Variable-parameter autoregressive model of an underwater acoustic channel in the frequency domain

CMMDL: Cross-modal multi-domain learning method for image fusion.

EPformer: Unlocking day-ahead electricity price forecasting accuracy using the time–frequency domain feature learning strategy considering renewable energy

Frequency domain iterative clustering for boundary-preserving superpixel segmentation

Frequency domain dynamic analysis method for steam network and leakage diagnosis

Spatiotemporal time series graph modeling method research based on the fusion of time–frequency domain analysis and deep learning: A case study of groundwater depth prediction in the Heihe River Basin, China

DrowsyDG-Phys: Generalizable driver drowsiness estimation in conditional automated vehicles using physiological signals.

ABSTRACT Due to climate policy uncertainty, the credit risk spillovers among industry bonds have become more dynamic and less predictable, which may increase the financing costs of corporations and undermine the implementation effectiveness of climate policies. This work quantifies the frequent credit risk spillovers among industry bonds by employing the time-varying parameter vector autoregression (TVP-VAR) frequency connectedness model. The causality and impact paths from climate policy uncertainty to frequent credit risk spillovers are further assessed using non-parametric causality-in-quantiles tests and the frequency minimum spanning tree (MST) model. The results indicate that: (i) industry bond credit risk spillovers are driven by short-term spillovers. (ii) The directional credit risk spillovers from cyclical industries to other industries are greater than those from consumption, service and support industries at different frequencies. (iii) There is a significant causality from climate policy uncertainty to directional credit risk spillovers of consumption and cyclical industries under at different frequencies and market conditions. (iv) Climate policy uncertainty triggers credit risk spillovers in the short and medium terms by directly affecting agriculture, forestry, animal husbandry and fishery industries. (v) The direct paths by which climate policy uncertainty affects credit risk spillovers increase in the long term.

Epilepsy is a chronic neurological disease that profoundly impacts patients' daily lives. Electroencephalography (EEG) serves as a crucial tool for the clinical diagnosis of epilepsy and other brain disorders. Current research methods primarily concentrate on the time domain of EEG signals, often preprocessing frequency domain information without thorough exploration or effective integration with the time domain. To overcome the limitations of traditional models in extracting comprehensive frequency domain information and fusing time and frequency data, this paper proposes a Time-Frequency Cross-Attention Network (TFCANet) based on the residual attention mechanism. This network converts time-domain features into frequency-domain features using a Fast Fourier Transform. Subsequently, four SE Residual modules are employed to extract features for the frequency domain branch, while a Residual Window Multi-head Self-Attention (ResWMSA) mechanism is utilized for the time domain branch. Finally, cross-attention is applied to achieve inter-modal feature fusion. The proposed model is experimentally evaluated on the HMS-Harmful Brain Activity Classification dataset from Kaggle's 2024 competition and a dataset from the University of Bonn, Germany. Our model achieved 96.15% accuracy on a five-category task using the HMS dataset and 93.63% accuracy on a five-category task using the University of Bonn dataset. These results demonstrate that our model fully integrates features from both time and frequency domains, highlighting the superiority of time-frequency feature fusion over single-modality approaches for epilepsy detection.

Initial afferent activation of V1, indexed by the C1 component of the human VEP, is often considered to be a rudimentary stage of visual processing, operating mostly as a conduit for later stages with limited cognitive penetrability. The full suite of visual analysis entails activity across several visual areas and feedback from later areas to earlier ones. This raises the question of whether the early sensory representation indexed by the C1 is read out for perceptual decisions or whether it is passed over in favour of more advanced representations. To address this question, we asked whether the C1 would predict time-pressured stimulus contrast comparisons independently of physical stimulus conditions, a phenomenon known as choice probability. We found that the C1 did this for a narrow range of response times, indicative of decision readout since the C1 is a transient signal. This effect could not be accounted for by stimulus differences, choice history, or any other choice-predictive signal that we could identify in either the time or frequency domain, either before or after target onset. It also preceded the onset of evidence-dependent decision formation estimated from the centroparietal positivity by tens of milliseconds, together providing an approximate timeline of early evidence readout and its delayed impact on the decision.

