Search Results (419)

Electroencephalogram (EEG) signals serve as the core information carrier for brain–computer interfaces (BCIs); however, their highly non-stationary nature, extremely low signal-to-noise ratio, and significant inter-individual variability pose considerable challenges for signal decoding. Existing deep learning methods struggle to strike a balance between multi-scale, fine-grained feature extraction and efficient long-range temporal modeling. To overcome this limitation, this study proposes a novel deep cascaded architecture, MSResBiMamba, which deeply integrates multi-scale spatiotemporal feature learning with cutting-edge long-sequence modeling techniques. The model first utilizes an enhanced multi-scale spatiotemporal convolutional network (MS-CNN) combined with a SE-channel attention mechanism to adaptively extract local multi-band features and dynamically suppress redundant artefacts. Subsequently, it innovatively introduces an enhanced bidirectional Mamba (Bi-Mamba) module to efficiently capture non-causal long-range temporal dependencies with linear computational complexity, whilst cascading multi-head self-attention mechanisms to establish global higher-order feature interactions. Extensive experiments on the BCI Competition IV-2a dataset demonstrate that MSResBiMamba achieves outstanding classification performance in multi-class motor imagery tasks, significantly outperforming traditional methods and existing state-of-the-art neural networks. Ablation studies and t-SNE visualisations further confirm the model’s robustness in feature decoupling and cross-subject applications, providing a high-precision, high-efficiency decoding solution for BCI systems. Full article

Background/Objectives: Alzheimer’s disease (AD) is an irreversible neurodegenerative disorder that imposes a profound burden on global public health. While resting-state functional magnetic resonance imaging (rs-fMRI)-based dynamic functional connectivity (dFC) analysis has demonstrated promise in capturing time-varying brain network abnormalities, existing deep learning methods suffer from three fundamental limitations: (1) an inability to model temporal dependencies across dynamic connectivity windows, (2) reliance on post hoc black-box explainability tools, and (3) misalignment between feature learning and classification objectives. Methods: To address these challenges, we propose MambaKAN, an end-to-end interpretable framework integrating a Variational Autoencoder (VAE), a Selective State Space Model (Mamba), and a Kolmogorov–Arnold Network (KAN). The VAE encodes each dFC snapshot into a compact latent representation, preserving nonlinear connectivity patterns. The Mamba encoder captures long-range temporal dynamics across the sequence of latent representations via input-selective state transitions. The KAN classifier provides intrinsic interpretability through learnable B-spline activation functions, enabling direct visualization of how latent features influence diagnostic decisions without post-hoc approximation. The entire pipeline is trained end-to-end with a joint loss function that aligns feature learning with classification. Results: Evaluated on the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset across five classification tasks (CN vs. AD, CN vs. EMCI, EMCI vs. LMCI, LMCI vs. AD, and four-class), MambaKAN achieves accuracies of 95.1%, 89.8%, 84.0%, 86.7%, and 70.5%, respectively, outperforming strong baselines including LSTM, Transformer, and MLP-based variants. Conclusions: Comprehensive ablation studies confirm the indispensable contribution of each module, and the three-layer interpretability analysis reveals key temporal patterns and brain regions associated with AD progression. Full article

As urbanization transitions from incremental expansion to the optimized utilization of existing construction land, the precise identification of land-use status and changes has become a core requirement for enhancing refined land resource management. However, in urban built environments characterized by dense object distributions and complex geometric contours, existing change detection methods often struggle to capture subtle boundaries, leading to edge blurring and loss of detail. To address these challenges, this study proposes an Edge-aware Change Detection Network for urban construction land change identification. The model features a shared Siamese encoding network based on MiT-B1, leveraging its hierarchical multi-scale attention mechanism to balance local detail extraction with long-range semantic dependency capture, thereby overcoming the limitations of monolithic feature extraction. Furthermore, a multi-level feature concatenation and fusion strategy is designed to align and interact with bi-temporal features along the channel dimension, significantly enhancing the saliency and discriminative representation of change areas. Experimental results on the Yongzhou building change detection dataset demonstrate that the proposed model outperforms state-of-the-art methods in both visual recognition and quantitative metrics. It effectively resolves the difficulty of boundary definition in complex urban scenarios, providing localized high-precision technical support for the assessment and dynamic monitoring of construction land within the study area. Full article

