Search Results (2,556)

This paper investigates the low-altitude penetration path planning problem for a fixed-wing UAVs in complex battlefield environments with mountainous terrain and radar threats. First, a radar detection probability model and a dynamic radar cross section (RCS) model are established to characterize target exposure risk. Then, a deep reinforcement learning-based (DRL) path planning framework is developed, in which the Soft Actor-Critic (SAC) algorithm is employed to solve the penetration task under dynamic RCS constraints. A multi-objective reward design is constructed to jointly account for target-reaching progress, radar-threat avoidance, obstacle avoidance, and terminal task completion. Simulations show dynamically feasible flight paths under the adopted models. HIL tests conducted in the corresponding five-radar and two-radar numerical scenarios support onboard path planning, waypoint transmission, and functional closed-loop execution. Comparisons reveal a trade-off among penetration time, modeled radar exposure, and computation, with SAC achieving the shortest mean penetration time among successful trials. Full article

Effective management of urban Asian water monitor (Varanus salvator (Laurenti, 1768)) populations requires precise individual identification, yet traditional physical-marking methods remain invasive and labor-intensive. This study developed a non-invasive, automated photographic re-identification (Re-ID) system using deep learning and computer vision to facilitate population monitoring in semi-urban environments. We evaluated seven deep learning configurations based on ResNet50 incorporating Squeeze-and-Excitation (SE), Convolutional Block Attention Module (CBAM), and Batch Normalization neck (BNNeck) optimizations and benchmarked them against traditional feature matching (HotSpotter) using an open-set evaluation dataset of 3311 images across 161 Side-IDs focusing on unique lateral head-scale patterns. HotSpotter demonstrated immediate field viability, achieving a Rank-1 accuracy of 99.88% and a mean Average Precision (mAP) of 90.40%. Among the deep learning architectures, the baseline ResNet50 achieved the highest Rank-1 accuracy of 75.39% and mAP of 55.97%. As a decision-support framework, the deep learning pipeline achieved over 87% Rank-5 accuracy, drastically reducing manual screening effort and cognitive load during capture–mark–recapture surveys. This non-invasive framework establishes a scalable, welfare-friendly protocol for long-term urban wildlife management and biodiversity monitoring. Full article

The time available for a counter-drone system to detect an unauthorized aircraft and determine an appropriate response can be limited. Although cloud processing remains useful for storage and offline analysis, communication delays may constrain its use in time-critical decision loops. This paper proposes EdgeTwin-DRL, an edge-assisted digital twin framework that integrates multimodal sensing and multi-agent deep reinforcement learning (DRL) for counter-drone detection and response optimization. The digital twin maintains a synchronized representation of the protected airspace using radar, electro-optical/infrared (EO/IR), radio-frequency (RF), and acoustic observations. This synchronized state is used by cooperative DRL actors to adjust computational-resource allocation, detection sensitivity, and candidate countermeasures, while a model-based forward-evaluation assesses proposed responses before they are passed to the simulated response pathway. The framework is evaluated in a simulation testbed in which the RF sensing models are calibrated and independently validated using publicly available datasets, while the remaining sensing components are parameterized using published experimental measurements. Within this calibrated simulation environment, EdgeTwin-DRL achieved a false-positive rate of 1.4% and reduced mean detection-to-response latency by up to 72% relative to the Cloud-DRL baseline and by 26% relative to the MAPPO baseline without calibrated, environment-dependent sensing under the communication and computational assumptions used in the simulator. The evaluation was conducted across modeled urban, suburban, and open-field conditions. These results demonstrate the comparative performance of the proposed architecture within the simulated environment and motivate further investigation of edge-assisted digital twins for counter-drone decision support. Hardware-in-the-loop and controlled field validation are required before operational deployment. Full article

