Search Results (48)

Long-term monitoring provides a unique opportunity to distinguish persistent ecological processes from short-term fluctuations in forest insect dynamics. However, the extent to which local insect populations persist through time and the mechanisms regulating their occupancy remain poorly understood. Using a 13-year dataset from the Polish ICP Forests network, we investigated temporal and spatial dynamics of insect occurrence in alder stands. We quantified persistence at tree and plot levels, evaluated signatures of density-dependent regulation, and assessed spatial structure using generalized additive models, logistic regression, transition analyses, and spatial autocorrelation metrics. Insect occurrence exhibited exceptionally strong temporal persistence. Occurrence in the previous year was the dominant predictor of current occurrence across all modeling approaches. Persistence patterns differed between plot and tree levels, with transition analyses revealing greater long-term stability at the level of monitoring plots, consistent with the long-term stability of local populations despite turnover among host trees. Occupancy dynamics were consistent with negative density-dependent regulation, with local populations fluctuating around a stable equilibrium occupancy of approximately 75%. Although year-to-year changes occurred regularly, most fluctuations were relatively small and rarely resulted in complete disappearance of populations from occupied locations. Insect occurrence also exhibited pronounced spatial structure. Occupancy remained significantly clustered throughout the study period, indicating persistent regional differences in local population abundance. In contrast, annual changes in occupancy showed only weak spatial autocorrelation, suggesting that short-term dynamics were driven primarily by local ecological processes rather than by highly synchronized regional fluctuations. Our results indicate that infestation patterns in alder stands are consistent with persistent and self-regulating local herbivore populations that remain associated with the same forest stands over extended periods. Long-term dynamics are governed primarily by temporal persistence, density-dependent feedback, and stable spatial structure rather than by repeated cycles of colonization and extinction. These findings highlight the importance of long-term monitoring for understanding population persistence and provide new insight into the processes maintaining herbivore populations in forest ecosystems. Full article

Earthquakes exhibit well-documented statistical regularities at the catalogue level, such as the Gutenberg–Richter magnitude–frequency relation and the Omori–Utsu aftershock decay, often interpreted as signatures of seismicity as a driven, dissipative system far from thermodynamic equilibrium. However, whether comparable signatures, such as scale invariance and anomalous diffusion, can be detected directly within individual ground motion recordings remains an open question. This work investigates whether acceleration, velocity, and displacement signals recorded during the 2009 $M_w$ 6.1 L’Aquila earthquake display statistical properties consistent with non-equilibrium complex systems, and whether different seismic phases carry distinct, reproducible statistical signatures. P- and S-wave onset times are estimated using AR-AIC, with adaptive search windows centred on theoretical arrivals from the CRUST1.0 velocity model. Coda onset is determined using three complementary criteria combined into a median ensemble, enabling the segmentation of each recording into up to five temporal windows. Displacement moment scaling is analysed for each window and signal type, within the framework of strong anomalous diffusion, yielding the scaling exponents $\zeta \left(q\right)$. Robustness is systematically assessed against the choice of coda onset method, the empirical thresholds defining coda onset and end, the sub-interval of $\tau$ used in the moment scaling fit, and the filter band applied to the ground motion signals. Evidence for anomalous diffusion is nuanced: both its sign and magnitude depend on the seismic phase, with only a subset of configurations remaining stable across all segmentation schemes tested. These results indicate that anomalous scaling signatures, when present, are not universal, and that systematic robustness analyses are essential to distinguish genuine physical effects from segmentation artefacts. Full article

Continuous monitoring of vineyard dynamics is essential for optimizing viticultural practices and assessing plant health. While the seasonal behaviors of satellite-derived vegetation indices are widely studied, robust parametric modeling of these temporal trends remains underexplored. Building upon initial clues derived from Italian vineyards, this study proposes a novel analytical framework based on the consistent parabolic temporal signature of optical and Synthetic Aperture Radar (SAR) indices. Focusing on the elevated Trinity Canyon Vineyards in Armenia, we model the yearly evolution and temporal aggregations of these indices using a parabolic fitting approach. Our results suggest that the parabola vertex, which we hypothesize corresponds to the absolute maximum of vegetative activity, remains remarkably stable across diverse vine types, satellite orbits, and years. While this stable behavior suggests an underlying phenological or structural consistency, distinct exceptions to this trend have also been identified and considered. Furthermore, to bridge the gap between remote sensing observables and agronomic traits, we investigated the relationship between the fitted parabolic parameters and the Winkler index, which is used here as an estimator of above-ground biomass (AGB). By correlating the vegetation indices’ temporal dynamics with biomass growth and by isolating specific anomalies driven by environmental or anthropogenic factors, this work offers a basis for a predictive methodology that enables tracking vineyard structural development. Full article

