Search Results (44)

Skeleton-based action recognition has attracted considerable research interest because skeleton data are inherently robust to illumination changes, viewpoint variation, background clutter, and camera motion. Nevertheless, extracting informative representations from skeleton sequences remains a challenging problem, as it requires capturing both the spatial co-occurrence patterns among body joints and the fine-grained kinematic cues that distinguish different actions. In this paper, we propose a single-stream architecture that constructs an action-specific skeleton graph directly from motion and processes it with a kinematic-aware Transformer. Rather than relying on a fixed skeleton topology, a motion-guided dynamic-graph construction module infers a per-frame adjacency matrix from short-term motion cues through a differentiable edge predictor and Gumbel-Softmax sparsification, allowing the model to discover action-driven connections between distant joints that lack direct bone connectivity (e.g., coordinated hand motion during clapping). Each joint is described by kinematic node features that combine its 3D position, instantaneous velocity, and limb-angle encodings within a single descriptor, so that both motion dynamics and higher-order limb configurations are available to the spatial encoder from the outset. A graph-attention network (GAT) encodes the spatial configuration of every frame over the learned graph, and the resulting sequence of frame descriptors is processed by a Transformer encoder that models long-range temporal dependencies; a learnable classification token aggregates the sequence, and a multi-layer perceptron (MLP) produces the final action classification. The entire model is trained end-to-end from action labels alone. We conduct a comprehensive ablation study and evaluate the proposed method on the large-scale NTU RGB+D 60 and NTU RGB+D 120 benchmarks, where the results demonstrate that our approach achieves competitive performance compared to state-of-the-art architectures. Full article

Stock market price prediction remains a persistent challenge owing to the non-stationarity, high noise content, and intricate spatiotemporal dependencies that characterize financial time series. Existing approaches typically excel at either local pattern extraction or long-range dependency modeling, yet seldom reconcile both within a unified framework. This paper introduces a CNN–Transformer dual-channel architecture equipped with a dynamic attention fusion module for stock price forecasting. The convolutional channel applies hierarchical dilated convolutions to distill fine-grained local patterns from multi-indicator sequences while suppressing high-frequency noise. Simultaneously, the Transformer channel employs multi-head self-attention to capture long-distance temporal correlations and regime-shift dynamics. A learnable gating mechanism then fuses the two feature streams by adaptively weighting local detail against global trend information according to market conditions. Experiments conducted on four real-world stock datasets spanning the S&P 500, CSI 300, NASDAQ Composite, and Hang Seng Index show that the proposed model reduces mean absolute error by 9.7–15.3% and root mean square error by 9.5–13.8% relative to competitive baselines including LSTM, CNN–LSTM, Informer, and PatchTST. Ablation studies further indicate that both channels and the fusion module contribute to prediction accuracy, and the architecture remains effective across markets with differing volatility profiles. Full article

Accurate state of health (SOH) estimation of lithium-ion batteries is critical to the reliability of electric vehicles. However, under dynamic operating conditions, conventional feature extraction based on fixed time window often exhibits poor generalization, as it fails to account for the multi-timescale parameter couplings inherent in the non-stationary voltage responses. To address this issue, this paper proposes a Bayesian Adaptive Time Window Optimization (BATWO) framework for feature extraction in battery SOH estimation. Within this framework, the time window length is treated as a learnable structural parameter and is adaptively optimized via Bayesian optimization to identify the most informative observation timescale for extracting degradation-sensitive statistical features under given operating conditions. Evaluations on a cycle-aging dataset containing 69 lithium-ion battery samples subjected to distinct dynamic operating profiles show that the optimal time window lengths vary significantly, ranging from 500 s to 27,630 s. The BATWO framework achieves an average root-mean-square error (RMSE) of 2.07% and a mean absolute error (MAE) of 1.45%, outperforming the best fixed time window strategy by reducing the RMSE and MAE by 2.35% and 2.68%, respectively. Moreover, compared with LSTM- and Transformer-based models without feature extraction, the BATWO framework reduces training time by over 97%. These results highlight the superior generalization capability and computational efficiency of the BATWO framework, demonstrating its great potential for practical deployment in battery management system. Full article