Abstract With the advancement of smart agriculture, the accuracy of fruit recognition and peduncle localization has become a key factor constraining the efficiency of automated harvesting. Precise fruit recognition determines the effective selection of operation targets by automated harvesting systems, while accurate localization and measurement of fruit peduncles directly affect the choice of cutting positions during harvesting and the overall harvesting efficiency. To address the challenges in open-field cultivation of Solanum muricatum, including dense fruit distribution, severe peduncle occlusion, and difficulty in accurately distinguishing fruit maturity, this paper proposes a fruit and peduncle detection algorithm based on frequency-domain multi-scale fusion, termed YOLO11-CCM. This algorithm introduces the CSP 2-convolution Wavelet Transform (C2WT) module to jointly model spatial- and frequency-domain features, thereby enhancing the representation of fine-grained information such as color and texture. In addition, a Slice Convolution Downsampling (SCD) module is incorporated to strengthen the extraction of local salient features while preserving feature resolution, effectively improving detection accuracy in dense scenes. Furthermore, to address the difficulty of detecting small-scale targets such as occluded pedicels, a Multi-scale Hybrid Attention (MSHA) module is designed to enhance cross-scale feature interaction, thereby improving small-object detection performance. This module combines multi-scale convolution with a dual attention mechanism, effectively enhancing the model’s ability to perceive small targets. Experimental results show that YOLO11-CCM significantly outperforms the original YOLOv11s model in the detection of pepino melon fruits and peduncles, with precision increased by 5.3%, recall increased by 2.8%, and mAP50 increased by 3.4%. Furthermore, the model achieves efficient real-time inference on both PC platforms and the NVIDIA Jetson Orin Nano platform. The findings provide strong technical support for the visual perception system of smart Solanum muricatum harvesting equipment and are of great significance for promoting the automated harvesting of specialized fruits and vegetables.

Arterial insufficiency is a key factor in chronic wounds, diabetes, and peripheral arterial disease, all of which impair vascular function. Accurate monitoring of tissue-level oxygenation and hemodynamics is critical for assessing outcomes in cases of vascular compromise. However, many existing tools only measure small, localized regions of tissue. This study evaluates spatial variation in oxygenation during a vascular occlusion test (VOT) using a wide-field imaging using spatial frequency domain imaging (SFDI). Tissue oxygenation and perfusion dynamics were assessed using SFDI to map oxygen saturation (StO₂), total hemoglobin in the papillary dermis (HbT1), and deeper tissue (HbT2) during a vascular occlusion test (VOT) in 13 subjects. Measurements were taken immediately before induction of the occlusion, after 4 minutes of occlusion and immediately after occlusion release. Two regions of interest (ROIs) were analyzed: 1) areas with larger subsurface vessels (macrovasculature), and 2) areas dominated by capillary networks (microvasculature). StO₂ values differed significantly between microvascular-only and macrovascular ROIs at all time points. Microvascular ROIs showed greater StO₂ changes during occlusion, indicating higher oxygen extraction. HbT1 concentrations did not differ significantly between ROIs at any time point. Spatial variation is critical when comparing tissue hemodynamics across time and subjects. Non-contact wide-field imaging enables assessment of heterogeneous tissue regions that are difficult to evaluate with probe-based methods.

Stroke remains a primary cause of disability globally. An in-depth comprehension of gait impairments following a stroke is vital for crafting effective therapeutic interventions, and vertical ground reaction forces (vGRF) provide valuable insights into these mechanics. Eighteen stroke survivors within 90days post-stroke and 18 healthy controls participated in this study, completing a 2-min walking trial on an instrumented treadmill to investigate vGRF signals using harmonic and power spectrum analysis. Significant interactions were observed in vGRF patterns between stroke survivors and controls, and between paretic and intact legs. The paretic leg of stroke survivors exhibited a significantly lower harmonic coefficient (paretic: 61.71±4.09% body weight vs. intact: 74.77±6.65% body weight), a lower essential number of harmonics (paretic: 7.57±1.12 vs. intact: 10.87±1.76), a lower 99.5% power frequency (paretic: 4.19±0.10Hz vs. intact: 4.71±0.31Hz), and a higher median power frequency (paretic: 0.43±0.03Hz vs. intact: 0.40±0.01Hz) compared to the intact leg. The paretic leg demonstrated a simplified waveform shape (inverse U) compared to the intact leg (M shape), which likely contributes to the reduced essential number of harmonics and 99.5% power frequency, and the paradoxical increase in median frequency. These findings quantify the mechanical consequences of post-stroke neuromuscular deficits and highlight the potential of frequency domain analysis as a diagnostic tool, offering important implications for developing personalized rehabilitation strategies to improve outcomes in subacute stroke survivors.