Valve leakage in marine oil and gas pipelines is a critical failure mode that threatens operational safety, ecological integrity and production economic benefits, creating an urgent demand for accurate, real-time and robust fault diagnosis systems. Acoustic Emission (AE) technology captures transient acoustic signatures generated by leakage to enable non-intrusive online monitoring, while deep learning supports intelligent analysis through automatic signal feature extraction. Nevertheless, traditional AE-based leakage diagnosis methods rely heavily on manual feature engineering and fixed signal processing rules. Existing AE-driven deep learning methods fail to simultaneously deliver high detection accuracy, low inference latency and strong noise immunity, hindering their practical deployment on offshore platforms. To address these limitations, this paper proposes a Parameter-free Star-shaped Attention Fusion Network (SAFNet) for lightweight valve leakage localization using AE signals. Centered on the Temporal Pyramid Encoder (TPE) and Progressive Lightweight Star-shaped Attention (PLSA) module, SAFNet integrates Dual Bilinear Star Mapping (DBSM), Energy-Driven Feature Refiner (EDFR) and Multi-Scale Gated Attention Fusion (MS-GAF) modules. This architecture achieves efficient multi-scale temporal feature extraction, parameter-free nonlinear enhancement, noise-resistant refined feature processing and adaptive hierarchical feature fusion. The proposed method is applicable to valve leakage diagnosis of marine oil and gas pipelines under variable pressure and complex marine noise conditions. Comprehensive experiments are conducted on a dataset constructed by combining laboratory controlled leakage signals with real marine background noise recorded from the Liwan 3-1 offshore platform. The experimental results reveal that SAFNet balances high detection accuracy, compact model size and low inference latency simultaneously. Specifically, the network maintains a stable detection accuracy above 95% under pipeline pressures ranging from 2 MPa to 5 MPa, and exhibits excellent stability under extreme heavy noise environments. Ablation experiments further validate the synergistic performance gain brought by all core modules. The presented network delivers an efficient lightweight solution for valve leakage localization under simulated marine acoustic conditions, promotes the development of intelligent monitoring technologies for marine pipeline systems, and comprehensively improves offshore operational safety and marine ecological protection capacity. Full article

Incremental automatic modulation recognition is essential for the awareness of complex electromagnetic environments but is prone to catastrophic forgetting. This is fundamentally precipitated by shortcut learning, a phenomenon where deep models prioritize stable but non-essential channel artifacts (e.g., noise, fading) over intrinsic modulation characteristics. Consequently, models rely on spurious correlations that collapse during incremental task updates or environmental shifts, leading to representation drift. To bridge this gap, we propose the ParalIMR framework, which integrates a parallel reconstruction architecture with the segment substitution (SS) strategy to decouple modulation signatures from environmental fingerprints. Specifically, the parallel branch utilizes a Denoising AutoEncoder (DAE) as a task-agnostic structural anchor, purifying feature representations and maintaining geometric consistency across varying signal-to-noise ratios without propagating noise-overfitting to the classifier. In the meantime, the SS strategy actively disrupts the temporal coupling between class labels and hardware fingerprints through random reorganization, forcing the model to extract modulation-invariant structural cues. Experimental results on the RML2016a datasets demonstrate that in a three-stage incremental setup, our method achieves an overall accuracy of 84.32% at 0 dB SNR, representing a 2.69% improvement over the iCaRL baseline. Notably, this advantage expanded to 4.46% on RML2018, demonstrating that ParalIMR effectively arrests catastrophic forgetting. Ultimately, this research provides a robust learning paradigm tailored for cognitive radio and electronic warfare in dynamic electromagnetic landscapes. Full article

