Search Results (370)

Surface electromyography (sEMG)-based gesture recognition has attracted considerable attention in intelligent prosthesis control, human–computer interaction, and rehabilitation assistance. However, practical deployment remains challenging because new users can usually provide only a few labeled samples for calibration. Under cross-subject and posture-varying conditions, this setting further aggravates distribution shifts and can destabilize target-domain adaptation. To address these issues, this paper proposes a cross-subject sEMG gesture-recognition framework integrating training-time augmentation, two-stage meta-transfer learning, and matched feature regularization. Model-agnostic meta-learning (MAML) is first used on source-subject data to learn an initialization for rapid transfer, after which target-subject fine-tuning is performed with an auxiliary class-consistent feature constraint. Experiments were conducted on a self-collected 11-subject multi-posture dataset using a leave-one-subject-out (LOSO) protocol under FT-1_full, which uses one complete calibration repetition, and FT-2_k10, which uses 10 windows per gesture per posture. The proposed MAML + FT + MATCHED method achieved higher average Accuracy and Macro-F1 than Source-pretrain + FT. Under FT-2_k10, the average improvements were 9.20 and 9.70 percentage points, respectively. Nevertheless, the primary subject-level paired Wilcoxon comparisons did not reach statistical significance (Accuracy: p = 0.2402; Macro-F1: p = 0.2402; Holm-adjusted p = 1.0000 for both). The results therefore indicate favorable average trends and positive effect sizes, while pairwise statistical superiority was not established in the present 11-subject cohort. Full article

(1) Background: Evoked potentials (EPs) are an elegant, non-invasive, and reliable technique for assessing the functional integrity of neural pathways. They are, however, often limited by the difficulty and time needed to consistently distinguish signal from noise. (2) Methods: We have recently proposed a novel technique based on spectral domain evoked-to-background ratio (EBR) that enables fast data acquisition with online feedback about actual signal quality utilizing state of the art analog-to-digital conversion. Furthermore, we have developed a novel model-based signal-to-noise management concept allowing for suppression of biological and technical interference (swallowing, stimulation artifacts, electropolarization, and powerline potentials) and for online assessment of signal-to-noise ratio (SNR) for EPs. In this work, we experimentally confirmed this concept in ten healthy volunteers by investigating cortical EPs and high-frequency oscillations (HFOs) following median nerve stimulation. (3) Results: Both mathematical model and human data demonstrate that spectral target-band EBR governs the progress in SNR with increasing sweep count. For cortical EPs, SNR exceeded 10 dB beyond 90 averages in all participants. An SNR > 20 dB documented excellent signal quality and reproducibility. For HFOs, the SNR shifted to lower values by 12 dB, displaying pronounced individual variation, however, with smaller variation of HFO-band background activity (1.9 vs. 7.6 dB between the 25% and 75% percentile). Thus, individual HFO responses are more important for actual signal extraction compared to background activity. In subjects displaying a high HFO amplitude, reproducibility was confirmed for less than 1000 sweeps. (4) Conclusions: The present investigations confirm that individual EBR is the major factor defining SNR. Background noise can be reduced to a negligible level. Online assessment of background activity will allow for the most accurate moment-to-moment visualization of raw signal quality. This will facilitate termination of the data acquisition and may be based on quantified signal quality rather than predefined sweep count. Full article

Vision-based traffic accident anticipation is critical for active vehicle safety systems, yet existing architectures frequently conflate performance gains with heavy parameter scaling optimized from scratch on compact domains. Consequently, during real-time inference, these frameworks suffer from severe prediction volatility and early triggering biases that induce dangerous control instability. To address these limitations, this paper shifts the research focus away from network modifications toward a highly controlled, strategy-driven training pipeline executed under a completely invariant spatial–temporal neural backbone. Our proposed paradigm establishes a robust framework through three decoupled milestones. First, an out-of-domain initialization strategy transferred generalized driving kinetics from a large-scale sequence domain (Mapillary) to serve as a stable temporal anchor. Second, a target-domain generative enrichment step injected synthetic nighttime scenes to decouple hazard features from low-light ambient noise. Third, progressive temporal supervision paradigm scaling targeted labels monotonically to align with continuous kinetic risk accumulation. Overall evaluations on the Car Crash Dataset (CCD) benchmark demonstrate that the fully integrated configuration (C4) pipeline achieves 69.89% in frame-level Mean Average Precision (mAP), which is an improvement of +22.81 percentage points over the baseline configuration. Continuous temporal measurements prove that our framework can adapt to tracking volatility, compressing Temporal Confidence Variance to 0.00328, and dropping the Prediction Instability Count to 0.66. While hyper-sensitive baselines report early raw latency averages driven by premature trigger noise, our model purposefully filters this early-frame variability to deliver a secure warning profile, achieving an absolute zero false alarm rate (FAR = 0.00%) across evaluated non-hazardous driving sequences, establishing the sequence-level trustworthiness required for practical autonomous deployment. Full article

