Search Results (2,699)

Inventory and condition assessment of vertical traffic signs constitutes an essential activity for road safety management, as these signs serve as the primary mechanism for regulating vehicular flow and alerting drivers to prevailing road conditions. However, current inspection procedures in Peru rely on field crews that evaluate each sign manually, thereby constraining the frequency, objectivity, and scalability of the process. This paper presents SISEVIR (Sistema de Supervisión de Señales Verticales en Infraestructura Vial), a three-stage deep learning pipeline for the automated diagnosis of vertical traffic sign condition. The first stage employs YOLO26s-seg for instance segmentation of 31 sign classes, achieving a test mAP50 of 0.9305 (box) and 0.9224 (mask). The second stage classifies each detected sign into seven deterioration states using EfficientNet-B0, optimized through a five-experiment ablation study that identified progressive offline augmentation as the most effective strategy for handling a 147:1 class imbalance (macro F1 = 0.8252; pairwise McNemar’s tests with Holm–Bonferroni correction did not confirm significance at the family-wise $\alpha =0.05$ level). The third stage integrates Qwen2-VL-2B-Instruct, a vision–language model, to generate natural-language descriptions of sign condition aligned with the MTC Manual of Traffic Control Devices for Streets and Highways. A structured evaluation by two independent raters on 35 descriptions yielded a correctness rate of 93.5% among valid responses (95% CI: 79.3–98.2%, Cohen’s $\kappa =1.00$). The system was trained and validated on a proprietary dataset of 5935 images and 6412 labeled crops collected along three routes in the Ayacucho Region (246.6 km total), with an inter-rater reliability of $\kappa =0.802$ (95% CI: 0.676–0.928). SISEVIR processes vehicular video at 30.7 FPS on an NVIDIA RTX 5080 GPU and assigns each sign a level within a four-tier condition scale (Optimal through Critical) linked to specific maintenance interventions, significantly reducing the time, cost, and personnel required compared with the manual inspection method established in the MSV-2016 Road Safety Manual. Full article

Reliable online condition monitoring is essential for maintaining process stability and guaranteeing machining quality in precision grinding. Against this background, this study proposes an electrostatic induction-based measurement strategy and further performs systematic comparisons with conventional force-based monitoring methods. First, an electrostatic sensor model is established to quantitatively characterize charge transfer behaviors during grinding interactions. A three-axis experimental grinding platform is deployed to correlate electrostatic responses and mechanical force signals with critical grinding variables, including grinding speed, depth of cut, feed rate, and wheel wear severity. To achieve quantitative and objective evaluation, a unit-free Dynamic Sensitivity Change Degree (DSCD) index is introduced. Comparative results based on the DSCD index reveal that electrostatic signals exhibit better performance than traditional force-based signals in tracking grinding wheel speed variations and progressive wear evolution under the tested conditions. Meanwhile, the DSCD fluctuation in electrostatic signals remains within 1% under varying cutting depths, indicating favorable linear stability within the scope of the present experiments. Furthermore, a Generalized Conditional Variational Auto-Encoder (G-CVAE) model is developed to augment insufficient wheel wear datasets. The generated synthetic signals exhibit high fidelity and enable accurate and robust classification of grinding wheel wear states. This study verifies the existence of explicit quantitative correlations between electrostatic induction signals and key grinding parameters as well as wheel wear conditions. The proposed monitoring method can provide high-quality, reliable data support for subsequent grinding condition assessment and intelligent process decision-making. Full article

Identifying interleukin-6 (IL-6) and interleukin-13 (IL-13)-inducing peptides is important for drug discovery targeting cancer, immune disorders, and infectious diseases. However, experimental screening is costly and time-consuming. Machine learning and deep learning models have been developed that distinguish functional peptides from no-function ones, but their performance is limited by the small number of experimentally validated peptides. In this study, we propose a generative AI-driven data augmentation framework, GDA, and its prediction system, GDA-Pred, to improve the performance of state-of-the-art (SOTA) classifiers under limited data. GDA generates peptide sequences using three generative models: generative adversarial networks, diffusion models, and variational autoencoders. The framework is controlled by four hyperparameters: generative model type, sequence identity cutoff, probability threshold, and augmentation ratio. Because optimizing these hyperparameters is difficult with small datasets, we used anti-inflammatory peptide (AIP) data as a proof-of-concept to identify an effective reference hyperparameter setting. We evaluated GDA using stratified 5-fold cross-validation with cluster-based partitioning and a hold-out benchmark test. The GDA with the AIP-derived reference hyperparameter setting was then applied to SOTA classifiers to identify IL-6 and IL-13-inducing peptides as a case study. GDA-Pred consistently improved prediction performance for both cytokine-inducing peptide datasets, demonstrating the potential of generative AI to overcome data scarcity in peptide prediction. Full article

