Search Results (722)

With the increasing complexity of engineering optimization and intelligent decision-making problems, traditional metaheuristic algorithms often suffer from premature convergence, loss of population diversity, and insufficient adaptability to complex fitness landscapes. To address these issues, this paper proposes a Multi-strategy Symmetry-Aware Improved Golden Sine Algorithm (MIGoldSA). The proposed algorithm introduces a symmetry-guided multi-strategy framework in which multiple complementary search operators are organized in a structurally balanced manner. Specifically, a strategy pool consisting of the original golden sine update rule, three differential evolution mutation strategies, and an elite-based quadratic interpolation local search operator is constructed. An adaptive strategy selection mechanism is further developed to dynamically regulate the selection probabilities of different strategies according to their historical success rates, forming a dynamic probabilistic symmetry that balances global exploration and local exploitation throughout the optimization process. The numerical performance of the resulting method is assessed using the CEC2014, 30-dimensional CEC2017, and 20-dimensional CEC2022 test collections. Comparative and statistical findings confirm that MIGoldSA generally delivers more accurate final solutions, more consistent outcomes across independent trials, and stronger convergence behavior than established algorithms and recently developed competitors. Its applicability is further examined in corporate insolvency forecasting by employing MIGoldSA to determine the hyperparameter configuration of a K-nearest neighbors classifier. Tests conducted on the Wieslaw financial database show that the resulting MIGoldSA-KNN system outperforms the selected reference models in classification accuracy, Matthews correlation coefficient, F1-score, and recall. These findings suggest that the proposed symmetry-inspired architecture offers an effective means of coordinating diversified search and intensive refinement, thereby providing a valuable computational approach for challenging global optimization and financial classification tasks. Full article

Power line component and insulator defect detection is an important task in intelligent transmission line inspection. However, UAV-based inspection images often contain small, densely distributed targets under complex backgrounds, making it difficult for existing detectors to balance accuracy and model compactness. To address these challenges, this paper proposes a lightweight YOLOv8-based detector named DSCMamba-TAD-YOLOv8. First, depthwise separable convolutions are introduced into the Neck to reduce parameters and computational cost. Second, DSCMambaNet replaces the original C2f module to enhance multi-scale feature representation by combining lightweight local feature extraction and cross-region contextual modeling. An embedded CBAM component is further integrated inside DSCMambaNet to strengthen informative channel responses and spatial regions. Finally, a Task-Aware Dynamic Detection Head, named TADetect, improves head adaptability through scale-aware and task-aware feature modulation. Experiments on the InsPLAD-det dataset show that DSCMamba-TAD-YOLOv8 achieves 91.86% Precision, 88.02% Recall, 91.83% mAP@0.5, and 74.82% mAP@0.5:0.95. Compared with YOLOv8n, the proposed model improves Precision, mAP@0.5, and mAP@0.5:0.95 by 4.09, 2.43, and 4.46 percentage points, respectively, while maintaining a comparable Recall level with a slight increase from 87.04% to 88.02%. Meanwhile, Params decrease from 3.209 M to 2.702 M and GFLOPs from 8.2 to 7.5. On the revised TPL-SOD held-out test subset, the proposed model improves Precision from 86.20% to 88.16%, mAP@0.5 from 87.09% to 88.81%, and mAP@0.5:0.95 from 68.44% to 70.13%, while Recall remains stable and slightly increases from 91.75% to 92.33%. These results demonstrate that DSCMamba-TAD-YOLOv8 improves detection accuracy and localization quality while maintaining a compact structure and stable recall performance. Full article

