Search Results (7,691)

Reliable fault diagnosis of milling machines under varying operating conditions remains challenging due to distribution shifts caused by speed variations, nonstationary dynamics, and limited labeled data in target domains. Conventional domain adaptation methods often assume equal reliability across samples and neglect the varying physical consistency of signals collected under different conditions. To address this limitation, this study proposes trust-aware domain adaptation network for cross-domain fault diagnosis that integrates physics-guided reliability estimation with deep representation learning. In the proposed framework, physically interpretable global and local features are first extracted from multi-channel vibration signals using energy, spectral, nonlinear, and impulsiveness descriptors. A dedicated Physics Trust Network is then introduced to estimate per-sample trust scores that quantify the physical reliability of each signal based on its physics feature consistency. These trust scores are explicitly embedded into representation learning through a trust-weighted feature encoder, ensuring that physically reliable samples contribute more strongly to the learned latent space. To address distribution mismatch between source and target conditions, a trust-weighted covariance alignment strategy is introduced, enabling domain adaptation to be guided by reliable samples instead of treating all data equally. In this way, the model simultaneously learns discriminative, transferable, and physically consistent features. The entire framework is trained end-to-end using labeled source data and unlabeled target data, enabling effective knowledge transfer under cross-speed conditions. Extensive experiments on a real milling machine dataset collected at different spindle speeds demonstrate that the proposed framework achieves an average accuracy of 98.07%, performing better than two recent state-of-the-art domain adaptation approaches by a significant margin. Ablation experiments further confirm that reliability estimation, trust-weighted representation learning, and trust-guided alignment each contribute independently to performance improvement. Full article

To address the insufficient representation of terrain constraints and surface resistance in traditional path planning for off-road environments, this study proposes an improved A* path planning method for unmanned ground vehicles. First, an off-road environment model is constructed using Digital Elevation Model (DEM) and land cover data, and environment–vehicle traversability is evaluated by integrating geometric and support-based traversability analyses. Geometric constraints are determined using slope thresholds, minimum ground clearance, and approach/departure angles, while support-based traversability is quantified through a surface velocity influence coefficient to reflect traversal-efficiency differences under various surface conditions. These terrain and surface constraints are incorporated into the actual cost function of the A* algorithm, and a direction-corrected heuristic function is designed to enhance goal-directed search. Experiments conducted in Jiancaoping District, Taiyuan, show that, compared with the traditional A* algorithm, the proposed method reduces cumulative travel time, maximum path slope, and expanded nodes by 15.3%, 22.9%, and 47.8%, respectively, with only a 2.4% increase in path length. The results demonstrate that the proposed method effectively avoids steep and high-resistance areas while achieving coordinated optimization of path length, traversal efficiency, and terrain safety. Full article

In this paper we study how external affective information should be integrated into a compact transformer-based text classifier. Rather than treating affective features as signals to be appended directly to the representation, we examine whether their contribution should be controlled through lightweight fusion mechanisms. The comparison focuses on scalar-gated fusion versus plain concatenation, using DistilBERT as the textual backbone and four affective resources: the NRC VAD Lexicon, VAD-BERT, Ekman-style emotion scores, and SenticNet. The evaluation is conducted on two English corpora with different label structures: a seven-class MentalHealth dataset and the fine-grained GoEmotions benchmark. Across both corpora, scalar gating consistently matches or outperforms concatenation in terms of Macro-F1. On MentalHealth, scalar gating improves all directly comparable configurations. On GoEmotions, it achieves the best overall Macro-F1 and improves most matched comparisons. Beyond static feature integration, we introduce affective flow (EmoFlow) representations derived from VAD-BERT, which model the evolution of valence, arousal, and dominance across segments of a text. These dynamic representations do not surpass the strongest static lexical resources in absolute performance, but they provide consistent improvements within the VAD-BERT family, particularly when combined with scalar gating or cross-attention. Our contribution is twofold. First, we show that a lightweight scalar gate provides an effective and interpretable mechanism for adaptively integrating low-dimensional affective side information into transformer-based classifiers. Second, we introduce affective flow representations that explicitly model how affect evolves within a document, enabling the analysis of both adaptive resource selection and intra-document affective dynamics. Together, these results suggest that the key issue is not only which affective resources to use, but also when and how they should influence the model. Full article