Accurate forecasting of wind and photovoltaic power remains challenging due to the strong nonlinearity, nonstationarity, and seasonal heterogeneity of renewable generation series. To address this issue, this study proposes a hybrid forecasting framework integrating time–frequency joint analysis (TFAA), temporal convolutional networks (TCN), long short-term memory (LSTM), and multi-head self-attention (MHSA). Wavelet transform is used to extract frequency-domain representations, which are jointly encoded with the original time-domain sequence through a dual-branch architecture and adaptively fused. The fused features are then processed by a TCN-LSTM backbone to capture both long-range dependencies and short-term dynamics, while MHSA is introduced to enhance global contextual modeling. Experiments on wind-farm and photovoltaic datasets from China, together with external validation on the NREL WIND Toolkit and the GEFCom2014 Solar benchmark, show that the proposed model achieves the best overall seasonal performance and maintains competitive improvements on public benchmarks. Additional ablation studies, repeated-run statistical validation, persistence-based skill-score analysis, prediction-interval evaluation, ramp-event assessment, meteorological-driver enrichment, permutation-based driver attribution, regime-conditioned error diagnostics, and transferability evidence analysis further confirm the effectiveness, robustness, physical consistency, and practical applicability of the proposed framework. The results indicate that the proposed model provides a reliable and operationally relevant solution for short-term wind and photovoltaic power forecasting. These findings further support sustainable renewable-energy integration, smart-grid dispatch, and low-carbon power-system operation. Full article

High-resolution remote sensing imagery encompasses complex land cover types and rich textural details, whilst temporal variations often manifest as subtle feature differences and unstable structural patterns. This renders traditional change detection methods ineffective at accurately characterizing genuine alterations, frequently leading to underdetection, false positives, and ambiguous boundaries. To address these challenges, this paper proposes a Transformer-Based Multi-level Adaptive Fusion Network. It is built upon the DeepLabV3+ encoder–decoder framework, in which a shared-weight ResNet-101 is adopted as the backbone for dual-temporal feature extraction, with the final residual block of layer 4 cropped to extract deeper semantic features at a higher spatial resolution. The Adaptive Window–Attention Feature Fusion Module (AWAFM) adaptively models local and global differences across temporal phases, enhancing sensitivity to genuine changes. The Dual Strip Pool Fusion Module (DSPFM) enhances sensitivity to directional structural variations through horizontal and vertical strip pooling. The Progressive Multi-Scale Feature Fusion Module (PMFFM) progressively aggregates deep and shallow features via semantic residual transmission. To further suppress misleading suppression caused by complex textures, the Transformer-Enhanced Reverse Attention Fusion Module (TRAFM) explicitly models long-range dependencies, effectively mitigating false change responses. On the LEVIR-CD dataset, it achieves state-of-the-art performance, with a PA and an IoU of 92.36% and 90.13%, respectively. On the SYSU-CD dataset, PA and IoU reach 88.96% and 86.15%, demonstrating TMAFNet’s stability and superiority in scenarios involving complex ground surface disturbances, weak textural variations, and large-scale structural changes. Full article

Colmation phenomena play a critical role in long-term gas flow through porous media, significantly influencing methane migration, mine ventilation efficiency, and emission control in both active and abandoned coal mines. In colmation modeling, three fundamental kinetic types are commonly distinguished, with the third kinetic providing a generalized nonlinear formulation capable of describing state-dependent and spatially variable permeability degradation. However, the strong nonlinearity of the coupled transport–colmation equations prevents the derivation of closed-form solutions, which necessitates the application of linearization techniques. In this study, gas flow with colmation governed by third-kinetics is analyzed with particular emphasis on methane migration in underground mining environments. Linearization of nonlinear kinetic terms is applied at the level of the coupled mass balance and colmation equations, resulting in an approximate form of Darcy’s law and an explicit analytical solution describing the evolution of the porous medium state. The primary objective of the study is to quantify the error introduced by the adopted linearization and to analyze its spatial and temporal propagation with respect to the nonlinear reference solution. A rigorous error estimation based on Taylor series truncation is developed, yielding an explicit criterion that defines the validity range of the linearized solution. The results demonstrate that the approximation remains reliable within the regime of weak colmation, while the associated error is locally generated and propagates through transport mechanisms without exhibiting uncontrolled growth. Full article

