Search Results (4,803)

To address the limitations of vision-only defect detection, image–semantic misalignment, and spatial-logic conflicts in complex substation inspection scenarios, this paper proposes a camera-sensor-based multimodal defect sensing framework with cross-modal feature mapping for substation equipment monitoring. The proposed framework integrates field inspection images acquired by camera sensors, defect textual descriptions, and equipment topology knowledge and establishes a unified domain-adaptive pre-training–bidirectional cross-modal mapping–hierarchical reasoning workflow. First, a Contrastive Language–Image Pre-training (CLIP)-based domain-adaptive pre-training strategy is developed to enhance the representation of equipment categories, defect attributes, and inspection-scene semantics. Second, a bidirectional cross-modal feature mapping network is constructed to model fine-grained interactions between candidate visual regions and textual semantics, where uncertainty-aware fusion and prototype constraints are introduced to improve semantic alignment and defect discrimination. Third, a hierarchical neuro-symbolic reasoning module incorporates equipment topology and spatial rules for posterior verification, logical consistency checking, and false-positive suppression. Experiments on a substation inspection image dataset demonstrate that the proposed method achieves 90.8% mAP@0.5, 68.7% mAP@0.5:0.95, and 89.4% F1-score, outperforming mainstream and recent detection models. Full article

Bridge decks are continuously subjected to high environmental exposure, traffic loading, and material aging, leading to progressive delamination which can negatively affect structural integrity and public safety. More specifically, subsurface delamination of concrete and corroded steel reinforcement must be repaired to keep the decks operational. Among non-destructive evaluation techniques, Ground-Penetrating Radar (GPR) and Infrared Thermography (IRT) offer complementary capabilities for detecting subsurface and near-surface defects; however, effective GPR-IRT data fusion remains challenging due to fundamental differences in sensing principles, spatial resolution and sensitivity. This study introduces a Physics-Enhanced Multi-Modal Fusion (PE-MMF) framework that integrates GPR and IRT data to improve delamination detection in reinforced concrete bridge decks. The proposed approach leverages transfer learning, cross-modal attention mechanisms, and gated fusion to enable robust learning from heterogeneous sensor inputs. Furthermore, a systematic feature selection protocol is integrated to identify physically meaningful indicators that remain consistent across different bridges, enhancing generalization capability. The framework is trained and validated using the publicly available SDNET2021 dataset, comprising co-registered GPR and IRT measurements from five in-service bridge decks with verified delamination ground truth. Results demonstrate substantial performance improvements, with average F1-score gains of up to 55% over IRT-based methods and 25% over GPR-based methods across all tested bridges. Comparative analysis against state-of-the-art methods confirmed the superior generalization capability of the proposed multi-modal approach over single-modality approaches. The findings highlight the potential of deep learning-based sensor fusion as a scalable and data-efficient decision-support tool to prioritize regions for detailed physical investigation during long-term infrastructure monitoring. Full article

Short-term traffic forecasting is a critical component of intelligent transportation systems. While deep learning architectures for this task have proliferated rapidly, the sensor-level data characteristics—zero-value prevalence, distributional heterogeneity, and cross-sensor correlation structure—that drive architecture-specific failure modes remain insufficiently understood, and their implications for evidence-based model selection in real deployments have not been systematically addressed. This study addresses that question through a sensor-network diagnostic framework applied to the METR-LA dataset (Metropolitan Los Angeles; 207 inductive loop detectors, 5-min resolution). The framework integrates systematic characterization of sensor data properties, a controlled benchmark of four representative architectures—Transformer, Spatio-Temporal Graph Convolutional Network (STGCN), Diffusion Convolutional Recurrent Neural Network (DCRNN), and Gated Temporal Convolutional Network (Gated TCN)—under a unified 12→3 prediction setting, and a novel per-sensor regression analysis that quantitatively links zero-value ratios to model-specific prediction errors across all 207 sensors. Building on these findings, this study further proposes Graph-Enhanced Transformer (GETFormer), a lightweight hybrid architecture that augments the Transformer with a single-hop Graph Convolutional Network (GCN) layer and a gated residual fusion module. The diagnostic findings and condition-dependent model-selection guidelines provide an empirically grounded foundation for principled hybrid architecture development in urban traffic sensing. Full article

