integrating reinforcement learning, the optimization process's convergence speed and accuracy are greatly increased while computing overhead is decreased. The proposed model has obtained Peak Signal Noise Ratio (PSNR) and Structural Similarity Index (SSIM) values of 30.587 and 0.9995 at threshold level 10. At threshold level 10, the proposed model has obtained Mean Square Error (MSE) and Feature Similarity Index (FSIM) values of 0.8531 and 0.96 248. For the Probabilistic Rand Index (PRI), the proposed model obtains a value of 0.72003, and for Variation of Information (VOI), it achieves 4.0298 at threshold level 10. The proposed approach outperforms existing methods regarding segmentation quality and computational efficiency, making it suitable for applications requiring high-accuracy image analysis, such as autonomous systems and medical imaging.

VANETs are essential for communication and coordination between autonomous vehicles, particularly in emergency scenarios where quick decisions are necessary. The proposed Deep Learning-based Control Approach for Autonomous Vehicles (DL-CA-AV) introduces a hybrid DL control framework that integrates Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) with an attention-driven decision-fusion mechanism for real-time maneuver control in VANET-enabled environments. The CNN module extracts spatial features from sensor and VANET communication data, while the LSTM network models temporal dependencies to predict dynamic vehicular states across time. These learned spatiotemporal representations are then passed to a reinforcement learning (RL) layer, where the actor–critic mechanism evaluates potential maneuvers and selects optimal control actions based on collision probability and situational awareness. The proposed approach encourages vehicles to autonomously choose the best emergency response (lane change, deceleration, or cooperative braking) while preserving stability and minimizing secondary risks. This study demonstrates the capability of DL-based VANET architectures to enable real-time autonomous driving control in hazardous environments, facilitating safer and more dependable intelligent mobility. The proposed framework achieves a collision probability of 29%, a latency below 225 ms, a detection accuracy above 85%, and a packet delivery ratio above 88%.

Vehicular networks are essential to the advancement of intelligent transportation systems by enabling continuous data exchange between vehicles and infrastructure to support safe mobility. However, the distributed and dynamic nature of these networks creates opportunities for adversarial threats. Existing attack-detection models primarily rely on centralized architectures, which often suffer from high latency, privacy risks, and limited robustness against advanced attacks. To address these challenges, this study proposes the LIFT-SFD model. This lightweight federated learning framework integrates Smooth Federated Dropout (SFD) with trust-weighted mask-aware aggregation for secure and resource-aware training in vehicular ad hoc networks (VANET). Each vehicle trains a masked submodel, while smooth dropout regularization ensures stable convergence and reduced communication overhead. Then, an integrated assault monitoring module detects and reduces malicious behavior by assigning anomaly scores at the vehicle level, adjusting trust weights during RSU-level aggregation, and gradually filtering out malicious participants. The simulation of the models is performed using the VeReMi dataset, demonstrating high detection accuracy of 99.98% at round 5. It acts as a stable and trustworthy global model for future vehicular networks.

Instantaneous interaction and autonomous modes of transport are facilitated by Vehicular Ad Hoc Networks (VANETs), which play a crucial role in the development of smart cities. However, the flexible and autonomous architecture of VANETs poses significant security risks, including susceptibility to cyberattacks, privacy breaches, and information manipulation. These challenges are exacerbated by high node mobility and the large volume of heterogeneous data exchanged in modern urban environments. This study proposes a novel security framework that leverages transfer learning-based threat detection and mitigation techniques to address these issues. By utilizing pre-trained machine learning models fine-tuned with VANET-specific datasets, the approach reduces training time and computational costs while enabling efficient detection of anomalies and attacks. The objectives are to enhance network security, safeguard data integrity, and minimize latency in threat detection processes. Research findings indicate that the proposed framework outperforms traditional machine learning models in terms of scalability, resilience, and the ability to detect malicious activities. By providing a secure communication infrastructure for VANETs, this research contributes to the development of reliable and efficient smart city systems.