Search Results (1,107)

Background: Metastatic lymph node (MLN) detection remains a major clinical challenge in head and neck cancer, as nodal involvement is strongly associated with poor prognosis and directly affects treatment planning. Previous approaches typically rely on cropped lymph node (LN) regions or tumor contours for MLN identification, requiring substantial expert annotation during preprocessing and relying solely on imaging information. As a result, small or low-contrast metastatic nodes may be missed, while benign lymph nodes may be incorrectly identified as metastatic due to overlapping imaging characteristics. To address these limitations, we propose a multimodal learning framework that integrates anatomical and metabolic features from head and neck PET/CT images with biochemical features derived from FLIm for patient-level MLN prediction, without requiring manual lymph node cropping or tumor contouring during inference. Methods: To enable robust imaging representation learning, a region-aware PET/CT network based on a merging-diverging architecture was first pretrained on the HECKTOR 2022 dataset and then fine-tuned on the institutional cohort. In parallel, FLIm point-wise measurements with clinical variables were encoded using a multilayer perceptron (MLP) and aggregated into subject-level representations. To effectively combine these modalities, two multimodal fusion strategies were evaluated at the decoder stage, including cube-based fusion and squeeze-and-excitation (SE)-based fusion. The proposed strategies were evaluated on a cohort of 53 patients. Results: Compared with the single-modality baselines, both multimodal fusion strategies achieved better patient-level MLN prediction. The PET/CT-only segmentation-driven model and FLIm-only model reached balanced accuracies of 0.815 and 0.665, with AUCs of 0.828 and 0.614, respectively. Cube-based fusion improved balanced accuracy and AUC to 0.827 and 0.850, respectively, while channel-wise SE-based fusion achieved the best overall performance, with a balanced accuracy of 0.839 and an AUC of 0.872. Conclusions: These results suggest that multimodal integration may improve patient-level MLN prediction compared with single-modality approaches. Given the limited sample size, these findings should be interpreted as hypothesis-generating and require validation in larger, independent patient cohorts. Full article

Traditional Unmanned Aerial Vehicle (UAV) oblique photogrammetry for 3D real-scene modeling of historic cultural districts suffers from data gaps, insufficient texture, and poor accuracy in complex alleyway environments, hindering the widespread adoption of UAV technology. To address these challenges, this paper establishes a distortion region identification algorithm based on image grayscale variation range parameters. Then, through fusing UAV oblique photogrammetry, close-range smartphone photogrammetry, and Real-Time Kinematic (RTK) positioning technology, it ultimately constructs a 3D real-scene reconstruction technical framework. To validate the method’s effectiveness and reliability, a field experiment was conducted in the Zaoerxiang Historic Cultural District of Zhanggong District, Ganzhou City, Jiangxi Province, China. The experimental results demonstrate that the proposed algorithm can effectively identify distortions in the modeling results from UAV images. After fusing smartphone images from distorted regions and RTK measurements from ground control points (GCPs), the discrepancies in X, Y, and Z coordinates between the results and verification points mostly fall within 10 to 25 mm, while the differences from the measured lengths using a steel tape measure and a leveling rod were within the range of 10 to 20 mm. Furthermore, compared to approaches that rely solely on UAV images or on the fusion of UAV and all ground-based images for modeling, the method proposed in this paper restores building texture information in occluded areas and improves the accuracy of 3D real-scene modeling while simultaneously reducing data-processing and storage requirements and enhancing operational efficiency. It provides a referenceable technical framework for digital preservation, restoration planning, and smart cultural tourism of historic districts. Full article

