Search Results (2,117)

Heart sound-based detection of cardiovascular diseases is a critical task in clinical diagnostics, where early and accurate identification can significantly improve patient outcomes. In this study, we investigate the effectiveness of combining traditional acoustic features and transformer-based Wav2Vec embeddings with advanced machine learning models for multi-class classification of five heart sound categories. Ten engineered acoustic features, i.e., Log Mel, MFCC, delta, delta-delta, chroma, discrete wavelet transform, zero-crossing rate, energy, spectral centroid, and temporal flatness, were extracted as regular features. Four model configurations were evaluated: a hybrid CNN + LSTM and XGBoost trained with either regular features or Wav2Vec embeddings. Models were assessed using a held-out test set with hyperparameter tuning and cross-validation. Results demonstrate that models trained on regular features consistently outperform Wav2Vec-based models, with XGBoost achieving the highest accuracy of 99%, surpassing the hybrid model at 98%. These findings highlight the importance of domain-specific feature engineering and the effectiveness of ensemble learning with XGBoost for robust and accurate heart sound classification, offering a promising approach for early detection and intervention in cardiovascular diseases. Full article

Remote sensing object detection is a critical task in Earth observation. Despite the remarkable progress made in general object detection, existing detectors struggle with remote sensing scenarios due to the prevalence of numerous small objects with limited discriminative cues. Cutting-edge studies have shown that incorporating contextual information effectively enhances the detection performance for small objects. Meanwhile, recent research has revealed that convolution in the frequency domain is capable of capturing long-range spatial dependencies with high efficiency. Inspired by this, we propose a Frequency-aware Feature Pyramid Framework (FFPF) for remote sensing object detection, which consists of a novel Frequency-aware ResNet (F-ResNet) and a Bilateral Spectral-aware Feature Pyramid Network (BS-FPN). Specifically, the F-ResNet is proposed to extract the spectral context information by plugging the frequency domain convolution into each stage of the backbone, thereby enriching features of small objects. In addition, the BS-FPN employs a bilateral sampling strategy and skipping connection to model the association of object features at different scales, enabling the contextual information extracted by the F-ResNet to be fully leveraged. Extensive experiments are conducted for object detection in the public remote sensing image dataset and natural image dataset. The experimental results demonstrate the excellent performance of the FFPF, achieving 73.8% mAP on the DIOR dataset without using any additional training tricks. Full article

High-Resolution Range Profile (HRRP) represents the electromagnetic backscattering distribution of targets and plays a pivotal role in remote-sensing-based Automatic Target Recognition (RATR). However, in non-cooperative sensing scenarios, acquiring sufficient measured data is severely constrained by operational costs and physical limitations, leading to data scarcity that hampers model robustness. To overcome this, we propose SpecM-DDPM, a spectral-aware Denoising Diffusion Probabilistic Models (DDPM) tailored for generating high-fidelity HRRPs that preserve physical scattering properties. Unlike generic generative models, SpecM-DDPM incorporates radar signal physics into the diffusion process. Specifically, a parallel multi-scale block is designed to adaptively capture both local scattering centers and global target resonance structures. To ensure spectral fidelity, a spectral gating mechanism serves as a physics-constrained filter to calibrate the energy distribution in the frequency domain. Furthermore, a Frequency-Aware Curriculum Learning (FACL) strategy is introduced to guide the progressive reconstruction from low-frequency structural components to high-frequency scattering details. Experiments on measured aircraft data demonstrate that SpecM-DDPM generates samples with high physical consistency, significantly enhancing the generalization performance of radar recognition systems in data-limited environments. Full article