Abstract Rolling bearing Remaining Useful Life (RUL) prediction plays a critical role in equipment health management. However, noise in practical engineering applications can submerge early weak degradation features, making it difficult to capture early degradation patterns. This constrains the application of prediction methods across all degradation stages.To address these issues, this paper proposes a rolling bearing RUL prediction method based on a Vibration Signal Reconstruction-enhanced Method (VSRM) and a Time-Adaptive Fusion Network (TAFN). First, VSRM performs decoupled learning and non-linear reconstruction of the signal's amplitude and phase spectra in the frequency domain. This effectively suppresses noise and significantly enhances the robustness and representational power of degradation features. Subsequently, TAFN introduces timestamp encoding as an explicit global temporal prior and designs a temporal-position-driven adaptive weighting mechanism. This achieves a dynamic fusion of global trends and local features, thereby resolving the lag problem that existing models face when capturing non-linear changes in degradation rates. In this work, the VSRM-TAFN framework is deeply integrated with several mainstream time-series prediction networks and comprehensively validated on two full-life-cycle rolling bearing datasets. The results demonstrate that the proposed VSRM-TAFN framework significantly improves the prediction accuracy of all mainstream time-series networks, achieving a minimum RMSE of 0.042 and a minimum MAE of 0.0327. This general architecture effectively overcomes the challenges posed by noise and non-linear degradation rate variations, providing an effective and universal solution for achieving high-robustness RUL prediction.

Recent advances in learning-based underwater image enhancement have achieved remarkable progress. However, the inherent diversity and complexity of underwater scenes still limit the ability of existing approaches to simultaneously restore fine structural details and global image layouts. To address this challenge, we propose a Resonant Fusion (ReFu) framework that explicitly leverages complementary information in both spatial and frequency domains. Specifically, we design a frequency decomposer and a spatial decomposer to capture high- and low-frequency cues from different perspectives. A resonant fuser is then introduced to adaptively integrate high-frequency resonances for detail refinement and low-frequency resonances for structural consistency. This fine-grained cross-domain fusion significantly improves structural preservation and detail enhancement, thereby generating visually more natural and perceptually friendly underwater images. Extensive quantitative and qualitative evaluations across diverse underwater benchmarks show that ReFu consistently surpasses state-of-the-art methods by a clear margin. Comprehensive ablation studies further validate the effectiveness of each module and prove the necessity of the proposed ReFu mechanism. Our code is available at https://github.com/CircleQa/ReFu-main.

Determination of torsion angles via recoupling of backbone HC and HN dipolar interactions is a well-known method in magic-angle spinning NMR spectroscopy. Torsion angle values can be obtained by comparing simulated and experimental signals, either in the frequency or time domains. Typically, all molecular orientations are assumed to have identical detected amplitudes at zero recoupling time. The changes in these amplitudes during the recoupling period define the dipolar coupling values and the torsion angles. Experimentally, however, orientations may exhibit different detected amplitudes due to additional cross-polarization (CP) blocks that connect different spins in multidimensional experiments. We numerically and experimentally investigate how CP blocks bias backbone φ torsion angle determination and propose CP conditions that minimize this effect, thereby improving accuracy. Applying these conditions in pseudo-4D (H)CANH experiments yields improved agreement of the extracted angles with X-ray crystallographic data for microcrystalline chicken α-spectrin SH3. For the influenza A M2 membrane protein, we identify an unexpected backbone dihedral angle for the I32 residue, which is consistent with TALOS-N predictions but deviates from ideal α-helical transmembrane geometry.

