Transactions on Emerging Telecommunications Technologies最新文献

Future generation Vehicular Ad-hoc Networks (VANETs) models primarily consider continuous power-aware communication to support Intelligent Transportation Systems (ITS) in exchanging diverse sensitive vehicular information. Existing systems often struggle to capture hidden misbehaviors, particularly in power-domain parameters such as abnormal power transmission, unexpected power changes, and malicious increases in signal strength. These threats disrupt sensitive vehicular communication and real-time vehicle coordination. This study aims to develop a lightweight, power-aware malicious-detection framework capable of identifying hidden, short-term power-domain attacks in resource-constrained VANET environments. To address these challenges, this study proposes a novel integrated machine learning (IML) framework that combines the strength of Feather Layer Perceptron (FLP) for feature encoding with a Dense Tree Module (DTM) for effective decision-making and Tiny-LSTM with the Endomode Sliding Window Approach to quickly analyze short-term changes in vehicle power signals. The suggested system is a future-expected green model that can efficiently analyze multidimensional power features in low-computation mode to identify malicious power patterns. The model is simulated using two distinct datasets: Secure VANET Vehicle Dataset and “The Indoor Localization Dataset,” both adapted from a public repository, which capture vehicular communication, power-related features, and motion information under everyday and attack scenarios. It also provides additional signal-based measurements and environmental features to enhance feature diversity. To improve the simulation, we consider additional synthetic features. Experimental results demonstrate that the proposed IML model achieves 99.7% test detection accuracy on standard P-OBU power data and maintains 97.6% accuracy under noisy power-domain inputs. By leveraging these advantages, the proposed framework effectively enhances power-domain security in VANETs by accurately detecting anomalies under realistic, noisy conditions. Also, it provides a scalable solution for next-generation intelligent vehicular networks.

Satellite-based video surveillance, sometimes known as “gazing,” is extremely useful for viewing, evaluating, and dynamically tracking developments on Earth. However, the tiny size and density of objects, overlapping targets, and unclear surroundings with variable illumination and complex backdrops make multi-object tracking in satellite movies particularly difficult. Limitations in bandwidth and computational capacity made accurate tracking and real-time processing increasingly challenging. Applications, including disaster response, traffic monitoring, defense, and security operations, depend on overcoming these obstacles. This work offers a novel framework to address these difficulties by merging sophisticated cross-frame connection techniques with frequency domain analysis using wavelet transform analysis. By separating high-frequency components and reducing noise, the wavelet transform improves the identification of small targets and makes it possible to recover fine-grained spatial and frequency data that are essential for reliable tracking. The system uses motion models and data association techniques to guarantee trajectory correctness and consistency, and it integrates cross-frame connections to create temporal continuity and preserve target identities over successive frames. The experimental results show significant increases in tracking performance, outperforming state-of-the-art methods in terms of multiple object tracking accuracy (MOTA), multiple object tracking precision (MOTP), and high tracking accuracy. These results demonstrate this proposed model's resilience and effectiveness in accurately identifying and following small targets in challenging satellite imagery settings. The accuracy achieved by the proposed method for the VISO, SATMTB, and Skysat-1 datasets is 96.82%, 95.26%, and 95.90%, respectively.

In Intelligent Transportation Systems, Vehicular Ad-hoc Networks (VANETs) provide real-time vehicle-infrastructure communication. VANETs may be attacked via denial-of-service, Sybil, and spoofing due to its open wireless medium and changeable topology. In dynamic vehicle contexts, signature-based intrusion detection systems (IDSs) fail to identify novel threats, while anomaly-based techniques have high false alarm rates and limited scalability. The Machine Learning-based Vehicular Intrusion Detection System (ML-VIDS) combines unsupervised clustering for zero-day anomaly detection and supervised learning for known attack categorization. The method improves detection accuracy and edge deployment efficiency via temporal traffic analysis and feature selection. On benchmark VANET datasets, ML-VIDS beats comparable IDS systems with 97.8% detection accuracy, 6% false positive rate, 15 ms latency, 65% lower computational overhead, and 30 W energy utilization. In next-generation intelligent transportation systems, ML-VIDS enables adaptive and resource-efficient intrusion detection for VANETs to provide strong performance and real-time security.

In this paper, the big data classification is done using the hybrid Squeeze-EfficientNet with Feature fusion in the spark framework. At first, the partitioning of big data is executed by employing Bayesian Fuzzy Clustering (BFC). Later, big data classification is accomplished in the spark architecture. At the slave node, the subsequent process is accomplished and the partitioned data are applied to the slaves, where they are pre-processed by Quantile normalization. Further, the feature fusion is carried out based on the Deep Kronecker Network (DKN) and Matusita Distance measure. At the Master node, all the features from the slave nodes are fused and the big data are classified by employing the Hybrid Squeeze-EfficientNet. The Hybrid Squeeze-EfficientNet is generated by integrating SqueezeNet and EfficientNet. The evaluation results show that the Squeeze-EfficientNet attained an accuracy of 0.904, a sensitivity of 0.916, and a specificity of 0.925. The high performance obtained by the devised model improves the big data classification task in real-time scenarios like targeted marketing, agriculture, smart transportation, efficient resource allocation, and personalized health monitoring systems. Thus, the integration of big data classification within daily life provides more informed decisions thereby improving the life quality.