Reliable fault diagnosis of rotating machinery is critical for averting serious failures in modern industrial systems. While data-driven deep learning has advanced condition monitoring, its success is fundamentally predicated on the availability of independent and identically distributed (I.I.D.) datasets. In realistic operational environments, machinery frequently experiences massive domain shifts induced by varying rotational speeds. Concurrently, acquiring high-fidelity fault instances is limited compared to abundant healthy baseline data, often resulting in a long-tailed distribution. Under such data-starved conditions, conventional few-shot domain adaptation (FSDA) methodologies often may be affected by distributional erasure; global alignment objectives are mainly driven by the healthy majority, causing sparse fault signatures to be erroneously absorbed as noise and leading to severe diagnostic performance degradation. To address this setting, this study develops a physics-guided dual-stream fusion framework for extreme few-shot cross-domain fault diagnosis. The method does not treat the Laplace wavelet, STFT, CNNs, or AdaBN as newly introduced techniques. Instead, it integrates these components into a unified diagnostic pipeline designed for long-tailed target support sets under large speed shifts. A learnable Laplace wavelet convolution is used in the temporal branch to emphasize transient impact responses, while STFT spectrograms provide a complementary time-frequency representation for the two-dimensional branch. The two feature streams are then fused for target fault classification. For domain adaptation, a Strict AdaBN strategy is applied using only the target support set, rather than the target test data or a large unlabeled target pool. Under the evaluated 50 healthy + 12 fault support condition, the healthy samples provide target-domain operating-background statistics for BN recalibration, while the limited fault samples are used for supervised classifier adjustment. Experiments on the HUSTbearing and Torino DIRG datasets show that the proposed integrated framework achieves stable performance under the evaluated few-shot cross-speed settings. These results suggest that combining physics-guided Laplace convolution, time-frequency representations, and support-set-restricted BN recalibration can be useful for bearing fault diagnosis when target fault samples are limited. Full article

Context: Accurate citrus crop mapping is essential for agricultural monitoring, production planning, and supply-chain management, particularly in Brazil, one of the world’s leading orange producers and the leading orange-juice exporter. Satellite image time series from Sentinel-2 provide rich spectral and temporal information for crop identification. However, citrus mapping remains challenging due to fragmented agricultural landscapes, cloud contamination, class imbalance, and spectral overlap with other vegetation classes. Problem: Conventional machine learning models often depend on handcrafted vegetation indices, while attention-based deep learning models may require larger datasets and can become unstable under geographically constrained conditions. Therefore, there is a need for a compact and robust deep learning architecture capable of extracting citrus phenological signatures directly from multispectral time-series data. Methods: This study evaluates a Spatio-Temporal Pixel-Set Encoder Convolutional Neural Network (PSE-CNN) for citrus crop classification in the immediate geographic regions of São João da Boa Vista and Mogi Guaçu, São Paulo, Brazil. MapBiomas Collection 10.1 data from 2019 to 2024 were used to derive reference polygons, and Sentinel-2 imagery was processed into cloud-masked, 15-day temporal composites using ten spectral bands. The proposed PSE-CNN was benchmarked against PSE-TAE, PSE-Transformer, Random Forest, and XGBoost using spatially grouped data partitioning and temporal test years. Results: The proposed PSE-CNN achieved the highest Unified F1-Score of $0.704$ and the lowest coefficient of variation of $3.03\%$, indicating stronger inter-annual stability across test years and random seeds among the evaluated models. It also outperformed classical models that relied on handcrafted vegetation indices and demonstrated greater overall stability than attention-based deep learning alternatives. Conclusions: The results indicate that combining pixel-set encoding with temporal convolution provides a resource-aware and stable framework for retrospective citrus crop mapping from Sentinel-2 satellite image time series. These findings suggest that PSE-CNN can support scalable agricultural monitoring, contributing to sustainable crop inventory systems in regions where labeled data and computational infrastructure are limited. Full article

Wild carrot (Daucus carota subsp. carota), the wild relative of cultivated carrot, is globally identified as an invasive weed that threatens hybrid carrot seed production through natural cross-pollination, resulting in compromised genetic purity. Manual identification across the large areas required to ensure genetic purity in carrot seed crops is impractical. Remote sensing offers an alternative; however, morphological similarities among wild carrot, cultivated carrot, and common weeds hinder reliable detection. Early identification, however, remains essential for preventing genetic contamination. This study evaluated leaf-level hyperspectral reflectance spectroscopy (400–2450 nm) with machine learning to discriminate wild carrot from hybrid carrots, parental lines, and 19 co-occurring weed species. Spectral data from 266 wild carrot plants across three New Zealand sites and six weeks (5–10 weeks after emergence) showed negligible spatial effects (R² = 0.034–0.055, pseudo-F = 1.46–2.39, p > 0.05) and moderate temporal variation (R² = 0.136–0.151, pseudo-F = 5.48–6.17, p < 0.001), indicating broadly stable spectral signatures suitable for model generalization. Savitzky–Golay filtering, with min–max normalization outperformed SNV, yielding high full-spectrum accuracies for wild carrot vs. other species (90.35%, κ = 0.80), wild carrot vs. weeds (96.03%, κ = 0.92), and a multi-class model (90.79%, κ = 0.88). After removing atmospheric water-absorption bands to follow airborne sensing, reduced-band models based on airborne-compatible wavelengths maintained strong performance, including 89.40% accuracy (κ = 0.79) for wild carrot vs. weeds using a 20-band Subspace Discriminant model (400–402, 527, 705–720 nm). These findings demonstrate that stable wild carrot spectra and carefully selected visible and red-edge bands can underpin cost-effective UAV/UGV-mounted hyperspectral or multispectral sensors for site-specific wild carrot management. Full article

