Search Results (2,195)

Distributed acoustic sensing (DAS) is increasingly used for third-party intrusion (TPI) detection in oil and gas pipeline monitoring, but labeled DAS data are often scarce, leading to overfitting, poor generalization, and increased false alarms and missed detections. Conventional data augmentation, GAN-based synthesis, and transfer learning may generate physically implausible samples or fail to cover the event feature space. To address this, we propose a physics-guided large language model (LLM) prompt-engineering framework for DAS data augmentation and pipeline intrusion detection. The framework establishes a physically grounded feature-indicator framework for DAS disturbance-event classification by mapping primary event mechanisms to measurable signal indicators, and then uses a standardized four-module prompt template to guide LLM-based synthesis-script generation. A two-stage iterative verification procedure is further introduced to constrain the generated samples in terms of physical-mechanism compliance and feature-parameter consistency. Synthetic data are combined with real data to train a lightweight PatchTransformer model for TPI detection, while an additional CNN is used to assess cross-architecture applicability. Using the public DAS1K benchmark with five-fold stratified cross-validation and a univariate controlled experiment (0–800 synthetic samples per category), the results show that the use of synthetic data improves detection performance overall. The configuration with 600 synthetic samples per category achieves 92.27% accuracy and 92.38% macro-F1, outperforming the conventional augmentation baseline by 4.74 and 4.86 percentage points, respectively. An additional CNN experiment also showed consistent performance gains across the tested augmentation settings, indicating that the benefit of the proposed synthetic data was not restricted to the PatchTransformer architecture. These findings indicate that LLM-assisted data augmentation can effectively improve the generalization of DAS-based pipeline intrusion detection when field-labeled samples are scarce. Full article

NASA’s Fire Information for Resource Management System (FIRMS) provides near-real-time thermal anomaly detections from VIIRS, but not all detections correspond to wildfire incidents: industrial heat, agricultural burning, and sensor artifacts produce false alarms that contribute to alert fatigue for emergency-management analysts. We study whether contextual machine learning (ML) features improve wildfire-incident classification from FIRMS detections, and—more importantly—whether reported gains survive leakage-controlled evaluation. We construct a labeled dataset by matching 521,395 VIIRS SNPP detections across CONUS in 2024 to 3766 NIFC 2024 wildfire perimeters, yielding 131,771 (25.3%) wildfire-matched and 389,624 candidate non-wildfire detections spanning 1067 distinct wildfire incidents. We benchmark five operational baselines and six classifiers under four validation regimes (random, event-aware, 5° spatial-block, and temporal holdout) with and without raw geographic coordinates. A naive random split inflates LightGBM to F1 $=0.985$, but a leakage-controlled event-aware split reduces it to F1 $=0.767$, and a spatial-block holdout to F1 $=0.627$. Feature attribution shows geographic coordinates account for 88.9% of model gain—the summed share of LightGBM’s total split-gain attributed to the three coordinate features within the full-feature model; removing coordinates improves spatial-block generalization from F1 $=0.627$ to $0.818$, demonstrating that raw coordinates drive memorization of where 2024 fires occurred rather than transferable discrimination. We further show that spatiotemporal clustering must be causal: a model using full-partition clustering appears strong (F1 $=0.908$) but leaks future detections, whereas a properly causal trailing-window version ties plain LightGBM in-distribution (F1 $=0.762$). Combining causal clustering with no raw coordinates is the most robust configuration under spatial transfer (spatial-block F1 $=0.868$ vs. $0.627$ for the coordinate model). Bootstrap 95% confidence intervals show these gaps far exceed statistical uncertainty, and sensitivity analyses show the conclusions are robust to the spatial-block size and to the clustering-window choice. Under natural class prevalence (14%), precision falls to 0.69, and results are sensitive to the labeling buffer. All ML models nonetheless far exceed FIRMS high-confidence thresholding (F1 $=0.128$). We argue that spatial leakage—not raw accuracy—is the central methodological issue for FIRMS wildfire-incident classification, and recommend coordinate-free, causal spatiotemporal-clustering features evaluated under spatial holdout. The system is intended as an analyst-prioritization decision-support layer, not autonomous incident confirmation. Full article

