Search Results (50,161)

Precise identification and monitoring of small agricultural water bodies are essential for sustainable water resources management in arid and semi-arid regions, where even limited water losses can significantly affect agricultural productivity and local water security. However, the accurate detection of small ponds remains a major challenge in remote sensing, to address this challenge, this study proposes an integrated three-step framework that combines high-resolution remote sensing imagery, machine and deep learning techniques, and hydrological analysis to identify agricultural ponds and quantify their evaporation losses in Bastam, Iran. In the first step, a dedicated annotated dataset comprising 1061 RGB satellite images, each with a spatial size of 256 × 256 pixels and a ground resolution of 0.5 m, was developed for model training and evaluation. Using this dataset, three deep learning models BiSeNet, UNet3+, and SegNet and four traditional supervised classifiers Maximum Likelihood, Neural Network, Mahalanobis Distance, and Minimum Distance were implemented and compared for pond detection. The results demonstrated that deep learning models consistently outperformed conventional classifiers in delineating small agricultural ponds. Among all evaluated methods, BiSeNet achieved the highest segmentation performance, with an IoU of 82.08%, an F1-score of 90.15%, a precision of 91.86%, and a recall of 88.50%. Among the conventional classifiers, Maximum Likelihood combined with a 5 × 5 spatial kernel produced the best performance, achieving an IoU of 76.90%, an F1-score of 86.93%, a precision of 90.87%, and a recall of 83.32%, whereas simpler classifiers such as Minimum Distance showed only marginal improvements after kernelization. In the final step, the detected ponds were used to estimate evaporation losses through the Meyer method. The hydrological analysis revealed a clear periodic pattern in evaporation and a cumulative water loss of 388,636.7 m³ over a nine-month period, highlighting the considerable impact of evaporation on the efficiency of small agricultural water storage systems in dry environments. Based on these findings, practical mitigation strategies, including evaporation-reducing chemical surface films and floating covers, are discussed as potential options for reducing water loss. Overall, the proposed framework demonstrates the clear advantage of deep learning for the accurate identification of small agricultural ponds and provides an integrated methodological basis for monitoring water bodies and evaluating associated evaporation losses. The study offers a practical and transferable approach for supporting agricultural water management and improving water-use efficiency in arid and semi-arid regions. Full article

Autonomous driving is undergoing a major architectural transition from modular, rule-based pipelines toward learning-based and increasingly end-to-end (E2E) driving systems. This paper examines this transition by tracing the evolution from classical sense–perceive–plan–control architectures to large driving models (LDMs) that integrate perception, prediction, planning, and control within unified learning frameworks. We review recent academic and industrial developments, including Tesla’s Full Self-Driving (FSD) V12–V14, Rivian’s Unified Intelligence platform, NVIDIA Cosmos, and emerging robotaxi deployments, with particular emphasis on Tesla FSD because it represents one of the most widely deployed supervised E2E driving systems currently available to consumers. The analysis focuses on architectural design, deployment pathways, safety challenges, and industry implications. Particular attention is given to the emerging category of supervised E2E driving, often described as FSD (Supervised) or L2++, in which the vehicle performs a substantial portion of the Dynamic Driving Task (DDT) while the human driver remains responsible for supervision and fallback intervention. We discuss the technical opportunities of these systems, including their potential to learn from large-scale fleet data and improve performance in long-tail driving scenarios, while also examining unresolved challenges related to validation, transparency, human–machine interaction, driver attention, liability, and regulatory assessment. The paper further argues that combining vision with range sensing can support continuous training and validation of camera-based depth and scene-understanding models. Finally, we consider how the architectural principles emerging in autonomous driving may extend to broader embodied AI systems, including humanoid robotics and other safety-critical autonomous platforms. Full article