Early diagnosis of oral squamous cell carcinoma (OSCC) remains challenging, with survival largely stage-dependent at presentation. Artificial intelligence (AI) promises to enhance detection and clinical decision-making across clinical photographs, radiology, optical imaging, and digital pathology. This narrative review synthesizes peer-reviewed PubMed-indexed English-language studies up to October 2025, prioritizing prospective designs, external validation, and clinically interpretable models. We focus on tasks relevant to clinicians: lesion triage from clinical images, prediction of nodal metastasis on CT/MRI/PET, margin assessment with optical modalities, and histopathology-based diagnosis/grading. We also discuss implementation issues: dataset shift, bias, and reporting standards. In clinical photographs, deep learning achieves high diagnostic accuracy for OSCC and oral potentially malignant disorders (OPMD) classification in single-center studies and shows promising generalization with multi-site external testing, yet performance still degrades on out-of-distribution images and under real-world artifacts. In radiology, radiomics and deep learning models improve risk stratification and prediction of cervical nodal metastasis beyond conventional imaging, particularly with multimodal feature fusion. Optical methods such as hyperspectral spatial frequency domain imaging and OCT combined with AI show feasibility for intraoperative margin assessment and in-clinic triage. Digital pathology models on whole-slide images approach expert-level classification for OSCC diagnosis and are beginning to predict malignant transformation risk in oral epithelial dysplasia; however, rigorous prospective validation remains scarce. AI systems for OSCC are maturing and clinically oriented. Before routine adoption, studies must demonstrate external validity, clinician-in-the-loop performance, calibration, and impact on time-to-diagnosis and patient outcomes. Pragmatic trials and transparent reporting are essential to move beyond proof-of-concept into equitable clinical benefit.

Low-complexity equalization of Zak-OTFS in the frequency domain

Abstract Traditional water pipeline leakage identification has several limitations, such as incomplete signal feature extraction and the underutilization of physical information. To overcome these limitations, this study proposes Multimodal Two-stream Feature-fusion with Two-stage Convolutional Neural Network (MTFT-CNN). The analysis reveals that local dynamic features in time-frequency spectrograms complement global static features in the time and frequency domains. Accordingly, a multimodal two-stream architecture is developed to extract global statistical features concurrently from vibration–acoustic signals and local dynamic features from spectrograms. The cross-modal bidirectional attention module enables the adaptive fusion of these heterogeneous representations. The two-stage classification strategy enhances recognition efficiency: in the first stage, a hard-gating mechanism filters out non-leakage samples, while in the second stage, a multi-task learning framework, guided by the physical correlation between pressure and aperture, achieves the fine-grained identification of operating conditions. The experimental results show accuracies of 99.15%, 97.75% and 96.63% for leakage state, pressure and aperture recognition, respectively, with an overall accuracy of 98.37%. Ablation and comparative analyses confirm that MTFT-CNN significantly improves the comprehensiveness, precision and engineering applicability of leakage information identification, offering an effective solution for high-precision and high-efficiency leakage information identification in water supply pipelines.

ABSTRACT Multi-variety and small-batch production is widely used in modern manufacturing. Monitoring the machining process in such systems is crucial for improving efficiency and reducing costs. However, most existing monitoring methods rely on intrusive sensors which often require equipment modifications and increase costs. Additionally, many methods can only determine the machining state after all operations on a machine are completed, making real-time monitoring difficult. The small sample size characteristic also limits the effectiveness of conventional machine learning approaches.Therefore, a real-time machining state monitoring method based on transfer learning and power signal is proposed. The input power of the machine tool is used as the original data, whose acquisition is non-intrusive. The power signal is segmented using the Bayesian online change point detection algorithm and converted into recurrence plots for state monitoring model. A transfer learning–based model is then developed to identify machining steps and processes using only a small amount of training data. Power signal features are extracted using frequency domain analysis and recurrence quantification analysis. Finally, machining anomalies are detected using a Z-score variant algorithm. Case studies show the proposed method achieves a monitoring accuracy of 97.1%, enabling effective real-time state monitoring in multi-variety and small-batch production systems.