The cost–performance relationship in mutual funds is a longstanding open question in financial economics, particularly when costs are assumed to exert a single, linear effect on returns. This study proposes an integrated causal machine learning framework to revisit this question using a panel of Hungarian open-ended public investment funds across all major asset classes—equity, bond, absolute yield, misc, money market, real estate, and commodity—covering 2017–2024. Six machine learning algorithms are benchmarked for return prediction, and Double Machine Learning, with fund-level cluster-robust inference and year fixed effects, is applied to estimate the effect of the Total Expense Ratio (TER) on next-year returns, under the identifying assumptions stated in the paper, while flexibly controlling for a set of observed fund-level confounders (size, NAV dynamics, volatility, past and cumulative performance, and fund age) without imposing a linear functional form. To move beyond average effects, a Causal Forest model—tuned using an out-of-fold, effect size-neutral selection criterion—estimates heterogeneous treatment effects across funds, and SHAP-based interpretation uncovers the mechanisms underlying this heterogeneity. The results show that, once the outcome is measured in the year following the one in which TER is observed and panel dependence is properly accounted for, the average TER effect is not robustly different from zero at the full-sample level; where a statistically robust effect emerges, it is negative rather than positive, concentrated in equity and absolute-yield funds, and largely confined to the period after 2022, which coincided with the war in Ukraine, rising interest rates, and heightened market volatility, although the research design does not identify which, if any, of these developments drove the change. Average-effect models are shown to conceal this heterogeneity, and the results are further shown to be sensitive to two methodological choices that might otherwise appear secondary—the timing convention linking cost and return, and the criterion used to select among competing heterogeneous-effects specifications—underscoring the importance of making such choices explicit. These findings demonstrate the added value of combining predictive and causal machine learning, together with identification-robust and panel-robust inference, for uncovering heterogeneity that conventional econometric approaches overlook and offer a transferable methodological template for causal machine learning applications in finance and other high-dimensional decision-making domains. Full article

Reliable detection of event-related potentials (ERPs) at the single-trial level remains a challenge due to low signal-to-noise ratio and high variability in electroencephalography (EEG) recordings. This work investigates the use of the Deep-Match framework (Deep-MF) for ERP detection. We examine whether incorporating prior knowledge of an ERP template into deep learning models improves detection performance. As a proof-of-concept study, the framework was evaluated on a single dataset with multi-channel EEG recordings during laser stimulation. The model was trained in two stages. First, an encoder–decoder architecture was trained to reconstruct input EEG signals in order to learn compact signal representations. In the second stage, the decoder was replaced with a detection module and the network was fine-tuned for ERP identification. Two model variants were evaluated: a standard model with randomly initialized filters and a Deep-MF model in which input kernels were initialized using ERP templates. Models performance was assessed on a single-trial ERP detection task during leave-one-out validation, and then compared with matched filter detector. The neural network models outperformed the matched filter detector and proposed that the Deep-MF model slightly outperformed the detector with standard kernel initialization for the majority of held-out subjects. Although both approaches exhibited substantial inter-subject variability, Deep-MF achieved a higher average F1-score (0.37) compared to the standard network (0.34), indicating improved robustness to cross-subject differences. Performance varied considerably across participants. The best performance obtained by Deep-MF reached an F1-score of 0.71, exceeding the maximum score achieved by the standard model (0.59). These results showed that ERP-informed kernel initialization provides improvements in single-trial ERP detection under subject-independent evaluation. These findings demonstrate that integrating domain knowledge with deep learning architectures can improve single-trial ERP detection. The proposed approach provides a step towards practical wearable EEG and passive brain–computer interface applications, as well as towards real-time monitoring of cognitive processes. Full article