Intrusion detection systems for 5G networks are typically reported at a single operating point, obscuring the trade-off between missed attacks and false alarms that governs real deployments. Building on a lightweight conditional counter-inspection pipeline, in which a global classifier is selectively validated by curriculum-biased experts under a unanimous dissent rule, we remove the constraint that all components share one learning algorithm, assigning decision trees, random forests, extremely randomized trees, and histogram-based gradient boosting independently to the global (G), malicious-biased (EM), and benign-biased (EB) roles. Across two datasets of contrasting difficulty, 5G-NIDD and UNSW-NB15, all 14 evaluated tree-based configurations reduce missed attacks, by 36.5–79.6% on 5G-NIDD, confirming that the recovery effect is a property of the architecture rather than of decision trees. The expert assignment also selects which error the system controls: the same pipeline can be steered toward fewer false alarms, fewer missed attacks, or higher aggregate F1 without retraining the first stage. The mechanism also rescues a weak linear filter: on 5G-NIDD it cuts false positives and false negatives by 92.8% and 95.8%, and on UNSW-NB15 it raises F1 from 0.903 to 0.934 while reducing missed attacks by 35.5%. These results reframe the pipeline as a configurable validation layer matched to a deployment’s cost structure. We further show, through direct measurement on both datasets, that the conditional routing evaluates at most four of seven models per record, keeping classifier inference below 0.1 ms per record and leaving the detection stage a small contributor to overall processing cost. Full article

Industrial Internet of Things (IIoT) networks use on-demand sensing and wireless power transfer (WPT) for self-sustainable operation. Existing scheduling frameworks are fundamentally limited because they react only after energy levels decline. Consequently, IoT nodes enter charging mode only when their residual energy falls below a threshold, causing energy outages, increased latency, and missed sensing tasks while preventing proactive WPT resource allocation. This paper proposes a Digital Twin (DT)-enabled proactive scheduling framework that transforms IIoT scheduling from reactive to proactive. The key innovation is a closed-loop virtual–real integration in which a DT layer, co-located with the control centre, maintains a Kalman filter predictor to forecast node energy over an H-slot horizon, enabling scheduling decisions before energy shortages occur. Physical layer security (PLS) constraints and DT-based anomaly detection protect against eavesdropping, energy depletion, and false data injection attacks. A multi-objective formulation jointly optimises sensing utility and WPT efficiency while accounting for DT synchronisation overhead and uplink bandwidth consumption. The resulting multi-slot Binary Integer Linear Programmes (BILP) are solved using branch-and-bound with a reliability branching rule, and a fast greedy heuristic is also developed. Simulation results over 50 Monte Carlo iterations show that the proposed framework reduces energy outage events by approximately 70% compared with the reactive baseline, activates less than 50% of available sensing nodes, and schedules less than 60% of energy transmitters for WPT. Ablation studies confirm that DT prediction is the primary contributor to the outage reduction. DT-based anomaly detection achieves a false alarm rate below 3% while maintaining a detection rate above 95%. The proposed framework improves the sustainability, efficiency, and security of IIoT networks with practical computational overhead, making it well suited for Industry 5.0 deployments. Full article

Overflow early warning during drilling requires timely interpretation of noisy, incomplete, and operation-dependent mud-logging streams. Single-threshold alarms are easy to deploy but are sensitive to pump changes and surface-tank operations, whereas generic anomaly scores may respond to operational disturbances that are not well-control events. This study develops a state-aware multi-evidence residual scoring method that combines flow imbalance, pit-volume variation, mud-density response, pressure response, gas response, and normal-response residuals into a bounded event risk score. Missing evidence is handled by weight renormalization, and operating-state confidence and surface-disturbance suppression are introduced to reduce false warnings during non-hazardous operations. Five field replay cases were analyzed, including three confirmed overflow events and two pseudo-overflow disturbances. The proposed method detected all three confirmed overflow events, reduced the pseudo-overflow false alarm rate from 1.000 for the rule-based, residual-only, and Isolation Forest baselines to 0.500, and yielded an average warning time difference of −8.44 min relative to the field-interpreted event time. Additional sensitivity analyses showed that the selected threshold–duration setting lay in a non-isolated high-recall region, but also that excessive score smoothing or simultaneous loss of density, gas, and pressure evidence reduced detection reliability. The results support a reproducible event-level workflow for overflow warning from real-time mud-logging data, with the current evidence bounded by the small number of replay cases and the absence of independent multi-well validation. Full article

