Search Results (2,659)

In this paper, we propose a novel integrated model architecture, called a level-of-detail (LoD)-aware convolutional vision transformer network (LCvT). It is designed to enhance digital twin (DT) synchronization by effectively integrating LoD awareness in DTs through hierarchical image classification. LCvT employs a vision transformer (ViT)-based backbone coupled with dedicated branch networks for each LoD. This integration of ViT and branch networks ensures that key features are accurately detected and tailored to the specific objectives of each detail level while also efficiently extracting common features across all levels. Furthermore, LCvT leverages a coarse-to-fine inference strategy and incorporates an early exit mechanism for each LoD, which significantly reduces computational overhead without compromising accuracy. This design enables a single model to dynamically adapt to varying LoD requirements in real-time, offering substantial improvements in inference time and resource efficiency compared to deploying separate models for each level. Through extensive experiments on benchmark datasets, we demonstrate that LCvT outperforms existing methods in accuracy and efficiency across all LoDs, especially in DT synchronization scenarios where LoD requirements fluctuate dynamically. Full article

Drone-based target detection using visible and thermal (RGB-T) images is critical in disaster rescue, intelligent transportation, and wildlife monitoring. However, persons typically occupy fewer pixels and exhibit more varied postures than vehicles or large animals, making them difficult to detect in unmanned aerial vehicle (UAV) remote sensing images with complex backgrounds. We propose a novel progressive target-aware network (PTANet) for person detection using RGB-T images. A global adaptive feature fusion module (GAFFM) is designed to fuse the texture and thermal features of persons. A progressive focusing strategy is used. Specifically, we incorporate a person segmentation auxiliary branch (PSAB) during training to enhance target discrimination, while a cross-modality background mask (CMBM) is applied in the inference phase to suppress irrelevant background regions. Extensive experiments demonstrate that the proposed PTANet achieves high accuracy and generalization performance, reaching 79.5%, 47.8%, and 97.3% mean average precision (mAP)@50 on three drone-based person detection benchmarks (VTUAV-det, RGBTDronePerson, and VTSaR), with only 4.72 M parameters. PTANet deployed on an embedded edge device with TensorRT acceleration and quantization achieves an inference speed of 11.177 ms (640 × 640 pixels), indicating its promising potential for real-time onboard person detection. The source code is publicly available on GitHub. Full article

Electrocardiogram (ECG) waveform delineation is fundamental to cardiac disease diagnosis. This task requires precise localization of key fiducial points, specifically the onset, peak, and offset positions of P-waves, QRS complexes, and T-waves. Current methods exhibit significant performance degradation in noisy clinical environments (baseline drift, electromyographic interference, powerline interference, etc.), compromising diagnostic reliability. To address this limitation, we introduce ECG-SCFNet: a novel dual-stream architecture employing selective context fusion. Our framework is further enhanced by a consistency training paradigm, enabling it to maintain robust waveform delineation accuracy under challenging noise conditions.The network employs a dual-stream architecture: (1) A temporal stream captures dynamic rhythmic features through sequential multi-branch convolution and temporal attention mechanisms; (2) A morphology stream combines parallel multi-scale convolution with feature pyramid integration to extract multi-scale waveform structural features through morphological attention; (3) The Selective Context Fusion (SCF) module adaptively integrates features from the temporal and morphology streams using a dual attention mechanism, which operates across both channel and spatial dimensions to selectively emphasize informative features from each stream, thereby enhancing the representation learning for accurate ECG segmentation. On the LUDB and QT datasets, ECG-SCFNet achieves high performance, with F1-scores of 97.83% and 97.80%, respectively. Crucially, it maintains robust performance under challenging noise conditions on these datasets, with 88.49% and 86.25% F1-scores, showing significantly improved noise robustness compared to other methods and demonstrating exceptional robustness and precise boundary localization for clinical ECG analysis. Full article