This paper proposes a novel multi-frequency extended Deep Learning (DL) model for electromagnetic image reconstruction under Transverse Magnetic (TM) wave incidence in layered media, inspired by conventional microwave imaging techniques that combine nonlinear inversion algorithms with neural networks to improve reconstruction performance. The proposed framework adopts a two-stage neural network architecture. In the first stage, a Deep Residual Convolutional Neural Network (DRCNN) is employed to extrapolate multi-frequency scattered fields from single-frequency input data, thereby enriching the frequency-dependent scattering information available for reconstruction. Subsequently, the extrapolated multi-frequency scattered fields are fed into a Deep Convolutional Encoder–Decoder (DCED) network to reconstruct an accurate dielectric constant distribution within the imaging domain. To validate the effectiveness of the proposed approach, two representative comparison methods are considered: (1) a hybrid framework combining the Back-Propagation Scheme (BPS) with a Convolutional Neural Network (CNN), and (2) a framework integrating the Dominant Current Scheme (DCS) with a CNN. In both approaches, conventional inversion algorithms are first utilized to generate coarse initial reconstructions, which are subsequently refined by the neural network. Numerical simulations and experimental results show that the proposed multi-frequency extension model achieves lower reconstruction error and higher structural similarity than the reference methods. These results confirm the effectiveness and potential of the proposed framework for advanced electromagnetic imaging applications. Full article

Machine learning has created new opportunities for gas adsorption research using nanoporous materials, but the field’s evolution remains insufficiently quantified. This study retrieved literature from the Web of Science Core Collection for 2010–2026 and retained 730 valid records from 1581 initial publications after screening. VOSviewer, CiteSpace, and R were used to analyze publication growth, collaboration networks, journal sources, and thematic evolution. Results show that annual output remained generally below 20 before 2019, then increased rapidly and reached approximately 280 publications in 2025, indicating accelerated integration of machine learning with adsorption simulation, material screening, and performance evaluation. The source distribution broadened from a limited set of chemistry and engineering journals to diverse venues, with recent high publication weights in Chemical Engineering Journal, Separation and Purification Technology, ACS Applied Materials & Interfaces, Microporous and Mesoporous Materials, and Journal of Materials Chemistry A. Collaboration analysis identified 10 compact author clusters, including groups associated with Randall Q. Snurr, Seda Keskin, Zhiwei Qiao, Qingyuan Yang, and Chongli Zhong, whereas the weak bridging links among clusters indicate that cross-community collaboration remains limited. Country and institutional analyses show that China, the United States, Canada, Iran, India, South Korea, and the United Kingdom are leading contributors, with Guangzhou University, Koç University, Northwestern University, the Chinese Academy of Sciences, Beijing University of Chemical Technology, and the United States Department of Energy occupying prominent positions. Keyword evolution reveals a shift from adsorption behavior and porous adsorbents toward data-guided material selection, high-throughput screening, deep learning, Bayesian optimization, and performance optimization, offering guidance for data-driven adsorbent discovery. Full article

Intelligent fault diagnosis of rotating machinery is essential for ensuring the safety and reliability of industrial systems. While hypergraph neural networks (HGNNs) have recently shown promise for modeling high-order dependencies beyond pairwise graph methods, most existing variants operate in Euclidean space, which is not explicitly aligned with hierarchical fault-response structure (root cause to fault mode to observed response). To address this limitation, we propose Hyperbolic Hypergraph Neural Network (H²GNN), a framework that integrates hyperbolic geometry with hypergraph neural networks for fault diagnosis. Specifically, H²GNN constructs fault-response-aware hyperedges over diagnostic views of vibration signals and performs message passing in the Poincaré ball model, a Riemannian manifold of constant negative curvature commonly used for hierarchical representation learning. We introduce Poincaré hyperedge aggregation via an iterative Fréchet-mean solver, a learnable curvature parameter for adaptive manifold fitting, and a tangent-space classification head. Experiments are conducted on two public benchmarks, namely the Case Western Reserve University (CWRU) bearing dataset and the Machinery Failure Prevention Technology (MFPT) bearing dataset, and report mean accuracies of 99.87% and 99.75%, respectively, outperforming six competing methods, including CNN, GCN, HGNN, dynamic-HGNN, contrastive-HGNN, and spatial-temporal HGNN. Ablation studies indicate that hyperbolic geometry and the adaptive curvature mechanism both contribute to the observed performance gain. Full article