Domain generalization under distribution shift remains a fundamental challenge in representation learning, where predictive models trained on multiple source domains are expected to generalize to previously unseen operating conditions. For rotating machinery fault diagnosis, variations in operating conditions often introduce domain-specific characteristics that interfere with fault-related representations, resulting in significant performance degradation across unseen domains. To address this issue, this paper formulates cross-condition fault diagnosis as a condition-disentangled representation optimization problem and proposes a Condition-Disentangled Representation Optimization Framework. The proposed framework jointly optimizes fault-discriminative and condition-related representations through a time-frequency collaborative architecture. Specifically, an asymmetric orthogonal constraint is introduced to encourage feature disentanglement between fault and operating-condition representations. A class-domain prototype regularization strategy is developed to improve intra-class compactness and inter-domain consistency, while a bidirectional prototype alignment mechanism further enhances cross-domain semantic correspondence. These objectives are integrated into a unified representation optimization framework and optimized jointly using a weighted objective function. Numerical experiments and comparative analyses on multiple benchmark datasets demonstrate that the proposed framework achieves the highest average accuracy on both datasets and the best performance among the compared methods on most cross-condition transfer tasks, while remaining competitive on the remaining task. The results indicate that optimizing condition-disentangled representations effectively improves robustness, feature separability, and generalization capability under distribution shift, providing an effective optimization framework for cross-condition intelligent fault diagnosis. Full article

Structural design codes provide safe procedures for verifying steel–concrete composite beams but do not directly guide engineers toward cost-effective configurations. This study aims to develop and evaluate a surrogate-assisted framework that combines a Multilayer Perceptron neural network with Bayesian Optimization for the preliminary flexural-resistance and material-cost optimization of simply supported steel–concrete composite beams designed according to the Brazilian code NBR 8800. A dataset containing 20,000 beam configurations and 15 input variables was generated using a Python-based analytical routine that implements the NBR 8800 provisions for the positive bending resistance of composite beams. The generated dataset was used to train a Multilayer Perceptron neural network to predict the design bending resistance. The trained surrogate model was then integrated with Bayesian Optimization to search a discrete design space comprising commercial steel profiles, concrete slab thicknesses, shear connector quantities, and connector diameters. The selected neural network architecture achieved validation MAE and RMSE values of 2.128 kN·m and 3.013 kN·m, respectively, with an $R^2$ of 0.9999. In ten benchmark scenarios, the BO–MLP framework identified candidate solutions using only 60 objective-function evaluations. This corresponds to 3% of the evaluation budget adopted for GA and PSO and approximately 0.057% of the configurations examined by exhaustive search. Despite this limited sampling budget, the resulting candidate solutions presented an average optimality gap of approximately 12.1% relative to the global reference. In computational terms, GA and PSO required approximately 4.1 and 4.9 times the execution time of BO–MLP, respectively, while exhaustive search required approximately 18.4 times the execution time. Overall, the proposed framework offers a computationally efficient means of exploring discrete composite-beam configurations and identifying cost-competitive candidate solutions. Direct NBR 8800 verification showed that seven of the ten selected candidates satisfied the resistance requirement, while three presented resistance-to-demand ratios slightly below unity. Therefore, the framework should be used as a preliminary screening tool, with the selected configurations subsequently verified using the complete code-based procedure. Within the restricted structural domain investigated, the framework can support preliminary decisions related to positive bending resistance and material cost. Full article