In recent years, research in sequential recommendation has primarily refined user intent by constructing sequence-level contrastive learning tasks through data augmentation or by extracting preference information from the latent space of user behavior sequences. However, existing methods suffer from two critical limitations. Firstly, they fail to account for how random data augmentation may introduce unreasonable item associations in contrastive learning samples, thereby perturbing sequential semantic relationships. Secondly, the neglect of temporal dependencies may prevent models from effectively distinguishing between incidental behaviors and stable intentions, ultimately impairing the learning of user intent representations. To address these limitations, we propose TCLRec, a novel temporal-aware and intent contrastive learning framework for sequential recommendation, incorporating symmetry into its architecture. During the data augmentation phase, the model employs a symmetrical contrastive learning architecture and incorporates semantic enhancement operators to integrate user preferences. By introducing user rating information into both branches of the contrastive learning framework, this approach effectively enhances the semantic relevance between positive sample pairs. Furthermore, in the intent contrastive learning phase, TCLRec adaptively attenuates noise information in the frequency domain through learnable filters, while in the pre-training phase of sequence-level contrastive learning, it introduces a temporal-aware network that utilizes additional self-supervised signals to assist the model in capturing both long-term dependencies and short-term interests from user behavior sequences. The model employs a multi-task training strategy that alternately performs intent contrastive learning and sequential recommendation tasks to jointly optimize user intent representations. Comprehensive experiments conducted on the Beauty, Sports, and LastFM datasets demonstrate the soundness and effectiveness of TCLRec, where the incorporation of symmetry enhances the model’s capability to represent user intentions. Full article

Traditionally, duct sizing in ventilation systems is based on balancing pressure losses across all branches, with fan selection performed subsequently. However, this sequential approach is inadequate for systems with distributed fans in the central duct network, where pressure losses can vary significantly. Consequently, when designing the system topology, fan placement and duct sizing must be considered together. Recent research has demonstrated that discrete optimisation methods can account for multiple load cases and produce ventilation layouts that are both cost- and energy-efficient. However, existing approaches usually concentrate on component placement and assume that duct sizing has already been finalised. While this is sufficient for later design stages, it is unsuitable for the early stages of planning, when numerous system configurations must be evaluated quickly. In this work, we present a novel methodology that simultaneously optimises duct sizing, fan placement, and volume flow controller configuration to minimise life-cycle costs. To achieve this, we exploit the structure of the problem and formulate a mixed-integer linear program (MILP), which, unlike existing non-linear models, significantly reduces computation time while introducing only minor approximation errors. The resulting model enables fast and robust early-stage planning, providing optimal solutions in a matter of seconds to minutes, as demonstrated by a case study. The methodology is demonstrated on a case study, yielding an optimal configuration with distributed fans in the central fan station and achieving a 5 reduction in life-cycle costs compared to conventional central designs. The MILP formulation achieves these results within seconds, with linearisation errors in electrical power consumption below 1.4%, confirming the approach’s accuracy and suitability for early-stage planning. Full article

In this paper, a target detection network algorithm based on a multi-branch convolution and attention improvement Cross-Stage Partial-Fusion Bottleneck with Two Convolutions (C2F) module is proposed for the difficult task of detecting small targets in unmanned aerial vehicles. A C2F module method consisting of fusing partial convolutional (PConv) layers was designed to improve the speed and efficiency of extracting features, and a method consisting of combining multi-scale feature fusion with a channel space attention mechanism was applied in the neck network. An FA-Block module was designed to improve feature fusion and attention to small targets’ features; this design increases the size of the miniscule target layer, allowing richer feature information about the small targets to be retained. Finally, the lightweight up-sampling operator Content-Aware ReAssembly of Features was used to replace the original up-sampling method to expand the network’s sensory field. Experimental tests were conducted on a self-complied mountain pedestrian dataset and the public VisDrone dataset. Compared with the base algorithm, the improved algorithm improved the mAP50, mAP50-95, P-value, and R-value by 2.8%, 3.5%, 2.3%, and 0.2%, respectively, on the Mountain Pedestrian dataset and the mAP50, mAP50-95, P-value, and R-value by 9.2%, 6.4%, 7.7%, and 7.6%, respectively, on the VisDrone dataset. Full article