The proliferation of rumors on social media has grown exponentially in recent years. Existing approaches focus on training models on single-domain datasets, resulting in limited generalization in cross-domain scenarios. Additionally, multi-domain rumor detection faces two critical challenges: bridging semantic gaps across heterogeneous domains and addressing modality-dependent dependencies across them. To address these issues, we propose a Multi-domain Aware Network (MDAN) specifically for multimodal rumor detection. MDAN synergizes adaptive domain embeddings with a domain self-augmentation mechanism to dynamically select expert knowledge while incorporating a multi-level knowledge fusion module that hierarchically integrates domain-shared and domain-specific representations via gated attention networks. These dual knowledge optimization strategies effectively address the aforementioned challenges, significantly enhancing model robustness in multi-domain detection scenarios. Extensive experiments on two public datasets demonstrate that MDAN outperforms state-of-the-art methods in accuracy and cross-domain generalization. Full article

Driving-fatigue-induced attentional decline and response retardation are critical contributors to traffic accidents. However, stably and precisely identifying fatigue states from noisy electroencephalogram (EEG) signals remains a challenging issue in intelligent driving safety. To address the dual deficiencies of traditional methods in fatigue feature extraction precision and noise robustness, this paper innovatively constructs a collaborative recognition framework that integrates an Improved Strip Attention Mechanism (ISAM) with a Deep Residual Shrinkage Network (DRSN). The core innovations of this framework are twofold: ISAM achieves precise localization and focused enhancement of fatigue-related rhythmic bands in EEG signals via row–column separable adaptive pooling and channel-wise attention augmentation; concurrently, the DRSN module introduces an improved soft-thresholding function, which adaptively generates filtering thresholds through channel attention to effectively suppress noise and artifact interference in physiological signals. The deep fusion of these two modules forms a closed-loop optimization chain of “targeted feature reinforcement–adaptive noise suppression,” enabling the model to stably extract highly discriminative fatigue representations from complex non-stationary EEG signals. Validation on two public datasets, SEED-VIG and SADT, demonstrates that the proposed method achieves recognition accuracies of 98.86% and 97.38%, respectively, outperforming mainstream methods such as the convolutional spatial-frequency network and multi-scale convolutional neural network by 17.38% and 17.76%. These results confirm the significant advantages of the proposed dual-module collaborative architecture in precise fatigue characterization and anti-interference capability, offering a highly reliable technical solution for real-time driver mental fatigue monitoring in real-world road scenarios. Full article

Cross-domain distribution shifts severely degrade the diagnostic performance of rolling bearing models under unseen variable operating scenarios. Single-source domain generalization (SDG) builds fault diagnosis models using only single-source vibration data, which fits the practical limitations of industrial data collection. Existing contrastive learning methods adopt uniform spectral perturbations for data augmentation, which easily corrupt fault harmonic characteristics and require massive, labeled training samples. To tackle these drawbacks, this paper proposes an unsupervised contrastive learning framework named FDACL. An adaptive frequency-domain augmentation (AFA) module equipped with learnable weights is designed to separate fault-critical frequency bands from noise components. Differentiated amplitude perturbations are applied to two categories of spectral signals to generate diverse pseudo-samples while retaining intrinsic fault information. A shared encoder is trained with combined InfoNCE contrast loss and classification loss to learn domain-invariant fault representations. Validations are carried out on three datasets, namely Case Western Reserve University (CWRU), Paderborn University (PU), and the industrial CRRC Qingdao Sifang railway wheelset bearing dataset acquired from physical test benches. FDACL achieves average cross-speed diagnostic accuracies of 92.68% and 77.85% on CWRU and PU, respectively, and maintains competitive performance on the Qingdao Sifang industrial dataset. It outperforms state-of-the-art baselines by 4.23–8.71% across all SDG transfer tasks. Ablation experiments and hyperparameter analysis verify the efficacy of the AFA module and contrastive learning scheme, providing an unsupervised diagnostic approach for railway bearings under unknown working conditions. Full article