Dynamic difficulty adjustment in Serious Games remains challenging because most adaptive approaches respond primarily to performance variables and insufficiently incorporate the player’s psychophysiological state in real time. Prior studies have shown the potential of biofeedback and heart rate variability based adaptation; however, many proposals rely on isolated indicators, limited temporal integration, or mechanisms that do not explicitly stabilize state transitions before modifying gameplay. In response, this article proposes a dynamic difficulty adjustment mechanism based on cardiac signals and Cellular Automata for Serious Games. The novelty of the proposal lies in combining individualized instantaneous heart rate ranges, time and frequency domain heart rate variability features, a hybrid multilayer caching for resolving discrepancies between short and longer window estimates, and a Cellular Automaton that introduces temporal inertia before applying adaptive changes to game parameters. Methodologically, the study follows a design and implementation research process in which the mechanism is integrated with a Polar H10 sensor, structured as a modular architecture, and evaluated through functional and architectural, including real time tests of arousal band assignment, adaptive response, and trace level coherence. The engine assigns three classes of operational arousal bands, defining individualized low, target, and high BPM/HRV control regions for DDA actuation. The results demonstrate technically stable operation and effective real time adaptation, with an average response latency of approximately two seconds, supporting the feasibility of the proposed mechanism. Overall, the findings indicate that cardiac signal driven adaptation combined with Cellular Automata based transition control constitutes a technically viable approach for Serious Games. Full article

Electromyography (EMG)-based sensing is an important tool for assessing neuromuscular activity and monitoring athlete development; its reliability depends on electrode placement, signal quality, and accurate identification of muscle activation periods. This study proposes an intelligent EMG sensing framework integrating preliminary electrode placement assessment, muscle activity detection, feature extraction, and regression-based progression prediction. A placement assessment indicated that positioning the electrode adjacent to the innervation zone produced the highest RMS under the tested conditions. A two-stage activity detection method based on clustering and probabilistic modeling achieved an average error of 1.5% and a temporal deviation of 19 ms. Nine time-domain EMG features extracted from the detected activity segments were used to characterize athlete progression and estimate the time required to reach a reference neuromuscular profile. Among the methods, Linear Regression provided the best fit to the data, obtaining R² = 0.987 and RMSE = 4.21 and suggesting a predominantly linear relationship between the EMG-derived features and training duration within the dataset. However, these results were obtained from only six longitudinal observation periods for a single representative athlete, with each period represented by a 90-dimensional EMG feature vector derived from the ten movement classes. Therefore, the results should be interpreted as preliminary, athlete-specific goodness-of-fit findings rather than evidence of generalizable predictive performance. Validation using larger longitudinal cohorts and independent datasets is required. The proposed framework is compatible with future IoT-enabled wearable and edge-computing architectures; however, hardware-level implementation was beyond the scope of this study. Full article

Low Earth orbit (LEO) satellite downlinks broadcast over wide ground footprints, exposing Earth-observation and remote-sensing sensor data to passive eavesdropping. Affine frequency division multiplexing (AFDM) is a candidate waveform for the doubly dispersive LEO channel and a natural integrated sensing and communication (ISAC) waveform whose delay–Doppler structure supports target parameter estimation; yet existing secure-AFDM schemes act only in the discrete affine Fourier transform (DAFT) parameter domain, leaving the transmitted waveform structurally recognizable. To address this gap, this paper applies time-domain waveform obfuscation to AFDM as physical layer encryption. Using a secret key, the transmitter permutes the inverse-DAFT samples and applies a phase rotation before chirp-periodic-prefix generation; the mask is unitary, so the peak-to-average power ratio is preserved exactly, and the key-holding receiver retains AFDM’s full delay–Doppler sensing capability, while a no-key receiver obtains a dense composite response that destroys target localization (sensing concentration drops from 0 dB to −16.6 dB). Secret pilot phases enable channel estimation at the legitimate receiver while blocking a naive composite-channel attack. Simulations at $N=64$ and 128 show that a wrong-key eavesdropper achieves uncoded BER within $0.01$ of $0.5$ across 0–20 dB and that blind Viterbi–Viterbi phase recovery is no more effective under QPSK (BER $0.46$–$0.48$), while the legitimate SNR penalty stays below $0.5$ dB. The mask also suppresses AFDM’s internal structure to the AWGN level under AFDM-aware processing. Time-domain obfuscation offers a complementary physical-layer security layer for confidential LEO remote-sensing data downlink and ISAC waveforms. Full article

