Search Results (8,375)

Human activity recognition (HAR) plays a critical role in intelligent wireless sensing and mobile edge computing. Compared with traditional vision-based and wearable-based approaches, WiFi channel state information (CSI) enables privacy-preserving and device-free activity perception. CSI encodes human-induced channel variations that serve as a natural basis for activity-oriented semantic analytics over wireless networks. However, existing WiFi CSI-based methods suffer from weak cross-scene generalization, high model complexity, and poor multi-task collaboration. To address these issues, this paper proposes UniSense-CSI, a unified multi-task framework that jointly learns dynamic gesture recognition, static posture classification, and fall detection. Specifically, the proposed framework converts CSI signals into pseudo-RGB images, extracts spatio-temporal features using a customized ConvNeXt backbone, and leverages an improved PerceiverIO module to compress high-dimensional features into a compact latent space. Based on the shared representation, task queries and adapters are introduced to enable parallel multi-task inference within a common architecture. Experiments on public datasets demonstrate that the proposed framework achieves accuracies of 99.64%, 99.66% and 96.09% on the three tasks, respectively, while maintaining low inference latency and favorable edge-deployment capability. Full article

Pathogenic fungi and their synergistic interactions with bark beetles, leading to vascular dysfunction, physiological stress, and eventual tree mortality, increasingly threaten forest ecosystems. Because fungal colonization often precedes visible macroscopic symptoms, early detection remains a critical yet challenging task in forest health monitoring. This study proposes a real-time deep learning-based object detection framework for identifying harmful fungi in proximity to host trees to support early intervention strategies. A custom dataset comprising 8900 images was constructed to represent two classes: Healthy and Unhealthy trees, where fungal presence is detected either directly on the tree or within its immediate ecological vicinity (e.g., near root systems). A fine-tuned YOLOv11 detection architecture is developed and augmented with a squeeze-and-excitation (SE)-like attention mechanism to enhance texture-sensitive feature representation. The model is trained and evaluated using precision, recall, F1-score and mean Average Precision (mAP). Experimental results demonstrate an overall mAP@0.5 of 0.825, with class-wise average precision values of 0.926 (Healthy) and 0.724 (Unhealthy). The Healthy class achieved classification accuracy of 0.92, while 0.71 of Unhealthy instances were correctly detected. F1-Confidence and recall-Confidence metrics indicate that optimal operational performance occurs within a confidence threshold range of 0.30–0.35, balancing false positives and false negatives. Despite the approximately balanced class distribution (50.6% Healthy and 49.4% Unhealthy), detection performance for the Unhealthy class was comparatively lower because of its greater intra-class variability, heterogeneous fungal appearance, and subtle visual manifestations. Findings demonstrate the feasibility of deploying real-time object detection models for early-stage fungal surveillance and highlight the importance of confidence calibration for operational disease monitoring systems. Full article

Functional near-infrared spectroscopy (fNIRS) has become an important signal source in motor imagery (MI) brain–computer interface research due to its non-invasive nature and high application flexibility. However, fNIRS signals exhibit significant inter-subject variability, complementary information from oxygenated hemoglobin (HbO) and deoxygenated hemoglobin (HbR), and complex spatiotemporal dynamics, making their efficient and robust classification challenging. To address these issues, this paper proposes a Dual-Stream Gated Fusion Network (DSGF-Net). This model employs a dual-branch architecture to perform complementary feature modeling of fNIRS signals: one branch focuses on extracting multi-scale temporal dynamic features, while the other learns the spatial distribution of hemodynamic features across channels, thereby effectively characterizing the signals from different perspectives. Upon this foundation, a gated fusion mechanism was designed to adaptively adjust the importance of different feature dimensions after the fusion of the two feature streams, thereby enhancing the discriminative power of the fused representation. On two public datasets, MI and UFFT, experimental results based on leave-one-subject-out (LOSO) cross-validation show that the proposed method achieves competitive performance across metrics such as classification accuracy, F1-score, and Kappa coefficient. Furthermore, a comparative analysis of performance under different network component configurations validates the contributions of the dual-branch structure and the gated fusion mechanism to performance improvements. Furthermore, complexity analysis results show that DSGF-Net achieves superior classification performance while maintaining a relatively small parameter size, striking a good balance between performance and computational complexity. DSGF-Net provides an effective, lightweight deep learning framework for offline fNIRS-based motor task classification, with potential applications in cross-subject BCI systems and brain signal decoding. Full article