Skeleton-based action recognition is a critical technology for intelligent sports analysis. Although the human skeletal structure exhibits inherent bilateral symmetry, sensor noise on resource-constrained edge devices frequently induces geometric distortion and topological asymmetry. Consequently, achieving a balance between high accuracy and real-time performance remains a significant challenge. To this end, we propose EMS-GCN, an Efficient Multi-scale Shift Graph Convolutional Network that integrates geometric priors. Specifically, we design a Gaussian kernel-driven topology refinement module to mitigate structural noise inherent in sensor data. By leveraging geometric symmetry and Gaussian distances among nodes, this module dynamically constrains graph topology learning, thereby effectively rectifying the structural asymmetry and ambiguity induced by noise. Furthermore, we construct a Multi-scale Shift Linear Attention (MSLA) module to replace computationally intensive temporal convolutions. Leveraging temporal shift invariance, this module captures multi-scale contexts via parameter-free shift operations. Furthermore, we introduce a linear temporal attention mechanism to model global temporal dependencies with linear complexity, effectively resolving the information asymmetry inherent in long-range interactions. Finally, EMS-GCN incorporates a dual-branch attention structure to adaptively calibrate feature responses. Extensive experiments demonstrate that our model maintains high recognition accuracy with only 0.56M parameters, representing a reduction of over 60% compared to mainstream baselines. These results validate the efficacy of leveraging geometric and temporal symmetries to enhance real-time sports analysis. Full article

This study proposes a unified multimodal temporal motion state perception framework for optical imaging-oriented biomedical applications, integrating visual skeleton sequences, inertial measurement unit (IMU) signals, and surface electromyography (EMG) signals. The framework utilizes modality-specific encoders and a cross-modal temporal alignment attention mechanism to explicitly model temporal offsets from heterogeneous sensing streams. A multimodal temporal Transformer backbone is introduced to capture long-range motion dependencies and cross-modal interactions, while an uncertainty-aware fusion module dynamically allocates weights based on modality confidence. Experimental results demonstrate that the proposed approach achieves an accuracy of 94.37%, an F1-score of 93.95%, and a mean average precision of 96.02%, outperforming mainstream baseline models. Robustness evaluations further confirm stable performance under visual occlusion and sensor noise. These results indicate that the framework provides a highly accurate and robust solution for rehabilitation assessment, sports training monitoring, and wearable intelligent interaction systems. Full article

The reliability of high-frequency (HF) frequency selection technology relies on the prediction accuracy of the Maximum Usable Frequency of the ionospheric F2 layer (MUF-F2). To improve its short-term prediction performance, a novel hybrid deep learning prediction model is proposed, which achieves accurate modeling of the complex spatiotemporal variation patterns of MUF-F2 by integrating a feature enhancement mechanism, a dual-branch feature extraction structure, and a bidirectional temporal dependency capture network. The hybrid prediction model integrates the Channel Attention mechanism (CA), Dual-Branch Convolutional Neural Network (DCNN), and Bidirectional Long Short-Term Memory network (BiLSTM). The model is trained and validated using MUF-F2 data from 5 communication links over China during geomagnetically quiet periods and 4 during geomagnetic storm periods, with the difference in the number of links attributed to experimental constraints and the disruptive effects of geomagnetic storms. Its performance is evaluated via multiple metrics, and a comparative analysis is conducted with commonly used prediction models such as the Long Short-Term Memory (LSTM) network. Experimental results show that during geomagnetically quiet periods, the proposed model achieves lower prediction errors (Root Mean Square Error (RMSE) < 1.1 MHz, Mean Absolute Percentage Error (MAPE) < 3.8%) and a higher goodness of fit (coefficient of determination (R²) > 0.94), with the average error reduction across all links ranging 8 from 6.2% to 46.9% compared with the baseline model. Under geomagnetic storm disturbance conditions, the model still maintains robust prediction performance, with R² > 0.89 for all communication links, as well as RMSE < 0.6 MHz, Mean Absolute Error (MAE) < 0.4 MHz, and MAPE < 3.3%. The study demonstrates that the proposed CA-DCNN-BiLSTM model exhibits excellent prediction accuracy and anti-interference capability under different geomagnetic activity conditions, which can effectively improve the short-term prediction accuracy of MUF-F2 and provide more reliable technical support for HF communication frequency decision-making. Full article