Industrial control systems (ICSs) are cyber–physical environments in which physical process data and network communication data are generated simultaneously. Existing studies have mainly focused on either sensor-based or network-based anomaly detection, making it difficult to capture diverse attack indicators and motivating the use of multimodal methods that can leverage complementary information from both modalities. In this paper, we propose an unsupervised multimodal anomaly detection framework for ICSs that jointly uses sensor and network modalities. For each modality, autoencoder-based single-modality models are trained in an unsupervised manner, and their anomaly scores and latent feature vectors are extracted. These outputs are temporally aligned to construct a time-aligned multimodal table, which is then used to implement and compare two fusion strategies: anomaly score fusion and latent feature fusion. In latent feature fusion, aligned modality-specific latent features are combined with canonical correlation analysis (CCA)-derived cross-modal correlation features. The experimental results showed that latent feature fusion achieved stable performance across multiple sensor–network encoder combinations. In particular, the gated recurrent unit–convolutional neural network (GRU–CNN) combination achieved the best F1-score of 0.9166 and ROC-AUC of 0.9795. In addition, the complementarity analysis showed that latent feature fusion recovered some missed detections by integrating complementary sensor and network evidence. These results demonstrate that latent feature fusion is an effective multimodal strategy for ICS anomaly detection. Full article

The growth of Intelligent Transportation Systems (ITS) has made high-quality perception data from multi-sensor setups essential. Pairing millimeter-wave (mmW) radar with a monocular camera is a common way to recover three-dimensional information about the environment, but aligning the two is difficult because sparse radar point clouds and dense camera images differ sharply in how they sense a scene. The problem grows more severe in roadside infrastructure, where the high mounting elevation introduces perspective distortion that vehicle-mounted systems rarely face. This paper presents a systematic review, conducted under the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, of radar-camera extrinsic calibration for fixed roadside infrastructure, organizing existing work into a taxonomy that separates traditional two-stage pipelines from recent end-to-end learning frameworks. Because methods designed specifically for roadside units remain scarce, the review also covers vehicle- and robot-mounted methods whose static-sensor formulation carries over to fixed roadside deployment. For the two-stage pipeline, the analysis covers target-based and targetless correspondence registration along with the optimization techniques and algorithmic assumptions behind parameter estimation. The end-to-end learning literature shows a clear shift toward self-supervised and fusion-based models, some of which report real-time performance. The review also compares the metrics and procedures used to quantify calibration accuracy. Progress is evident, but robustness in cluttered urban environments remains an open challenge, and the paper closes by outlining future directions, arguing that standardized roadside benchmarks are needed before scalable, targetless calibration can mature. Full article

Background/Objectives: Deep learning models for digital mammography sensor data are increasingly deployed across hospitals using different X-ray detector technologies and patient populations. Whether models trained on one sensor platform and population maintain accuracy when transferred to another has not been tested for the latest generation of mammography-specific foundation models under one controlled protocol. Methods: We fine-tuned five backbone architectures (ResNet-50, DINOv2-B14, Rad-DINO, Mammo-CLIP B5, and Mammo-FM) on CBIS-DDSM (film-digitized, USA, n = 714 validation) with three seeds, ablated a density-aware focal loss across three auxiliary weights, and evaluated transfer to three external sensor cohorts: CMMD (full-field digital, China, n = 1032), DMID (mixed digital, India, n = 509), and MIAS (film-digitized, UK, n = 322). Significance used paired DeLong z-tests with Benjamini–Hochberg FDR correction; temperature scaling tested post hoc recalibration at all transfer targets. Results: Within this single-source three-seed evaluation, ResNet-50 outperformed all four foundation models on CBIS-DDSM (AUC 0.867 vs. 0.847, 0.846, 0.813, and 0.703; all gaps p_adj < 0.05). The density-aware focal loss degraded both AUC and calibration at every weight tested. At transfer, every model lost 0.165 to 0.320 AUC points relative to in-distribution performance, with sensitivity at 95% specificity collapsing from 0.31 to 0.47 in-distribution to 0.11 to 0.22 across the three external targets. A per-seed Stouffer meta-analysis confirms that Mammo-CLIP B5 and Mammo-FM significantly outperformed ResNet-50 on DMID and Mammo-CLIP on CMMD, after BH-FDR; MIAS comparisons remained directional only. In the extremely dense subgroup (BI-RADS D4), Mammo-FM reached AUC 0.870 versus ResNet-50 at 0.842, a directional observation whose 95% CIs overlap heavily at the n = 140 sample size and which we do not interpret as a statistically supported advantage. Conclusions: In this single training-source, three-seed protocol, mammography-specific pretraining did not deliver the in-distribution AUC premium reported in the originating papers, and no architecture reached a level at which transfer deployment without local validation would be defensible. We frame these as observations specific to the present protocol rather than as broader conclusions about foundation models for mammography classification. The findings argue for sensor-stratified and population-stratified external validation and for local recalibration as practical prerequisites before clinical use. Code and weights are released under MIT license. Full article