In today's digital world, with increasing threats of cyberattacks and unauthorized data access, protecting confidential information demands more robust security mechanisms. Cryptography and steganography are two prominent techniques employed to secure data. However, conventional steganography suffers from reduced embedding capacity and the risk of image distortion. To overcome these challenges, this research proposes a novel hybrid framework to enhance data security and embedding efficiency. The approach comprises two phases including the embedding phase and the extraction phase. In the embedding phase, both the cover and secret images undergo a 3-level discrete wavelet transform (DWT) using the Daubechies wavelet. Region selection in the transformed cover image is optimized by extracting and leveraging various features, including color, shape, deep features, and local Gabor transitional pattern (LGTrP) features. These features are processed via a modified Bidirectional Long Short-Term Memory (Bi-LSTM) model, enhanced with architectural improvements, which boost feature learning. Simultaneously, the secret image undergoes transformation using a modified Arnold map integrated with a Bernoulli map allows faster execution. The modified Arnold function's outcome is subjected to an encryption process. The embedding process is done after the encryption process; the modified Blowfish algorithm is used for the decryption process. Subsequently, the inverse Bernoulli map is utilized, with the resultant output given to the inverse Arnold map. Finally, an inverse 3-level DWT reconstructs the original secret image. Comparative evaluations demonstrate the proposed framework attains lower KPA and KCA rates of 0.12 and 0.15, respectively, which underscores the innovation of integrating a steganography-cryptography model in securing sensitive data against sophisticated attacks.

Healthcare data holds immense potential to improve diagnostics and treatment, but centralized collection risks patient privacy and suffers from limited labeled samples. Existing methods fall short in securely leveraging distributed data while reducing annotation costs. To address this, we propose FTAL-QNC—a novel Federated Transfer Active Learning framework with Quantized Neural Cryptography—that enables privacy-preserving, communication-efficient, and label-efficient collaborative model training. FTAL-QNC integrates four core components in a unified architecture: (1) Federated learning for decentralized model training without data sharing, (2) Transfer learning to handle data heterogeneity across hospitals, (3) Active learning to minimize manual labeling by prioritizing informative samples, and (4) Quantized neural cryptography to ensure secure, low-overhead exchange of encrypted model updates. Empirical evaluations on real-world datasets, including Lung CT and ChestX-ray14, demonstrate that FTAL-QNC enhances segmentation and classification accuracy to 97.3% and 97.0% recall for Lung CT, respectively, while significantly reducing annotation effort compared to standard federated learning and other sampling methods. Our contributions include a privacy-preserving and communication-efficient collaborative framework, an integrated active learning mechanism for efficient data labeling, and a secure aggregation protocol via quantized neural cryptography. These results demonstrate FTAL-QNC's potential to advance safe, collaborative medical research and improve patient outcomes.

The rise in network intrusions has led to significant consequences, including privacy violations, financial losses, and unauthorized data transfers. Attackers exploit vulnerabilities in network systems, compromising security and disrupting services. Traditional intrusion detection systems (IDS) often face challenges such as false positives, delayed threat identification, and poor detection of minority attack classes. To address these issues, this research proposes advanced deep reinforcement learning with deep learning-based NIDS for improved threat detection and mitigation. Data from IoT-2023, BoT-IoT, CIC-IoT 2023, and RT-IoT 2022 datasets are preprocessed through null value handling, data cleaning, one-hot encoding, and Min-Max normalization. To enhance the detection of minority attacks, the Tabular Auxiliary Classifier Generative Adversarial Network (TACGAN) is employed for synthetic data augmentation. Feature extraction is performed using the Graph Sample and Aggregate Attention Network (GSAAN), which captures basic, content, and traffic-based features. Significant features are selected using the Mountaineering Team-Based Optimization (MTBO). Attack classification is carried out using a novel ensemble of the Improved Double Deep Q-Network (IDDQN) and Deep Autoregression Feature Augmented Bidirectional LSTM (DAF-BiLSTM), which is termed the OptIDQDBiLSTM approach, ensuring robust learning of spatial and temporal dependencies. Hyperparameter tuning is optimized using the Boosted Wild Horse Optimization Algorithm (BWHOA). Experimental results show that the proposed approach outperforms existing IDS methods, achieving higher detection rates, improved accuracy, and a reduced false alarm rate while maintaining computational efficiency. While comparing with existing state of the art approaches, the proposed approach surpasses existing methods with over 99.64% accuracy, 99.34% precision, and 99.42% recall. These findings demonstrate the effectiveness of deep reinforcement learning in enhancing network security against evolving cyber threats.

The rapid expansion of digitalization has intensified cybersecurity risks, exposing critical network vulnerabilities despite significant advances in encryption and intrusion detection systems (IDS). Many existing deep learning–based IDS still struggle with high false-positive rates, misclassification, and limited adaptability, reducing their effectiveness in real-time defense scenarios. To address these limitations, this study proposes a Polymorphic Graph Gudermannian Neural Network integrated with Adaptive Chaotic Satin Bowerbird Optimization (PG-GNN-AC-SBO), complemented by a lightweight encryption mechanism. The framework incorporates a Fuzzy K-Top Matching Value (FKTMV) module for robust preprocessing and normalization, along with a Hybrid Cat Hunting Sea-Horse Optimizer (H-CHO-SHO) for efficient and interpretable feature selection. The PG-GNN classifier employs graph-based learning and a Gudermannian nonlinear activation function to effectively capture complex traffic behavior, while AC-SBO dynamically tunes hyperparameters to enhance stability and classification accuracy. To ensure data confidentiality, a Synchronously Scrambled Diffuse Encryption (SSDE) scheme is applied, delivering strong security with low computational overhead. Experimental evaluations on the NSL-KDD and CICIDS2017 datasets demonstrate the superiority of the proposed approach, achieving up to 99.82% accuracy and outperforming state-of-the-art methods. The encryption and decryption times of 3.50 and 3.55 ms further confirm the model's lightweight design. Overall, the proposed system provides high throughput with minimal latency, demonstrating strong potential for real-time and large-scale cybersecurity deployments.