Background/Objectives: Existing motor imagery (MI) EEG decoding is often limited by small datasets, affecting generalization reliability. This study aims to robustly decode multi-limb MI intentions. Methods: We collected an MI-EEG dataset from 292 participants (242 young adults, 50 older adults) performing left/right-arm and left/right-leg imagery. After extracting time-, frequency-, and channel correlation features, we proposed an SE-EEG-Inception model for classification. Results: Evaluated under a strict intra-subject cross-validation protocol, the model achieved a mean 4-class accuracy of 89.4%. For binary tasks, accuracies reached 88.3% (left vs. right arm) and 90.0% (left vs. right leg). Conclusions: The model successfully distinguishes predictive EEG features across different and symmetric limbs. Crucially, this high classification performance demonstrates data-driven predictive utility rather than mechanistic proof of neural differences, providing an offline proof-of-concept for multi-limb BCI control. Full article

Construction material descriptions in engineering projects are often heterogeneous due to inconsistent naming conventions, supplier-specific terminology, and incomplete specification information, making it difficult to automatically map non-standard descriptions to standardized material dictionaries. Existing rule-based approaches rely heavily on manually constructed rules and predefined dictionaries, limiting their adaptability, whereas direct large language model (LLM)-based methods often lack domain-specific constraints, resulting in less reliable material classification and attribute decomposition. To address these challenges, this paper proposes LLM-MSMA, an LLM-based multi-agent framework for construction material classification and attribute decomposition. The framework adopts a multi-scale task decomposition strategy that integrates hybrid retrieval, logic validation, and template-guided attribute decomposition. Hybrid retrieval generates candidate material categories, which are subsequently refined through LLM-based semantic reasoning. A Logic Validation Agent further verifies classification results using engineering rules, while an Attribute Decomposition Agent extracts structured material attributes according to classification-specific templates. The proposed framework is evaluated using a real-world bill-of-materials dataset containing 2912 material records collected from a large-scale engineering project as a case study. Experimental results show that, compared with the rule-based baseline, LLM-MSMA improves material classification accuracy by 43.39 percentage points and increases the attribute decomposition F1-score by 68.89 percentage points. Compared with the direct LLM baseline, the proposed framework further improves material classification accuracy by 8.24 percentage points and the attribute decomposition F1-score by 7.31 percentage points, while achieving an Exact Match Accuracy of 82.83%. These results indicate that combining hybrid retrieval with domain-specific structured knowledge and multi-agent LLM reasoning improves the reliability of construction material classification and attribute decomposition for heterogeneous material descriptions within the evaluated engineering context. Full article

Medical-image annotation is costly and limits the use of fully supervised learning. This study proposes the Self-Supervised Stacked Masked Denoising Autoencoder (S²MDAE) for four-class brain MRI classification. During pre-training, three convolutional encoder–decoder blocks perform layer-wise masked reconstruction under a 75% mask ratio and Gaussian corruption; the pretrained encoder stack is then fine-tuned for classification. On the primary dataset, S²MDAE achieved 87.207% Accuracy, while ablation studies supported the contributions of pre-training, masking, noise injection, and layer-wise reconstruction. Representation analysis further showed increased class separability in deeper blocks after fine-tuning. Without retraining, external evaluation on a duplicate-screened independent dataset achieved 67.254% Accuracy and 66.009% Macro-F1, indicating partial transfer under domain shift. These findings support the task-specific value of the proposed framework while showing that broader clinical generalization requires further validation. Full article