Adaptive and explainable systems are two critical concepts in personalized learning that enhance the customized experience based on the adjusted content according to the individual learning of a learner. In this paper, we present a modular and interpretable architecture for the development of a System for Adaptive and Explainable Feedback, which is a potential solution for the above-mentioned issues in personalized learning environments. The adaptive feedback provided by the System for Adaptive and Explainable Feedback is in real time. Still, even more importantly, it is justified clearly and transparently so that both the learner and instructor understand the why behind the recommendations for action given by the system. It consists of modular technologies for data collection (word and phrase occurrence, time tracking), analysis (latency, process mining), feedback generation (adaptive after-action review), and result presentation, making a modular, flexible, and scalable system to suit many educational scenarios. To evaluate the system’s functional performance, the SAEF pipeline was applied to a dataset derived from the ASSISTments platform, a well-established educational dataset widely used in learning analytics research. On a cohort of 500 student profiles, the SAEF achieved an overall recommendation accuracy of 83.6%, a weighted F1-score of 82.9%, and a mean system response time of 42.1 ms, demonstrating both the internal computational consistency and efficiency of the adaptive pipeline. These results indicate strong agreement with score-derived difficulty categories and support the internal computational consistency of the recommendation pipeline as a proof-of-concept system, though they do not constitute independent evidence of instructional appropriateness. The SAEF is designed to support learner engagement, personalize learning pathways, and foster transparency in AI-driven educational environments; a full empirical evaluation involving real-world deployment is identified as the primary direction for future work. The perceived understandability, trustworthiness, and pedagogical usefulness of the SAEF’s explanations by learners and instructors represent a complementary dimension yet to be empirically explored. The SAEF’s architecture is designed with the explicit objective of supporting learner engagement and fostering transparency; however, these pedagogical benefits are architectural design goals rather than empirically demonstrated outcomes in the present study, which focuses exclusively on computational validation. Full article

Urban rail transit infrastructure inspection increasingly relies on Light Detection and Ranging (LiDAR) due to its capability for efficient and high-precision 3D data acquisition. However, robust rail tread segmentation in ballastless metro environments remains challenging due to boundary leakage, interference from geometrically similar structures, and the heavy dependence of existing methods on Red-Green-Blue (RGB) imagery, trajectory priors, or template matching. To address these limitations, this study proposes a lightweight rail tread extraction framework for ballastless track LiDAR point clouds based on multi-stage filtering and curvature-guided region growing. First, intensity thresholding and cloth simulation filtering are leveraged to prune tunnel walls, track beds, and other large-scale non-target structures, thereby reducing computational overhead. Subsequently, local normal vectors and curvature features are estimated via Principal Component Analysis (PCA). A curvature-ranked seed selection strategy and a dual-constrained region growing mechanism, integrating normal consistency and curvature thresholds, are then introduced to suppress excessive growth near rail boundaries and enhance regional homogeneity. Experimental results on field data collected from Shanghai Metro Line 10 demonstrate that the proposed method achieves a recall of 92.23%, a precision of 95.32%, and an F1-score of 93.7%, outperforming conventional Euclidean clustering and standard region growing algorithms. Compared with deep learning approaches, the proposed framework requires no large-scale annotated training data and is independent of RGB information or trajectory priors, making it better suited for lightweight engineering deployment in practical urban rail transit maintenance. Full article

To address the challenge of satisfying strict Service-Level Agreement (SLA) requirements for concurrent smart video surveillance tasks in heterogeneous edge computing environments, an SLA-aware adaptive scheduling algorithm for heterogeneous computing collaboration is proposed. First, a mixed-task flow model is constructed, and a finite-state Markov chain is utilized to dynamically model the time-varying wireless channel. Second, a Dueling Double Deep Q-Network (Dueling DDQN) scheduling algorithm based on SLA awareness and channel adaptation is proposed, with a designed SLA action-masking mechanism. This mechanism advances hard delay constraints to the decision-generation stage, dynamically prunes the action space based on real-time channel conditions and node loads, and filters out actions predicted to violate the SLA before execution. Experimental results show that the proposed algorithm coordinates heterogeneous computing resources between the cloud center and the edge and exhibits earlier empirical reward stabilization and lower task-violation rates than the compared learning-based baselines under the tested workload conditions. Full article