The accuracy of numerical model outputs strongly depends on the quality of the initial wind field, yet ground observation data are typically sparse and provide incomplete spatial coverage. More importantly, many current mainstream correction models rely on reanalysis grid datasets like ERA5 as the true value, which relies on interpolation calculation, which directly affects the accuracy of the correction results. To address these issues, we propose a new deep learning model, PPWNet. The model directly uses sparse and discretely distributed observation data as the true value, which integrates observation point positions and a physical consistency term to achieve a high-precision corrected wind field. The model design is inspired by biological intelligence. First, observation point positions are encoded as input and observation values are included in the loss function. Second, a parallel dual-branch DenseInception network is employed to extract multi-scale grid features, simulating the hierarchical processing of the biological visual system. Meanwhile, PPWNet references the PointNet architecture and introduces an attention mechanism to efficiently extract features from sparse and irregular observation positions. This mechanism reflects the selective focus of cognitive functions. Furthermore, this paper incorporates physical knowledge into the model optimization process by adding a learned physical consistency term to the loss function, ensuring that the corrected results not only approximate the observations but also adhere to physical laws. Finally, hyperparameters are automatically tuned using the Bayesian TPE algorithm. Experiments demonstrate that PPWNet outperforms both traditional and existing deep learning methods. It reduces the MAE by 38.65% and the RMSE by 28.93%. The corrected wind field shows better agreement with observations in both wind speed and direction, confirming the effectiveness of incorporating position information and a physics-informed approach into deep learning-based wind field correction. Full article

Aiming at the problem that the well depth parameters in existing intelligent drilling technology can not be obtained underground, a multi-branch parallel neural network is proposed to solve the problem of downhole well depth tracking, and its effectiveness is verified by a field example. After analyzing and correcting the quality of the logging data collected on site by using DBSCAN (a density clustering algorithm), five parameters of WOB, rotating speed, displacement, pump pressure, and torque are selected to predict and calculate the downhole mechanical ROP. Adjust the structure of a traditional artificial BP neural network and design a multi-branch parallel neural network, change the basic architecture of the original hierarchical operation, make full use of the operation efficiency of a computer to realize parallel operation, and adopt the method of point-to-point depth comparison when evaluating the well depth tracking effect. The results indicate that the MAE and mechanical drilling rate evaluation values obtained were 1.18 and 0.873, respectively. The multi-branch parallel neural network achieved a 66.55% improvement in MAE compared to the original BP neural network, while the R² evaluation method showed a 61.82% increase. The average point-by-point comparison error in the example calculation was 0.012 m, with a maximum error of 0.268 m. This result can serve as a fundamental basis for judging changes in well depth during the drilling process. Full article

Inertial measurement unit (IMU)-based gait biometrics have attracted increasing attention for unobtrusive identity recognition. While recent studies often fuse signals from multiple sensor positions and time–frequency features, the actual contribution of each sensor location and signal modality remains insufficiently explored. In this work, we present a comprehensive quantitative analysis of the role of different IMU placements and feature domains in gait-based identity recognition. IMU data were collected from three body positions (shank, waist, and wrist) and processed to extract both time-domain and frequency-domain features. An attention-gated fusion network was employed to weight each signal branch adaptively, enabling interpretable assessment of their discriminative power. Experimental results show that shank IMU dominates recognition accuracy, while waist and wrist sensors primarily provide auxiliary information. Similarly, the contribution of time-domain features to classification performance is the greatest, while frequency-domain features offer complementary robustness. These findings illustrate the importance of sensor and feature selection in designing efficient, scalable IMU-based identity recognition systems for wearable applications. Full article