Hyperparameter optimization is a critical determinant of the performance and generalizability of convolutional neural networks (CNNs), yet selecting optimal configurations remains a complex challenge due to the large, non-linear search space. While numerous optimization techniques have been explored in the literature, most comparative studies have relied on benchmark datasets, limiting insights into their practical effectiveness under real-world constraints. This study evaluates four widely used hyperparameter optimization methods (i.e., Grid Search, Optuna, Hyperopt, and DEAP) across five CNN architectures (i.e., AlexNet, GoogLeNet, LeNet, MobileNet, and ResNet18) under two distinct data regimes: the Modified National Institute of Standards and Technology (MNIST) benchmark and a custom SkinCancer image dataset. On MNIST, all configurations achieved strong classification performance, with test accuracies ranging from 95.45% to 97.61%, led by ResNet18 optimized via Grid Search (97.61%). In this simpler setting, DEAP provided the shortest optimization runtimes, while Optuna and Hyperopt offered an effective balance between performance and computational cost. In contrast, the SkinCancer dataset exhibited substantially greater variability in performance, highlighting the stronger influence of both model architecture and optimization strategy. ResNet18 achieved the highest overall test accuracy (81.50%) when optimized with Hyperopt. Although no single optimization method consistently outperformed all others, adaptive approaches frequently achieved competitive or superior predictive performance while requiring considerably less computation than exhaustive search. These findings demonstrate that hyperparameter optimization dynamics depend heavily on dataset complexity, where computational efficiency is the primary differentiator for simpler classification tasks, but optimization architecture selection becomes critical for navigating challenging medical imaging applications. Full article

Early identification of individuals with abnormal glucose metabolism is essential for timely intervention and prevention of diabetes-related complications. Routine laboratory testing generates large amounts of clinical data that may support automated glycemic classification through machine learning approaches. This study aimed to develop and evaluate a machine learning framework for the classification of HbA1c-defined glycemic status using routinely available clinical laboratory features. A retrospective dataset of 1434 individuals with available glycemic measurements was analyzed. Participants were categorized into HbA1c-defined normoglycemic, prediabetic-range, or diabetic-range groups. Three concurrent classification tasks were examined: HbA1c-defined dysglycemia classification, diabetic-range HbA1c classification, and multiclass HbA1c-defined glycemic-status classification. Demographic, biochemical, and hematological variables were used as predictors. Data preprocessing included missing-value handling, feature filtering, and outlier treatment. Several supervised learning algorithms were evaluated, including Logistic Regression, Random Forest, Gradient Boosting, Support Vector Machine, and Multinomial Logistic Regression. Model performance was assessed using train–test validation and cross-validation with accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve. For dysglycemia, Gradient Boosting achieved the highest AUC (0.848), while Random Forest achieved the highest accuracy (0.801) and sensitivity (0.908). For diabetic-range HbA1c, Random Forest achieved the highest AUC (0.864), whereas SVM achieved the highest accuracy (0.794). In multiclass classification, Random Forest achieved the highest accuracy (0.610), while Gradient Boosting achieved the highest macro-AUC (0.796) and macro-F1 score (0.603). Pairwise comparisons showed no statistically significant superiority of any classifier after Holm correction. Clinical-baseline and ablation analyses demonstrated that fasting glucose accounted for a substantial proportion of discrimination, with only modest incremental value from additional laboratory variables. These findings support cautious interpretation of routine laboratory-based classification models pending further validation and clinical-utility assessment. Full article

Quantum machine learning is often proposed for richer feature representations, yet most evidence rests on idealized simulation rather than real noisy intermediate-scale quantum (NISQ) hardware. This research presents a controlled comparison of classical and quantum-enhanced diagnostic pipelines on three clinical binary classification tasks: Mammographic Mass, Anemia, and Diabetic Retinopathy. All pipelines share standardized preprocessing, principal component analysis (PCA), and a fixed extreme gradient boosting (XGBoost) classifier, so differences arise only from the feature representation. Four quantum encodings (angle, phase, basis, and the ZZ feature map) are each run on two backends: a noiseless simulator and the real IBM Heron r2 processor (156 qubits). Across nine performance metrics, compared with the classical pipeline, the performance decreases in the quantum hardware execution for all datasets, with a more significant reduction observed for the Anemia dataset. In contrast, compared with the simulated pipeline, the hardware execution shows a slight performance decrease for the Mammographic and Diabetic Retinopathy datasets. The exceptions are the angle and phase encodings for the Mammographic dataset, where the hardware result improves slightly compared with the simulator. For the Anemia dataset, the transition from simulation to real quantum hardware results in a considerable performance reduction. These findings show that current quantum feature maps through the encode–measure–boost pipeline on NISQ hardware do not yet outperform a well-designed classical pipeline. Full article