The increase in the number and complexity of interconnected systems requires new methods to identify potential threats in today’s hyperconnected world. This trend affects systems ranging from smart homes and Internet of Things (IoT) devices to critical infrastructure which must be equipped with the corresponding cyber defense methods. Given the new landscape, it is more difficult for classical cybersecurity systems to stay up to date with novel threats, as well as to keep track of all interconnected devices defined by various protocols and behaviors. Artificial intelligence (AI) represents a strong candidate to complement traditional cyber defense methods due to its adaptability to variation and capability to identify complex data patterns, which has led researchers to develop state-of-the-art anomaly detection systems. The current critical review aims to analyze the scientific literature on three dimensions including used algorithms and datasets, domain challenges hindering AI deployment in productive environments, and the capability of explainable artificial intelligence (XAI) to support cyber security experts with insights into the model’s inner workings and decision rationale. Compared to existing scientific reviews, this paper moves beyond algorithmic comparison by providing a methodological interpretation of AI anomaly detection landscape, demonstrating how data availability, learning paradigms, and explainability collectively influence the evolution of cyber defense research towards operational deployment. This approach revealed that AI development for cyber defense is highly heterogenous, and that the available datasets strongly influence the algorithm of choice, rather than the models being chosen methodologically based on proven performance. The analysis further indicates that operational deployment remains challenging, as the literature continues to report substantial limitations related to data quality, computational requirements, and model interpretability. Full article

Urban-forest monitoring increasingly requires intelligent sensor-driven systems capable of characterizing short-term tree responses while operating efficiently within Internet of Things (IoT) and edge-computing environments. This study proposes a fusion-based artificial intelligence framework that integrates Quantization-Aware Training (QAT)-optimized PointNet++ models with machine-learning regression to predict a short-term dendrometer-derived stem-diameter response expressed in biomass-equivalent units. The framework combines 1024-point LiDAR tree representations, geometric measurements, and environmental sensor data through three components: QAT-optimized PointNet++ models for 34-species classification and trunk–crown part segmentation, frozen model-based prediction and geometric feature extraction, and MLP and XGBoost regression models for prediction of the short-term target. The dataset contained 2694 trees from five regions of South Korea, with the target derived from dendrometer-based stem-diameter measurements recorded over a 14-day interval between 8 September 2022 and 22 September 2022. Importantly, this short-term signal reflects both structural and reversible water-status-related stem dynamics and is therefore not interpreted as direct dry-biomass accumulation or carbon sequestration. The QAT-optimized models retained 92.52% segmentation accuracy (82.67% mIoU) and 80.46% species-classification accuracy, while the regression model reached R² = 0.9663 and RMSE = 0.4437 kg for the defined biomass-equivalent target. Quantization reduced the saved model size of both encoders by approximately 10.5× (21 MB → 2 MB) and accelerated CPU inference by up to 4.1×. These efficiency measurements were obtained on an ×86 desktop CPU and therefore characterize computational compression benefits rather than completed deployment or field validation on a low-power embedded device. These results demonstrate the computational feasibility of combining compressed point-cloud perception with multimodal prediction of short-term dendrometer-derived stem dynamics. Validation over seasonal and multi-year periods using independent biomass-reference measurements would be required before extending the framework to long-term biomass accumulation or carbon-sequestration assessment. Full article

Combinatorial optimization problems (COPs), including the Knapsack Problem (KP), the Traveling Salesman Problem (TSP), and regional variants such as the Close-Enough Traveling Salesman Problem (CETSP), constitute fundamental models for addressing complex decision-making tasks in modern computational systems. Their computational difficulty has motivated the development of a broad range of exact, heuristic, metaheuristic, learning-based, and hybrid optimization approaches. This work presents a structured survey and problem-structure-oriented comparative analysis of these methods, covering Genetic Algorithms (GAs), Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), Simulated Annealing (SA), Tabu Search (TS), Artificial Bee Colony (ABC), Graph Neural Networks (GNNs), and learning-enhanced hybrid approaches. Particular attention is given to how differences in problem structure, formulation characteristics, feasibility requirements, computational cost, and evaluation settings influence the suitability of different optimization paradigms. The practical relevance of these approaches is discussed through two case studies based on our prior work: Virtual Machine Placement (VMP) in cloud computing, examined through a simplified single-resource 0–1 KP abstraction, and UAV trajectory optimization for data collection in Wireless Sensor Networks (WSNs), examined through TSP-, CETSP-, and neighborhood-based routing models. The reviewed evidence indicates that metaheuristic approaches can provide flexible search mechanisms in computationally challenging settings, while learning-based models may support rapid inference, prediction, initialization, or search guidance when appropriate training data and generalization conditions are available. Hybrid learning–optimization frameworks may be beneficial when their components address complementary limitations of the underlying problem, although their effectiveness remains dependent on problem structure, training requirements, computational budget, feasibility handling, and implementation design. In cloud environments, knapsack-based VMP formulations provide useful abstractions for resource-allocation decisions, whereas in UAV-assisted WSNs, routing models such as the TSP and CETSP provide structured representations of trajectory and data-collection decisions. Overall, this survey highlights the complementary strengths and limitations of modern optimization paradigms and emphasizes that method selection should be guided by problem structure, operational requirements, feasibility considerations, and computational constraints. Full article