Posture recognition is critical in modern educational and office environments for preventing musculoskeletal disorders and maintaining cognitive performance. Existing methods based on human keypoint detection typically rely on convolutional neural networks (CNNs) and single-scale features, which limit representation capacity and suffer from overfitting under small-sample conditions. To address these issues, we propose MSBN-SPose, a Multi-Scale Bayesian Neuro-Symbolic Posture Recognition framework that integrates geometric features at multiple levels—including global body structure, local regions, facial landmarks, distances, and angles—extracted from OpenPose keypoints. These features are processed by a multi-branch Bayesian neural architecture that models epistemic uncertainty, enabling improved generalization and robustness. Furthermore, a lightweight neuro-symbolic reasoning module incorporates human-understandable rules into the inference process, enhancing transparency and interpretability. To support real-world evaluation, we construct the USSP dataset, a diverse, classroom-representative collection of student postures under varying conditions. Experimental results show that MSBN-SPose achieves 96.01% accuracy on USSP, outperforming baseline and traditional methods under data-limited scenarios. Full article

Ground deformation monitoring in mining areas is essential for hazard prevention and environmental protection. Although interferometric synthetic aperture radar (InSAR) provides detailed phase information for accurate deformation measurement, its performance is often compromised in regions experiencing rapid subsidence and strong noise, where phase aliasing and coherence loss lead to significant inaccuracies. To overcome these limitations, this paper proposes JOTGLNet, a guided learning network with joint offset tracking, for multiscale deformation monitoring. This method integrates pixel offset tracking (OT), which robustly captures large-gradient displacements, with interferometric phase data that offers high sensitivity in coherent regions. A dual-path deep learning architecture was designed where the interferometric phase serves as the primary branch and OT features act as complementary information, enhancing the network’s ability to handle varying deformation rates and coherence conditions. Additionally, a novel shape perception loss combining morphological similarity measurement and error learning was introduced to improve geometric fidelity and reduce unbalanced errors across deformation regions. The model was trained on 4000 simulated samples reflecting diverse real-world scenarios and validated on 1100 test samples with a maximum deformation up to 12.6 m, achieving an average prediction error of less than 0.15 m—outperforming state-of-the-art methods whose errors exceeded 0.19 m. Additionally, experiments on five real monitoring datasets further confirmed the superiority and consistency of the proposed approach. Full article

With the widespread deployment of deep neural networks in real-world physical environments, assessing their robustness against adversarial attacks has become a central issue in AI safety. However, the existing two-dimensional adversarial methods often lack robustness in the physical world, while three-dimensional adversarial camouflage generation typically relies on high-fidelity 3D models, limiting practicality. To address these limitations, we propose CAM3D, a cross-domain 3D adversarial camouflage generation framework based on single-view image input. The framework establishes an inverse graphics network based on the Mamba architecture, integrating a hybrid non-causal state-space-duality module and a wavelet-enhanced dual-branch local perception module. This design preserves global dependency modeling while strengthening high-frequency detail representation, enabling high-precision recovery of 3D geometry and texture from a single image and providing a high-quality structural prior for subsequent adversarial camouflage optimization. On this basis, CAM3D employs a progressive three-stage optimization strategy that sequentially performs multi-view pseudo-supervised reconstruction, real-image detail refinement, and cross-domain adversarial camouflage generation, thereby systematically improving the attack effectiveness of adversarial camouflage in both the digital and physical domains. The experimental results demonstrate that CAM3D substantially reduces the detection performance of mainstream object detectors, and comparative as well as ablation studies further confirm its advantages in geometric consistency, texture fidelity, and physical transferability. Overall, CAM3D offers an effective paradigm for adversarial attack research in real-world physical settings, characterized by low data dependency and strong physical generalization. Full article