Digital-twin-enabled human–machine collaboration (HMC) has increasingly been proposed as a system-level approach for connecting human operators, robots, sensors, artificial intelligence modules, and manufacturing resources. However, the literature varies substantially in what is called a digital twin, how physical and virtual models are coupled, whether models are updated from physical data, and how far systems have progressed beyond simulation or controlled laboratory demonstrations. This structured integrative review examines the conditions under which a digital twin can function as an integration layer for HMC in sustainable smart manufacturing, rather than assuming that such integration is already established industrial practice. The literature corpus was assembled through searches of the Web of Science Core Collection, Scopus, and IEEE Xplore, complemented by Google Scholar-based citation tracking and backward and forward citation tracing. The core search focused on studies published from 1 January 2020 to 5 August 2026, while earlier seminal studies were retained to support definitions and historical context. Studies were screened using explicit criteria for manufacturing relevance, physical–virtual coupling, state synchronization or model updating, feedback capability, and validation setting, and were critically coded by model type, integration mechanism, deployment maturity, and sustainability evidence. The review compares multimodal perception and human-state modeling, intention understanding and augmented interaction, task allocation and shared planning, digital-twin architectures, adaptive control and safety verification, and human–AI decision-making. The evidence indicates that digital twins are promising as coordination and verification layers, but many reported systems remain conceptual, simulation-based, or limited to controlled physical prototypes. Key barriers include model fidelity, online model updating, real-time synchronization, cross-platform interoperability, safety assurance, human-data governance, and the limited availability of directly measured sustainability outcomes. Future work should prioritize validated hybrid models, traceable model-update mechanisms, staged virtual-to-physical deployment, interoperable data contracts, and longitudinal evaluation of technical, human, economic, and environmental performance. Full article

Earth Observation (EO) analysis increasingly relies on large and heterogeneous satellite datasets, yet developing EO workflows often requires specialized expertise in data selection, geospatial programming, and cloud-based processing. Recent advances in Large Language Models (LLMs) offer new opportunities for natural-language interaction with EO systems, although challenges related to transparency, reproducibility, and domain-specific reasoning remain. This study presents SeaScope, an explainable AI framework that integrates LLMs, Retrieval-Augmented Generation (RAG), scientific knowledge retrieval, and Google Earth Engine (GEE) to transform natural-language requests into transparent and executable EO workflows. The framework combines knowledge retrieval, code generation, cloud execution, provenance tracking, and interactive visualization within a unified environment. A pilot implementation is demonstrated through maritime and coastal monitoring applications, including oil spill detection, vessel monitoring, water quality assessment, floating debris detection, and air quality analysis. Multiple state-of-the-art LLMs are evaluated under both RAG and non-RAG configurations using representative EO case studies. The results indicate substantial differences among model families and show that retrieval augmentation can significantly improve workflow generation quality and reliability for capable models, while providing more limited benefits for smaller models. The proposed framework demonstrates the potential of explainable AI agents to support transparent, reproducible, and scalable EO analysis. Full article

Accurately forecasting waypoint-level tropical cyclone (TC) intensity, defined as the local wind speed at specific maritime route waypoints under TC influence, is crucial for navigation safety and voyage planning. Conventional studies mainly focus on the central intensity of TC systems and underutilize the complementary value of multimodal meteorological data with inconsistent sampling intervals. To address these challenges, this study proposes a multimodal-augmented conditional diffusion model (MADiff) for waypoint-level TC intensity prediction. To exploit the potential of multimodal inputs, we first design a temporal-adaptive dynamic convolution module (TDConv) to capture multi-timescale features, mitigating multimodal sampling discrepancies without rigid temporal alignment. Second, we develop a discriminative cross-fusion module (DisCF) to aggregate multi-timescale features across diverse modalities, quantifying multimodal heterogeneity and integrating valuable modality-specific features while suppressing noise interference. Fused features are fed into a diffusion model with physics-informed regularization to generate final intensity forecasts. Extensive experiments on four Western North Pacific datasets show that MADiff achieves average MAE and RMSE values of 2.08 kt and 2.37 kt, respectively, for 12 h intensity forecasting. Compared with the state-of-the-art baseline (TC-Clouds-DP), MADiff yields substantial performance improvements, reducing MAE by 16.3% and RMSE by 10.6% on average. This study provides an effective framework for fine-grained TC intensity forecasting, offering valuable insights for extreme marine weather early warning and intelligent navigation decision-making. Full article