Autonomous ship maneuvering in harbor environments is increasingly supported by advances in model predictive control, reinforcement learning, digital twins, multi-sensor fusion, berth allocation, and multi-agent coordination. However, these developments are often studied as separate technological domains, while real harbor autonomy requires coordinated operation across vessels, port infrastructure, regulatory systems, cybersecurity mechanisms, and human supervisory processes. This study presents an architecture-oriented critical review of autonomous ship and harbor maneuvering research published between 2015 and May 2026. The review synthesizes literature from control engineering, maritime artificial intelligence, sensor fusion, digital twins, port logistics, cyber-physical systems, regulation, cybersecurity, and human–AI supervision. The analysis introduces two conceptual contributions: a layered cyber-physical taxonomy and an integration maturity model. The taxonomy organizes autonomous harbor maneuvering into seven interdependent layers: physical dynamics, perception and sensor fusion, prediction and state estimation, control, decision and coordination, digital twin federation, and regulatory–supervisory governance. The maturity model distinguishes isolated vessel autonomy, assisted coordination, shared digital synchronization, agent-based coordination, and fully federated maritime cyber-physical autonomy. The reviewed evidence shows substantial progress in individual layers, especially control, perception, digital twins, and berth allocation. However, major gaps remain in cross-layer synchronization, semantic interoperability, regulation-aware decision-making, cybersecurity integration, and validated ship–shore federation. To address these gaps, this study proposes a Federated Maritime Cyber-Physical Architecture for autonomous harbor maneuvering. The architecture integrates vessel autonomy cores, port intelligence cores, semantic federation middleware, agent-based negotiation, regulatory verification, cybersecurity safeguards, and human supervisory interfaces. This review argues that future progress in autonomous harbor operations depends not only on stronger algorithms, but on interoperable, explainable, regulation-aware, and cyber-resilient ship–shore ecosystems. Full article

Safety-critical Industrial Internet of Things (IIoT) sensor networks deployed in disaster scenarios require intelligent routing mechanisms that prioritize mission-critical packets without relying on centralized coordination. Federated learning on resource-constrained edge nodes presents three primary challenges: the absence of an interpretable supervisory signal, the inability to act conservatively based on per-inference confidence, and vulnerability to partial node availability. The proposed FedCARE framework addresses these issues by employing a Mamdani Fuzzy Inference System to generate traceable criticality labels from multi-modal sensor telemetry, a dropout-aware aggregation protocol that normalizes over only reachable nodes, and a confidence-gated resolver that defers to symbolic fuzzy classification when model confidence is insufficient, otherwise applying an auditable maximization rule to prevent under-prioritization of safety-critical data. Evaluation on 50-, 100-, and 200-node Watts–Strogatz topologies under fault rates up to 50%, using the Edge-IIoTset and WUSTL-IIoT-2021 benchmarks, demonstrates 99.00% critical recall and up to $1.8\times$ higher overall-packet delivery compared to RPL-RP under severe fault conditions. Routing improvements are primarily attributed to fuzzy criticality labeling and multi-path replication. These findings indicate that fuzzy-supervised federated inference offers a practical and interpretable solution for safety-critical IIoT routing, with an observed energy overhead of 7.8% per delivered packet. Full article