Multimedia data have been continuously increasing in magnitude, and so has the sophistication of manipulation methods, thereby making the digital forensic investigation process more complicated. The easy access to sophisticated image editing software and AI-generated materials has brought up the issue of information integrity, the reliability of legal evidence, and public trust. Traditional image forensics methods are usually concerned with either the detection of visual artifacts based on convolutional neural networks (CNNs) or based on metadata analysis, frequently independently of each other. This paper presents a multi-modal fusion paradigm, comprising visual feature-based feature extraction and metadata inconsistency-based detectors, to improve the classification strength. A two-stream design is used, comprising a high-level visual artifact capturing the transfer learning-based MobileNetV2 network and an XGBoost classifier that analyses EXIF metadata discrepancies. The heterogeneous representations are merged in a feature-level fusion strategy to generate a final authenticity prediction. It was tested on individual datasets and a compiled dataset of 26,023 images from CoMoFoD, CG-1050 and CASIA v1 and v2. The suggested approach had an overall accuracy of 83.85%, which was higher than the visual-only (68.61%) and metadata-only (75.85%) baselines. These findings show that complementary visual and metadata cues are much more useful in detection, while the use of a lightweight backbone enables efficient, high-throughput forensic analysis suitable for real-world deployment. Full article

To address the limited edge-preserving capability of traditional guided filtering caused by fixing the filtering window center at the target pixel, an edge-preserving guided filtering algorithm based on edge tangent flow and side window (EWGF) is proposed. Rather than relying on local intensity statistics, the proposed method exploits local geometric structures through the integration of edge-aware weighting, tangent direction estimation, and adaptive side window filtering. Firstly, an edge-aware weighting strategy is introduced to construct a weighted gradient representation, enabling characterization of edge intensity and local structural features. Secondly, based on the weighted gradient field, a structure tensor constrained by local geometric information is established, and edge tangent flow is estimated through eigenvalue decomposition to capture local edge orientations. Furthermore, a tangent-guided one-dimensional side window guided filtering mechanism is developed, in which the filtering direction is adaptively aligned with local edge structures to suppress noise and textures while preserving edge sharpness and structural continuity. Finally, comparative experiments are conducted on the BSD500, Set5, and Set14 datasets, with PSNR and SSIM employed as evaluation metrics. Experimental results demonstrate that, compared with the traditional guided filtering algorithm, the proposed method improves PSNR by an average of 5.69 dB and SSIM by an average of 0.11, validating its performance in structure preservation and noise suppression. Full article

Steel manufacturers currently rely on optical sensing systems that capture surface images for quality control, but these systems still face practical challenges in industrial environments. Early machine vision methods used handcrafted features, and their accuracy dropped when lighting changed or the background texture became complex. Deep learning improved detection, yet small defects that blend into background noise or share visual patterns with other classes still cause missed detections and false positives on factory floors where hardware resources are tight. GEA-YOLO integrates different refinement strategies into the Backbone, Neck, and training stage. C2DGA replaces C2PSA in the Backbone. Deformable attention adapts to defect structures that deviate from fixed sampling patterns, and a dynamic gate fuses global contextual information with local texture features. EMA modules are inserted into the Neck, where they recalibrate features independently at each scale and reduce the influence of background interference. DetectAux provides auxiliary supervision for hard samples during training. Unlike approaches that introduce attention at a single fixed position, GEA-YOLO places each module at the stage where it can improve the corresponding representation. We evaluated GEA-YOLO on the NEU-DET and GC10-DET datasets. On NEU-DET, the model reached 80.3% mAP@0.5, 2.6 percentage points above YOLOv11s, while keeping a reasonable balance between accuracy and computational cost and maintaining an inference speed of 169.5 FPS. Cross-dataset validation on GC10-DET further confirmed generalization, yielding 74.9% mAP@0.5, 2.9 percentage points above YOLOv11s, showing strong potential for real-time steel surface inspection. Ablation results confirm that each modification fixes a different weakness in the detection pipeline, but the full gain only appears when all three are used together. These results indicate that GEA-YOLO is promising for real-time optical inspection in controlled benchmark settings. Full article