Multi-modal image matching (MIM) remains a challenging problem due to nonlinear radiometric variations and geometric distortions across heterogeneous sensors. This paper proposes a robust feature-based matching framework that reduces reliance on intensity information while enhancing structural representation. The filter with local normalization is applied to transform the input images into a common intermediate domain. A block-based strategy is then employed to enforce a uniform spatial distribution of keypoints using the ORB (Oriented FAST and Rotated BRIEF) detector. To further suppress intensity variations and improve discriminability, a novel Max-Index-based HOG (MIHOG) is developed. This descriptor integrates multi-scale feature representations and encodes dominant structural information through discrete max-index mapping. Finally, correspondences are established using a brute-force matching strategy. Extensive experiments are conducted on two multi-modal datasets covering eight diverse scenarios. The proposed method achieves an average NCM of 224.52, RMSE of 3.4788, and SR of 92%. MIHOG obtains the highest NCM on 4/8 test scenarios and improves the average NCM by 18.3% compared with the second-best method. Meanwhile, it maintains competitive computational efficiency, with an average running time of 10.20s. These results demonstrate that MIHOG can provide dense and reliable correspondences under complex cross-modal radiometric and geometric variations. Full article

Micro-Electro-Mechanical System (MEMS) gyroscopes serve as core components for attitude-sensing systems in low-speed unmanned vehicles and mobile robots. The long-term dynamic stability of MEMS gyroscopes under actual continuous low-speed vehicle operation is significantly inferior to the nominal performance derived from laboratory-based static calibration. This paper adopts a navigation-grade fiber optic gyroscope as the high-precision angular velocity reference. Angular velocity error sequences between the MEMS gyroscope and fiber optic gyroscope are established using field test data collected from a low-speed vehicle experiment lasting approximately 3.3 h. Three analytical approaches are applied to systematically characterize noise evolution features of the MEMS gyroscope under static and dynamic conditions from multiple dimensions. These approaches include time-domain drift analysis, angle random walk evaluation via Allan Variance, and frequency-domain interpretation based on Welch power spectral density. The test results reveal that vehicle motion significantly degrades the 10 s averaged bias stability of the MEMS gyroscope. The bias stability values under dynamic conditions increase by 2.1, 2.7 and 7.7 times compared with static states respectively. Root mean square analysis through band segmentation integration of Welch power spectral density indicates that vehicle motion induces the most obvious rise in noise energy in the middle frequency band. The amplification factor of noise along the Y axis reaches 24.7 times. The joint analytical framework proposed in this paper takes fiber optic gyroscope measurements as the reference and integrates time-domain analysis, Allan Variance and frequency-domain methods. It can provide sufficient experimental evidence and technical support for dynamic error compensation of MEMS gyroscopes deployed on low-speed mobile platforms. Full article

With the continuous penetration of renewable energy and power electronic equipment into modern power systems, the occurrence frequency of composite power quality disturbances has increased significantly. The accurate classification of various composite disturbances under strong noise remains a critical technical challenge. The existing single time–frequency transformation methods cannot simultaneously capture transient time-domain details and fine frequency-domain features of steady-state harmonics, while mainstream deep learning classification networks contain redundant parameters and introduce excessive computational overhead, failing to meet the real-time deployment requirements of power edge terminals. To address these limitations, a lightweight Coordinate Attention ResNet network named ResNet–LCA is proposed based on the dual-modal time–frequency fusion of the Continuous Wavelet Transform and Short-Time Fourier Transform. First, the two transforms are implemented separately to generate two groups of complementary time–frequency maps, which are concatenated along the channel dimension to fully extract the coupling features between the steady-state harmonics and the transient impulses. Second, a Haar wavelet subband mean aggregation module is designed for dimensionality reduction with negligible information loss. This module eliminates the channel redundancy introduced by the multimodal fusion and reduces the overall computational overhead at the input stage. Finally, a lightweight residual network integrated with Coordinate Attention is constructed, with Grouped Half-Convolution adopted to compress the model parameters. CA offsets the feature attenuation induced by the lightweight structural design and further improves the model’s noise immunity. A simulation verification was carried out on a simulated dataset covering 25 types of single and superimposed composite disturbances. At a signal-to-noise ratio of 20 dB, the proposed method achieved an average classification accuracy of 97.92%, with only 5.32 M total parameters and a single-sample GPU inference latency of 0.33 ms. Compared with standard ResNet-18 under 20 dB noisy conditions, the total parameter volume was reduced by 52.7%, the inference latency was shortened by 0.13 ms, and the classification accuracy was improved by 0.60 percentage points. The proposed method achieves coordinated optimization of classification accuracy, noise immunity and inference efficiency, and it can provide lightweight technical support for online intelligent power quality monitoring at the edge nodes of microgrids and islanded power systems. Full article