Emergency stroke assessment using non-contrast computed tomography (NCCT) remains challenging because early ischemic signs may be subtle and anatomical variability across cranial scans can affect model performance. This study presents AGTS-Net (Anatomically Guided Two-Stage Network), an explainable deep learning framework designed to support stroke classification from complete NCCT studies. The framework was developed using a proprietary cohort of 99 patients, comprising 3440 NCCT slices annotated as hemorrhagic stroke, posterior ischemic stroke, anterior ischemic stroke, or no stroke. For each region-specific diagnostic classifier, data were split using the same scheme: 80% for training and 20% for testing, followed by a 20% validation split from the training subset. AGTS-Net first assigns each slice to a predefined anatomical region and then applies a region-specific classifier. Internal evaluation showed 0.99 anatomical-routing accuracy, and transfer learning experiments confirmed that anatomical regionalization improved classification over global models. A public Kaggle dataset of 7012 NCCT images was used only for external cross-domain evaluation of the anatomical pre-classifier, as compatible territorial diagnostic labels were unavailable. Grad-CAM maps visualized regions contributing to predictions. AGTS-Net is intended as a decision-support tool, providing actionable slice-level predictions and visual guidance during urgent stroke assessment. Full article

Network intrusion detection systems (NIDSs) deployed in dynamic environments face concept drift from evolving attacks and traffic patterns, causing model reliability to degrade over time. Continual learning (CL) offers an adaptive solution, yet many methods misalign drift detection, memory updating, and optimization: drift is often judged with low-dimensional statistics, while adaptation occurs in representation space, limiting consistency under buffer constraints. To address concept drift in non-stationary network traffic and catastrophic forgetting during online intrusion detection updates, we propose a continual-learning framework built upon SSF that combines feature-space Gaussian-kernel Maximum Mean Discrepancy (MMD) drift detection with gradient-matching coresets for memory admission. The proposed framework retains strategic forgetting and steady-state distillation while replacing low-dimensional drift tests with feature-space MMD and mask-based selection with gradient-matching coresets, thereby improving incremental updates under a limited memory budget. On NSL-KDD and UNSW-NB15 under a unified multi-seed streaming protocol, the proposed method improves detection performance and knowledge retention. Experimental results demonstrate that gradient-matching coreset selection is the primary contributor to the observed performance improvements, while the effectiveness of MMD-based drift scheduling and strategic forgetting depends on the underlying data distribution and drift-trigger threshold. The proposed framework employs batch-level MMD scheduling to coordinate memory admission and online optimization, providing a practical path toward robust continual intrusion detection. Full article

High-speed maneuvering flight vehicles operating in the subsonic-to-transonic regime (250–500 m/s) pose severe challenges to defense interception systems due to their rapid and unpredictable maneuvering behaviors. Accurate short-term trajectory prediction is essential for effective terminal-phase interception guidance. This paper proposes DSGF-Net (Dual-Stream Gated Fusion Network), a hybrid deep learning architecture for 3D trajectory prediction that simultaneously exploits frequency-domain and temporal-domain information through independent parallel streams. DSGF-Net employs two Temporal Convolutional Networks (TCNs) as parallel encoders: a frequency stream processes Wavelet Packet Decomposition (WPD) features (24-dimensional, db4 wavelet, level-3 decomposition), and a temporal stream processes raw 3D coordinates. An adaptive sigmoid gating module dynamically fuses the two independently encoded streams at each time step and feature dimension, followed by an LSTM sequence learner and a single-step fully connected decoder. Experiments on a simulated dataset covering five representative maneuvering modes (cruise, dive, climb, serpentine, composite) reveal a three-level performance hierarchy. First, incorporating raw 3D coordinates alongside WPD features substantially improves clean-data accuracy over WPD-only baselines: DSGF-Net achieves ADE = 3.476 ± 0.010 m (5 random seeds) versus TCN-LSTM (WPD-only) at 3.938 ± 0.103 m (11.7% improvement). Second, a single-stream concatenation baseline (Concat-TCNLSTM) using identical inputs achieves comparable clean-data accuracy (3.333 ± 0.009 m), confirming that input information—rather than fusion mechanism—drives clean-data gains. Third, and most critically, DSGF-Net’s independently encoded dual-stream architecture enables adaptive suppression of degraded sensor inputs: under multi-sensor noise (complementary radar/GPS profiles), DSGF-Net achieves ADE = 13.91 m versus TCN-LSTM’s 21.32 m (34.9% advantage)—a substantially larger margin than on clean data—a capability structurally unavailable to concatenation-based models. With 300K parameters and a 4.96 ms inference time on an A100 GPU, DSGF-Net meets real-time terminal interception requirements (<10 ms). Full article