Sequential decision-making is a critical process that governs the loading and unloading operations of container terminals through a series of instructions. The quality of these batch-level dispatching decisions directly determines operational efficiency and the competitive position of the terminal. This paper proposes an intelligent instruction sequencing model based on a Parallel Priority-driven Deep Q-Network (ppDDQN) algorithm, acting as a centralized instruction dispatcher. The proposed ppDDQN extends standard Double DQN by incorporating (i) a parallel priority mechanism that integrates five domain-specific decision features—yard crane movement, container flipping, loading sequence, operational conflicts, and task completion potential—into the experience replay prioritization, and (ii) a wake–sleep feature learning architecture for structured state representation. A comprehensive simulation study with 100 validation cases across three scales (29, 50, and 100 containers) demonstrates that ppDDQN achieves a 20.0% improvement in normalized objective value over the genetic algorithm baseline and a 9.0% improvement over the step-activation variant (with statistical significance p < 0.01), while maintaining feasibility in over 96% of test cases across all scales. The proposed method effectively mitigates yard crane travel distance, limits container rehandling, and resolves operational conflicts, providing a robust batch-level sequencing solution under multi-equipment operational constraints. Full article

Precipitation nowcasting plays an important role in mitigating the impacts of extreme weather events on social production and daily life. However, existing methods still face two major limitations. (1) Convolutional neural network-based methods are insufficient in modeling the temporal dependencies of radar echo sequences, which may lead to information loss in the prediction results. (2) Most existing methods enlarge the receptive field by stacking convolutional layers. This strategy makes it difficult to obtain a truly global receptive field and effectively model global dependencies, resulting in limited accuracy in heavy rainfall prediction. In addition, spatiotemporal information at different time steps is not fully integrated, and the multi-scale directional features of rainbands are often ignored. To address these issues, this paper proposes ST-TriMambaUNet, which consists of an encoder, a decoder, and a feature enhancement module. First, a Spatiotemporal Fusion Attention (STFA) was designed, including global spatial attention and temporal attention. It can effectively learn long-range spatial correlations and capture the temporal dependencies of radar echo sequences in a parallel manner. Second, a Multi-Scale Interaction Mamba (MSIM) module was developed with three branches. The first branch leverages Mamba to model global spatiotemporal dependencies with linear complexity. The second branch promotes spatiotemporal information interaction through channel shuffle and further combines Mamba to model global spatiotemporal dependencies. The third branch designs Multi-Scale Directional Convolution (MSDC) to learn the multi-scale directional features of rainbands. Finally, the features from the three branches are dynamically fused through the designed adaptive gated fusion mechanism. This enhances the model’s representation capability for strong-echo core regions and multi-scale precipitation band structures. Experimental results on two public datasets, SEVIR and CIKM, demonstrated that the proposed ST-TriMambaUNet achieved clear advantages in both overall prediction accuracy and heavy rainfall scenarios. In particular, under the high-threshold precipitation scenarios of SEVIR (160, 181, and 219), the CSI was improved by up to 10.19%. In the heavy rainfall scenario of CIKM at 40 dBZ, CSI, POD, and HSS were improved by 5.29%, 8.07%, and 4.65%, respectively. Full article

Traditional “hórreos”, vernacular granaries widely distributed across northern Spain, constitute a highly valuable form of cultural heritage due to their historical, architectural, and ethnographic significance. However, their progressive deterioration, dispersion in rural contexts, and limited maintenance resources pose significant challenges for their long-term conservation. This research article explores the potential of the Historic/Heritage Building Information Modelling (HBIM) methodology as an innovative and effective tool for the documentation, analysis, conservation, and management of “hórreos” as cultural heritage assets. The study proposes an HBIM-based workflow adapted to the specific characteristics of “hórreos”, integrating data acquisition techniques such as laser scanning, photogrammetry, and historical archival research with parametric modelling of traditional construction elements. The resulting HBIM models are conceived not only as geometric representations, but as comprehensive digital repositories that store historical data, construction techniques, materials, conservation states, and recorded pathologies. The research analyses how HBIM supports decision-making in restoration planning and enables preventive maintenance strategies over time. Furthermore, the article discusses the role of HBIM in improving heritage management at a territorial scale, enabling standardised inventories and supporting institutional protection policies. The potential of HBIM for heritage dissemination, education, and digital preservation is also examined. The results highlight HBIM as a powerful and adaptable methodology that contributes to a more sustainable, informed, and holistic approach to the conservation of “hórreos”, enhancing both their physical preservation and their transmission as living cultural heritage. Full article