Millimeter-wave (mmWave) radar point clouds offer a privacy-preserving solution for Human Activity Recognition (HAR), but their inherent sparsity and noise limit single-modal performance. While multimodal fusion mitigates this issue, existing methods often suffer from severe negative transfer during visual degradation and incur high computational costs, unsuitable for edge devices. To address these challenges, we propose Tac-Mamba, a lightweight cross-modal state space model. First, we introduce a topology-guided distillation scheme that uses a Spatial Mamba teacher to extract structural priors from visual skeletons. These priors are then explicitly distilled into a Point Transformer v3 (PTv3) radar student with a modality dropout strategy. We also developed a Trust-Aware Cross-Modal Attention (TACMA) module to prevent negative transfer. It evaluates the reliability of visual features through a SiLU-activated cross-modal bilinear interaction, smoothly degrading to a pure radar-driven fallback projection when visual inputs are corrupted. Finally, a Lightweight Temporal Mamba Block (LTMB) with a Zero-Parameter Cross-Gating (ZPCG) mechanism captures long-range kinematic dependencies with linear complexity. Experiments on the public MM-Fi dataset under strict cross-environment protocols demonstrate that Tac-Mamba achieves competitive accuracies of 95.37% (multimodal) and 87.54% (radar-only) with only 0.86M parameters and 1.89 ms inference latency. These results highlight the model’s exceptional robustness to modality missingness and its feasibility for edge deployment. Full article

Dew constitutes a component of the near-surface water balance, but its large-scale fractal dynamical properties remain poorly documented across Africa. This study estimates dew amounts and investigates their fractal and multifractal behavior under African climatic conditions using gridded ERA5 datasets from 1993 to 2022. The Rescaled-Range (R/S) method, Multifractal Detrended Fluctuation Analysis (MFDFA), and the Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) algorithm are used. Hurst exponent (Hu) and the multifractal spectrum width (ω) are evaluated at daily and monthly scales over the full period and two sub-periods (1993–2007 and 2008–2022). The results reveal pronounced spatial heterogeneity in dew distribution. Daily mean amounts range between 0 and 0.18 mm, corresponding to annual accumulations reaching up to ~85 mm·yr⁻¹ in humid coastal, equatorial, and sub-equatorial regions, while remaining below 0.5 mm·yr⁻¹ in hyper-arid deserts. The continental mean annual amount is ~35.5 mm·yr⁻¹. The Hurst exponent exhibits values between zero and one, indicating region-dependent persistent and anti-persistent behaviors. This suggests that prediction schemes based on preceding values may be suitable for dew time series prediction in African regions exhibiting persistent characteristics. The multifractal spectrum width (ω), reaching values of up to 10, highlights strong scaling heterogeneity, particularly at the monthly timescale. These findings indicate that African dew dynamics exhibit significant long-range dependence and multifractal variability, providing new insights into the intrinsic temporal structure of dew and into appropriate approaches for its forecasting. Full article

Millimeter-wave radar has gradually gained popularity as a sensor mode for Human Activity Recognition (HAR) in recent years because it preserves the privacy of individuals and is resistant to environmental conditions. Nevertheless, the fast inference of high-dimensional and sparse 4D radar data is still difficult to perform on low-resource edge devices. Current models, including 3D Convolutional Neural Networks and Transformer-based models, are frequently plagued by extensive parameter overhead or quadratic computational complexity, which restricts their applicability to edge applications. The present paper attempts to resolve these issues by introducing RadarSSM as a lightweight spatiotemporal hybrid network in the context of radar-based HAR. The explicit separation of spatial feature extraction and temporal dependency modeling helps RadarSSM decrease the overall complexity of computation significantly. Specifically, a spatial encoder based on depthwise separable 3D convolutions is designed to efficiently capture fine-grained geometric and motion features from voxelized radar data. For temporal modeling, a bidirectional State Space Model is introduced to capture long-range temporal dependencies with linear time complexity $O\left(T\right)$, thereby avoiding the quadratic cost associated with self-attention mechanisms. Extensive experiments conducted on public radar HAR datasets demonstrate that RadarSSM achieves accuracy competitive with state-of-the-art methods while substantially reducing parameter count and computational cost relative to representative convolutional baselines. These results validate the effectiveness of RadarSSM and highlight its suitability for efficient radar sensing on edge hardware. Full article