Collaborative intelligent sensing in the Social Internet of Things (Social IoT) relies on distributed AI-enabled sensors to support complementary information sharing, multimodal perception, and real-time autonomous decision-making. Under high-load conditions, mismatches between resource provisioning and sensing quality of experience (QoE) can significantly degrade system performance in applications such as smart cities. To address this issue, this paper proposes a service priority-aware collaborative sensing support framework based on a joint next-generation passive optical network (NG-PON) and cooperative intelligent service-based radio access network (CIS-RAN) architecture. The framework enables edge AI-driven inference and distributed sensor collaboration in heterogeneous Social IoT environments. Service-slice-specific priority weights are assigned to optical network units (ONUs) and wavelengths according to the QoE requirements and latency sensitivity of sensing tasks, allowing dynamic wavelength tuning that prioritizes high-impact collaborative services. The utility of a centralized intelligent processing pool is formulated to achieve priority-consistent and efficient resource coordination under collaborative constraints. In addition, a multi-agent AI-driven optimization framework is employed to derive adaptive resource allocation strategies that incorporate service priorities while satisfying stringent service-level agreements (SLAs). Simulation results show that the proposed framework improves system-level proxy metrics, including total utility, wavelength satisfaction, and resource utilization, compared with representative baseline schemes. Full article

Smart charging stations require mobile charging robots to respond to dynamically arriving charging requests with heterogeneous priorities, varying travel costs, and uneven workloads while maintaining online scheduling feasibility. Conventional single-layer approaches often optimize task assignment or route ordering separately, which limits their ability to coordinate allocation quality, route efficiency, and workload regulation under real-time constraints. This study proposes a hierarchical improved consensus-based bundle algorithm–ant colony optimization (ICBBA-ACO) framework for dynamic multi-robot task allocation. The upper ICBBA layer combines deterministic task clustering, intra-cluster greedy bundling, conflict resolution, and feedback-guided workload-aware reassignment, while the lower ACO layer refines the visiting order of unstarted tasks under fixed ownership using the same normalized four-objective scheduling cost. Complete decision time is evaluated separately against a $200\mathrm{ms}$ online requirement, and estimated motion energy is retained only as a distance-derived auxiliary indicator. In a five-method comparison over 100 paired scenarios, ICBBA-ACO achieves a mean composite objective of $J=0.663052$, a mean decision time of $33.07\mathrm{ms}$, and 100% deadline compliance. GA-MRTA obtains a lower unconstrained mean objective of $J=0.615790$, but requires approximately $2199.30\mathrm{ms}$ on average and satisfies the $200\mathrm{ms}$ requirement in only $8.89\%$ of the evaluated updates. Thus, ICBBA-ACO provides the lowest mean objective among the compared methods that maintain full deadline compliance, demonstrating a favorable quality–runtime trade-off within the tested operating range. ROS-based engineering verification further completes all 15 repeated trials and all 48 verification tasks with no recorded invariant violations. Full article

This paper proposes a simulation-driven hierarchical Stackelberg–distributed model predictive control framework (SHS-DMPC) for intercepting multi-wave UAV swarm attacks under limited defensive resources. The interaction between the defender and the attacker is modeled as a Stackelberg leader–follower game. At the strategy layer, a finite-response approximation of Stackelberg decision making is constructed under incomplete information: the attacker’s response type is inferred online from swarm-level motion features, and candidate defender strategies are subsequently evaluated through state-dependent short-horizon rollout simulations. This formulation avoids requiring explicit knowledge of the attacker’s utility function while retaining anticipatory leader–follower strategy evaluation. At the task-allocation layer, target value, threat level, spatial bias, and a reassignment penalty are incorporated into the allocation cost to translate the selected defense strategy into dynamic defender–attacker assignments. At the control layer, each defending UAV solves a local DMPC problem to generate continuous control inputs while satisfying kinematic, inter-UAV separation, and airspace-boundary constraints. Simulation results show that SHS-DMPC achieve a higher interception success rate, a lower value-weighted target loss rate, and fewer minimum-separation violations than the comparison methods under multi-wave heterogeneous attack scenarios, demonstrating the benefits of closed-loop coupling among response inference, strategy-conditioned allocation, and constraint-aware distributed trajectory optimization. Full article