A comprehensive multiregional characterization of the spectral response of cassava leaves across different ontogenetic stages was performed. For this, ultraviolet (UV), visible (VIS) and shortwave near-infrared (UV-VIS-NIR; 200–900 nm) regions were used to identify spectral signatures and indices for their potential use as biomarkers of leaf development and physiological status of plants under induced senescence conditions. Manihot esculenta Crantz (HMC-1 variety) was used as a model. Spectral signatures were obtained from leaves at two phenological stages (4 and 6 months after planting) using UV-VIS-NIR spectroscopy by the diffuse reflectance technique. Classical and experimental spectral indices were evaluated, and their discriminatory power through different ontogenies was assessed using ANOVA/Kruskal–Wallis and post hoc tests. Senescence effects were further examined by postharvest monitoring (1–20 days), with temporal, ontogenetic, and interaction effects validated using linear mixed models (LMMs), while multivariate structure and spectral convergence were explored via principal component analysis and hierarchical clustering (PCA-HCA). Functionally Enhanced Derivative Spectroscopy (FEDS), comparative analysis, and spectral correlation mapping allowed signal’s selective enhancement and the identification of phenolic compounds, photosynthetic pigments, and structural molecular components. Results showed high ontogenetic stability of UV-associated phenolic signals (~210–220 nm), whereas the VIS region (420–600 nm) clearly differentiated young leaves. The NIR region was stable across ontogeny but highly sensitive to temporal degradation, reflecting changes in water status and internal structure. UV-VIS-NIR indices effectively differentiated young leaves and changes by stress. It is concluded that multiregional characterization of the spectral response supported by FEDS allows the extraction of robust indices with strong potential as biomarkers of leaf maturation and senescence in cassava. Full article

Background: Sleep deprivation is one of the major public health and performance risk factors, with documented effects on vigilance, executive function, emotional regulation, and safety-critical behaviour. This review examines how event-related potentials (ERPs)—which provide millisecond-level resolution of cognitive processing stages—can clarify which neural processes are most affected by sleep loss, from early sensory encoding to later evaluative and control-related stages. Materials and Methods: This study was conducted as a systematic review of human studies on sleep deprivation and ERPs. Eligible studies included human participants, focused primarily on acute/total sleep deprivation, and reported ERP outcomes (e.g., amplitude, latency, topography, or related event-locked EEG measures). Searches were performed in major biomedical/psychology databases using sleep deprivation and ERP terms, with additional forward/backward citation searching. Data was extracted in a structured format (participant characteristics, deprivation protocol, ERP methods, behavioural outcomes, ERP findings, and recovery/countermeasure effects). Due to substantial heterogeneity in paradigms, protocols, and ERP methods, findings were synthesised narratively rather than meta-analysed. Risk of bias was assessed with RoB 2 and ROBINS-I. Results: The search identified 854 records, of which 82 studies were included following deduplication, screening, full-text review, and citation chasing. Samples were typically small, highly selected, and dominated by healthy young adults, with frequent attrition related to prolonged wakefulness and EEG data-quality constraints. Across studies, sleep deprivation produced stage-specific and task-dependent ERP effects rather than a single uniform pattern. The most consistent findings involved mid-to-late components. These components typically showed prolonged latency and reduced amplitude. In some cases, amplitude increases were observed and interpreted as compensatory recruitment. Early sensory/pre-attentive components (e.g., P1/N1/MMN/P50) were often relatively preserved, but showed selective vulnerability, including latency slowing, reduced filtering/gating, or decreased phase locking. A recurring observation was a behaviour–ERP dissociation, where ERP abnormalities were detectable even when behavioural impairment was modest, indicating covert neural inefficiency or compensation. Recovery sleep, naps, and countermeasures (e.g., modafinil, caffeine) produced partial, component-specific recovery, with amplitude and latency often recovering at different rates. Conclusions: The evidence indicates that sleep deprivation primarily disrupts higher-order, late-stage, and temporally coordinated neural processing, while earlier sensory processing is often preserved but becomes slower and less stable. Among ERP markers, the P300/P3 family is the most robust and informative signature of sleep loss effects and recovery. ERPs are therefore a sensitive tool for detecting neural dysfunction and compensation under sleep deprivation, including changes that may precede overt behavioural decline. Future research must improve the generalisability and reproducibility of ERP findings by employing larger, more diverse samples, alongside more standardised methodological, recording, and reporting practices. Full article