This study compares the effectiveness of two approaches to speech emotion recognition for three affective states in Polish: sad, neutral, and happy. Both a set of acoustic features—capturing prosodic, phonatory, temporal, spectral, and cepstral properties—and representations learned by self-supervised models (wav2vec 2.0 and WavLM) were analyzed. Experiments were conducted on the nEMO corpus, comprising 2327 recordings from nine speakers, using a rigorous leave-one-subject-out protocol to evaluate cross-speaker generalization. In the feature-based approach, 107 acoustic features were used, and classification was performed with logistic regression and, additionally, SVM variants. In the deep learning approach, the wav2vec2-base and WavLM-base models were fine-tuned for the three-class task. The best results were achieved by the self-supervised models: WavLM reached a global balanced accuracy of 0.727 and a macro-F1 score of 0.710, while wav2vec 2.0 achieved 0.722 and 0.695, respectively. Both outperformed the feature-based approach (BAcc = 0.627, macro-F1 = 0.584). Confusion matrix analysis showed that the greatest difficulty lies in distinguishing the neutral class from the sad and happy classes, whereas sad and happy classes are more clearly separable. Feature utility analysis (SFS under the LOSO protocol) indicated the significant role of cepstral features (MFCCs and their derivatives), complemented by selected prosodic and temporal features. An additional comparison of SVM classifiers suggested that the main limitation of this approach lies in the signal representation itself rather than solely in the choice of classifier. Explainability analyses of the deep models, using layer-wise probing and integrated gradients, showed that affective information is best represented in intermediate layers, and that model decisions rely on locally salient segments of the signal. Furthermore, a speaker adaptation experiment demonstrated that personalization significantly improves classification performance, highlighting the potential of such methods for long-term monitoring of affective expression changes in the same individual. Full article

The Internet of Things (IoT) has increased the cyber-attack surface by bringing together a variety of different devices, sensors, and services in critical digital infrastructure. Federated learning (FL) is a solution that enables local devices to train together without sharing raw traffic data; thus, it can be used for intrusion detection without compromising privacy. Nevertheless, traditional FL aggregation techniques are still susceptible to non-IID client distributions, data imbalance, unreliable local updates, and poor representation learning approaches. This study introduces a novel method, called Trust-Aware Contrastive Federated Learning for IoT intrusion detection, TACMA Fed. The framework extends AMAFed and combines trust-aware client scoring, aggregation based on similarity of updates, supervised contrastive representation learning, adaptive focal–Dice loss, and rare-class-aware weighting into a lightweight 1D convolutional model. The ten simulated IoT clients and the non-IID Dirichlet partition are used in experiments with the ToN-IoT train_test_network dataset. TACMA Fed achieves an accuracy of 0.9957, an F1 score of 0.9937, an ROC-AUC of 0.9991, a PR-AUC of 0.9997, and a false-positive rate of 0.0087. Robustness analysis also shows stability parameters in the presence of Gaussian noise and feature masking, as well as varying levels of client heterogeneity. The outcomes of these experiments prove that, in the context of federated IDS (FIDS) for a heterogeneous IoT network, the integration of trust-aware aggregation with contrastive representation learning and imbalance-aware optimization can enhance performance. Full article

Fine-scale identification of urban building functions is essential for understanding urban spatial structure, socioeconomic organization, and sustainable urban development. However, large-scale building function mapping remains constrained by reliance on proprietary data, insufficient representation of geographical context, and limited cross-city generalization. To address these challenges, this study proposes a multi-source geographical knowledge fusion framework for fine-scale building function classification using exclusively open-source data. In addition to conventional morphological, POI-based, and spectral features, the framework systematically integrates open-source geographical environmental features to characterize accessibility, infrastructure relationships, ecological surroundings, and environmental conditions at the building level. Beijing is selected as the training area, while Tianjin is used for independent cross-city validation. Ten representative models, including deep learning, ensemble learning, and traditional machine learning methods, are systematically evaluated for classifying six building function types. Results show that the systematic integration of geographical environmental features improves classification performance and interpretability. Deep learning and ensemble models outperform traditional methods, with CNN achieving the highest accuracy of 87.07% in Beijing and 69.83% in Tianjin. Feature contribution analysis further indicates that geographical environmental features play a dominant role in functional discrimination, while POI features provide important socio-semantic information. Overall, this study provides a reproducible open-data framework for urban building function mapping, supporting sustainable urban planning, land-use optimization, infrastructure allocation, and smart city governance. Full article