We present ESNformer, a hybrid architecture that couples an Echo State Network (ESN) reservoir with a Transformer encoder for classification of structured, multi-indicator assessment data: a fixed-order vector of complementary indicators per assessment instance rather than a repeated-measures time series. The reservoir acts as a fixed nonlinear feature map over the indicator vector, while self-attention, made position-aware over the fixed column order, learns how each indicator’s evidence contributes to the final decision, so the two components, together, capture local, indicator-level detail and global, cross-indicator interactions within a single, end-to-end trainable model. Interpretability is treated as a first-class design requirement rather than an afterthought: the architecture is paired with an explainability layer combining SHAP feature attribution (reported both globally and per class), the model’s own attention weights, a deletion/insertion faithfulness test that quantitatively verifies which inputs the model actually relies on, and counterfactual maps that translate a prediction into an actionable, inspectable recommendation. We evaluate the architecture on a concrete case study, classifying Braille-literacy instructional recommendations from 15 pedagogical indicators grouped into three categories (Mangold’s, ABKL, and Progresar), using a benchmark of 900 real assessment instances (630 used, together with a class-conditional augmentation procedure, to build a 2100-instance training set) with validation and test partitions (135 instances each) kept exclusively real. On this benchmark, the tuned model reached 85.33% accuracy, 85.90% macro-precision, 85.33% macro-recall, an F1 score of 85.25%, and an AUC of 0.95 on the real test set. SHAP attribution, attention weights, and the faithfulness test converge on the same two dominant indicators (response time and error count): removing them alone collapses accuracy to chance, while retaining only them recovers most of the model’s accuracy. We report this transparently alongside a comparison against ESN-only, Transformer-only, and tabular baselines (logistic regression, decision tree, random forest, XGBoost, and an MLP) on the same data and discuss what the hybrid architecture and its explainability pipeline add beyond what the two dominant indicators already explain and how the approach generalizes to other tabular and mixed-granularity assessment settings that require both predictive accuracy and a verifiable account of what drove each decision. Full article

Industry 4.0 has produced one of the fastest-growing bodies of engineering and management research; much of this output remains siloed by technology domain. This study addresses that fragmentation through a structured critical review (a review-of-reviews), synthesising 70 peer-reviewed review articles and foundational sources (2003–2026) spanning 14 technology domains. The review introduces the I4.0-STS framework, an original four-layer structure organising evidence across physical, cyber, cognitive, and socio-organisational dimensions. Five principal findings emerge. The physical and cyber layers show consistent evidence of maturity. Industry-reported lighthouse IIoT deployments show 20–30% energy and up to 39% lead-time reductions. AI-driven predictive maintenance shows 30–50% unplanned-downtime reductions. The cognitive layer (LLM-augmented digital twins and generative AI interfaces) is technically feasible but outpaces its governance frameworks. Cybersecurity remains insufficiently governed, with documented ransomware incidents in manufacturing OT environments underscoring the risks of OT–IT convergence. SME adoption and developing-economy manufacturing transformation remain comparatively under-addressed. Finally, 12 research gaps are assessed as of June 2026, five rated Open, with future research directions proposed for each, framed against the emerging Industry 5.0 agenda. All findings are second-order interpretations from the source reviews, and their limitations are stated explicitly. Full article

This paper presents a robust attitude-tracking control architecture for rigid spacecraft subject to model mismatches and external disturbances. Quaternions are utilized for attitude representation to prevent the gimbal lock associated with Euler angles. While conventional nonlinear dynamic inversion (NDI) relies on Newtonian mechanics and input–output linearization—which inadvertently generates internal zero dynamics and encounters severe control derivative discontinuities at the $q_0=0$ singularity—this study proposes a novel NDI framework derived strictly from Udwadia’s Lagrangian formulation. This approach realizes an exact input-state linearization directly on the 6-degree-of-freedom active holonomic constraint manifold, completely eliminating internal zero dynamics and mathematical singularities. To ensure robustness against physical uncertainties, the singularity-free NDI is augmented with a nonlinear disturbance observer (DOBC) and an outer-loop linear quadratic (LQ) tracking controller. A rigorous composite Lyapunov stability analysis is conducted for the complete closed-loop architecture. The analysis formally guarantees that both the isolated disturbance estimation error and the fully interconnected dual-loop NDI-DOBC system are Uniformly Ultimately Bounded (UUB), even in the presence of realistic, time-varying disturbances with non-vanishing derivatives ($\dot{\mathit{d}}\ne \mathbf{0}$). Comprehensive numerical simulations, parameterized by a physical spherical air-bearing testbed subject to state-dependent gravitational imbalance torques, validate the architecture’s exceptional tracking precision, smooth transient response, and robust disturbance rejection. Full article