The accumulation of training and competition loads represents a critical determinant of musculoskeletal injury risk in professional team sports, yet contemporary monitoring systems remain limited by their reliance on single-domain temporal analysis that overlooks the multi-scale rhythmic patterns inherent in athletic workload signals. This study introduces FDTM (frequency-aware dual-stream temporal model), a deep learning framework that jointly encodes time-domain dependencies and frequency-domain spectral signatures from digital athlete monitoring streams to predict individual injury risk over a forward-looking seven-game horizon. The framework integrates a stacked bidirectional long short-term memory branch augmented with temporal self-attention pooling, a spectral encoding branch employing discrete Fourier transform decomposition across high-frequency (weekly), mid-frequency (bi-weekly), and low-frequency (seasonal) bands, and a cross-modal gated attention fusion module that adaptively balances temporal and spectral representations conditioned on player context. We evaluate FDTM on three heterogeneous public sports datasets spanning basketball (NBA game-log corpus 2013–2023), Australian rules football (AFL Player Workload Dataset), and soccer (SoccerMon open monitoring corpus), comprising 612 athletes and 247,830 player-game observations across ten competitive seasons. FDTM achieves AUC-ROC values of 0.858, 0.833, and 0.821 on the three datasets respectively, outperforming the strongest deep-learning baseline (FEDformer) by 2.0 to 3.3 percentage points and the strongest non-spectral baseline (TCN) by 3.2 to 4.5 percentage points while maintaining a Brier score below 0.04. Ablation studies confirm that the spectral branch contributes 5.1 percent to overall discriminative performance. SHAP attribution analyses identify high-frequency weekly components as the dominant injury-relevant signal, followed by low-frequency seasonal trends and the cumulative acute-to-chronic workload temporal feature, with gating-weight visualizations revealing dynamic modality contributions consistent with established sports science theory. Direct spectral analysis of the raw workload signal confirms that injury-preceding windows exhibit significantly elevated weekly-band power across all three datasets (Mann–Whitney U test, p < 1 × 10⁻⁷), and the architectural advantage is shown to be robust across 30 independent training seeds. These findings suggest that frequency-aware modeling may serve as a transferable methodology for sports engineering applications in injury prevention, return-to-play planning, and individualized rehabilitation, pending further external validation in female athletes and additional team sports. Full article

The passive nature of gravity matching navigation, along with its concealment and freedom from error accumulation over time, is essential for reducing inertial navigation system (INS) errors and enabling high-precision autonomous underwater positioning. The current paper provides a systematic review of major technologies in the field, including the development of underwater gravimeters, construction of gravity reference maps, suitable area selection, optimization of matching algorithms, gravity–inertial integrated navigation, and path planning. We discuss hardware developments, including classical sensors, gradiometers, and quantum sensors, as well as methodological concepts such as multi-source sensor data fusion, intelligent area selection, algorithm optimizations, connections between multiple filters, and intelligent trajectory design. Despite a relatively well-developed technical infrastructure, several bottlenecks remain, including the low engineering maturity of high-end hardware, poor algorithmic performance under extreme conditions, over-reliance on simulation, and weak module integration. Future research should focus on hardware miniaturization, cross-domain intelligent adaptive algorithms, multi-condition real-world validation, and the transition from loosely coupled to tightly coupled architectures to achieve improved accuracy and robustness. Full article

Distributed-drive all-wheel steering (AWS) six-axle vehicles possess distinct advantages in power performance, maneuverability, and environmental adaptability. However, when navigating tight curves under sudden low-friction road conditions, their inherent long wheelbase and strong inter-axle coupling typically lead to compromised spatial maneuverability, trajectory decoupling between the vehicle nose and tail, and lateral dynamic instability. To resolve these critical issues, this paper proposes a geometry–dynamics coupled lateral control scheme with adaptive speed planning for six-axle vehicles under confined spatial and low-friction conditions by seamlessly fusing a dual-point preview mechanism with multi-mode steering mappings. First, a three-degree-of-freedom nonlinear vehicle dynamic model incorporating longitudinal, lateral, and yaw motions is constructed, alongside the formulation of extended Ackermann kinematic steering manifolds for three distinct modes: rear-axle steering, center steering, and crab steering. To rectify the kinematic under-constrained deficiency inherent in conventional single-point preview path-tracking architectures, a joint front-and-rear dual-point preview constraint mechanism is established. This framework permits the quantitative derivation of a spatial geometric reconstruction method for the instantaneous center of rotation (ICR), which algebraically maps the ideal ICR trajectory requirements onto the physical constraints of the selected steering modes. Consequently, complete geometric constraints on both the front and rear trajectories are achieved, enabling active compression of the vehicle’s turning radius. Furthermore, to handle sudden low-friction disturbances, road adhesion limits and vehicle lateral stability boundaries are explicitly incorporated to design a multi-scale adaptive preview distance dynamic scaling mechanism driven by dynamic safety margin corrections. By adaptively scaling the spatial constraint at the geometric layer, this mechanism proactively mitigates nonlinear tire sideslip force saturation via feedforward action, thereby preventing tracking divergence and catastrophic sideslip instability under physical adhesion limits. Co-simulations based on the high-fidelity TruckSim-Simulink platform demonstrate that, in standard curves, the proposed dual-point preview manifold fusion strategy reduces the minimum turning radius by 9.6–10.1% and shortens the cornering transit time by 7.5% compared with the traditional single-point preview mechanism. By actively constraining the front and rear trajectories, the trajectory decoupling between the vehicle nose and tail is effectively resolved. Under narrow-lane scenarios, the maximum lateral error is restricted within 0.78 m, representing a 37.6% reduction relative to the single-point preview, while the maximum steering angle of the front axle is compressed by approximately 18%, thereby significantly improving spatial passability and preventing intermediate body interference. Most notably, under low-friction surface disturbances, the dynamic-margin-corrected adaptive preview adjustment mechanism exhibits remarkable robustness, constraining the maximum lateral tracking error to within 0.68 m. The proposed geometry–dynamics coupled lateral control strategy successfully elevates the tight-curve maneuverability of heavy transport vehicles while concurrently reinforcing their lateral dynamic stability under limit combined spatial and adhesion constraints. Full article