Accurate wave detection is essential for reliable ship motion prediction and the safety of offshore operations. Wave buoys are widely deployed as key instruments for capturing wave characteristics. However, buoys drift due to the waves and currents, resulting in errors in reconstructed wave elevation. To address this challenge, a fixed-position wave-elevation reconstruction method is proposed in this paper. First, a temporal convolutional network (TCN) module is integrated with a gated recurrent unit (GRU) network to efficiently capture the nonlinear relationship between buoy motion and wave elevation, enabling simultaneous wave elevation reconstruction and dynamic deviation compensation. Second, a static deviation compensation algorithm developed from wave theory is introduced to convert the spatial deviation into temporal misalignment. The proposed method is evaluated in both time and frequency domains across various sea conditions. Results demonstrate that the proposed method effectively compensates for deviations and achieves accurate reconstruction of wave elevation at the target position. In higher sea states, accurate reconstruction is maintained even at large static deviations, with relative errors typically within 10–15%. Frequency-domain analysis shows that coherence approaches 1 near the spectral peak and below 0.3 at higher frequencies, indicating that the dominant wave components are accurately reconstructed and that high-frequency noise has a limited impact on overall accuracy. Full article

Respiratory diseases pose a significant threat to intensive pig farming, and cough recognition serves as a key indicator for early intervention. However, its practical application is constrained by the scarcity of labeled samples and the complex acoustic conditions of farm environments. To address these challenges, this study proposes a lightweight pig cough recognition method based on a pre-trained model. By freezing the backbone of a pre-trained audio neural network and fine-tuning only the classifier, our approach achieves effective knowledge transfer and domain adaptation with very limited data. We further enhance the model’s ability to capture temporal–spectral features of coughs through a time–frequency dual-stream module. On a dataset consisting of 107 cough events and 590 environmental noise clips, the proposed method achieved an accuracy of 94.59% and an F1-score of 92.86%, significantly outperforming several traditional machine learning and deep learning baseline models. Ablation studies validated the effectiveness of each component, with the model attaining a mean accuracy of 96.99% in cross-validation and demonstrating good calibration. The results indicate that our framework can achieve high-accuracy and well-generalized pig cough recognition under small-sample conditions. The main contribution of this work lies in proposing a lightweight fine-tuning paradigm for small-sample audio recognition in agricultural settings, offering a reliable technical solution for early warning of respiratory diseases on farms. It also highlights the potential of transfer learning in resource-limited scenarios. Full article

This work explores late-time gravitational collapse using timelike thin-shell methods in classical general relativity. A junction surface separates a regular de Sitter interior from a Schwarzschild or Schwarzschild–de Sitter exterior in a post-transient regime with fixed exterior mass M (ADM for ${\mathrm{\Lambda }}_{+}=0$), modelling a vacuum–energy core surrounded by an asymptotically classical spacetime. The configuration admits a natural thermodynamic interpretation based on a geometric area functional $S_{\mathrm{shell}}\propto R^2$ and Tolman redshift, both derived from classical junction conditions and used as an entropy-like coarse-grained quantity rather than a fundamental statistical entropy. Key results include (i) identification of a deceleration mechanism at the balance radius $R_{\mathrm{thr}}={(3M/{\mathrm{\Lambda }}_{-})}^{1/3}$ for linear surface equations of state $p=w\sigma$; (ii) classification of the allowable radial domain $V\left(R\right)\le 0$ for outward evolution; (iii) bounded curvature invariants throughout the shell-supported spacetime region; and (iv) a mass-scaled frequency bound $f_c{R}_S\le \xi /\left(3\sqrt{3}\pi \right)$ for persistent near-shell spectral modes. All predictions follow from standard Israel junction techniques and provide concrete observational tests. The framework offers an analytically tractable example of regular thin-shell collapse dynamics within classical general relativity, with implications for alternative compact object scenarios. Full article