Fog computing enables low-latency processing for Internet of Things (IoT) applications by moving computation closer to end devices. However, fog infrastructures are often deployed on commodity and geographically distributed resources, making them vulnerable to correlated node failures caused by shared power systems, network switches, cooling units, or physical proximity. Most existing fault-tolerant scheduling methods treat node failures as independent events, which limits their ability to anticipate multi-node outages in shared-infrastructure fog environments. This paper presents CorrFault-GNN, a topology-aware fault-tolerant scheduling framework for predicting and mitigating correlated failures in fog computing. The framework models the fog infrastructure as a dynamic weighted graph that captures power, network, and geographic dependencies among fog nodes. A Temporal Graph Convolutional Network (T-GCN) learns spatial and temporal failure patterns and predicts node-level failure risks one scheduling epoch ahead. These predictions drive a proactive migration module that moves tasks away from high-risk nodes, while a Criticality-Aware Reactive Fallback handles unexpected failures. The framework is evaluated in three-tier IoT–Fog–Cloud simulations with correlated failure traces derived from cloud failure data. The results show that CorrFault-GNN improves task success, latency, energy efficiency, and deadline satisfaction compared with representative reactive, proactive, and learning-based baselines, and that its advantage grows as infrastructure sharing and failure correlation increase. Full article

The rapid emergence of large language models (LLMs) is reshaping Computer Science (CS) education. Yet, little is known about how students engage with these tools as both technical and ethical learning partners in authentic project-based environments. This qualitative study investigates how 29 undergraduate CS students used LLMs while developing open-ended AI applications in a project-based course. Analysis of project documentation, reflection logs, and presentation transcripts revealed three interconnected forms of student–LLM interaction. First, we observed a technical partnership, in which students leveraged LLMs for code generation, debugging, architectural planning, and API integration while working on complex development challenges. Second, there was an ethical-reflective partnership, as students negotiated transparency, originality, bias, and the risks of over-reliance, demonstrating elements of critical AI literacy. Third, students reported perceived shifts in their learning practices, including increased confidence, more intentional problem-solving, and changes in how they sought support from peers and instructors, suggesting a reconfiguration of self-regulatory learning practices in interaction with LLMs. Together, these findings suggest that LLMs can be understood not merely as productivity tools but as multifaceted partners involved in the technical, ethical, and self-regulatory dimensions of learning in CS education. The study offers theoretical and practical implications for designing AI-enabled curricula that cultivate responsible, reflective, and critically engaged use of LLMs. Full article

Tea bud detection is a critical technology in the development of intelligent tea-picking robots. To address the issues of high false negative and false positive rates under complex environment as well as poor generalization performance across varieties, weather conditions, years, etc., this study proposes a method named DMWE-YOLOv8n based on dual-branch multi-scale wavelet enhancement for tea bud detection. In this method, a multi-scale wavelet feature branch is incorporated in parallel into the model’s backbone and multi-level wavelet features are extracted through two-dimensional discrete wavelet transform, followed by dynamic wavelet feature enhancement. Then, the spatial features and wavelet features with the same scale are then enhanced by a gated attention mechanism and fused in a bidirectional guided manner. Finally, a novel wavelet cosine loss is introduced to boost the model’s learning capability and generalization. Experimental results demonstrate improved performance of DMWE-YOLOv8n on the ZC108-23 tea bud dataset and five cross-domain tea bud datasets. Specifically, in terms of AP₅₀, it outperforms the similarly sized YOLOv11n by 1.6% on ZC108-23 and by 7.65%, 9.83%, 10.67%, 5.15%, and 6.80% on the five cross-domain datasets, respectively. In conclusion, the proposed method achieves an excellent balance between detection accuracy, computational cost, and generalization performance. Full article