Background: Convolutional neural networks (CNNs) perform well in histopathological classification of oral squamous cell carcinoma (OSCC), but their robustness across distinct tissue domains remains insufficiently studied. This study assessed whether a CNN trained only on primary oral and oropharyngeal squamous cell carcinoma images remained transferable to an independent metastatic lymph node histopathology dataset. Methods: Three public datasets containing 14,760 OSCC/OPSCC and 8530 normal oral mucosa images were combined for model development. An ImageNet-pretrained EfficientNetB0 backbone was used as a fixed feature extractor with a task-specific binary classification head. Performance was first assessed on an independent internal testing subset and subsequently evaluated on 20,000 H&E-stained normal and metastatic lymph node patches from a separate public dataset. Results: The model achieved an internal testing accuracy of 91.48%, with 92.95% sensitivity, 88.92% specificity, 93.56% precision, a 93.25% F1-score, and a Youden’s J index of 0.819. External validation resulted in a substantial decrease in overall classification performance, with an accuracy of 56.91%, specificity of 33.32%, precision of 48.71%, F1-score of 63.37%, balanced accuracy of 61.99%, MCC of 0.279, and a Youden’s J index of 0.240. Nevertheless, sensitivity for metastatic tissue remained high at 90.66%, indicating a markedly asymmetric external error profile characterized predominantly by false-positive classifications. Conclusions: The marked performance decrease during external validation demonstrates the limitations of direct cross-domain transfer between substantially different histopathological environments. However, the preserved sensitivity suggests that some discriminative information remained transferable beyond the development domain. These findings support partial rather than universal cross-domain generalization and emphasize the importance of independent out-of-distribution evaluation when assessing deep learning robustness in computational pathology. Full article

Accurately predicting transportation accident severity is critical for targeted risk governance, yet research based on accident investigation reports is often hampered by small sample sizes and skewed class distributions. This study develops a dual-track feature fusion and interpretable prediction framework to overcome these constraints. More than 1000 candidate documents were screened, yielding a reconstructed analytical sample of 157 eligible accident investigation reports for a three-class accident severity classification task. The methodology integrates HFACS-Lite vertical hierarchy and DEMATEL-Lite horizontal coupling to construct high-order fused features, employing the TabPFN foundation model as the backbone learner alongside SMOTENC and post hoc dual-threshold adjustments. Empirical results show that the final SMOTENC-enhanced dual-track TabPFN achieved an accuracy of 0.783 and a Macro-F1 of 0.717, while post hoc dual-threshold adjustment increased major-and-above recall to 0.6429. SHAP attribution indicates that high-consequence accidents are associated with joint patterns of micro-level operations, operating scenarios, and safety governance. The proposed framework supports association-based severity classification and risk screening under constrained data conditions. Full article

Temperature sensor systems are critical for nuclear power plant (NPP) condition monitoring, whose reliability underpins unit safety and stability. Fault localization and diagnosis are essential to sustain their stable service. Conventional Principal Component Analysis (PCA) and Graph Neural Network (GNN) methods suffer clear drawbacks: PCA is vulnerable to noise and cannot classify fault types accurately, while GNNs struggle to quantify correlations among temperature data. This paper fuses PCA’s anomaly representation capability and GNN’s structural feature extraction capacity to propose an Abnormal Feature-GCN method for joint fault localization and diagnosis. First, an Adaptive PCA (APCA) model fed with multi-dimensional sensor features computes abnormal features. These features are then transformed into edge weights to construct a weighted graph. A dual-branch GCN is finally trained via a joint loss function for parallel multi-task learning to simultaneously locate faulty sensors and identify fault types. Validated on a nuclear primary circuit temperature sensor system under constant-, rising-, and falling-temperature working conditions, the proposed method realizes accurate fault localization and classification. The mean overall accuracy of the proposed method surpasses mainstream baselines by 2.23%, 1.62%, and 1.89% for the three typical working conditions, respectively. Full article