Introduction: The preferred PK/PD goal for vancomycin is now AUC_0–24/MIC-guided dosing, which requires two blood samples, additional time, increased cost, and possibly specialized staff, potentially limiting its use. Our study compared vancomycin target attainment using trough versus single-sample Bayesian AUC_0–24/MIC in critically ill patients, using a validated online calculator. Methods: This retrospective cohort study included adults aged ≥18 years with stable renal function who were receiving vancomycin. Exclusions were age under 18, unstable renal function, hemodialysis, missing data, or non-steady-state vancomycin. Steady-state troughs were obtained, and AUC_0–24/MIC was calculated using ClinCalc. The primary endpoint was discordance in target attainment between trough (15–20 mg/L) and AUC_0–24/MIC (400–600 mg·h/L). The secondary endpoint was AKI within 48 h of initiating vancomycin. Results: The mean trough was 14.8 ± 8.0 mg/L, and the mean AUC_0–24/MIC was 499.7 ± 201.3 mg·h/L. Trough levels were within target in 20%, and AUC_0–24/MIC in 67%. The 3 × 3 cross-tabulation showed a significant association (χ² = 27.33, p < 0.001), but only fair agreement (Cohen’s κ = 0.215) and a biased disagreement pattern (p < 0.001). Thirty-eight patients (38%) had “false alarming” results: trough < 15 mg/L despite AUC_0–24/MIC ≥ 400 mg·h/L. Three patients (3.0%) showed “falsely reassuring” results: trough 15–20 mg/L with AUC_0–24/MIC > 600 mg·h/L. The odds of supratherapeutic AUC_0–24/MIC > 600 mg·h/L were tenfold higher in patients with trough 15–20 mg/L than in those below 15 mg/L (OR 10.24, p = 0.048). Conclusions: In critically ill ICU patients, calculator-derived AUC_0–24/MIC showed significant discordance with single-trough vancomycin monitoring, primarily indicating apparent under-exposure by trough criteria in patients with adequate AUC_0–24/MIC levels, which may lead to unnecessary dose escalation. Full article

Clinical artificial-intelligence systems are starting to act across a course of care, not to answer one question at a time. Their safety is checked by methods that sample the input space: a test suite tries some inputs, a language-model reviewer reads some cases, a physician panel audits some cases. A sampling check can pass a safety rule and still miss the rare input that breaks it, such as a documented obligation dropped several encounters later. This study measures that gap and releases CIV-Bench, a public benchmark of 832 clinical rule sets with safety properties across eight whole-person domains, in single-encounter and longitudinal forms, plus a computational stress tier, each with independently established ground truth. We compare formal verification, which uses a satisfiability-modulo-theories (SMT) solver to check every possible input at once, against the methods used in practice: random unit testing, language-model judges, and a blinded physician panel. Formal verification detected all 612 violations, raised no false alarm, and returned no unsound verdict; for each item it returned either a proof that the rule holds over every input or one concrete input that breaks it. A frontier language-model judge matched this detection, but it returned a pass rate over sampled cases rather than a guarantee, at three orders of magnitude more compute per item. The general open-weights judge returned unsound verdicts on the computational stress tier; the medically fine-tuned judge was unsound far more widely, collapsing on the longitudinal properties despite strong single-encounter medical detection, so medical fine-tuning did not close the gap. Unit testing and the physician panel missed the deep, cross-encounter violations that hold a course of care together. Formal verification is set apart not by a higher detection rate but by the kind of evidence it returns: a proof over the whole input space, a replayable counterexample, or an explicit statement that it cannot decide. The guarantee holds for the decisions placed in this layer, and it depends on the safety rule being specified correctly. Full article

UAV-based weed detection in precision agriculture is constrained by asymmetric error costs: a missed weed patch causes herbicide under-treatment and yield loss, whereas a false alarm only prompts an unnecessary spot treatment. Cost-sensitive active learning (cAL) addresses this through an asymmetric misclassification penalty $r^{+}$, but the optimal value is scene-dependent and cannot be determined without domain expertise or advance knowledge of scene difficulty—a fundamental barrier to autonomous UAV monitoring workflows. We propose AdaRisk-Agent, the first LLM-orchestrated framework for adaptive $r^{+}$ calibration in cAL-based UAV weed detection. We validate the framework on four UAV multispectral scenes from two public datasets—WeedsGalore (Germany, five-band maize) and WeedyRice (Vietnam, four-band paddy)—spanning two crop types, two sensor configurations, and weed prevalence from 3.1% to 30.5%. Adaptive calibration reduces the false-negative rate (FNR) by up to $80\%$ relative to symmetric-cost baselines across all scenes. The deterministic surrogate (AdaRisk-Rule) surpasses the fixed-policy oracle (cAL $r^{+}=7$) on two of four scenes without advance scene knowledge, achieving a $50\%$ FNR reduction on the most spectrally challenging scene. A context-feature ablation confirms that budget urgency is the primary calibration signal and that test-set-independent deployment is feasible. Each calibration decision is accompanied by a natural-language justification, enabling auditable deployment in operational precision agriculture workflows. Future work will extend AdaRisk-Agent to multi-class weed species detection and multi-scene meta-learning for compact offline surrogate policies. 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