With advances in biomedical technologies and the continued expansion of experimental resources, biological data are growing rapidly in both scale and complexity. Contrastive learning provides an effective framework for integrating heterogeneous biological information. However, many protein–protein interaction (PPI) prediction methods still represent protein sequences and annotations as flat features and do not explicitly model hierarchical biological relationships among protein families, clans, and functional annotations. Here, we introduce HIPPO (HIerarchical Protein–Protein interaction prediction across Organisms), a hierarchical contrastive learning framework for PPI prediction. HIPPO aligns protein sequence representations with structured biological attributes. Across intra-species benchmark PPI datasets, HIPPO improves the average micro-F1 by 2.9% compared with the best baseline across the evaluated splits. In the host–pathogen interaction benchmark, HIPPO achieves the highest AUROC under the standard split (0.731) and the second-best AUPRC (0.332). Under leave-one-virus-family-out evaluation, HIPPO obtains the best AUROC on Papillomaviridae (0.603) and Retroviridae (0.612), while also showing family-dependent transfer behavior. Ablation experiments support the contribution of hierarchical feature integration, and attention-based residue attribution provides preliminary evidence that the learned representations highlight interface-related residues. Together, these results suggest that structured biological knowledge can improve representation learning for PPI prediction across diverse and imbalanced datasets. Full article

Conveyor belts constitute critical components of bulk material handling systems, and their reliable operation directly affects process continuity, operational safety, and maintenance costs in industrial environments. Increasing requirements regarding system reliability and predictive maintenance have stimulated the development of advanced diagnostic methods for conveyor belt condition monitoring. This review presents a comprehensive analysis of conveyor belt damage detection and diagnostic approaches, with particular emphasis on sensing technologies and signal-based methodologies. The paper discusses major conveyor belt degradation mechanisms and analyzes their representation in diagnostic data obtained using different sensing modalities. Current developments in machine vision systems, magnetic methods based on magnetic flux leakage, ultrasonic techniques, and X-ray imaging are critically reviewed together with signal preprocessing procedures, feature extraction strategies, and damage classification approaches. Particular attention is devoted to the transition from conventional signal processing techniques toward machine learning and deep learning methods enabling automated feature representation and fault identification. The analysis indicates that despite substantial progress in sensing technologies and artificial intelligence, most existing solutions remain strongly sensor-specific and limited to individual data modalities. Key research gaps include the lack of unified damage representation frameworks, limited benchmark datasets, and the insufficient integration of multimodal sensing information. Future progress will likely depend on the development of integrated diagnostic ecosystems combining heterogeneous sensing technologies, advanced feature representation methods, and intelligent decision-support systems. Full article

With the widespread deployment of IoT devices, deploying efficient network traffic classification models on resource-constrained edge nodes is critical for real-time boundary security. However, traditional lightweight models primarily rely on macro-level structural pruning, which often sacrifices crucial feature extraction capabilities when handling complex heterogeneous traffic, leading to a severe imbalance between parameter compression and detection accuracy. To overcome this bottleneck, we propose TinyFlowNet, an ultra-lightweight multi-module fusion architecture. To prevent the parameter explosion inherent in combining CNN, LSTM, and Transformer modules, TinyFlowNet innovatively adopts an extreme operator-level reconstruction strategy. By introducing debiased computations, affine-free normalization, and a customized micro-self-attention mechanism, it comprehensively strips away underlying redundant parameters. Simultaneously, an integrated parameter-free regularization mechanism is introduced to compensate for the representational capacity lost under this extreme compression, ensuring robust spatio-temporal feature fusion. Comprehensive evaluations on the custom X-IDS-20 balanced dataset alongside the complex CICDarknet2020 and ToN_IoT public datasets demonstrate that TinyFlowNet achieves exceptional accuracies of 95.31 percent, 99.53 percent, and 97.13 percent, respectively. Furthermore, it exhibits formidable robustness against extreme class imbalances by securing a peak Matthews Correlation Coefficient of 0.9465 and an outstanding PR-AUC of 0.9834, all while strictly confining the parameter count to merely 74,600. Crucially, actual on-device hardware profiling on a commercial edge device corroborates its deployment viability, exhibiting a minimal dynamic memory footprint of 8.26 MB, an average inference latency of 0.79 ms, and a processing throughput exceeding 1200 FPS. Compared to a standard heavy Hybrid CNN-LSTM-Transformer baseline, TinyFlowNet achieves superior detection accuracy while drastically reducing the parameter footprint by over 99.3% and computational FLOPs by 95.8%. Furthermore, against mainstream lightweight benchmarks like DistilBERT and heavy baselines such as LSTM, TinyFlowNet reduces parameters by 61.4% to 94% while simultaneously achieving absolute accuracy leaps and accelerating inference speed by nearly 4× over MobileNetV2, establishing a highly efficient new paradigm for intelligent edge defense. Full article