Power system blackouts represent typical manifestations of instability in complex systems, whose evolution often exhibits non-stationarity, long-range correlations, and nonlinear scaling behavior. Most reliability assessment methods widely used in engineering practice are built on the core assumptions of event independence and light-tailed distribution, which will inevitably lead to systematic underestimation of extreme tail risks when blackouts actually present long-range memory and power-law heavy-tailed characteristics. Based on long-cycle historical blackout records of China’s power grid spanning 1981–2025, this paper develops an integrated framework combining Self-Organized Criticality (SOC) theory, Hurst exponent analysis, symbolic time-series methods, and Multifractal Detrended Fluctuation Analysis (MFDFA). This study systematically characterizes the evolution law and inherent dependence structure of blackout events from four dimensions: statistical scaling, temporal correlation, nonlinear structure, and multi-scale fractal spectrum. The results show that both the load-loss magnitudes and inter-event intervals of blackouts follow strict power-law distributions, with the system exhibiting scaling behavior consistent with SOC theory. The blackout event sequence presents significant long-range positive correlation and self-similarity, confirming a persistent long-term memory effect in the system evolution. Symbolic analysis further reveals the nonlinear fluctuation patterns and burst clustering behavior of the blackout process, reflecting the intermittency and complexity of blackout risks. MFDFA results verify that the blackout sequence has a broad-spectrum multifractal structure across different temporal scales, and Monte Carlo shuffle tests demonstrate that this multifractality mainly arises from intrinsic long-range temporal correlations, rather than being driven solely by heavy-tailed distribution. This study confirms that blackouts in China’s power grid are not random independent events, but present fractal statistical characteristics consistent with the self-organized critical mechanism. The findings provide a novel fractal perspective and quantitative framework for the statistical characterization, operational security assessment, and multi-scale early-warning modeling of blackout risks in China’s large-scale power systems. Full article

Against the backdrop of large-scale integration of renewables into the power grid, frequent load-following operation of thermal power units substantially increases the difficulty of controlling boiler NOx emissions. Accurate forecasting of boiler NOx emissions is crucial for guiding efficient and clean operation under such flexible operating conditions. However, under frequent load-following conditions, NOx dynamics are highly nonlinear and non-stationary, making it challenging to achieve accurate prediction using only time-domain information. To address these issues, we propose DTTE-Net, a time–frequency feature fusion framework for predicting SCR-inlet NOx concentration in coal-fired boilers. DTTE-Net consists of three components: a time-domain branch, a frequency-domain branch, and a gated feature fusion module. The time-domain branch captures short-term fluctuations and long-range temporal dependencies, while the frequency-domain branch extracts complementary spectral representations to enhance the characterization of non-stationary fluctuations. The gated feature fusion module then adaptively integrates the two-domain features by using a gated mechanism and produces the NOx concentration forecast. In addition, a Gaussian kernel-based loss is introduced to improve robustness to nonlinear error structures. Experiments on real distributed control system data from a 660 MW ultra-supercritical coal-fired unit show that DTTE-Net outperforms existing baseline models, achieving lower forecasting errors and higher R². Full article

The rapid advancement of deepfake generation technologies presents significant challenges to the verification of digital video authenticity. These time-dependent artifacts are difficult to detect using conventional frame-based detection approaches. This paper introduces AT-HSTNet, an Action-Transformer-based Hierarchical Spatiotemporal Network designed for robust and computationally efficient deepfake video detection. The proposed framework adopts a multi-stage hierarchical architecture in which frame-level visual features are extracted using an EfficientNet-B0 backbone, short- and medium-range temporal patterns are modeled through Bidirectional Long Short-Term Memory (BiLSTM) networks, and long-range temporal dependencies are captured using an action-aware Transformer operating on temporally aggregated representations. Unlike conventional video transformers that apply self-attention directly to raw frame-level features, the proposed action-aware attention mechanism reduces redundant computation and improves stability in temporal reasoning. Extensive experiments on the balanced FFIW-10K dataset demonstrate that AT-HSTNet achieves an accuracy of 98.7%, with 98.0% precision, 96.0% recall, and a 96.9% F1-score, outperforming representative CNN–BiLSTM and CNN–Transformer baseline architectures. In addition, AT-HSTNet is highly efficient, requiring only 0.45 GFLOPs and achieving an inference speed of approximately 30 FPS on consumer-grade GPU hardware. As a result of this study, we found hierarchical temporal modeling more effective when combined with action-aware attention for any deepfake video detection. Full article