Glow-discharge plasma (GDP) poling is revisited as an electrode-free method for activating piezoelectricity in poly(vinylidene fluoride) (PVDF) films. Although this method was proposed several decades ago, its effect on the properties of PVDF films has remained poorly understood. In this work, we demonstrate that GDP enables efficient poling of oriented PVDF films without pre-deposited electrodes and investigate the relationship between plasma treatment time, structural evolution, and piezoelectric response. Commercially available 25 μm-thick oriented PVDF films (PolyK) were treated in a DC glow discharge for 15 s to 15 min and characterized using FTIR, DSC, piezoresponse force microscopy, UV–Vis–NIR spectrophotometry, quasi-static d₃₃ measurements and water contact-angle measurements. GDP poling produced a side-averaged piezoelectric coefficient d₃₃ of up to ~25 pC/N within 1–5 min, with local maxima at approximately 1, 2.5, and 5 min. This behavior was accompanied by pronounced changes in the domain structure, including an increase in the ferroelectric domain size from 86 to 552 nm, while the crystallinity and electroactive phase fraction changed only moderately. Plasma treatment also increased the wettability of the plasma-facing surface, reducing the water contact angle from about 85° to 42° within 3 min. At longer treatment times (>5 min), however, the piezoelectric response decreased and the optical transparency deteriorated because of increased haze and turbidity, most likely associated with plasma-induced chemical modification of the surface layers. These results indicate that GDP poling has an effective processing window of 1–5 min. The proposed approach provides a vacuum-compatible and electrode-free route for preparing PVDF films with increased surface wettability for flexible piezoelectric sensors, wearable electronics, and integrated polymer-based devices, because it is compatible with electrode deposition on an already activated polymer surface within a single vacuum cycle. Full article

Electro-hydraulic actuator systems are widely used in precision motion-control applications; however, their displacement regulation remains challenging because fast response, low overshoot, and high steady-state accuracy must be achieved simultaneously under strongly dynamic operating conditions. In this study, a proportional-integral-derivative (PID) controller augmented with a Gompertz function (PID-G) is proposed for the position control of a four-way valve-controlled linear actuator, and its parameters are tuned by the recently introduced dholes-inspired optimizer (DIO). First, a control-oriented mathematical model of the electro-hydraulic actuator system is established by combining the valve and actuator dynamics. Then, the PID-G structure is formulated by incorporating a nonlinear Gompertz-based term into the conventional PID framework, and the resulting seven-parameter tuning problem is cast as an optimization task using a composite objective function that accounts for overshoot, steady-state error, rise time, and settling time. The effectiveness of DIO is evaluated comparatively against flood algorithm (FLA), covariance matrix adaptation evolution strategy (CMA-ES), and particle swarm optimization (PSO) under identical simulation conditions. The results show that DIO provides the best optimization performance, yielding the lowest best, average, and standard-deviation values of the objective function among the compared algorithms. In the time domain, the DIO-based PID-G controller achieves the most favorable overall response with a rise time of 0.079511 s, a settling time of 0.099326 s, an overshoot of 0.15110%, and a steady-state error of 0.089317%. The superiority of the DIO-based design is further confirmed by lower values of error based performance metrics (IAE, ISE, ITAE, and ITSE), improved convergence characteristics, and statistically significant advantages in the Wilcoxon test. Additional comparisons with different (PI, PID, 2DOF-PID, and FOPID) controllers also demonstrate that the proposed PID-G structure provides markedly better transient and error-based performance when tuned by DIO. Frequency-domain and varying-setpoint results further indicate satisfactory stability margins, robust tracking ability, and bounded control effort. Overall, the study shows that combining DIO with a Gompertz-augmented PID structure constitutes an effective strategy for high-performance electro-hydraulic actuator displacement control. Full article