With the development of smart classrooms and digital learning platforms, multimodal learning behavior data provide a new sensing basis for understanding students’ cognitive states and psychological risk warnings. However, existing educational data mining methods mainly focus on performance prediction, dropout warning, or surface-level emotion recognition, while continuous and interpretable modeling of deeper cognitive biases and related psychological risks remains insufficient. To address this issue, we propose MLBS-Net, a multimodal learning behavior sensing network for teaching feedback that jointly models students’ textual expressions, behavioral sequences, classroom interactions, and psychological auxiliary signals. MLBS-Net integrates theory-guided textual cognitive bias encoding, temporal behavioral state modeling, and reliability-aware cross-modal fusion to capture psychologically interpretable cognitive patterns, characterize dynamic learning-state changes, and adaptively integrate multimodal information according to data quality and task contribution while providing interpretable feedback for teachers. Experimental results show that MLBS-Net achieves a Macro-F1 of 0.855 for cognitive bias recognition and an AUC of 0.891 for psychological risk warning, outperforming traditional machine learning, unimodal deep learning, and standard multimodal methods. Ablation results further support the effectiveness of theory-guided semantic encoding, temporal behavioral modeling, reliability estimation, and multitask learning. These findings demonstrate that MLBS-Net can jointly characterize cognitive biases and potential psychological risks from multisource learning behaviors, providing a feasible approach for learning-state sensing, risk warning, and interpretable teaching support in smart education. Full article

Accurate estimation of pasture biomass components from imagery is essential for sustainable grazing management and precision agriculture. Conventional methods such as destructive harvesting, rising plate meters, and remote sensing are limited by scalability, reliability, or the ability to disaggregate biomass by species. We propose a parameter-efficient multi-output regression framework predicting five biomass components (dry green, dry dead, dry clover, green dry matter, and total dry biomass) from high-resolution top-view pasture images. It employs a pretrained DINOv3 Vision Transformer backbone adapted via a dynamic, depth-aware Low-Rank Adaptation (LoRA) strategy, in which the adaptation rank and scaling factor increase exponentially with layer depth: early layers encoding generic visual primitives are minimally perturbed, while deeper layers receive stronger task-specific adaptation. This schedule is effective in low-data regimes, where uniform adaptation or full fine-tuning overfits. To handle rectangular image geometry, each image is split into two square halves processed as a dual-view stream with a contrastive alignment loss. The system ensembles ViT-Large and ViT-Huge backbones with test-time augmentation across five-fold cross-validation. On the CSIRO Image2Biomass benchmark, the full pipeline attains a cross-validated weighted R-squared of 0.81, indicating that depth-aware, parameter-efficient adaptation of large vision models is effective for non-invasive biomass estimation under data scarcity. Full article

Cross-border trade security risks are often embedded in inconsistencies among trade documents, logistics trajectories, hardware sensor states, and financial settlement activities. Existing methods primarily rely on structured declaration fields, making it difficult to verify digital declarations against actual physical processes or to generate complete evidence suitable for regulatory review. To address these challenges, TradeSense-EQA is proposed as a cross-border trade security anomaly detection and evidence-grounded English question-answering framework. Multisource sensing information, including trade documents, GPS/AIS trajectories, RFID records, electronic seal events, port weighing data, temperature and humidity measurements, vibration signals, container door states, and visual images, is jointly modeled within the framework. The reliability-aware representation module dynamically adjusts sensing-channel weights according to data missingness, sampling intervals, device health states, and communication quality. The trade-process-constrained module identifies anomalies across declaration, packing, transportation, transshipment, arrival, and customs clearance stages and generates process-consistent evidence chains. The evidence-grounded question-answering module answers English trade risk questions on the basis of verified documentary fields and sensor records, while confidence estimation and abstention mechanisms are incorporated to reduce factual hallucinations. Experimental results demonstrate that TradeSense-EQA achieved an Accuracy of $0.918$, a Precision of $0.909$, a Recall of $0.897$, a Macro-F1 of $0.903$, and a ROC-AUC of $0.958$ on the cross-border trade anomaly detection task, outperforming baseline methods including XGBoost, LightGBM, TCN, Transformer, BERT, CLIP, and VisualBERT. On the English trade risk question-answering task, Exact Match, Token-level F1, BLEU, ROUGE-L, and BERTScore reached $0.782$, $0.851$, $0.668$, $0.801$, and $0.934$, respectively. Ablation results further confirmed the effectiveness of hardware sensing input, reliability-aware weighting, declaration–fact alignment, process-graph reasoning, and evidence-constrained generation. The proposed framework provides a reliable, interpretable, and auditable artificial intelligence-driven sensing solution for customs supervision, port security, international logistics review, and trade-background investigation. Full article