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

Wind turbine towers operate in complex outdoor environments, where visible surface anomalies such as cracks, pitting, and honeycombing can develop. Field-acquired visible-light images are commonly affected by illumination variation, shadows, local reflections, surface textures, and structural joints, which makes pixel-level anomaly segmentation difficult. This study proposes a task-oriented lightweight U-Net, termed LPAU-Net, that combines DWConv–PWConv feature extraction, parallel channel–spatial attention with learnable scalar fusion, grouped multi-level feature aggregation, and multi-scale decoding. Defect-free images are included during training, and defective and defect-free samples are evaluated separately to distinguish anomaly segmentation from false-positive suppression. The dataset contains 762 original field images collected from the same nine wind turbine towers at one wind farm during five time-separated acquisition campaigns. Campaigns 1–3 were used for training, Campaign 4 for validation, checkpoint selection, and threshold determination, and Campaign 5 for final evaluation. Thus, Campaign 5 is a later acquisition batch from the same towers and site, rather than unseen-tower or cross-wind-farm validation. Across three independent random seeds, LPAU-Net achieved 89.84 ± 0.08% Precision, 89.24 ± 0.08% Recall, 89.54 ± 0.08% F1-score, and 81.07 ± 0.13% Defect IoU on defective Campaign 5 images, with 4.34 M parameters, 14.8 G FLOPs, and 32.81 FPS under the reported desktop-GPU benchmark. On the 30 defect-free Campaign 5 images, the average false-positive area ratio was 0.42 ± 0.03%, and the image-level false-alarm rate was 10.00 ± 3.33%. The results indicate a balanced accuracy–complexity trade-off within the evaluated cross-time-campaign setting. However, strong light, low light, shadows, and reflections were not evaluated as independent subsets, so condition-specific robustness improvement cannot be quantified. Because all visible anomalies were merged into one binary defect class, the model localizes anomalous regions but does not classify cracks, pitting, honeycombing, or other defect types. Full article

Accurate detection in the early stages of a fire is a crucial prerequisite for the efficient implementation of fire suppression and emergency rescue operations. Its accuracy and timeliness directly affect the control of disaster loss severity. Traditional fire detection methods mainly include three categories, which are manual inspection, sensor detection, and visual recognition. However, manual inspection is restricted by labor costs and time efficiency, making it difficult to achieve large-scale, high-frequency and real-time fire monitoring. Sensor detection is easily interfered by environmental factors such as temperature, humidity, and dust, leading to frequent false alarms and missed alarms. Visual recognition technology has shortcomings in aspects such as detailed feature perception, dynamic scene modeling, and reasoning robustness in complex environments, making it difficult to meet the requirements of high-precision detection. To address these issues, this study innovatively proposes a lightweight fire and smoke detection model based on semantic enhancement and frequency domain perception modeling, which is named the SemaFire you only look once (SemaFire-YOLO) model. The model constructs a large language and vision assistant (LLaVA) semantic guidance module, which uses a large language model to understand and guide the semantic features of images, thereby enhancing the saliency representation intensity of small and weak target regions. Then, a Haar wavelet-based downsampling module is adopted, which compresses spatial information while preserving high-frequency features such as flame edges and smoke textures, improving the accuracy of target recognition. Next, the convolution modulation mechanism is introduced to replace the traditional attention mechanism, enhancing the overall modeling efficiency and reducing computational overhead. Finally, a Dynamic Tanh normalization module is adopted to replace the batch normalization module in the traditional YOLO algorithm, strengthening the model’s representation stability and reasoning robustness under unstable input distributions. Experimental results show that the SemaFire-YOLO model achieves a mean average precision (mAP@0.5) of 64.30% on the fire image dataset, which is 0.8, 2.0, 0.6, and 3.8 percentage points higher than that of mainstream models such as YOLOv5n, YOLOv8n, YOLOv11n, and YOLOv12n, respectively. It exhibits better boundary detection capability and practical deployment potential. Through visual analysis, the results indicate that the improved SemaFire-YOLO model achieves more accurate detection and higher confidence in actual complex scenarios, further verifying the model’s robustness and accuracy in complex scenarios such as low contrast and dynamic fire conditions. Full article