Accurate prediction of missing or potential links is crucial for understanding complex network dynamics and supporting applications such as social recommendation and infrastructure planning. To effectively exploit both global and local structural information, this study proposes a spectral-adaptive fusion (SAF) algorithm. SAF first constructs a spectral embedding matrix by retaining a subset of spectral components, from which a row-column normalized matrix and a Gaussian kernel matrix are derived. These matrices are then adaptively fused to produce link scores, using a common-neighbor-based mechanism that dynamically balances their contributions, capturing both local and global network features while mitigating the influence of highly central nodes. Energy retention and spectral gap analyses set the truncated ratio to 5%, resulting in an average runtime reduction of 71.0% across eight datasets. Under the AUC index, SAF achieves an average relative improvement of 2.22% over advanced graph neural network methods and 10.65% over matrix factorization approaches. Importantly, even at low training ratios, SAF maintains AUPR values above 0.91 on four networks and exhibits stable performance on recall, confirming its robustness and effectiveness for link prediction. Full article

Intermittent sales patterns, prevalent in fast-fashion retail, pose a critical challenge for conventional forecasting methods. This study empirically compares one-stage and two-stage machine learning (ML) frameworks with classical benchmarks (Croston, SBA). The two-stage approach uses a Random Forest classifier for demand occurrence, followed by regression models (RF, GBM, XGBoost, LightGBM) for magnitude. Models are evaluated using weekly sales data from an Iraqi fashion retailer, incorporating rich exogenous features like product attributes, pricing, weather, and special events across 64 unique attribute-defined product group time series. Performance is assessed via a fixed 13-week holdout and rolling-origin cross-validation, with LSTM and Temporal Fusion Transformer (TFT) serving as deep learning benchmarks. Empirical findings show that machine learning configurations achieve superior WRMSSE accuracy, with two-stage models often outperforming one-stage counterparts, and both significantly surpassing classical and deep learning baselines. The Two-Stage XGBoost yielded the lowest WRMSSE, establishing the feature-engineered two-stage framework as the strongest overall for this intermittent retail setting. Furthermore, a detailed SHAP analysis elucidated the distinct feature contributions to demand occurrence versus demand magnitude, providing actionable insights for inventory management. This rigorous benchmarking analysis offers practical implications for inventory planning and demand management in volatile markets, highlighting the effectiveness of explicit demand occurrence modeling. Full article

With the advancement of smart city technologies and the proliferation of Mobility as a Service (MaaS), realizing seamless navigation that continuously connects heterogeneous mobility modes and indoor–outdoor spaces has emerged as a critical challenge. However, existing navigation services operate in a fragmented, siloed manner, segmented by transport mode and spatial environment, and thus possess fundamental limitations in supporting continuous mobility. This study establishes an analytical framework comprising the four core components of navigation systems (positioning, navigation data, route planning, and guidance) and critically reviews 108 prior studies identified through purposive sampling from Web of Science, Scopus, and Google Scholar to evaluate the technical requirements and the level of seamless integration achieved for each component. The analysis reveals that while each component has reached a high level of maturity within its individual domain, four critical technical gaps persist across all components: positioning handover discontinuities at indoor–outdoor transition zones, structural and semantic inconsistencies between heterogeneous spatial datasets, static route planning that fails to account for transition-space uncertainties, and guidance systems whose context resets upon changes in transport mode. These gaps originate not from insufficient performance of individual technologies but from a systematic lack of research at the interface points between components. Overcoming these challenges necessitates a comprehensive redesign of the integrated system architecture, encompassing dynamically adaptive multi-sensor fusion positioning, hierarchical heterogeneous data integration models, probabilistic cost modeling for transition spaces, and adaptive guidance systems based on automatic context handover. Full article