Autonomous navigation of unmanned surface vehicles (USVs) in complex obstacle scenarios remains challenging due to redundant perception inputs, unstable value estimation, and inefficient policy convergence. To address these problems, this paper proposes LDA-D3QN, an improved deep reinforcement learning method for USV autonomous navigation. The proposed method constructs a compact navigation state representation by combining target-related information with local obstacle features, allowing the agent to retain key decision-making information while reducing unnecessary environmental redundancy. Based on this representation, an enhanced value-learning framework is developed to improve the stability of navigation decisions in cluttered environments. Moreover, a reward-guided and staged training strategy is introduced to help the agent gradually adapt to increasingly complex navigation tasks. The proposed method was evaluated on a Unity–ROS–MATLAB integrated simulation platform. Experimental results show that LDA-D3QN achieves superior overall navigation performance compared with several representative reinforcement learning algorithms. Specifically, the proposed method achieves a final training success rate of 91.4%, outperforming PPO (82.3%), Dueling DQN (78.5%), Double DQN (79.8%), and Rainbow DQN (86.5%). Additional tests in complex multi-obstacle and multi-target scenarios further demonstrate that the learned policy can generate safe, stable, and effective navigation behaviors. Preliminary validation using real-USV sensor data also confirms the feasibility of the LiDAR and GPS data processing procedures, providing a basis for future closed-loop autonomous navigation experiments and multi-sensor fusion deployment. Full article

Introduction: Pedestrians lying on the road—collapsed through medical emergency, intoxication, or displacement following a prior collision—represent a disproportionately lethal and underaddressed category in road traffic safety. Forensic database analyses derived from Japan’s national police records document a fatality rate of 33.0% for collisions involving pedestrians lying on the road, more than double the rate for upright pedestrian collisions. Standard Advanced Driver-Assistance Systems (ADAS) yield a True Positive Rate (TPR) of only 21.4% for detecting pedestrians lying on the road under night conditions—a classification gap of 73.3 percentage points. Methods: In simulation trials, we evaluated the Advanced Falling Object Detection System (AFODS—where “falling object” denotes the low-profile human form at road level, distinguishing the prone pedestrian from the upright postures addressed by conventional ADAS) on a composite dataset of 3200 annotated fall events and 12,000 negative samples (training/validation), with 320 independent controlled simulation trials used for performance evaluation, spanning real-world, forensic-reconstruction, and Total Human Body Model for Safety (THUMS)-validated synthetic scenarios. No physical prototype has been evaluated; all performance data are derived from simulation, and 37.5% of positive samples are synthetically generated. These simulation conditions represent a first feasibility demonstration pending real-world hardware validation. This paper introduces three original contributions absent from prior work: a three-stage quantitative injury-risk model, a formal ISO 26262 Hazard Analysis and Risk Assessment (HARA), and a medicolegal SHAP interpretability framework. The injury-risk model translated detection latency via impact velocity to Head Injury Criterion (HIC) and estimated fatal injury probability (AIS ≥ 5); these model outputs should be interpreted as exploratory estimates pending ATD validation. Reporting follows principles consistent with the TRIPOD statement. Results: Under clear daytime conditions, AFODS demonstrated a TPR of 98.2% (95% CI: 97.4–98.8%) in simulation, decreasing to 95.6% under night dry-road conditions and 89.4% under night rain. The system achieved an AUC of 0.981 and a mean end-to-end latency of 46.5 ms, representing a 76.8 percentage-point improvement in simulation over the monocular RGB baseline (p < 0.001). The injury-risk model projects a reduction in estimated fatal head injury probability from 66.2% (Monte Carlo mean) (no detection, 50 km/h full-speed impact) to 0.7% under AFODS worst-case night/rain conditions, and to ≈0% under clear daytime simulation conditions. Conclusions: A 73.3 percentage-point classification gap places pedestrians lying on the road outside the effective detection envelope of current ADAS, compounded by the systematic exclusion of non-upright postures from regulatory test protocols and benchmark datasets. AFODS supports proof-of-concept feasibility under simulation conditions. Three translational steps are required: prototype validation on real-world hardware using instrumented Anthropomorphic Test Devices (ATDs); prone-posture biomechanical injury modelling using HIC and BrIC criteria; and regulatory extension of pedestrian AEB test standards to non-upright scenarios. Full article