Forecasting outcomes in high-stakes competitive spectacles like the Olympic Games, World Cups, and professional league championships has grown increasingly vital, directly impacting strategic planning, resource allocation, and performance optimization across a multitude of fields. However, accurate forecasting remains challenging due to complex, nonlinear interactions inherent in high-dimensional time-series data, further complicated by socioeconomic indicators, historical influences, and host-country advantages. In this study, we propose a comprehensive forecasting framework integrating structured time-series modeling with ensemble learning. We extract key structural features via two novel indices: the Advantage Index (measuring a competitor’s dominance in specific areas) and the Herfindahl Index (quantifying performance outcome concentration). We also evaluate host-country advantage using a Difference-in-Differences (DiD) approach. Leveraging these insights, we develop a dual-branch predictive model combining an Attention-augmented Long Short-Term Memory (Attention-LSTM) network and a Random Forest classifier. Attention-LSTM captures long-term dependencies and dynamic patterns in structured temporal data, while Random Forest handles predictions for unrecognized contenders, addressing zero-inflation issues. Extensive stability and comparative analyses demonstrate that our model outperforms traditional and state-of-the-art methods, exhibiting strong resilience to input perturbations, consistent performance across multiple runs, and appropriate sensitivity to key features. Our key contributions include the development of a novel integrated forecasting framework, the introduction of two innovative structural indices for competitive dynamics analysis, and the demonstration of robust predictive performance that bridges technical innovation with practical strategic application. Finally, we transform our modeling insights into actionable strategic insights. This translation is powered by interpretable feature importance rankings and stability analysis that rigorously validate the robustness of key predictors. These insights apply across multiple dimensions—encompassing advantage assessment, resource distribution, strategic simulation, and breakthrough potential identification—providing comprehensive decision support for strategic planners and policymakers navigating competitive environments. Full article

Background: The accurate classification of brain tumor subtypes from MRI scans is critical for timely diagnosis, yet the manual annotation of large datasets remains prohibitively labor-intensive. Method: We present SSPLNet (Semi-Supervised Pseudo-Labeling Network), a dual-branch deep learning framework that synergizes confidence-guided iterative pseudo-labelling with deep feature fusion to enable robust MRI-based tumor classification in data-constrained clinical environments. SSPLNet integrates a custom convolutional neural network (CNN) and a pretrained ResNet50 model, trained semi-supervised using adaptive confidence thresholds ($\tau = 0.98 \to 0.95 \to 0.90$) to iteratively refine pseudo-labels for unlabelled MRI scans. Feature representations from both branches are fused via a dense network, combining localized texture patterns with hierarchical deep features. Results: SSPLNet achieves state-of-the-art accuracy across labelled–unlabelled data splits ($90:10$ to $10:90$), outperforming supervised baselines in extreme low-label regimes (10:90) by up to 5.34% from Custom CNN and 5.58% from ResNet50. The framework reduces annotation dependence and with 40% unlabeled data maintains 98.17% diagnostic accuracy, demonstrating its viability for scalable deployment in resource-limited healthcare settings. Conclusions: Statistical Evaluation and Robustness Analysis of SSPLNet Performance confirms that SSPLNet’s lower error rate is not due to chance. The bootstrap results also confirm that SSPLNet’s reported accuracy falls well within the 95% CI of the sampling distribution. Full article

The curing of silicone networks from dimethylsilanediol and methylsilanetriol chainbuilder–crosslinker precursor mixtures is investigated from combined quantum/molecular mechanics simulations. Upon screening different crosslinker content from 5 to 15%, we provide a series of atomic-resolution bulk models all featuring 98–99% curing degree, albeit at rather different arrangement of the chains and nodes, respectively. To elucidate the nm scale alignment of the polymer networks, we bridge scales from atomic simulation cells to graph theory and demonstrate the analyses of 3-dimensional percolation of -O-Si-O- bonds, polydimethylsiloxane branching characteristics and the interpenetration of loops. Our findings are discussed in the context of the available experimental data to relate heat of formation, curing degree and elastic properties to the molecular scale structural details—thus promoting the in-depth understanding of silicone resins. Full article