The architecture, engineering, and construction (AEC) sector operates increasingly under persistent macroeconomic and supply chain volatility, requiring project organizations to simultaneously achieve delivery efficiency and design innovation. While leadership theory attributes this ambidextrous capability to “integrated leaders” who can enact competing behavioral repertoires in parallel, this assumption remains untested in highly volatile, high-pressure project environments. This study investigates the boundary conditions of integrated leadership theory and examines structurally enabled temporal sequencing as an alternative explanation for organizational ambidexterity. Using an explanatory sequential mixed-methods design in the Turkish AEC sector (Phase 1 Survey: N = 278; Phase 2 Interviews: N = 32), we examined how building information modeling (BIM)-enabled digital governance capabilities and organizational tenure predict organizational ambidexterity, including a moderation analysis examining whether industry volatility conditions the effect of integrated leadership. Phase 1 ordinary least squares (OLS) regression and exploratory machine learning stress testing revealed that self-reported integrated leadership profiles had limited predictive utility for ambidexterity (β = 0.035, p = 0.751) and that industry volatility did not significantly moderate this relationship. Conversely, digital governance capability (β = 0.462, p < 0.001) and organizational tenure (β = 0.205, p < 0.01) emerged as consistent structural predictors even after controlling for firm size, industry volatility, project type, and project duration. Phase 2 qualitative findings help explain this predictive pattern by suggesting that, under persistent volatility, attempts to enact simultaneous behavioral integration may generate cognitive overload and execution friction. Instead, participants described project-level ambidexterity as emerging through temporal sequencing and an alternating focus between exploration and exploitation, supported by digital coordination systems that externalize workflows and relational networks associated with longer organizational tenure. Taken together, the findings suggest structurally enabled temporal sequencing as a theoretically derived, qualitatively supported, and interpretive explanation of how organizational ambidexterity may emerge under persistent volatility, rather than as an empirically established mechanism. Full article

Water leakage in urban water distribution networks (WDNs) poses significant challenges for sustainable resource management and infrastructure reliability. Traditional detection methods are often reactive and difficult to scale in modern sensor-rich environments. This paper proposes a hybrid data-driven framework for early leak detection that integrates physics-informed simulation with machine learning and explainable analytics. A region-aware EPANET-style simulator is developed to generate realistic hydraulic data under varying demand patterns, environmental conditions, and pressure-dependent leak scenarios. To enhance generalizability, the synthetic dataset is combined with a BATADAL-inspired benchmark, enabling both in-domain and cross-domain evaluation. A feature engineering pipeline is introduced to capture temporal, spatial, and hydraulic relationships, expanding raw sensor signals into a high-dimensional representation. Six machine learning models, including Random Forest, Gradient Boosting, Support Vector Machine, Logistic Regression, Isolation Forest, and a PCA-Based Autoencoder, are systematically evaluated under constrained false-positive requirements. The results show that tree-based ensemble models achieve strong detection performance while maintaining low false-alarm rates (FPR ≤ 0.05). Importantly, cross-domain experiments demonstrate that models trained on simulated data retain competitive performance when applied to benchmark datasets, indicating robust transferability. Finally, explainability analysis reveals that pressure-based temporal statistics and spatial gradients are key indicators of leakage, providing interpretable insights for system monitoring. The proposed framework offers a scalable and generalizable approach for intelligent leak detection in modern water distribution systems. Full article

Autonomous Underwater Vehicles (AUVs) face significant challenges in path-following control due to strong environmental disturbances and model uncertainties. To address these issues, this paper proposes a model-free deep reinforcement learning framework, named ILLT (Improved LOS-LSTM-TD3), which integrates an integral line-of-sight (LOS) guidance law with the twin delayed deep deterministic policy gradient (TD3) algorithm. The framework treats the LOS look-ahead distance as a learnable optimization variable and incorporates an LSTM network to capture temporal motion dependencies. A progressive unfreezing transfer learning strategy, combined with attention-based feature–current fusion, is designed to enhance domain adaptation under varying ocean currents. Simulation results demonstrate that ILLT reduces the average cross-track error by 48.5% compared to the baseline ILT algorithm and by 66.4% compared to traditional PID control, while achieving significantly faster convergence in target domains. Physical experiments in tank and lake environments further validate the algorithm’s feasibility and robustness, with tracking errors approaching simulation results under moderate current conditions. These findings confirm the effectiveness of the proposed framework for underactuated AUV path-following tasks. Full article