Wearable inertial measurement units (IMUs) provide an accessible means of monitoring fatigue-related changes in running biomechanics, yet most existing methods rely on limited feature sets, lack personalization, or fail to generalize across individuals. This study introduces a bioinformatics-inspired stride sequence modeling framework that integrates spectral–entropy features, sample entropy, frequency-domain descriptors, and mixed-effects statistical modeling to detect fatigue using a single lumbar-mounted IMU. Nineteen recreational runners completed non-fatigued and fatigued 400 m runs, from which we extracted stride-level features and evaluated (1) population-level fatigue classification via global leave-one-participant-out (LOPO) models and (2) individualized fatigue detection through supervised participant-specific models and non-fatigued-only anomaly detection. Mixed-effects models revealed robust and multidimensional fatigue effects across key biomechanical features, with large standardized effect sizes (Cohen’s d up to 1.35) and substantial variance uniquely explained by fatigue (partial R² up to 0.31). Global LOPO machine learning models achieved modest accuracy (55%), highlighting strong inter-individual variability. In contrast, personalized supervised Random Forest classifiers achieved near-perfect performance (mean accuracy 97.7%; mean AUC 0.997), and NF-only One-Class SVMs detected fatigue as a deviation from individual baseline patterns (mean AUC 0.967). Entropy and stride-to-stride variability metrics further demonstrated consistent fatigue-linked increases in movement irregularity and reduced neuromuscular control. These findings show that IMU stride sequences contain highly informative, fatigue-sensitive biomechanical signatures, and that combining bioinformatics-inspired sequence analysis with hybrid statistical and personalized AI models enables both robust population-level insights and highly reliable individualized fatigue monitoring. The proposed framework supports future integration into sports analytics platforms, digital coaching systems, and real-time wearable fatigue detection technologies. This highlights the necessity of personalized fatigue-monitoring strategies in wearable systems. Full article

Coastal wetland vegetation exhibits pronounced spectral mixing, complex mosaic spatial patterns, and small target sizes, posing considerable challenges for fine-grained classification in high-resolution UAV imagery. At present, remote sensing classification of ground objects based on deep learning mainly relies on spectral and structural features, while the frequency domain features of ground objects are not fully considered. To address these issues, this study proposes a vegetation classification model that integrates spatial-domain and frequency-domain features. The model enhances global contextual modeling through a large-kernel convolution branch, while a frequency-domain interaction branch separates and fuses low-frequency structural information with high-frequency details. In addition, a shallow auxiliary supervision module is introduced to improve local detail learning and stabilize training. With a compact parameter scale suitable for real-world deployment, the proposed framework effectively adapts to high-resolution remote sensing scenarios. Experiments on typical coastal wetland vegetation including Reeds, Spartina alterniflora, and Suaeda salsa demonstrate that the proposed method consistently outperforms representative segmentation models such as UNet, DeepLabV3, TransUNet, SegFormer, D-LinkNet, and MCCA across multiple metrics including Accuracy, Recall, F1 Score, and mIoU. Overall, the results show that the proposed model effectively addresses the challenges of subtle spectral differences, pervasive species mixture, and intricate structural details, offering a robust and efficient solution for UAV-based wetland vegetation mapping and ecological monitoring. Full article

The flexible positioning platform is a critical structural component in the ultra-high acceleration macro–micro motion platform, enabling precise positioning across multiple scales. However, under high-frequency start–stop cycles and prolonged multi-condition operation, it is prone to fatigue damage induced by random vibrations, which poses a threat to system reliability. This study proposes a method for evaluating and optimizing the platform’s performance under random vibration based on power spectral density (PSD) analysis. In accordance with the IEC 60068-2-64 standard, representative load spectra from Tables A.8 and A.6 were selected as excitation inputs. Frequency-domain analyses of stress, strain, and displacement were conducted using ANSYS Workbench 2022R1 in conjunction with the nCode platform, incorporating the Gaussian three-sigma probability interval. The results reveal that stress and deformation are highly concentrated in the hinge region, indicating a structural vulnerability. Fatigue life predictions were carried out using the Dirlik method and Miner’s linear damage rule under various PSD loading conditions. The findings demonstrate that hinge stiffness is a key factor influencing vibration resistance and service life. This research provides theoretical support for the design optimization of flexible structures operating in complex random vibration environments. Full article