Wearable healthcare technologies are transforming the healthcare landscape by enabling remote, real-time health data collection, supporting early diagnosis, personalizing treatment plans, and reducing healthcare costs and medical burdens. Central to these advancements are wearable sensors, which continuously capture physiological data such as heart rate, temperature, activity levels, and biomarker concentrations. However, the large volume and complexity of this data demand effective processing to extract meaningful medical insights. Artificial intelligence (AI) and machine learning (ML) have significantly enhanced the capabilities of wearable sensors by enabling advanced data analysis, pattern recognition, and predictive modeling. AI-enhanced wearable sensors can detect early signs of health issues, such as heart attacks, chronic diseases, and mental health conditions like stress, often before clinical symptoms become apparent. This review examines the integration of AI/ML models with wearable sensors across physical activity recognition, stress assessment, cardiovascular monitoring, personal exposure monitoring, and sweat biomarker detection. Unlike prior application-centered reviews, we emphasize methodological and translational evaluation by comparing task formulations, sensing modalities, dataset scale, validation protocols, performance metrics, and deployment constraints across domains. We further discuss advanced architectures, multimodal fusion, explainable AI, edge deployment, privacy and regulatory considerations, and the translational gap between research prototypes and clinically deployable wearable AI systems. Full article

The Vortex Ring State (VRS) poses a catastrophic aerodynamic threat to coaxial dual-rotor unmanned aerial vehicles (UAVs). Traditional reactive detection mechanisms provide insufficient altitude for recovery, while existing data-driven diagnostics are severely bottlenecked by data leakage, extreme class imbalance, and a lack of physical interpretability. To bridge these gaps, this paper proposes a physics-informed multimodal deep learning framework that transitions from post-occurrence detection to proactive early warning. We establish a 1.5 s precursor window—creating a three-class ordinal state space—to provide the flight control system with critical intervention time for differential rotor recovery. We developed a novel ViT-LSTM architecture (MTSF-Net) to fuse continuous seven-channel onboard-recorded data (comprising three-axis acceleration, three-axis angular velocity, and barometric vertical velocity), which are subsequently transformed into Continuous Wavelet Transform (CWT) spectrograms. To ensure real-time unidirectional inference while preserving absolute physical vibration scales across heterogeneous sensors, a Calibrated Benchmark Normalization (CBN) strategy is introduced. Furthermore, a Hybrid Ordinal Loss is proposed to mitigate the extreme sample imbalance (<0.5%) of the precursor state by penalizing asymmetric aerodynamic degradation. Evaluated under a strict sortie-based isolation protocol, the proposed system achieves an exceptional test accuracy of 98.26% and an unprecedented precursor recall of 100%. Notably, it completely eliminates fatal missed detections (VRS predicted as Normal) and false-positive VRS predictions triggered by precursor states. Finally, Gradient-weighted Class Activation Mapping (Grad-CAM) is utilized to verify that the multimodal sensor processing pipeline successfully anchors onto authentic physical vibration frequencies rather than artifactual noise, laying a rigorous, interpretable foundation for intelligent aviation safety systems. Full article