Background: Oral potentially malignant disorders (OPMDs) have a complex, but not consistently consistent, risk of transformation to oral squamous cell carcinoma (OSCC). The histopathologic grading of dysplasia is the traditional means of prognosis although there is considerable inter-observer variability and the tool has poor predictive value. Thus, image-based AI-based methods such as computational pathomics (CP), clinical-photograph deep learning (CDL), and optical or spectroscopic image analysis via machine learning (ML) or deep learning (DL) have been proposed as potential non-invasive risk stratification methods. Aim of the review is to identify the current state of the evidence for AI applications in the field of image-based diagnosis and treatment of OPMDs, their methodological characteristics and predictive accuracy, and priorities for future research. Methods: the scoping review was conducted using Joanna Briggs Institute methodology and PRISMA-ScR guidelines. The PubMed/MEDLINE, Scopus, Web of Science and Embase databases were searched between January 2018 and March 2026. Inclusion criteria were studies that used quantitative image analysis, ML or DL to diagnose, prognosticate, or stratify risk of OPMD in OPMD cohorts. A pre-piloted form was used to extract data which were then synthesized descriptively. Results: the 423 records identified included 24 (16 primary image-based studies and 8 contextual reviews) that met the inclusion criteria. The majority of studies were retrospective (15/16), and computational pathomics (n = 10), clinical-photograph deep learning and optical/spectroscopic imaging (n = 4) were emphasized. There was no study that used the conventional radiologic radiomics (CT, MRI, CBCT, or PET/CT) in OPMD cohorts. Overall, predictive performance was good, with AUROC values ranging from 0.73 to 0.96 for the detection of malignant transformation, 0.94 to 0.97 for OPMD versus OSCC discrimination, and 0.90 to 0.97 for dysplasia grading. Only 44% contained external validation, only 6% were prospective, and only limited adherence was made to IBSI, TRIPOD-AI and CLAIM standards. Conclusions: AI-based image recognition for OPMDs shows good predictive performance, and it has yet to be developed to a mature stage of the process. Future multicentre study, image standardisation, external validation and better reporting standards are needed. Remarkably, radiologic radiomics is an important and unexplored research gap in OPMD risk prediction. Full article

Urban traffic congestion prediction is an important problem in smart city sensing and intelligent traffic governance. Existing methods mostly rely on single-source traffic flow sensing data or static road topology, making it difficult to sufficiently characterize the dynamic congestion propagation process driven by multisource sensing information, such as traffic flow, vehicle trajectories, road images, public transportation, meteorological conditions, and sudden events. To address this issue, a spatiotemporal causal graph learning framework based on multisource urban sensing data is proposed for urban traffic state prediction, congestion identification, and explainable early warning. In this framework, traffic flow detector data, GPS trajectories, roadside camera data, public transportation data, weather data, and event records are first fused through a multisource urban sensing data collaborative encoding module, and the influence of low-quality or missing sensing modalities is suppressed using a reliability-aware attention mechanism. Subsequently, time-varying causal propagation relationships among road segments are adaptively learned from historical traffic states, road topology, and external disturbances through a dynamic spatiotemporal causal graph learning module. Finally, spatial diffusion and temporal evolution are jointly modeled by a causality-explanation-driven congestion prediction module, and key congestion sources, propagation paths, and inducing factors are outputs. Experimental results based on multisource traffic sensing data from the main urban area of Hangzhou show that the proposed method achieves MAE values of $3.21$, $3.79$, and $4.48$ in 15-min, 30-min, and 60-min traffic state prediction tasks, respectively, outperforming ARIMA, XGBoost, LSTM, Transformer, STGCN, Graph WaveNet, GMAN, Multimodal Transformer, and the Causal Temporal Graph Network. In the ablation study, the complete model achieves an Accuracy of $0.914$, a Precision of $0.902$, a Recall of $0.889$, an F1 of $0.895$, and an AUC of $0.956$. For congestion identification and early warning under complex scenarios, F1 values of $0.927$, $0.904$, and $0.893$ are achieved under peak-hour, rainy-weather, and traffic-event scenarios, respectively; the corresponding AUC values reach $0.966$, $0.957$, and $0.948$; and the false alarm rate (FAR) values are reduced to $0.061$, $0.072$, and $0.081$. The results indicate that the proposed method can effectively improve traffic state prediction accuracy, congestion early warning reliability, and model interpretability under multisource urban sensing conditions, thereby providing an effective technical pathway for AI-driven intelligent traffic sensing. Full article