Running shoe midsoles are designed to attenuate impact forces while maintaining or improving performance. However, the literature is equivocal, likely due to measurement systems, whereas in vitro testing is conclusively favourable. This study investigated three densities of ATPU foam, comparing in vitro mechanical properties with in vivo plantar force spectral characteristics derived from individualised pressure distributions during treadmill running at varied speeds. In vitro results of slab foam and shoes showed strong positive relationships between impact variables normalised to total impact energy and foam density (r² > 0.90), and strong negative relationships for time-domain variables normalised to deformation (mm) as density increased (r² > 0.89). During running, lower midsole density increased ground contact time across speeds (p = 0.041), while spatially resolved high-frequency PSD and peak impact force both decreased (p = 0.043; p = 0.030). However, there were no differences between total vertical force and midsole density (p = 0.232). Relationships between in vitro Peak G and high-frequency PSD were strong across all speeds (r² = 0.63–0.91). Conversely, reducing midsole density increased active peak force across speeds (p = 0.003), which was strongly related to in vitro energy return (r² > 0.89). Therefore, plantar force spectra and spatially resolved analyses demonstrate how foam density properties translate from in vitro to in vivo treadmill running, with lower-density foam improving impact attenuation but elevating propulsive forces. Future work needs to verify this in an outdoor setting. Full article

To address the problem that transformer inrush currents under no-load and energization conditions can easily trigger misoperations of differential protection, this paper proposes a multi-feature identification method for transformer inrush current based on adaptive variational mode decomposition. Traditional methods typically rely on fixed physical features or single criteria, making them sensitive to operating condition variations and prone to misclassification or missed detection under complex disturbances, with limited generalization capability. The proposed method first performs adaptive VMD decomposition of current waveforms under different operating conditions. On this basis, time-domain, frequency-domain, and nonlinear features are extracted to comprehensively characterize the signal’s amplitude, spectral, and complexity information. Then, by combining the ReliefF algorithm with forward stepwise feature selection, the method reduces feature dimensionality while maintaining high discriminative power and low redundancy. Using the VMD-ReliefF-EEFO-SVM classification model, the approach achieves efficient and accurate discrimination between inrush currents and fault currents. Simulation results demonstrate that the proposed identification method adapts well to various operating conditions and exhibits strong robustness and versatility. Full article

The performance of deep learning-based hydrological forecasting is highly sensitive to input quality, yet existing studies lack a systematic framework to evaluate the impact of high-frequency noise based on hydrological characteristics. To address this, we propose a frequency-based framework to assess the robustness of LSTM runoff prediction models. We define three hydrologically meaningful noise types—long-term trend, short-term event, and transient interference—and employ a synthetic noise injection strategy on the CAMELS dataset. Furthermore, we introduce an adaptive exponentially weighted moving average (AEWMA) algorithm that dynamically adjusts smoothing based on local signal variability. Results from dual-domain evaluation (time and frequency) indicate that model accuracy deteriorates significantly when high-frequency noise exceeds 30% of the total signal energy. Moderate adaptive smoothing (e.g., $\alpha =0.9\&0.6$) effectively preserves hydrological signals while mitigating performance loss, whereas aggressive smoothing suppresses meaningful variations. This study underscores the necessity of noise-type-specific preprocessing and suggests spectral energy ratios as quantitative thresholds for adaptive data quality control in hydrological modeling workflows. Full article

This study focuses on the calibration and verification of a large-scale coupled numerical model to simulate the complex hydrodynamic–wave–sediment transport processes in the coastal and estuarine regions of the Mekong River Delta (MRD), Vietnam. Using the MIKE 21/3 modeling system, the research integrates Hydrodynamics (HD), Spectral Wave (SW), and Mud Transport (MT) modules across a computational domain of 270 × 300 km. The models were rigorously tested using field measurement data from three distinct periods: May 2004 (dry season calibration), September 2017 (first verification), and June 2024 (second verification). The results from the hydrodynamic model demonstrated high accuracy in predicting water levels, with the average Root Mean Square Error (RMSE) values ranging between 4.4% and 5.8%. The wave spectral model showed reliable performance, with the average RMSE values for wave height ranging from 15.1% to 18.0%. Furthermore, the Mud Transport module successfully captured suspended sediment concentrations (SSC), yielding average RMSE values between 26.0% and 32.1% after the fine-tuning of site-specific parameters such as critical shear stress for erosion and deposition. The study highlights the critical importance of utilizing site-specific sedimentological parameters to accurately predict morphological changes in highly dynamic estuarine environments. This validated model provides a robust tool for assessing coastal erosion and developing protection measures in regions that are increasingly vulnerable to climate change and human activities. Full article