Circadian regulation has emerged as an important modulator of ocular growth; however, its role in organizing retina–choroid transcriptomic responses during myopia development remains incompletely understood. In this study, we reanalyzed publicly available retinal and choroidal RNA-seq datasets from chick models of form-deprivation myopia using a multistage machine learning framework. A biologically motivated ZT8/12 circadian window was defined from prior published time-of-day transcriptomic evidence and evaluated using feature selection, cross-tissue and cross-stage validation, and external validation in an independent retinal dataset. Machine learning models classified the ZT8/12 window with high performance across onset and progression datasets, and control analyses indicated that this signal reflects a broad transcriptome-wide temporal state rather than a pattern unique to the 53-gene signature. The final gene signature is therefore interpreted as a stable representative subset of the ZT8/12-associated expression state. Cross-species functional enrichment and ortholog mapping suggested hypothesis-generating functional relationships between chicken genes and human orthologs. Overall, this work provides a computational framework for evaluating time-associated expression patterns in myopia and highlights circadian timing as a candidate component of retina–choroid biology requiring further functional validation. Full article

When a fuel briquette is pressed using solar electricity in summer and burned for heating in winter, the briquette functions as a seasonal energy store—without batteries, self-discharge, or capital investment in storage infrastructure. This paper quantifies such “indirect energy storage” at an operating briquette production facility in Sumy, Ukraine, using 2024 operational data from a 34 kW hybrid solar power plant integrated into the production process without battery storage under continental climate conditions (50°55′ N) and full-scale military conflict. The objective was to determine the contribution of the solar power plant (SPP) to energy supply, analyse the structure of electricity consumption, and quantify the mechanism of indirect accumulation of renewable energy through transformation into solid biofuels. The study tested two hypotheses: (H1) that integration of a solar power plant into industrial daytime operation (6:00–22:00) achieves a self-consumption rate close to 100%, displacing grid electricity without curtailment or storage losses; and (H2) that the solar fraction embedded in produced briquettes constitutes a quantifiable mechanism of indirect seasonal energy storage despite a temporal mismatch between solar peaks (summer) and product demand (winter). Methods included statistical analysis of monthly and intraday operational data; Pearson correlation analysis between solar generation and production cycles; energy audit of production processes; decomposition of specific consumption into pressing and packaging components; and a simple economic assessment (NPV, IRR, LCOE, payback) with sensitivity analysis. Annual production reached 1222.975 t of briquettes. Total specific electricity consumption (including two short packaging campaigns in June and July only) was 141.3 ± 12.6 kWh/t (CV = 8.9%). After deducting 4962 kWh of dedicated packaging electricity (2.9% of annual consumption), the specific consumption for briquette pressing alone was 136.7 ± 5.0 kWh/t (CV = 3.7%)—within the European benchmark range of 80–150 kWh/t for wood densification, with tight monthly variation indicating a stable, well-tuned pressing operation throughout the year. The SPP supplied 18.3% of total annual electricity, peaking at 33.06% in May and averaging 29.95% from March to August. Intraday analysis of 530 five-minute intervals confirmed a 100% self-consumption rate across all seasons (H1 supported). A total of 223.4 t of briquettes containing accumulated solar energy were produced during the spring–summer period. A weak negative correlation (r = −0.28) between monthly SPP generation and briquette production was observed but did not reach statistical significance (p = 0.385); this descriptive—rather than causal—relationship is consistent with the expected temporal shift between summer surpluses and winter demand, and is itself a signature of indirect rather than direct energy coupling (H2 supported in a descriptive sense). The compound efficiency along the solar-to-stored-fuel chain was estimated at approximately 68%, providing a quantitative indicator for the indirect-storage concept. Economic analysis yielded a simple payback period of about 3 years, NPV (20 yr, 12%) ≈ 1.15 million UAH, IRR ≈ 33%, and LCOE ≈ 3.28 UAH/kWh—61% below the prevailing industrial tariff of 8.45 UAH/kWh—with sensitivity analysis showing positive NPV across ±20% variation in electricity price and ±15% in CAPEX. To the best of the authors’ knowledge, this is the first empirical quantification of biomass-solar integration as a seasonal energy buffer operating without battery storage. The solar energy accumulated in briquettes is sufficient to heat 56–74 households for a full winter season. Regional scaling of the present configuration—under explicit assumptions of comparable facility sizes and operating regimes—could in principle provide fuel for 15,000–20,000 households (8–12% of regional heating needs during energy crises). These findings are directly relevant to post-conflict energy recovery and to regions where attacks on energy infrastructure have left solid biofuels as the primary available heating source. Full article