Accurate non-contact human height estimation is vital for large-scale growth monitoring in schools but remains challenging for monocular RGB sensors due to scale ambiguity and keypoint jitter. This study proposes a robust temporal SegPose framework for high-precision height measurement in unconstrained outdoor playground environments. We develop a multi-task deep learning model using a MobileNetV4 backbone and a novel Height-Aware Boundary Refinement (HABR) module, which utilizes nose-spatial priors to refine cranial vertex localization. To resolve scale issues, a three-way orthogonal calibration system is established using existing playground marking lines and goalposts to dynamically estimate ground plane metric factors. A linear Kalman filter is integrated to smooth keypoint trajectories, suppressing gait-induced oscillations and reducing high-frequency jitter by 64.92%. Validated on a dataset of 95 volunteers (53 males, 42 females) at distances of 6–12 m, the proposed system achieves a mean absolute error (MAE) of 1.42 cm and a mean absolute percentage error (MAPE) of 0.84%, significantly outperforming recent Transformer-based state-of-the-art methods. The framework operates at 42.7 FPS, ensuring real-time performance while adhering to a privacy-preserving protocol that decouples biometric records from individual identities. These results demonstrate that our framework effectively overcomes boundary ambiguity and distance-dependent resolution loss, providing a reliable, efficient, and ethical solution for automated physical health assessments in educational settings. Full article

Ambient Assisted Living (AAL) environments aim to enable older people to remain active and lead an autonomous and independent life for as long as possible. Among the technologies that can be incorporated into these settings, socially assistive robots (SAR) seek to establish a more natural and intuitive means of interaction with people, whilst helping them to carry out everyday tasks. One of the main challenges facing the design of these robots is how to enable them to undertake more complex tasks. Recent advances in Large Language Models (LLMs) have opened new avenues for flexible robot deliberation, yet their integration into real-time robotic systems remains challenging due to latency constraints, reasoning reliability, and the complexity of coordinating multi-step tasks. This paper proposes a hierarchical multi-agent architecture for robot deliberation that addresses these challenges by combining LLM-based planning with structured execution mechanisms within the ROS 2 ecosystem. The proposed architecture employs a supervisor agent that decomposes high-level natural language instructions into prioritised subtasks, enabling a priority-driven execution model that dynamically adapts to task relevance, temporal constraints, and environmental feedback. Subtasks are delegated to a set of Single-Purpose Agents (SPAs), orchestrated via LangGraph state machines and coordinated through a priority-aware scheduling mechanism. A key design principle is the use of Behaviour Trees (BTs) as high-level callable tools through the Model Context Protocol (MCP), encapsulating closed-loop control strategies while enabling preemptive and priority-consistent execution. This reduces the number of LLM inference steps required per task and improves robustness under dynamic conditions. A further contribution concerns the deployment of fine-tuned, lightweight LLMs—on the order of 0.6 billion parameters—specifically adapted for both the supervisor and the individual SPA roles through parameter-efficient low-rank adaptation (LoRA). These models are trained on role-specific tool-calling datasets to specialise in constrained reasoning patterns and task-specific decision-making, enabling efficient, low-latency inference directly on edge hardware. The combination of fine-tuning and hierarchical priority control enhances both the determinism and responsiveness of the system while mitigating error propagation across agent interactions. The paper presents the full software architecture, a formal characterisation of the system as a priority-aware hierarchical policy over a graph of agent workflows, and an experimental evaluation in an Ambient Assisted Living scenario assessing task success rate, inference efficiency, responsiveness under competing priorities, and overall user experience. Because SPA execution is decoupled from the supervisor’s own reasoning loop, the architecture is designed to keep accepting, processing, and queuing new user queries while previously dispatched SPAs are still executing their tasks. Full article