This study addresses the issue of accurate and rapid aboveground biomass estimation in rangeland ecosystems, as traditional grazing methods are labor-intensive, while modern remote sensing techniques often require expensive equipment and controlled conditions. The goal of this work is to develop an efficient and accessible approach for biomass estimation of natural pastures based on ground-level RGB images captured with smartphones. For this purpose, a dataset consisting of 1196 field images and corresponding biomass values collected from 40 districts in southern Kazakhstan was used, and a wide + tiles architecture based on the DINOv3 model of Vision Transformer was proposed. The model utilized attention pooling and feature fusion mechanisms to integrate both global and local features, and various preprocessing and augmentation strategies were comparatively examined. Experimental results demonstrated that the proposed method exhibits high accuracy (with the best result being R² = 0.733, MAE ≈ 0.779 c/ha), where the DINOv3 model showed clear advantages over ConvNeXtV2. Furthermore, the impact of preprocessing strategies was minimal, and the importance of high-resolution images was clearly established. The obtained results show that the proposed method performs consistently under heterogeneous field conditions and allows for reliable biomass estimation without the need for specialized equipment. This makes it a practical tool for monitoring pastures, planning forage supply, and supporting agronomic decision-making. Full article

Transitions of control between automated driving systems and human drivers remain safety-relevant and cognitively demanding moments in human–automation interaction. Recent studies show that transition performance depends not only on takeover timing or response speed but also on traffic complexity, driver readiness, automation limitations, trust calibration, and situational-awareness recovery. As in-vehicle interaction evolves toward conversational and agentic AI assistance, takeover support also becomes a problem of governing how natural-language AI systems communicate with the driver under uncertainty. This paper proposes a digital-twin-mediated framework for human-aware takeover support in automated driving. In this framework, the companion AI is treated as an assumed LLM-based in-vehicle conversational or agentic assistant used as an advisory interaction component. The contribution is defined at the architectural level: human, vehicle, and context/road digital twins provide structured semantic state abstractions through a semantic state interface exposing confidence, freshness, provenance, and consistency metadata, while a trustworthy companion AI (TCAI) layer grounds, constrains, validates, and governs companion AI output proposals before HMI delivery. Building on the research on driver-state monitoring, adaptive HMI, trust calibration, explainability, conversational assistance, and human assistance systems (HASs), the framework coordinates advisory interaction across vigilance support, contextual explanation, trust-calibrating communication, and directive handover guidance. The TCAI layer combines bounded reasoning, human-factor-derived guardrails, state-consistency management, dynamic explanation-depth control, trust-dynamics modeling, graded watchdog veto handling, mandatory access-control assumptions, and deterministic fallback. Safety-critical vehicle-control and minimum risk condition (MRC) functions remain assigned to the deterministic vehicle-control stack, while the authorized output path of the TCAI layer is validated HMI delivery. The paper concludes with a validation agenda and technical roadmap covering planned transitions, urgent handovers, degraded or adversarial conditions, temporal fusion of driver-state evidence, phase-sensitive HMI policies, trust-calibration trajectories, driver veto and partial-disabling mechanisms, and staged simulator-to-vehicle evaluation. Although motivated by SAE Level 3 automation, the framework may also inform fallback-related Level 4 scenarios in which human and automated agency must be managed under uncertainty. Full article

Computed tomography angiography (CTA)-based segmentation of the superficial temporal arteries (STAs) and facial vessels (FVs) is important for neurosurgical and reconstructive planning. Nevertheless, segmentation of STAs and FVs remains challenging because of their small caliber, tortuous courses, and proximity to high-intensity bony structures. This study aims to develop a deep learning framework for accurate automated segmentation of these craniofacial vessels. A single-input 3D nnU-Net v2 model was trained using raw CTA volumes, while a Fusion-based Vesselness Map (FVM) was constructed from multiscale vessel-enhancement filters to emphasize small vascular structures and suppress irrelevant regions such as the skull and skin. Instead of being used as an additional input channel, the FVM was incorporated into the loss function as a spatial prior to guide the network toward vessel boundaries and distal branches. In 72 clinical cases, the FVM-guided model improved segmentation accuracy compared with a baseline model trained with Dice Focal Loss, particularly in boundary delineation. For the STAs, the Average Symmetric Surface Distance decreased from 6.543 mm to 2.941 mm. Qualitative evaluation further showed reduced segmentation noise and fewer false positives near bone and distal branches. These findings suggest that integrating classical vessel enhancement into deep learning supervision can improve morphologically consistent craniofacial vessel segmentation and support preoperative surgical planning. Full article