Immersive virtual reality has been increasingly adopted in higher education, particularly in computer science and programming education, and shows promising results in addressing challenges such as low student motivation and difficulties in understanding abstract and complex concepts. However, the pedagogical foundations underpinning the use of immersive virtual reality in programming education have not been sufficiently examined. In particular, the literature reveals limited integration among pedagogical theories, instructional design, immersive affordances, and assessment practices, which hinders the systematic design and evaluation of immersive virtual reality-based learning environments grounded in sound pedagogical principles. This study presents a critical review of the pedagogical foundations of immersive virtual reality for programming education in higher education and proposes an integrative conceptual framework. The review analyzes learning theories, pedagogical strategies, instructional design approaches, and assessment methods reported in empirical studies, with particular attention to their alignment with intended learning outcomes. Based on a synthesis of the literature, a pedagogical taxonomy of immersive virtual reality for programming education is developed to organize the main pedagogical elements identified in current research. These elements are subsequently integrated into a conceptual framework that establishes relationships between pedagogical theories, instructional design, immersive affordances, and assessment. The proposed framework provides a structured representation of current pedagogical practices, identifies gaps and fragmentation in the literature, and offers orientation for the design, implementation, and evaluation of future immersive virtual reality-based programming learning environments in higher education. Full article

Artificial Intelligence (AI) literacy has emerged as a critical 21st-century competence, yet its integration within interdisciplinary STEAM (Science, Technology, Engineering, Arts, and Mathematics) education remains fragmented. This study addresses this gap by investigating how AI literacy is conceptualized, implemented, and evaluated within STEAM learning environments. Following the PRISMA guidelines, a systematic literature review was conducted on empirical studies published between 2015 and 2025 to examine pedagogical approaches, methodological designs, AI technologies, and arts disciplines. An analysis of 32 studies revealed key insights into the development of AI literacy within STEAM education: (1) AI literacy is most often fostered through active and interdisciplinary learning rather than through lecture-based instruction alone, (2) AI serves both as a subject of study and as a tool for learners to create, design, investigate, and solve problems, (3) arts integration in AI-STEAM education typically support broader learning objectives and less frequently is assessed as distinct learning outcome, (4) ethical and societal aspects of AI literacy receive less attention than technical and computational skills, (5) the predominance of non-formal settings, short interventions, and context-specific studies indicates limited evidence on long-term progression, scalability, and sustained outcomes. The review contributes to the emerging field of AI-STEAM education by mapping empirical research from the past decade and developing two analytical tools: a five-cluster framework for learning objectives and a four-cluster framework for arts integration in AI-STEAM education. It concludes with a multi-level synthesis that identifies key educational and technological implications for the design, implementation, and future development of the field. Full article

Modern smart agriculture increasingly relies on edge computing for real-time, high-concurrency tasks such as wide-area drone-based crop monitoring. However, highly volatile workloads and severe environmental noise in agricultural Internet of Things (IoT) networks often lead to resource congestion and high latency when relying on traditional reactive scheduling. To address these challenges, this paper proposes a hybrid prediction-driven proactive resource scheduling method for edge computing. We construct a Variational Mode Decomposition-Convolutional Neural Network-Attention-Bidirectional Long Short-Term Memory (VMD-CNN-Attention-BiLSTM) model to filter environmental noise and accurately capture the spatio-temporal features of bursty traffic. Furthermore, a deep reinforcement learning scheduling algorithm based on Proximal Policy Optimization (PPO) incorporates future workload trends into its state space, dynamically optimizing task offloading. To evaluate the proposed Predictive Computational Scheduling Framework (PCSF), we developed a custom edge computing simulation environment and synthesized a hybrid dataset combining real-world server logs from the Alibaba Cluster Trace with deep learning inference workloads derived from a Wheat Plant Diseases image repository. Simulations demonstrate that the prediction model achieves a Root Mean Square Error of 0.030 and a Mean Absolute Error of 0.0215. Compared to static and reactive baselines, the PCSF reduces average task timeout violations to 2.2 and total system energy consumption by nearly 40%. This proactive mechanism effectively overcomes decision-making lags, enabling efficient, low-latency computing resource allocation for modern agricultural facilities. Full article