Smart-grid sensing digital twins require dynamic resource allocation and collaborative scheduling to keep status updates from feeder segments, substation equipment areas, distributed-energy-resource access points, and alarmed devices fresh enough for cyber-physical synchronization. The difficulty is not only transmitting more data, but coordinating limited wireless resource blocks, feasible resource-block occupancy, and edge-computing capacity so that critical grid states are delivered and processed before they become stale. This paper studies hierarchical freshness-aware scheduling using Age of Information (AoI) as the main timeliness metric. Dynamic regional priorities are modeled as inputs supplied by the grid monitoring and event-management system; no external mobility-domain dataset is used to validate smart-grid sensing. The core freshness scheduling method combines value-network-assisted communication-resource budgeting, masked policy-gradient resource-block scheduling, and priority-aware computation offloading, while selective redundancy is treated as an optional enhancement for high-priority tail-risk tasks. The evaluation is conducted using a scenario-based smart-grid simulation covering normal monitoring, localized alarms, concurrent high-priority events, and priority migration. The results show that the core ValueNet-MaskedPG scheduling solver reduces mean priority-weighted AoI by about 12.6–15.9% compared with uniform first-come-first-served scheduling. When selective redundancy is enabled, high-priority-zone freshness is improved in event-driven scenarios, but the benefit for mean and peak AoI is scenario-dependent and comes at the cost of additional computation copies. The results support the usefulness of hierarchical scheduling under the considered scenario-based settings, while field SCADA/PMU or hardware-in-the-loop validation remains necessary before practical deployment. Full article

Fake news often manipulates fine-grained knowledge elements such as entities, events, claims, temporal expressions, attributes, and source credibility. Existing information extraction methods usually require task-specific annotations or focus on generic named entities, making them less effective for open-domain fake news scenarios where labeled data are scarce and logical inconsistencies are subtle. This paper proposes HPDO-KEE, a hierarchical prompting framework with dynamic optimization for knowledge element extraction and feature enhancement in fake news detection. The method first defines a fake-news-oriented schema covering entities, events, claims, attribute–value pairs, relations, contradictions, and user authority. It then designs a four-layer prompt consisting of task description, core information, structure awareness, and demonstration assistance. The revised implementation distinguishes offline prompt-template rewriting from input-adaptive demonstration retrieval and automatic schema-validation retries during inference. Domain-aware demonstration selection, strict JSON constraints, redundancy removal, contradiction-candidate verification, and type correction are incorporated to improve extraction accuracy, format compliance, and stability. Experiments on CoNLL03, ACE2005, and DuEE2.0 show that HPDO-KEE achieves F1 scores of 88.9%, 82.6%, 80.3%, and 78.6% on named entity, entity, event, and Chinese event extraction tasks, respectively. Full article