Joint detection of strawberry growth stages and defective fruit is needed for harvest planning and quality screening, but field images make this task difficult because stage-related visual differences are subtle, flowers and early fruits are often small and densely distributed, and occlusion weakens localization reliability. This study develops Multi-Scale Refined Detection and Classification YOLO (MRDC-YOLO), a lightweight detector based on the YOLO11s framework, for this fine-grained detection scenario. The backbone, neck, and detection head are redesigned with three modules: a Multi-Scale Adaptive Edge Enhancement Module (MAEM), a Reparameterized Progressive Feature Aggregation (RPFA) module, and a Decoupled Cross-Scan Head (DCSH). MAEM strengthens boundary and texture responses for visually similar categories, RPFA reduces redundant multi-scale fusion while maintaining features for dense small targets, and DCSH introduces task-aware classification and regression branches with cross-scan-inspired spatial modeling for occlusion-sensitive localization. Experiments on a five-class strawberry dataset containing 5114 images show that MRDC-YOLO achieves 95.63% mAP@0.5 and 82.39% mAP@0.5:0.95. Over YOLO11s, the model yields a 2.06-percentage-point gain in precision and 1.34- and 1.53-percentage-point gains in mAP@0.5 and mAP@0.5:0.95, together with 10.7% fewer parameters and 8.9% lower GFLOPs. These results suggest that MRDC-YOLO improves fine-grained category discrimination and localization while retaining a smaller model size than the YOLO11s baseline. Full article

Vessel port stay time is a key indicator for berth allocation, crane planning, and short-term operational coordination in container terminals. However, existing prediction approaches often rely mainly on numerical operational data and assume complete information availability, limiting their reliability when localized visibility constraints or incomplete sensing occur. This study develops and evaluates an availability-aware multimodal prediction framework for vessel port stay time estimation. The framework adapts cross-attention-based fusion to integrate structured operational variables, numerical marine weather observations, and image-derived visibility information extracted from port monitoring images under incomplete monitoring image availability. In the framework, operational and numerical weather variables form the structured predictive state, whereas image-derived visibility information is conditionally incorporated as an auxiliary visual signal only when a matched and usable monitoring image is available. The proposed approach was evaluated using long-term vessel call data from a major container terminal. Compared with commonly used machine learning and deep learning baselines, the proposed model improved prediction accuracy, while residual analyses indicated reduced systematic prediction bias. These findings suggest that the proposed framework can support more reliable short-term berth planning under practical data-collection constraints. Full article

The detection and localization of large underwater targets are important for maritime security, marine resource exploration, and underwater situational awareness, while the increasing acoustic stealth of underwater vehicles has limited conventional acoustic methods. This review provides a systematic overview of non-acoustic detection and localization technologies for large underwater targets, with emphasis on their relevance to unmanned aerial, surface, and underwater platforms. Wake-based detection, magnetic anomaly detection (MAD), and gravity anomaly detection (GAD) are reviewed as three representative non-acoustic routes. A bibliometric analysis is first conducted to summarize research trends, major contributors, and emerging hotspots. Wake-based methods are discussed in terms of wake signatures, modeling approaches, sensing platforms, and localization potential. MAD is analyzed from the perspectives of magnetic dipole modeling, target-based detection, noise-based detection, artificial intelligence (AI)-based detection, and magnetic localization. GAD is discussed with respect to physical feasibility, gravity-gradient target modeling, inversion methods, and engineering constraints. The review shows that wake-based methods are suitable for wide-area search and trajectory inference, MAD is relatively mature for short-range confirmation and localization, and GAD remains promising but less mature. Future research should focus on onboard sensors, platform stability, weak-signal extraction, background suppression, quantitative evaluation metrics, multi-source fusion, autonomous mission planning, and multi-platform collaboration. Full article

Unmanned aerial vehicles (UAVs) and deep learning (DL) have introduced a new framework for intelligent building façade defect detection, yet existing studies often focus on isolated technical components and lack a systematic evaluation of the entire pipeline. To address this gap, this paper conducts a systematic literature review of 135 peer-reviewed journal articles retrieved from the Web of Science database over the period 2021–2026. This review investigates four key domains: (1) UAV inspection path planning and data acquisition; (2) multi-modal data fusion; (3) DL-driven defect detection algorithms; and (4) 3D reconstruction and digital twin integration. Our analysis reveals the following main findings. Real-time perception-aware planning is central to UAV path planning, yet most studies lack robustness evaluations under real-world deployment conditions. Multi-modal data fusion improves detection across multiple defect types, yet edge deployment requires balancing lightweight design with recognition stability. Defect recognition algorithms increasingly adopt task-driven architectures, but limited edge-device resources demand joint optimization of efficiency and accuracy. In digital twins, systematic research is still lacking on semantically integrating recognition results into BIM for O&M decision-making, leaving the closed loop from defect detection to maintenance unresolved. This review aims to help researchers and practitioners advance UAV-based inspection from an auxiliary tool to a fully autonomous, reliable intelligent agent for refined management of the urban built environment. Full article