Search Results (234)

Piano fingering estimation remains a complex problem due to the combinatorial nature of hand movements and no best solution for any situation. A recent model-free reinforcement learning framework for piano fingering modeled each monophonic piece as an environment and demonstrated that value-based methods outperform probability-based approaches. Building on their finding, this paper addresses the more complex polyphonic fingering problem by formulating it as an online model-free reinforcement learning task with a novel training strategy. Thus, we introduce a novel Elite Episode Replay (EER) method to improve learning efficiency by prioritizing high-quality episodes during training. This strategy accelerates early reward acquisition and improves convergence without sacrificing fingering quality. The proposed architecture produces multiple-action outputs for polyphonic settings and is trained using both elite-guided and uniform sampling. Experimental results show that the EER strategy reduces training time per step by 21% and speeds up convergence by 18% while preserving the difficulty level and result of the generated fingerings. An empirical study of elite memory size further highlights its impact on training performance in solving piano fingering estimation. Full article

This paper presents a novel data-driven framework for real-time State of Charge (SOC) estimation in lithium-ion battery systems using a data-driven Prognostics and Health Management (PHM) approach. The method leverages an optimized bidirectional Long Short-Term Memory (Bi-LSTM) network, trained with enhanced datasets filtered via exponentially weighted moving averages (EWMAs) and refined through SHAP-based feature attribution. Compared against a Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF) across ten diverse drive cycles, the proposed model consistently achieved superior performance, with mean absolute errors (MAEs) as low as 0.40%, outperforming EKF (0.66%) and UKF (1.36%). The Bi-LSTM model also demonstrated higher R² values (up to 0.9999) and narrower 95% confidence intervals, confirming its precision and robustness. Real-time implementation on embedded platforms yielded inference times of 1.3–2.2 s, validating its deployability for edge applications. The framework’s model-free nature makes it adaptable to other nonlinear, time-dependent systems beyond battery SOC estimation. Full article

In this work, we introduce IK-AUG, a unified algorithmic framework for kinematics-driven data augmentation tailored to sign language recognition (SLR). Departing from traditional augmentation techniques that operate at the pixel or feature level, our method integrates inverse kinematics (IK) and virtual simulation to synthesize anatomically valid gesture sequences within a structured 3D environment. The proposed system begins with sparse 3D keypoints extracted via a pose estimator and projects them into a virtual coordinate space. A differentiable IK solver based on forward-and-backward constrained optimization is then employed to reconstruct biomechanically plausible joint trajectories. To emulate natural signer variability and enhance data richness, we define a set of parametric perturbation operators spanning spatial displacement, depth modulation, and solver sensitivity control. These operators are embedded into a generative loop that transforms each original gesture sample into a diverse sequence cluster, forming a high-fidelity augmentation corpus. We benchmark our method across five deep sequence models (CNN3D, TCN, Transformer, Informer, and Sparse Transformer) and observe consistent improvements in accuracy and convergence. Notably, Informer achieves 94.1% validation accuracy with IK-AUG enhanced training, underscoring the framework’s efficacy. These results suggest that algorithmic augmentation via kinematic modeling offers a scalable, annotation free pathway for improving SLR systems and lays the foundation for future integration with multi-sensor inputs in hybrid recognition pipelines. Full article