Concrete placing surface flatness is an important quality indicator affecting thickness control, finishing efficiency, and subsequent construction quality. Although point-cloud-based flatness inspection has been widely studied for completed floors, walls, slabs, and precast components, near-real-time detection of small local unevenness on temporary concrete placing surfaces during construction remains insufficiently investigated, especially under the effects of point-cloud noise, local slope, incomplete coverage, and baseline false alarms. To address this gap, this study proposes a LiDAR-based automatic workflow that combines multi-frame median fusion, local reference-plane fitting, threshold-based deviation judgment, and visualized output. Indoor validation was conducted using an empty-ground baseline, 20 cm × 20 cm square plates, and 50 cm × 5 cm strip plates with thicknesses of 3, 5, and 10 mm. The average over-limit point ratio was 2.165% ± 0.192% for the empty-ground baseline. For 3, 5, and 10 mm square plates, the ratios were 2.536% ± 0.370%, 3.811% ± 0.638%, and 4.657% ± 0.850%, respectively; for strip plates, they were 2.269% ± 0.059%, 2.937% ± 0.443%, and 3.347% ± 0.435%, respectively. Compared with the baseline, the 5 and 10 mm square plates increased the ratio by 1.646 and 2.492 percentage points, while the 5 and 10 mm strip plates increased it by 0.772 and 1.182 percentage points. These results show that the workflow provides a clear thickness-dependent response for square targets and a detectable but weaker response for narrow strip targets. Scientifically, the study demonstrates how local threshold-scale unevenness can be distinguished from baseline point-cloud fluctuations. In application, it provides preliminary perception support for future online flatness inspection and automatic screeding assistance. Full article

Real-time estimation of the safety distance between power operation workers and energized equipment is essential for preventing electric shock, arc flash injury, and equipment accidents during live or near-energized operations. However, large-scale deployment remains difficult because conventional binocular ranging systems require accurate offline calibration, while monocular vision suffers from metric-scale ambiguity. This paper proposes a weakly calibrated multi-camera vision framework for safety distance estimation in power operation scenes. First, a cross-camera tracking strategy combining mask-guided ReID features, Kalman filtering, spatio-temporal constraints, and Hungarian matching is developed to maintain globally consistent worker identities. Second, a region-wise self-supervised depth estimation model is introduced: photometric consistency is imposed within worker regions, while cross-view feature reprojection is imposed within rigid equipment regions. Intrinsic parameter regularization and scene-inherent dimensional priors are then used to recover metric scale without placing additional calibration targets. Third, the YOLOv8-seg backbone is enhanced with a CNN-Transformer C2f_CT module to improve pixel-level segmentation of workers, energized equipment, and structural components under cluttered backgrounds. Experiments on three representative power operation scenarios show that the proposed method achieves 92.3% MOTA, 94.2% Rank-1 accuracy, 82.0% mAP, 8.7% relative distance error, and 22 FPS end-to-end speed. When using five-frame smoothing and uncertainty threshold, the proposed method can achieve 98.7% warning precision, 0.9% missed alarm rate and 1.3% false alarm rate. The results indicate that the proposed framework provides a practical balance between deployment cost, geometric accuracy, and real-time warning capability. Full article

Monitoring the health of bearings in industrial rotating machines is a major challenge for ensuring the reliability and continuous operation of installations. Conventional fault detection methods, based on multivariate control charts such as Hotelling’s T², multivariate exponentially weighted moving average, or multivariate cumulative sum control chart, are limited by the complex nonlinear dynamics of the system. In this article, we propose an innovative monitoring approach based on the Koopman operator, allowing the linearization of a nonlinear system in an observed space and the application of drift detection techniques via an extended T² control chart. The study is based on two experimental approaches: one using controlled simulated data to analyze the responsiveness and robustness of the model, and the other applied to real data from an industrial turbogenerator monitoring the vibrations, temperatures, and speeds of the front and rear bearings. Comparative results show that the Koopman-based T² map detects defects earlier, with better accuracy under noise and a reduced false alarm rate compared to conventional methods. The integration of wavelet preprocessing, statistical feature extraction by sliding windows, and PCA representation of the trajectories enhances the robustness and interpretability of the model. Full article