This study proposes a lightweight framework for transferring pretrained land cover classification architectures without additional training. The system utilizes French IGN imagery and Korean UAV and aerial imagery. It employs FLAIR U-Net models with ResNet34 and MiTB5 backbones, along with the AI-HUB U-Net. The implementation consists of four sequential stages. First, we perform class mapping between heterogeneous schemes and unify coordinate systems. Second, a quadratic polynomial regression equation is constructed. This formula uses multispectral band statistics as hyperparameters and class-wise IoU as the dependent variable. Third, optimal parameters are identified using the stationary point condition of Response Surface Methodology (RSM). Fourth, the final land cover map is generated by fusing class-wise optimal results at the pixel level. Experimental results show that optimization is typically completed within 60 inferences. This procedure achieves IoU improvements of up to 67.86 percentage points compared to the baseline. For automated application, these optimized values from a source domain are successfully transferred to target areas. This includes transfers between high-altitude mountainous and low-lying coastal territories via proportional mapping. This capability demonstrates cross-regional and cross-platform generalization between ResNet34 and MiTB5. Statistical validation confirmed that the performance surface followed a systematic quadratic response. Adjusted R² values ranged from 0.706 to 0.999, with all p-values below 0.001. Consequently, the performance function is universally applicable across diverse geographic zones, spectral distributions, spatial resolutions, sensors, neural networks, and land cover classes. This approach achieves more than a 4000-fold reduction in computational resources compared to full model training, using only 32 to 150 tiles. Furthermore, the proposed technique demonstrates 10–74× superior resource efficiency (resource consumption per unit error reduction) over prior transfer learning schemes. Finally, this study presents a practical solution for inference and performance optimization of land cover semantic segmentation on standard commodity CPUs, while maintaining equivalent or superior IoU. Full article

Background/Objectives: We aim to develop an image linearization process and a program capable of quantifying vertical and left–right asymmetries observed in macular scans. We then sought to verify its applicability in clinical settings. Methods: In this single-center cross-sectional study, we examined 37 consecutive patients with unilateral open-angle glaucoma and analyzed paired data (glaucomatous eye vs. fellow normal eye). Spectral-domain OCT images were automatically processed by a custom program to align the disc–fovea axis and temporal raphe, and the following parameters were evaluated: (1) mean inner retinal thickness difference (superior–inferior), (2) Vertical Asymmetry Score, and (3) Quadrantal Asymmetry Score. Results: We analyzed 37 healthy eyes and 37 POAG eyes. After linearization, the mean inner retinal thicknesses for the normal and POAG groups were 93.4 µm (interquartile range [IQR]: 90.1–98.5) and 80.3 µm (IQR: 77.3–85.0), respectively. The Vertical Asymmetry Score was 6.80 (IQR: 6.15–7.25) for healthy eyes and 9.69 (IQR: 9.16–11.58) for POAG eyes. The Quadrantal Asymmetry Score was 6.35 (IQR: 5.94–7.19) for healthy eyes and 8.47 (IQR: 8.11–9.63) for POAG eyes. Significant differences were found between groups for all parameters (p < 0.001). The Vertical Asymmetry Score (AUC = 0.967, p < 0.001) and Quadrantal Asymmetry Score (AUC = 0.946, p < 0.001) demonstrated significantly greater accuracy in detecting glaucoma compared to the mean inner retinal thickness (AUC = 0.743). Conclusions: The developed linearization program and asymmetry scores have shown promise as parameters for objectively quantifying macular asymmetry using spectral-domain OCT. External validation in independent cohorts, including bilateral disease, is warranted. Full article