Robotic-assisted therapy is an increasingly vital approach for upper-limb rehabilitation, offering consistent, high-intensity training critical to neuroplastic recovery. However, current control strategies often lack robustness against uncertainties and external disturbances, limiting their efficacy in dynamic, real-world settings. Addressing this gap, this study proposes a novel control framework for the iTbot—a 2-DoF end-effector rehabilitation robot—by integrating differential flatness theory with a derivative-free Kalman filter (DFK). The objective is to achieve accurate and adaptive trajectory tracking in the presence of unmeasured dynamics and human–robot interaction forces. The control design reformulates the nonlinear joint-space dynamics into a 0-flat canonical form, enabling real-time computation of feedforward control laws based solely on flat outputs and their derivatives. Simultaneously, the DFK-based observer estimates external perturbations and unmeasured states without requiring derivative calculations, allowing for online disturbance compensation. Extensive simulations across nominal and disturbed conditions demonstrate that the proposed controller significantly outperforms conventional flatness-based control in tracking accuracy and robustness, as measured by reduced mean absolute error and standard deviation. Experimental validation under both simple and repetitive physiotherapy tasks confirms the system’s ability to maintain sub-millimeter Cartesian accuracy and sub-degree joint errors even amid dynamic perturbations. These results underscore the controller’s effectiveness in enabling compliant, safe, and disturbance-resilient rehabilitation, paving the way for broader deployment of robotic therapy in clinical and home-based environments. Full article

Hybrid beamforming plays a critical role in evaluating wireless communication technology, particularly for millimeter-wave (mmWave) multiple-input multiple-out (MIMO) communication. Several hybrid beamforming systems are investigated for millimeter-wave multiple-input multiple-output (MIMO) communication. The deployment of huge grant-free transmission in the millimeter-wave (mmWave) band is required due to the growing demands for spectrum resources in upcoming enormous machine-type communication applications. Ultra-high data speed, reduced latency, and improved connection are all promised by the development of 5G mmWave networks. Yet, due to severe route loss and directional communication requirements, there are substantial obstacles to transmission reliability and energy efficiency. To address this limitation in this research we present an intelligent transmission control scheme tailored to 5G mmWave networks. Transport control protocol (TCP) performance over mmWave links can be enhanced for network protocols by utilizing the mmWave scalable (mmS)-TCP. To ensure that users have the stronger average power, we suggest a novel method called row compression two-stage learning-based accurate multi-path processing network with received signal strength indicator-based association strategy (RCTS-AMP-RSSI-AS) for an estimate of both the direct and indirect channels. To change user scenarios and maintain effective communication constantly, we utilize the innovative method known as multi-user scenario-based MATD3 (Mu-MATD3). To improve performance, we introduce the novel method of “digital and analog beam training with long-short term memory (DAH-BT-LSTM)”. Finally, as optimizing network performance requires bottleneck-aware congestion reduction, the low-latency congestion control schemes (LLCCS) are proposed. The overall proposed method improves the performance of 5G mmWave networks. Full article

In lunar exploration missions, rovers must navigate multiple waypoints within strict time constraints while avoiding dynamic obstacles, demanding real-time, collision-free path planning. This paper proposes a digital twin-enhanced hierarchical planning method, A*-D3QN-Opt (A-Star-Dueling Double Deep Q-Network-Optimized). The framework combines the A* algorithm for global optimal paths in static environments with an improved D3QN (Dueling Double Deep Q-Network) for dynamic obstacle avoidance. A multi-dimensional reward function balances path efficiency, safety, energy, and time, while priority experience replay accelerates training convergence. A high-fidelity digital twin simulation environment integrates a YOLOv5-based multimodal perception system for real-time obstacle detection and distance estimation. Experimental validation across low-, medium-, and high-complexity scenarios demonstrates superior performance: the method achieves shorter paths, zero collisions in dynamic settings, and 30% faster convergence than baseline D3QN. Results confirm its ability to harmonize optimality, safety, and real-time adaptability under dynamic constraints, offering critical support for autonomous navigation in lunar missions like Chang’e and future deep space exploration, thereby reducing operational risks and enhancing mission efficiency. Full article

