ABSTRACT In recent years, the development of distributed learning systems and advancements in deep learning have led to significant improvements in music style transfer techniques. However, these improvements face significant challenges when implemented in space‐air‐ground integrated network (SAGIN) environments, due to issues such as high latency, limited bandwidth, and privacy concerns. This research explores a distributed, cross‐domain music style transfer model based on SAGIN environments, proposing a federated learning (FL) approach to mitigate these challenges. The proposed method facilitates efficient music style transformation while maintaining high content and style fidelity, ensuring privacy protection by keeping sensitive data localized to edge devices. We analyze and compare the performance of several models on different music datasets (including classical, jazz, and rock genres), demonstrating that our method outperforms traditional centralized models in terms of latency, communication efficiency, and privacy preservation. Moreover, we present several ablation experiments, illustrating the contribution of each component in the model. The proposed method demonstrates its applicability in distributed, real‐time environments, offering a solution for scalable and privacy‐preserving music style transfer applications in the SAGIN framework.

ABSTRACT This paper introduces an AI‐driven soccer training optimization method based on the space–air–ground integrated network, named SAGIN‐Play, which integrates real‐time multimodal data from ground sensors, aerial surveillance (drones), and cloud‐based processing. The system is designed to optimize player performance, enhance tactical positioning, and provide real‐time feedback during soccer training and matches. By leveraging wearable motion sensors, drone‐based aerial surveillance, and cloud computing, the method enables precise tracking of player actions and interactions, facilitating personalized performance improvements. This paper evaluates the proposed method across various datasets, including SoccerNet, Kaggle Football, UCF101 Sports, and the player event system (PES), highlighting the effectiveness of SAGIN‐Play in action recognition, tactical positioning, and real‐time feedback precision. The method outperforms traditional techniques, such as object detection models, multi‐object tracking, and reinforcement learning (RL), in key metrics, demonstrating its potential in dynamic soccer training environments. Additionally, an ablation study reveals the critical contributions of each SAGIN‐Play component, particularly aerial surveillance and cloud‐based processing, in optimizing player positioning and tactical execution. The study further demonstrates the system's ability to improve player performance through personalized feedback, showing significant progress in key skill areas like passing, shooting, and defense. Simulated live training sessions demonstrate substantial improvement in coordination, decision‐making, and overall performance. The results underscore the importance of integrating multilayered data for real‐time tactical adjustments in soccer training. Experimental results show that SAGIN‐Play achieves 94.2% action recognition accuracy on SoccerNet and 92.8% tactical positioning accuracy on Kaggle Football, while reducing the average feedback latency from 650 ms to 240 ms compared with baseline models.

ABSTRACT 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.

ABSTRACT 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.

ABSTRACT Object detection has gradually developed into a popular research area because of the rife use of Remote Sensing Images (RSIs) in military and civil domains. However, complex backgrounds and problems with small items are the two biggest obstacles in object detection. Deep learning techniques have a significant advantage for object detection over conventional methods that rely on manually derived characteristics, and they have a lot of attention. Due to the wide range of objects and complex visual backgrounds in RSIs, current deep‐learning algorithms in the area of RSI object detection leave much to be desired. For this case, algorithms need to be specifically optimized. In this paper, a novel optimization‐driven Enhanced Faster Region Convolutional Neural Network (FRCNN) is proposed for object detection. This approach has five modules, namely pre‐processing, data augmentation, segmentation, feature extraction, and object detection. The detection process is performed using Enhanced FRCNN and it is trained by the proposed Gazelle White Shark Optimization Algorithm (GWSOA). Extensive experiments in high‐resolution RSI data sets have exposed the efficacy of the proposed approach. The novel approach achieved better accuracy of 97.31%, Mean Average Precision (MAP) of 97.85%, precision of 97.76%, recall of 96.16%, F ‐score of 95.78%, and error rate of 2.69.

ABSTRACT 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.

ABSTRACT Smart agriculture uses Internet of Things (IoT) and sensors to increase productivity, optimize resource use, and enhance decision‐making. However, the large and varied agricultural data sets pose challenges for traffic management and network security. Traditional software‐defined networking (SDN) offers centralized, flexible traffic control, but it cannot detect malicious or abnormal activities in real‐time. To fill this gap, we introduce an Artificial Intelligence (AI)‐based anomaly detection and traffic management framework for SD‐IoT networks in smart agriculture. This system combines a machine learning (ML) Intrusion Detection System (IDS) with the SDN controller, enabling real‐time identification of abnormal traffic and adaptive prioritization of critical data flows. The IDS detects unusual traffic patterns, while the SDN controller allocates resources and routes flows based on quality of service (QoS) and security needs. Simulations demonstrate that our system achieves high accuracy in anomaly detection, reduces latency for emergency flows, enhances throughput, and reduces packet loss compared to traditional methods. This work highlights the importance of integrating AI‐powered IDS with SDN traffic management to enhance security and communication in smart agricultural networks.

ABSTRACT Underwater Acoustic Sensor Networks (UASNs) are essential for offshore engineering, military surveillance, environmental monitoring, and oceanographic exploration. However, challenges including high latency, congested networks, low bandwidth, and energy limitations make it difficult to communicate effectively in UASNs. In order to tackle these problems, this study suggests a new Energy‐Efficient Double‐Level Deep Reinforcement Learning with Chaotic Chimp Optimization (E2D2RL‐ChCo) model that is intended to improve localization precision, reduce packet loss, and effectively handle network congestion. The proposed E2D2RL‐ChCo model uses transformer‐based Markov Decision Processes (MDPs) for node localization at the top level and congestion‐aware routing at the bottom level to guarantee efficient task scheduling and energy‐efficient data transfer. By integrating Chaotic Chimp Optimization (ChCo), learning parameters are adjusted to enhance convergence and solution quality. Simulation results demonstrate up to 33.8% reduction in energy consumption, 61.5% decrease in end‐to‐end delay, and over 52% improvement in successful packet delivery compared to baseline models. The proposed model also improves localization error to 0.0075, outperforming existing strategies. This work fills gaps in existing literature by integrating deep reinforcement learning and metaheuristics to optimize both localization and congestion in real‐time UASN scenarios. This approach enables more sustainable and reliable UASN operations, paving the way for enhanced performance in real‐world aquatic environments.

ABSTRACT In smart cities, AI‐driven traffic flow prediction and anomaly detection play a crucial role in optimizing urban mobility and reducing congestion. Traditional traffic management systems (TMS) often struggle with real‐time adaptability, leading to inefficiencies in traffic prediction and failure to detect anomalies accurately. To address these limitations, we propose the Traffic Flow Prediction using Multi‐Agent Approach (TFP‐MAA) framework, which leverages multiple intelligent agents to analyze real‐time traffic data, predict congestion patterns, and detect anomalies with high precision. The framework integrates deep learning models with a decentralized multi‐agent system to enhance prediction accuracy and response efficiency. The proposed method enables adaptive decision‐making for traffic authorities by providing real‐time congestion forecasts and identifying unusual traffic patterns, such as accidents or roadblocks. This enhances proactive traffic management and reduces delays. Experimental results demonstrate that TFP‐MAA significantly outperforms existing models in terms of accuracy (94.7%), response time (96.2%), and anomaly detection efficiency (95.9%). The findings suggest that integrating multi‐agent AI systems into smart city infrastructure can lead to improved traffic flow (97.5%), reduced congestion (28.6%), and enhanced road safety.