Precision feeding is an important foundation for improving production efficiency in aquaculture, reducing feed waste, mitigating water pollution, and promoting the intelligent development of aquaculture. Conventional feeding practices remain heavily dependent on operator experience and are typically executed at predetermined times or fixed ration levels. Such approaches frequently result in extensive feeding management, poor adaptability, low feed utilization efficiency, and delayed responses to environmental changes. Advances in machine vision, the Internet of Things, machine learning, deep learning, and automatic control have progressively shifted aquaculture feeding research beyond standalone automatic feeders toward integrated systems encompassing demand perception, intelligent decision-making, precise control, and equipment coordination. This paper reviews the state of the art in precision feeding technologies and equipment in aquaculture. At the technical level, it summarizes advances in feeding demand perception, intelligent feeding decision-making, and precise control and execution. At the equipment level, it reviews the main types, design features, and field application status of precision feeding equipment in intensive aquaculture, pond aquaculture, and offshore aquaculture scenarios. Despite the considerable progress achieved, the practical deployment of precision feeding still faces several limitations. Environmental disturbances, water turbidity, illumination variation, and sensor drift may compromise the reliability of feeding demand perception. Existing decision-making models frequently exhibit limited generalizability across species, growth stages, and aquaculture scenarios. Moreover, insufficient integration of sensing, decision-making, and execution restricts the development of fully closed-loop feeding systems. High initial investment, maintenance costs, and the shortage of skilled personnel further constrain the adoption of precision feeding equipment, particularly in resource-limited regions. On this basis, the main challenges including sensing accuracy, model practicability, closed-loop control, equipment reliability, and standardization, are examined. Future development trends are also discussed, covering multi-source information fusion, synergy between mechanistic models and data-driven methods, system-level closed-loop control, equipment modularization, and industrial application. This review is expected to provide a reference for subsequent research and engineering applications. Full article

Air quality forecasting and environmental health research at urban and regional scales depend on the combination of measurements from heterogeneous sensor networks, yet the construction of integrated multi-source datasets is rarely described or released as a self-contained deliverable. This paper presents an open dataset that combines four sensor-derived sources covering the whole of Spain over the period from 2022 to 2024: hourly air quality observations from the 588 stations of the national network operated by the Ministerio para la Transición Ecológica y el Reto Demográfico (MITECO), daily meteorological records from the Agencia Estatal de Meteorología (AEMET), daily mobility indicators derived from anonymised mobile telephony events published by the Ministerio de Transportes y Movilidad Sostenible (MITMA) at the municipality level, and a calendar of national and Autonomous Community public holidays. The processing pipeline harmonises sources that differ in temporal resolution, spatial codification and quality regime into a tidy hourly table indexed by station and timestamp, with a fixed feature schema of 56 variables per record. Air quality stations are paired with their nearest AEMET station through a three-tier distance rule, and the daily exogenous features are aligned to the air quality time axis through a two-variant temporal-alignment scheme (lag-and-expand to the hourly grid for the hourly release, same-calendar-day join for the daily release). A complementary daily resolution variant of the dataset is also released, with 72 columns and the same feature schema except for the air quality block, which is aggregated to daily mean, minimum and maximum. The integrated dataset contains approximately 15 million hourly records across the 588 stations and is released on Zenodo (DOI 10.5281/zenodo.20196221) under a Creative Commons Attribution 4.0 International (CC BY 4.0) licence. It is intended as a substrate for research on air quality forecasting, environmental epidemiology and multi-source data fusion at the nationwide scale. Full article

Offshore wind turbine blades are chronically exposed to complex marine environments with high humidity, salt spray, strong wind, waves, and intense radiation. Under such conditions, blade defects often exhibit small sizes, weak visual features, and heterogeneous visible infrared manifestations. Conventional single-sensor monitoring and empirically weighted fusion methods are insufficient for reliable defect diagnosis and risk grading. To address this problem, this paper proposes a cross-sensor consistency-guided dual-spectrum fusion framework, termed CG-DSF, for offshore wind turbine blade defect diagnosis and risk assessment. First, visible-light images and infrared thermal images are acquired by UAV-mounted imaging sensors, and sensor-specific branches are constructed to extract surface structural features and thermal anomaly responses. Second, visible and infrared features are aligned at the feature token level, and cross-sensor evidence is evaluated for spatial consistency, diagnostic semantic consistency, and anomaly consistency. A reliability-aware fusion strategy is then used to suppress low-quality or conflicting observations and construct a unified defect representation. Finally, a series of representative simulation case studies are carried out to comprehensively assess the overall performance and practical applicability of the constructed model. Experimental results reveal that the proposed framework possesses evident advantages in blade defect identification for offshore wind turbines, offering a feasible solution for advancing proactive and intelligent condition-based operation and maintenance of offshore wind assets in complex marine environments. Full article