Traditional methods for assessing honey storage in beehives predominantly rely on manual visual inspection, which often leads to inconsistencies and inefficiencies. This study presents an automated deep learning approach utilizing the YOLOv11 model to detect, classify, and quantify honey cells within Apis mellifera frames across monthly sampling periods. The model’s performance varied depending on image resolution and dataset partitioning. Using the free version of YOLOv11 with high-resolution images (960 × 960 resolution) and a dataset split of 90:5:5 for training, validating, and testing, the model achieved a mean average precision at IoU threshold of 0.5 (mAP@0.5) of 83.4% for uncapped honey cells and 80.5% for capped honey cells. A strong correlation (r = 0.94) was observed between the 90:5:5 and 80:10:10 dataset splits, indicating that increasing the volume of training data enhances classification accuracy. In parallel, the study investigated the relationship between the physical properties of honey and image-based honey storage detection. Of the four tested properties, electrical conductivity (R² = 0.19) and color (R² = 0.21) showed weak predictive power for honey storage area estimation, with even weaker associations found for pH and moisture content. The honey storage areas via 90:5:5 and 80:10:10 datasets moderately correlated (r = 0.44–0.46) with increasing electrical conductivity and color. Especially, electrical conductivity exhibited statistically significant correlations with dataset performance across different dataset splits (p < 0.05), suggesting some potential influence of chemical composition on model accuracy. Our findings demonstrate the viability of image-based honey classification as a reliable technique for monitoring beehive productivity. Additionally, the research on image-based honey detection can be a non-invasive solution for improved honey production, beehive productivity, and optimized beekeeping practices. Full article

Background/Objectives: The Tropis^® ID device (PharmaJet^®), a needle-free injection system, is a World Health Organization prequalified, hand-held device, which delivers intradermal injections without the use of needles and has previously been used for the delivery of fractional doses of inactivated polio vaccine (fIPV) in campaign and house-to-house settings. This implementation research study aimed to comparatively evaluate the vaccine coverage, cost, feasibility, and acceptability of using Tropis for fIPV for routine immunizations in two states in Nigeria (Kano and Oyo). Methods: The study included: (i) a cluster randomized trial (22 intervention facilities using Tropis for fIPV and 30 control facilities using the standard of care [SoC—full-dose IPV]) to assess the effectiveness in terms of improving the coverage of two doses of IPV, using a coverage survey involving 3433 children (aged 3–12 months); (ii) a pre- and post-implementation micro-costing evaluation involving the intervention facilities to estimate the costs; and (iii) mixed methods assessments (post-training assessment, provider survey, key informant interviews, and focus group discussions) to assess the feasibility and acceptability of fIPV delivery using Tropis. Results: The intention-to-treat analysis among the 3433 children surveyed did not show any difference between the intervention and control groups, primarily due to low compliance (approximately 50% of target beneficiaries reported Tropis use). The more relevant per protocol analysis, adjusting for lower compliance, showed that among those vaccinated with Tropis, second dose IPV coverage was 11.2% higher than the SoC. The delivery of fIPV using Tropis compared to the SoC resulted in incremental program cost savings, ranging from USD 0.07 to USD 1.00 per dose, administered across the scenarios evaluated. High acceptability was seen amongst caregivers (94%), and 95% of healthcare workers preferred Tropis over the SoC. Conclusions: Tropis is effective, feasible, acceptable, and saves costs when used as part of routine immunization programs. Full article

As the integration of distributed energy resources (DERs) continues to rise, the simultaneous import and export of energy can lead to excessive voltage violations. Therefore, calculating dynamic operating envelopes (DOEs), which represent time-varying export restrictions, is essential for ensuring the safe operation of distribution networks. Traditional methods for calculating DOEs rely on complete distribution network parameters for power-flow calculations. However, acquiring accurate parameters and network topology is often challenging, which limits the practical implementation of these traditional approaches. This paper proposes an interpretable model-free DOE calculation method that leverages smart meter data to address this issue. We train a CNN-LSTM-Attention neural network for voltage estimation, where we employ the whale optimization algorithm (WOA) to adjust hyperparameters automatically. Additionally, this paper employs the SHAP algorithm to interpret the deep learning model, providing insights into the relationship between the bus voltage and the condition of each bus, which enhances the model’s transparency and helps identify the key factors influencing voltage levels. The proposed method is validated through simulations on the IEEE 33−bus distribution network model, demonstrating favorable results. Full article