Accurate air pollutant concentration prediction is crucial for public health and sustainable urban development. Existing methods predominantly rely on single-modal data, resulting in inadequate representation of pollutant spatiotemporal evolution, poor prediction accuracy, and limited generalization capabilities. To address these challenges, this research proposes a novel PM_2.5 prediction framework termed MFD-DF that integrates ground-station time series and satellite remote sensing images. In feature extraction, learnable decomposition and deformable convolution are introduced, and a Cross-Modal Slot Attention module explicitly decomposes features to resolve information blurring. Subsequently, a dynamic cross-modal alignment mechanism is designed alongside a learnable Time-Expansion Network (TEN) to ensure fine-grained interaction. Furthermore, a local-global attention feature fusion mechanism is proposed to optimize data integration efficacy. Experimental results demonstrate that in single-step PM_2.5 prediction tasks, the proposed MFD-DF achieves significant improvements of approximately 10–20% in MAE, RMSE, and MAPE compared to state-of-the-art baselines. In multi-step PM_2.5 prediction, it effectively alleviates the error accumulation problem in long-sequence forecasting, demonstrating superior robustness and accuracy. Full article

Blind image watermarking models such as HiDDeN have laid an important foundation for end-to-end watermarking. Nevertheless, they still suffer from three major limitations: single-scale feature extraction, fixed fusion weights, and slow training convergence. To address these issues, this paper proposes an adaptive multi-scale watermarking algorithm based on collaborative dual-attention mechanisms. The algorithm designs an adaptive multi-scale feature fusion module (MA-FFM) with a dynamic gating network in the encoder, which flexibly combines local multi-scale textures with global contextual information, overcoming the limitation of fixed fusion weights. In the decoder, a multi-level channel attention module is embedded to strengthen the extraction of watermark signals. The two attention modules work synergistically: the encoder focuses on adaptive feature fusion while the decoder leverages channel attention to selectively enhance watermark-related features, forming a dual-attention synergy that balances robustness and imperceptibility. Moreover, the dynamic gating network adaptively adjusts the contribution of local versus global features via learnable weights, whose evolution from approximately 0.51 to about 0.89 improves model interpretability. Experiments are conducted on the COCO 2017 dataset. Compared with HiDDeN, the proposed algorithm reduces the bit error rate (BER) from 0.1696 to 0.1538 under no attack with a relative reduction of 9.3%, increases PSNR by 0.61 dB, and improves SSIM from 0.9058 to 0.9077. Under various attacks—including JPEG compression, Gaussian noise, salt-and-pepper noise, and brightness/contrast adjustments—the BER remains consistently lower than that of HiDDeN. Ablation studies confirm the effectiveness of each module. Overall, the proposed algorithm preserves visual quality, improves the accuracy of watermark embedding and extraction, and exhibits strong generalization robustness against common image distortions. Full article

The linearization of wideband power amplifiers is critical for modern communication systems, yet modeling their severe nonlinearities and dynamic memory effects presents a significant data engineering challenge. Traditional polynomial models suffer from the curse of dimensionality, whereas deep neural networks entail high computational complexity and unstable convergence. To address these limitations, this paper proposes a novel physics-informed lightweight architecture, termed DPD-BLS, which integrates a block-oriented time delay structure with the Broad Learning System. Recognizing the distinct physical behaviors of radio frequency signals, the proposed model initially extracts temporal memory features and structurally decouples the signal magnitude and phase. To overcome the precision constraints of purely stochastic mapping in standard broad learning, we introduce an attentive dual-stream mapping module. This bifurcated architecture combines frozen random nodes for expansive state-space exploration with adaptive learnable nodes for precise error compensation, dynamically aggregating the most effective basis functions. Furthermore, an adaptive gating mechanism is incorporated to regulate nonlinear feature fusion, ensuring robust training stability. Comprehensive experiments demonstrate that the DPD-BLS achieves superior linearization performance while maintaining strict structural simplicity, offering a highly efficient data modeling paradigm for real-time edge deployment. Full article