Maritime object detection remains challenging because of complex sea-surface backgrounds, adverse illumination conditions, and severe class imbalance, especially when safety-critical targets such as search-and-rescue vessels are sparsely represented. To address these challenges, we propose a cascaded hybrid DINOv2-YOLOv8n detection framework for maritime scenes. Rather than relying only on supervised learning from raw RGB images, the proposed method introduces semantic priors from a frozen DINOv2 encoder and projects them into a compact representation for a YOLOv8n-based detector. To improve robustness under diverse maritime conditions, the framework uses a sea-state-aware online augmentation strategy and is trained with the standard YOLO detection objective. Experiments on the Maritime Target Data Sharing Project (MTDSP) dataset show that the proposed framework achieves strong detection performance. Specifically, it obtains an overall mAP@0.5 of 89.4% and a precision of 100% on the test set. For sparse search-and-rescue vessels and rigid offshore structures, it achieves mAP@0.5 scores of 99.5% and 96.3%, respectively. These results indicate that combining foundation-model semantic priors with a lightweight detector can improve the reliability of maritime object detection under complex sea-surface conditions. Full article

Pansharpening aims to fuse low-resolution multispectral (LRMS) and panchromatic (PAN) images to generate high-resolution multispectral (HRMS) imagery. However, models trained on data from a specific sensor often generalize poorly to unseen sensors, resulting in significant performance degradation. Existing solutions can be categorized into full model retraining and zero-shot adaptation. The former requires substantial computational resources and labeled target-domain data, making it impractical for rapid deployment. The latter avoids retraining but typically suffers from considerable performance degradation under cross-sensor domain shifts. To address the issues, we propose a Confidence-Aware Selective Test-time Adaptation (CSTTA) framework for cross-sensor pansharpening. Specifically, test-time augmentation is employed to estimate prediction uncertainty and construct confidence-aware weighting maps, which guide adaptation toward more reliable regions. In addition, a multi-patch sensitivity-based parameter selection strategy is introduced to update only a small subset of sensor-sensitive parameters, thereby reducing optimization cost while maintaining adaptation effectiveness. Spatial, spectral, and output-consistency constraints are further incorporated to stabilize the adaptation process and balance structural preservation with spectral fidelity. Extensive experiments on multiple cross-sensor pansharpening benchmarks demonstrate that CSTTA consistently improves the performance of various backbone networks and achieves state-of-the-art results compared with existing transfer methods. Full article

Digital ad fraud continuously evolves through device spoofing, behavior simulation, and coordinated traffic attacks, making the detection of sophisticated fraudulent activities a persistent challenge. In real-world settings, only a small fraction of fraudulent traffic receives reliable labels, while vast amounts of unlabeled data may still contain latent fraud. This severely limits the generalization ability of conventional supervised learning models. To address this issue, we propose DiffVC—a novel contrastive learning framework tailored for ad fraud detection under the Positive-Unlabeled (PU) learning paradigm. DiffVC employs a multi-granularity diffusion augmentation strategy that builds upon a denoising diffusion probabilistic model to generate semantically augmented samples at three granularities: weak, medium, and strong. This strategy expands the latent fraud feature space while preserving diverse semantic information. We further introduce a diffusion-distance-calibrated similarity that dynamically adjusts constraints between augmented samples based on their diffusion distance, thereby improving the discrimination of unlabeled fraud. In addition, we design a dual-gated residual Transformer encoder that adaptively captures high-order feature interactions via gated residual connections and a channel recalibration mechanism. Experimental results on multiple real-world ad fraud detection datasets demonstrate that DiffVC consistently outperforms state-of-the-art methods and achieves stronger generalization. Our results confirm the effectiveness and practical applicability of DiffVC for ad fraud detection under label-scarce conditions. Full article