Nocturnal monitoring of continuous, cuffless blood pressure (BP) can unleash a whole new world for the prognostication of cardiovascular and other diseases due to its strong predictive capability. Nevertheless, the lack of an accurate and reliable method, primarily due to confounding variables, has prevented its widespread clinical adoption. Herein, we demonstrate how optimized machine learning using the Catch-22 features, when applied to the photoplethysmogram waveform and personalized with direct BP data through transfer learning, can accurately estimate systolic and diastolic BP. After training with a hemodynamically compromised VitalDB “calibration-free” dataset (n = 1293), the systolic and diastolic BP tested on a distinct VitalDB dataset that met AAMI criteria (n = 116) had acceptable error biases of −1.85 mm Hg and 0.11 mm Hg, respectively [within the 5 mm Hg IEC/ANSI/AAMI 80601-2-30, 2018 standard], but standard deviation (SD) errors of 19.55 mm Hg and 11.55 mm Hg, respectively [exceeding the stipulated 8 mm Hg limit]. However, personalization using an initial calibration data segment and subsequent use of transfer learning to fine-tune the pretrained model produced acceptable mean (−1.31 mm Hg and 0.10 mm Hg) and SD (7.91 mm Hg and 4.59 mm Hg) errors for systolic and diastolic BP, respectively. Levene’s test for variance found that the personalization method significantly outperformed (p < 0.05) the calibration-free method, but there was no difference between three machine learning methods. Optimized multimodal Catch-22 features, coupled with personalization, demonstrate great promise in the clinical adoption of continuous, cuffless blood pressure estimation in applications such as nocturnal BP monitoring. Full article

Due to access to the source data during the transfer phase, conventional domain adaptation works have recently raised safety and privacy concerns. More research attention thus shifts to a more practical setting known as source-data-free domain adaptation (SFDA). The new challenge is how to obtain reliable semantic supervision in the absence of source domain training data and the labels on the target domain. To that end, in this work, we introduce a novel Gradual Geometry-Guided Knowledge Distillation (G2KD) approach for SFDA. Specifically, to address the lack of supervision, we used local geometry of data to construct a more credible probability distribution over the potential categories, termed geometry-guided knowledge. Then, knowledge distillation was adopted to integrate this extra information for boosting the adaptation. More specifically, first, we constructed a neighborhood geometry for any target data using a similarity comparison on the whole target dataset. Second, based on pre-obtained semantic estimation by clustering, we mined soft semantic representations expressing the geometry-guided knowledge by semantic fusion. Third, using the soften labels, we performed knowledge distillation regulated by the new objective. Considering the unsupervised setting of SFDA, in addition to the distillation loss and student loss, we introduced a mixed entropy regulator that minimized the entropy of individual data as well as maximized the mutual entropy with augmentation data to utilize neighbor relation. Our contribution is that, through local geometry discovery with semantic representation and self-knowledge distillation, the semantic information hidden in the local structures is transformed to effective semantic self-supervision. Also, our knowledge distillation works in a gradual way that is helpful to capture the dynamic variations in the local geometry, mitigating the previous guidance degradation and deviation at the same time. Extensive experiments on five challenging benchmarks confirmed the state-of-the-art performance of our method. Full article

We propose a novel geometry-constrained learning-based method for camera-in-hand visual servoing systems that eliminates the need for camera intrinsic parameters, depth information, and the robot’s kinematic model. Our method uses a cerebellar model articulation controller (CMAC) to execute online Jacobian estimation within the control framework. Specifically, we introduce a fixed-dimension, uniform-magnitude error function based on the projective homography matrix. The fixed-dimension error function ensures a constant Jacobian size regardless of the number of feature points, thereby reducing computational complexity. By not relying on individual feature points, the approach maintains robustness even when some features are occluded. The uniform magnitude of the error vector elements simplifies neural network input normalization, thereby enhancing online training efficiency. Furthermore, we incorporate geometric constraints between feature points (such as collinearity preservation) into the network update process, ensuring that model predictions conform to the fundamental principles of projective geometry and eliminating physically impossible control outputs. Experimental and simulation results demonstrate that our approach achieves superior robustness and faster learning rates compared to other model-free image-based visual servoing methods. Full article