Background: Cervical cancer remains one of the leading causes of cancer-related morbidity among women worldwide. The Papanicolaou (Pap) smear is widely used for early detection; however, its manual interpretation is time-consuming, requires substantial expertise, and is often affected by inter-observer variability, particularly in cases with dense and overlapping cells. Methods: This study developed an optimized deep learning framework for cervical cell instance segmentation, specifically targeting the separation of overlapping cells in Pap smear images. The proposed framework was based on Mask R-CNN with a ResNet-50 backbone and Feature Pyramid Network. A public development dataset of 460 cervical smear images was used for model development and internal evaluation, while an independent 210-image dataset collected from King Abdullah University Hospital was reserved as an external held-out clinical assessment set. To improve small-cell detection and mask refinement, the Region Proposal Network was adapted using biologically informed anchor scales (16, 32, 64, 128, 256), followed by soft calibration-based post-processing. Results: The proposed framework achieved an AP50 of 89.90% and a Mask IoU of 90.44%. Small-cell performance, measured using APs under the COCO small-object convention, reached 22.80%, while APm and APl reached 68.45% and 81.42%, respectively. Clinical cell-count evaluation on the independent KAUH held-out set was performed using the mean count of five physicians as the reference standard and achieved an overall clinical detection accuracy of 93.00%. Conclusions: The optimized Mask R-CNN framework improved the detection and separation of overlapping cervical cells in Pap smear images and may serve as a supportive tool for cytopathology workflows. The results suggest that biologically informed anchor optimization and soft calibration can improve cell-level instance segmentation, particularly for small and overlapping cells. Further validation on larger multi-center datasets remains necessary before routine clinical deployment. Full article

The accurate and nondestructive quantification of subgrade moisture content is a core demand for highway construction quality control and long-term performance maintenance. In order to study the response relationship between subgrade moisture content and ground-penetrating radar (GPR) signal eigenvalues, this study constructs the volumetric moisture content–dielectric constant relationship of Guangxi high-plasticity clay and carries out gprMax forward numerical simulations. Fourteen radar signal eigenvalues are extracted from preprocessed signals via time-domain waveform analysis, Hilbert transform analysis, and power spectral density analysis. Seven key eigenvalues are screened out through Pearson correlation coefficient-based dimensionality reduction. Three machine learning algorithms—artificial neural network (ANN), random forest (RF), and light gradient boosting machine (LightGBM)—are adopted to optimize the subgrade moisture-content inversion model, which is finally validated through indoor model box tests and field subgrade tests. The results show that: (1) The linear fitting formula is the most suitable for describing the volumetric moisture content–dielectric constant relationship of Guangxi clay, with a coefficient of determination (R²) of 0.979 and a mean absolute error (MAE) of 0.31. (2) The feature matrix after dimensionality reduction effectively alleviates the degradation of model generalization ability and interpretability. (3) The LightGBM model achieves the highest prediction accuracy for clay volumetric moisture content, with an R² of 0.99926 and an MAE of 0.172%. (4) For gravimetric moisture-content inversion, the maximum relative error is 1.6% in the indoor model box test and 1.7% in the field test, both within the 2% tolerance of engineering requirements. This study verifies the feasibility of the proposed subgrade moisture-content identification method based on GPR signal eigenvalues. The proposed method provides an efficient technical path for the large-area and nondestructive detection of subgrade moisture and has promising application prospects in subgrade construction quality acceptance, daily maintenance monitoring and hidden disease early warning. Full article