Tracking multiple targets within a group is a challenging task in the radar field, especially for a bird flock. Targets in a group are usually closely spaced and exhibit similar characteristics. Additionally, the tracking radar typically employs a narrow beam to achieve a high range–angular resolution, resulting in incomplete measurements within the limited beamwidth. These factors lead to false association and track fragmentation in target tracking. However, in addition to kinematic characteristics, birds exhibit temporally correlated micro-Doppler signatures because of their wingbeat behavior, which can be utilized in target tracking. Therefore, this paper proposes an adaptive TPHD tracking method using Doppler features. First, a Doppler temporal contrastive network is designed to learn the micro-Doppler representation for the association of birds. Then, the learned feature is fused with kinematic parameters, using XGBoost to guide the weight update in the filter. Moreover, adaptive mechanisms are incorporated into the TPHD filter to achieve stable tracking under incomplete measurements. Simulation and experimental results verified the effectiveness of the proposed method and showed better tracking performance than the competing method. Full article

Bolt loosening is a critical precursor to structural failure in major industrial and transportation equipment. Although acoustic non-destructive testing (NDT) offers a cost-effective diagnostic solution, its practical deployment is often hindered by low signal-to-noise ratios (SNRs) and the limited ability of conventional models to isolate fine-grained transient acoustic signatures from complex background interference. To address these challenges, this paper proposes PCT-Net, a multi-scenario noise-adaptive fusion network for bolt-state recognition. First, an Adaptive Spectral Masking mechanism is introduced as a data augmentation strategy. Instead of rigid zero-padding, it dynamically blends local spectral energies to encourage the learning of more robust and noise-invariant representations. Furthermore, rather than simply concatenating multiple modules, PCT-Net adopts a synergistic feature extraction framework to decouple complex acoustic signatures. A perceptual frontend is used to establish acoustically meaningful representation priors. To handle the highly dispersed characteristics of loosening signals, cascaded convolutional modules progressively suppress redundant environmental interference while capturing high-frequency local transient impacts. Meanwhile, to overcome the limited receptive field of convolutional operations, an embedded Transformer mechanism is introduced to model long-range temporal dependencies and low-frequency structural variations throughout the tapping cycle. By integrating local fine-grained transient modeling with global structural dependency modeling, the proposed network can better distinguish subtle decision boundaries among different loosening states. Extensive experiments show that PCT-Net achieves a classification accuracy of 97.12% under standard conditions and maintains stable performance under severe noise scenarios. These results demonstrate the effectiveness of the proposed method and highlight its potential for intelligent industrial safety monitoring. Full article

Intelligent fault diagnosis of rolling bearings under high-noise industrial conditions remains a significant challenge. Traditional attention-based deep learning models often rely on global average pooling, which may inadvertently smooth out high-frequency transient impulses essential for fault identification, potentially leading to degraded performance in low signal-to-noise ratio (SNR) environments. To address this, we propose a Multi-Scale Temporal Coordinate Attention Network (MS-TCANet). The framework introduces a Peak-Aware Coordinate Attention (PACA) mechanism that combines max-pooling and average-pooling along directional coordinates. This dual-pooling design aims to better preserve transient impact features while maintaining a stable global representation, thereby mitigating the feature over-smoothing issue common in conventional attention modules. Additionally, an asymmetric multi-scale convolution block is incorporated to capture both short-term impacts and long-range periodic signatures. Experiments on three benchmark datasets (CWRU, Paderborn University, and XJTU-SY) indicate that the proposed MS-TCANet achieves favorable diagnostic accuracy compared to several representative and advanced methods, particularly under severe noise conditions (e.g., −10 dB SNR). t-SNE and Grad-CAM visualizations further suggest that the model can capture fault-related signatures more reliably than standard architectures in noisy environments. Full article