Accurate estimation of the state of energy (SOE) in lithium-ion batteries is crucial for determining the output power and driving range of electric vehicles. However, in series-connected battery packs, inconsistencies among individual cells pose significant challenges for precise SOE estimation. This issue is particularly pronounced for lithium iron phosphate (LFP) batteries. Their relatively flat open-circuit voltage (OCV) curve makes the classic method of directly weighting the SOE of representative cells—commonly used for ternary batteries—ineffective. This is because the traditional method relies heavily on a linear relationship between the SOE and the voltage, which is not present in LFP batteries. To address this challenge, a novel SOE estimation approach based on the online construction of the battery pack’s comprehensive OCV curve is proposed in this paper. In this new approach, the weighting of representative cells shifts from a result-oriented mode to a key-parameter-oriented mode. By adopting this mode, the whole pack’s comprehensive OCV can be obtained training free and the pack’s SOE can be estimated online within an equivalent circuit model framework. The experimental results demonstrate that the proposed method effectively controls the SOE estimation error within 3% for series battery packs composed of cells with varying degrees of aging. Full article

Objectives: Evaluating middle ear function is essential for interpreting screening results and prioritizing diagnostic referrals for infants with hearing impairments. Wideband Acoustic Immittance (WAI) technology offers a comprehensive approach by utilizing sound stimuli across various frequencies, providing a deeper understanding of ear physiology. However, current clinical practices often restrict WAI data analysis to peak information at specific frequencies, limiting its comprehensiveness. Design: In this study, we developed five machine learning models—feedforward neural network, convolutional neural network, kernel density estimation, random forest, and support vector machine—to extract features from wideband acoustic immittance data collected from newborns aged 2–6 months. These models were trained to predict and assess the normalcy of middle ear function in the samples. Results: The integrated machine learning models achieved an average accuracy exceeding 90% in the test set, with various classification performance metrics (accuracy, precision, recall, F1 score, MCC) surpassing 0.8. Furthermore, we developed a program based on ML models with an interactive GUI interface. The software is available for free download. Conclusions: This study showcases the capability to automatically diagnose middle ear function in infants based on WAI data. While not intended for diagnosing specific pathologies, the approach provides valuable insights to guide follow-up testing and clinical decision-making, supporting the early identification and management of auditory conditions in newborns. Full article

Phase denoising constitutes a critical component of the synthetic aperture radar interferometry (InSAR) processing chain, where noise suppression and detail preservation are two mutually constraining objectives. Recently, deep learning has attracted considerable interest due to its promising performance in the field of image denoising. In this paper, a Neighbor2Neighbor denoising network (NBDNet) is proposed, which is capable of simultaneously estimating phase and coherence in both single-look and multi-look cases. Specifically, repeat-pass PALSAR real interferograms encompassing a diverse range of coherence, fringe density, and terrain features are used as the training dataset, and the novel Neighbor2Neighbor self-supervised training framework is leveraged. The Neighbor2Neighbor framework eliminates the necessity of noise-free labels, simplifying the training process. Furthermore, rich features can be learned directly from real interferograms. In order to validate the denoising capability and generalization ability of the proposed NBDNet, simulated data, repeat-pass data from Sentinel-1 Interferometric Wide (IW) swath mode, and single-pass data from Hongtu-1 stripmap mode are used for phase denoising experiments. The results demonstrate that NBDNet performs well in terms of noise suppression, detail preservation and computation efficiency, validating its potential for high-precision and high-resolution topography reconstruction. Full article