Point cloud segmentation is a core technology in remote sensing, enabling the extraction of rich semantic information from complex scenes. Existing methods struggle with semantic inconsistency across multiple heterogeneous datasets in complex urban environments. To address semantic inconsistencies, we propose SemAlign3D, a novel multimodal framework for point cloud segmentation that combines learnable class prompts with a multi-scale feature attention module. We integrate five large-scale datasets (SensatUrban, STPLS3D, WHU3D, SemanticKITTI, Semantic3D) to construct a unified training framework, ensuring label consistency by recalibrating semantic labels. The learnable class prompt mechanism dynamically adapts to dataset-specific semantics, enhancing the semantic consistency across multiple datasets of point cloud segmentation. Additionally, the Multi-scale K-Nearest Neighbor Feature Attention Enhancement module integrates local and global features, improving semantic discriminability in complex scenes. Within a single unified training framework, our method effectively aligns semantic labels from multiple heterogeneous datasets, achieving gains of +1.61% mIoU on WHU3D and +0.98% mIoU on SemanticKITTI. These results demonstrate the effectiveness of our framework in improving semantic consistency and robustness across heterogeneous point cloud datasets. Full article

Encrypted network traffic classification is an important research topic in the field of network security. Although deep learning-based methods have made progress, they still face three main challenges: first, the semantic information in encrypted traffic is inadequately represented, making it difficult for existing methods to effectively capture the hierarchical interaction relationships between packet-level and flow-level features; second, models rely on large amounts of labeled data for supervised training, resulting in high training costs and limited generalization ability in new scenarios; third, in existing self-supervised methods, the functions of the encoder and decoder are coupled, which restricts the full potential of the encoder’s representation learning. To address these issues, this paper proposes an Enhanced Traffic Classifier (ETC) based on self-supervised feature learning. The model first constructs a multi-level interactive traffic representation matrix, converting raw traffic into structured grayscale images that fuse packet-level and flow-level temporal features, thereby addressing the problem of missing semantic information. On this basis, an improved Masked Image Modeling Vision Transformer architecture is adopted. Through a three-stage decoupled design of encoder–regressor–decoder, the encoder focuses solely on feature extraction, the regressor performs masked representation prediction, and the decoder is only responsible for image reconstruction, thereby fully unleashing the encoder’s feature learning capability. Furthermore, during the fine-tuning stage, an Attentive Probing classification mechanism is introduced to replace the traditional linear classification head. By using learnable class query vectors to dynamically focus on semantic regions relevant to the classification target, the model’s recognition accuracy and robustness are further improved. Experiments are conducted on five public datasets, including USTC-TFC2016 and CICIoT2022, as well as a self-built Human-Internet dataset. The results show that ETC significantly outperforms mainstream methods such as YaTC and ET-BERT in core metrics including accuracy and F1-score, while also demonstrating strong generalization in few-shot scenarios. Full article

Pedestrian detection from unmanned aerial vehicles (UAVs) holds significant value in security surveillance and emergency response applications. While visible-infrared (RGB-IR) fusion technology demonstrates potential in handling complex lighting conditions through cross-modal information complementarity, current mainstream fusion mechanisms still suffer from two evident shortcomings: (1) Existing approaches insufficiently account for the significant differences in noise distribution between infrared and visible images under varying imaging conditions, leading to unstable feature representations and posing fundamental challenges to subsequent effective fusion; and (2) Existing fusion strategies lack dynamic adaptability to features from different modalities, making it difficult to fully exploit complementary key information across modalities. To address these issues, this paper proposes a novel Laplacian Frequency Enhanced DETR (LF-DETR). The core innovations are threefold: (1) A Laplacian of Gaussian feature enhancement module is designed to independently enhance features in the visible and infrared branches at the early stage of feature extraction, effectively improving the representation quality of each modality. (2) A learnable frequency-domain fusion module is constructed to achieve adaptive complementary fusion of cross-modal features. (3) A dual-domain collaborative framework is proposed to integrate the above modules within a unified DETR architecture for RGB-IR pedestrian detection. Experimental results on the public RGBTDronePerson, VTUAV-det and DVTOD datasets demonstrate that LF-DETR achieves state-of-the-art performance, with particularly significant detection gains in challenging scenarios such as nighttime and low-light conditions, validating the effectiveness and superiority of the proposed method. Full article