Enumerating all inclusion-minimal hitting sets is a fundamental combinatorial task that arises in areas such as model-based diagnosis, hypergraph dualization, and data mining. In conflict-driven model-based diagnosis, exact enumeration becomes a computational bottleneck when both the component set and the diagnosis family are large. Existing exact methods mainly exploit conflict reuse or conflict-family structure, while component equivalence has rarely been integrated into search, output representation, and partition maintenance as a unified mechanism. This paper presents EC-MHS, an equivalence-aware framework for minimal hitting set enumeration in diagnosis. It combines Static Twin Compression (STC), which exploits coverage-signature symmetry among components to reduce the representative search space before enumeration, Dynamic Twin Compression (DTC) to merge candidates whose residual-state symmetry renders them interchangeable during search, a Compact Diagnosis Family Representation (CDFR) for exact weight-based aggregation and lossless class-wise expansion at the STC level, and iSTC to maintain the static partition under monotonic conflict addition. Experiments on ISCAS-85 benchmark circuits show that static equivalence appears in at least 96% of the dataset instances and that STC reduces the number of search candidates by 69–87% before search. On redundancy-rich benchmark instances, DTC reduces runtime by up to 75%, and iSTC maintains the STC partition 24 times faster than a full rebuild. Full article

Geospatial data provide an important basis for urban governance, resource management, disaster assessment, and public health analysis by linking spatial locations, attribute information, and dynamic processes. However, complex geospatial analysis still requires substantial expertise in spatial databases, spatial SQL, and GIS computation tools. Although large language model-based Text-to-SQL methods have lowered the barrier to natural language-driven data querying, most existing approaches rely on single-step SQL generation and remain unstable for spatial tasks that involve attribute retrieval, spatial relationship evaluation, geometric operations, and statistical aggregation. To address this limitation, this paper proposes a Code Generation-Driven Query–Computation Separation Framework (CGD-QCSF). The framework is based on the separation of query and computation, and decomposes complex geospatial analysis into a staged execution process. CGD-QCSF coordinates intent understanding, schema pre-filtering, planning, execution state management, SQL generation, and spatiotemporal computation. A structured planner and an execution state manager coordinate task decomposition, capability-aware routing, and evidence-based recovery. A SQL Code Generation Agent (SCGA) handles database access, attribute filtering, and intermediate data extraction, while a Spatiotemporal Computation Agent (STCA) performs out-of-database spatial computation and statistical aggregation in an isolated Python sandbox. We construct a benchmark of 200 tasks, covering easy, medium, and hard spatial tasks. In the main experiment with Qwen3.7-Plus as the foundation model, CGD-QCSF achieves a Strict Structured Accuracy (SSA) of 90.5%. Removing the Planner reduces SSA to 84.5%, while removing the STCA reduces it to 70.5%. The ablation experiments show that removing either the Python sandbox or the Planner Agent degrades performance on complex tasks. These results indicate that CGD-QCSF extends complex geospatial analysis from single-step SQL generation into a multi-staged execution process. By explicitly separating query and computation, the framework reduces interference between spatial computation logic and database schema information, thereby improving the stability and success rate of natural language-driven geospatial analysis. Full article

Determining the parameters of a manipulator for optimal performance is a challenging task. This is primarily due to possible conflicting objectives, various tasks that should be considered, and the highly non-linear behavior that is involved. This work proposes an enhanced version of the concept-based design space exploration (C-DSE) approach for the design of manipulators. According to the C-DSE approach, prior to the search, the designers divide the set of feasible solutions into meaningful subsets, which are termed concepts. The design space exploration involves a simultaneous search for optimal solutions within each of the pre-defined concepts. This enhanced framework integrates the following: (1) kinematics, dynamics, and control co-design, and the simultaneous optimization of manipulator morphology and controller parameters; (2) surrogate-assisted optimization using Gaussian process (GP) and neural network (NN) models to reduce computational cost; (3) approximately 30 performance metrics spanning kinematic, dynamic, structural, control, and task performance domains; (4) task-aware feasibility verification applying a multi-level hierarchy; (5) a generative AI integration pathway using diffusion models and LLM-guided concept generation (proposed in this preliminary investigation). The results demonstrate a 95.7% reduction in high-fidelity function evaluations (50,000 to 2150), corresponding to a 23.3 times reduction in evaluation count and a 6.6 times reduction in wall-clock computation time (25 h to 3.8 h). Co-design yields up to a 35% improvement in energy efficiency and a 28% reduction in tracking error compared to sequential morphology-only optimization. Full article