The propagation of uncertainties in structural dynamic responses, arising from variations in material properties, geometry, and boundary conditions, is of critical concern to researchers in a variety of engineering instances. Conventional methods like high-fidelity Monte Carlo simulation are computationally prohibitive, while existing surrogate models can improve efficiency at the expense of accuracy. To achieve a trade-off between accuracy and efficiency, a Physics-Informed Conditional Variational Autoencoder (PI-CVAE) model is proposed. It integrates a novel dual-branch encoder for time-frequency feature extraction, a learnable frequency-filtering decoder, and a holistic physics-informed loss function so as to enable efficient generation of dynamic responses with high accuracy and adequate physics consistency. Comprehensive numerical analysis of plate structures demonstrates that the proposed approach achieves remarkable accuracy (maximum FRF error < 0.2% and $R^2$ > 0.99) and a computational speedup of 8–11 times in comparison with conventional simulation techniques. By maintaining high accuracy while efficiently propagating uncertainties, the PI-CVAE model provides a practical framework for probabilistic vibration analysis, especially during the acoustic design phase. Full article

Reciprocating compressors, which serve as core equipment in the petrochemical and natural gas transmission sectors, operate under prolonged variable loads and high-frequency impact conditions. Critical components, such as valves and piston rings, are prone to failure. Existing fault diagnosis methods suffer from inadequate spatio-temporal feature extraction and neglect spatio-temporal correlations. To address this, this paper proposes a spatio-temporal feature fusion-based fault diagnosis method for reciprocating compressors. This method constructs a spatio-temporal feature fusion model (STFFM) comprising three principal modules: First, a spatio-temporal feature extraction module employing a multi-layered stacked bidirectional gated recurrent unit (BiGRU) with batch normalisation to uncover temporal dependencies in long-term sequence data. A graph structure is constructed via k-nearest neighbours (KNN), and an enhanced graph isomorphism network (GIN) is integrated to capture spatial domain fault information variations. Second, the spatio-temporal bidirectional attention-gated fusion module employs a bidirectional multi-head attention mechanism to enhance temporal and spatial features. It incorporates a cross-modal gated update mechanism and learnable weight parameters to dynamically retain the highly discriminative features. Third, the classification output module enhances the model’s generalisation capability through multi-layer fully connected layers and regularisation design. Research findings demonstrate that this approach effectively integrates spatio-temporal coupled fault features, achieving an average accuracy of 99.14% on an experimental dataset. This provides an effective technical pathway for the precise identification of faults in the critical components of reciprocating compressors. Full article

Surface defect evaluation in steel production demands both high inference speed and accuracy for efficient production. However, existing methods face two critical challenges: (1) the diverse dimensions and irregular morphologies of surface defects reduce detection accuracy, and (2) computationally intensive feature extraction slows inference. In response to these challenges, this study proposes an innovative network based on dual-branch feature enhancement and downsampling (DFED-Net). First, an atrous convolution and multi-scale dilated attention fusion module (AMFM) is developed, incorporating local–global feature representation. By emphasizing local details and global semantics, the module suppresses noise interference and enhances the capability of the model to separate small-object features from complex backgrounds. Additionally, a dual-branch downsampling module (DBDM) is developed to preserve the fine details related to scale that are typically lost during downsampling. The DBDM efficiently fuses semantic and detailed information, improving consistency across feature maps at different scales. A lightweight dynamic upsampling (DySample) is introduced to supplant traditional fixed methods with a learnable, adaptive approach, which retains critical feature information more flexibly while reducing redundant computation. Experimental evaluation shows a mean average precision (mAP) of 81.5% on the Northeastern University surface defect detection (NEU-DET) dataset, a 5.2% increase compared to the baseline, while maintaining a real-time inference speed of 120 FPS compared to the 118 FPS of the baseline. The proposed DFED-Net provides strong support for the development of automated visual inspection systems for detecting defects on steel surfaces. Full article

Underwater image enhancement remains challenging due to wavelength-dependent absorption, spatially varying scattering, and non-uniform illumination, which jointly cause severe color distortion, contrast degradation, and structural information loss. To address these issues, we propose UCS-Net, a degradation-aware dual-path framework that exploits the complementarity between global and local representations. A spatial color balance module first stabilizes the chromatic distribution of degraded inputs through a learnable gray-world-guided normalization, mitigating wavelength-induced color bias prior to feature extraction. The network then adopts a dual-branch architecture, where a hierarchical Swin Transformer branch models long-range contextual dependencies and global color relationships, while a multi-scale residual convolutional branch focuses on recovering local textures and structural details suppressed by scattering. Furthermore, a multi-scale attention fusion mechanism adaptively integrates features from both branches in a degradation-aware manner, enabling dynamic emphasis on global or local cues according to regional attenuation severity. A hue-preserving reconstruction module is finally employed to suppress color artifacts and ensure faithful color rendition. Extensive experiments on UIEB, EUVP, and UFO benchmarks demonstrate that UCS-Net consistently outperforms state-of-the-art methods in both full-reference and non-reference evaluations. Qualitative results further confirm its effectiveness in restoring fine structural details while maintaining globally consistent and visually realistic colors across diverse underwater scenes. Full article

Wild ginkgo, as an endangered species, holds significant value for genetic resource conservation, yet its practical applications face numerous challenges. Traditional field surveys are inefficient in mountainous mixed forests, while satellite remote sensing is limited by spatial resolution. Current deep learning approaches relying on single-source data or merely simple multi-source fusion fail to fully exploit information, leading to suboptimal recognition performance. This study presents a multimodal ginkgo crown dataset, comprising RGB and multispectral images acquired by an UAV platform. To achieve precise crown segmentation with this data, we propose a novel dual-branch dynamic weighting fusion network, termed dual-branch cross-modal attention-enhanced UNet (DCA-UNet). We design a dual-branch encoder (DBE) with a two-stream architecture for independent feature extraction from each modality. We further develop a cross-modal interaction fusion module (CIF), employing cross-modal attention and learnable dynamic weights to boost multi-source information fusion. Additionally, we introduce an attention-enhanced decoder (AED) that combines progressive upsampling with a hybrid channel-spatial attention mechanism, thereby effectively utilizing multi-scale features and enhancing boundary semantic consistency. Evaluation on the ginkgo dataset demonstrates that DCA-UNet achieves a segmentation performance of 93.42% IoU (Intersection over Union), 96.82% PA (Pixel Accuracy), 96.38% Precision, and 96.60% F1-score. These results outperform differential feature attention fusion network (DFAFNet) by 12.19%, 6.37%, 4.62%, and 6.95%, respectively, and surpasses the single-modality baselines (RGB or multispectral) in all metrics. Superior performance on cross-flight-altitude data further validates the model’s strong generalization capability and robustness in complex scenarios. These results demonstrate the superiority of DCA-UNet in UAV-based multimodal ginkgo crown recognition, offering a reliable and efficient solution for monitoring wild endangered tree species. Full article

As the third generation of neural networks, Spiking Neural Networks (SNNs) simulate the event-driven processing mode of the brain, offering superior energy efficiency and biological interpretability compared to traditional deep learning. Combining the architectural strengths of Transformers with SNNs has recently demonstrated high accuracy and significant potential. SNNs process binary spikes and rich temporal information, resulting in lower computational complexity and making them particularly suitable for neuromorphic datasets. However, neuromorphic data typically involve dynamic edges and high-frequency pixel intensity changes. Capturing this frequency information is challenging for traditional spatial methods but is critical for event-driven vision. To address this, we investigate the integration of the Fast Fourier Transform (FFT) into SNNs and propose the Adaptive Spiking Fourier Neural Operator Transformer (ASFNOformer). This architecture adapts the Adaptive Fourier Neural Operator (AFNO)—originally validated in Artificial Neural Networks (ANNs)—specifically for the spiking domain. Unlike standard AFNOs, our module applies FFT across both spatial (H, W) and temporal (T) dimensions, followed by a Multi-Layer Perceptron structure (MLP) mechanism with a block-diagonal weight matrix. This design effectively captures both spatial features and temporal dynamics inherent in event streams. Furthermore, we incorporate Leaky Integrate-and-Fire (LIF) neurons optimized with Learnable Weight Parameters (LWP-LIF) to enhance temporal feature extraction and adaptivity. Experimental results on standard benchmarks indicate that our method reduces the parameter count by approximately 25%. In terms of recognition accuracy, ASFNOformer is comparable to mainstream models on static datasets and demonstrates superior performance on neuromorphic datasets by efficiently capturing frequency features. Notably, ablation studies confirm the model’s generalizability, and when using QKformer as a baseline, our method achieves state-of-the-art (SOTA) performance on the CIFAR10-DVS dataset. This work advances frequency-domain analysis in SNNs, paving the way for efficient deployment on neuromorphic hardware. Full article