Cross-domain distribution shifts severely degrade the diagnostic performance of rolling bearing models under unseen variable operating scenarios. Single-source domain generalization (SDG) builds fault diagnosis models using only single-source vibration data, which fits the practical limitations of industrial data collection. Existing contrastive learning methods adopt uniform spectral perturbations for data augmentation, which easily corrupt fault harmonic characteristics and require massive, labeled training samples. To tackle these drawbacks, this paper proposes an unsupervised contrastive learning framework named FDACL. An adaptive frequency-domain augmentation (AFA) module equipped with learnable weights is designed to separate fault-critical frequency bands from noise components. Differentiated amplitude perturbations are applied to two categories of spectral signals to generate diverse pseudo-samples while retaining intrinsic fault information. A shared encoder is trained with combined InfoNCE contrast loss and classification loss to learn domain-invariant fault representations. Validations are carried out on three datasets, namely Case Western Reserve University (CWRU), Paderborn University (PU), and the industrial CRRC Qingdao Sifang railway wheelset bearing dataset acquired from physical test benches. FDACL achieves average cross-speed diagnostic accuracies of 92.68% and 77.85% on CWRU and PU, respectively, and maintains competitive performance on the Qingdao Sifang industrial dataset. It outperforms state-of-the-art baselines by 4.23–8.71% across all SDG transfer tasks. Ablation experiments and hyperparameter analysis verify the efficacy of the AFA module and contrastive learning scheme, providing an unsupervised diagnostic approach for railway bearings under unknown working conditions. Full article

Safe rail transit operation is essential for socioeconomic development and the protection of public life and property. Traditional railway foreign object intrusion detection methods suffer from insufficient detection accuracy and poor real-time performance under complex scene conditions. While deep learning has achieved remarkable performance in general object detection tasks, existing lightweight detectors still face prominent challenges in railway scenarios, including severe background clutter, drastic variations in target scales, and constrained edge computing resources. To tackle the above issues, this paper proposes YOLOv11n-LSL, an improved lightweight and high-precision detector based on YOLOv11n. Specifically, a C2PSA-SWSA shifted-window self-attention module is designed to suppress background interference and improve the feature representation of small targets; an SPPF-LSKA large-kernel attention module is introduced to construct a large and adaptive receptive field, thereby improving the detection capability for foreign objects of different scales; in addition, the lightweight adaptive decoupled head (LADH) is introduced to alleviate feature conflicts between the classification and regression branches and reduce network parameter redundancy. Comparative experiments on a self-built railway foreign object intrusion dataset show that the proposed YOLOv11n-LSL outperforms the original YOLOv11n, achieving 84.6% mAP@0.5, 88.1% precision, and 79.5% recall, which are 3.0, 6.0, and 7.8 percentage points higher than the baseline, respectively. The model only contains 2.522 M parameters and achieves a single-frame latency of 12.3 ms. Compared with the original YOLOv11n with 2.583 M parameters, the parameter count is reduced by 2.4%, and the inference latency is decreased by 41.1%. The experimental results show that the proposed method effectively improves detection accuracy in complex railway scenes while maintaining lightweight inference performance. Full article

Background/Objectives: Adverse drug reactions (ADRs) caused by drug–drug interactions (DDIs) represent an important problem in pharmacotherapy, especially in patients receiving multiple medications. Most computational approaches to DDI-associated ADR prediction formulate the task as a binary classification, but they do not explicitly consider the severity of adverse reactions. Our study aims to develop structure-based models that predict the severity of ADRs associated with specific drug pairs. Methods: Datasets were generated using DrugMAP as the source of drug pair–ADR associations with annotated severity categories, and TwoSides was used as an additional source to generate conditionally negative examples. The drug pairs were represented using PoSMNA descriptors, which encode pair-specific structural features derived from the molecular structures of both compounds. Predictive models were built using PASS DDI software. Model performance was evaluated using a modified cross-validation procedure that excluded compound-level overlap between the training and test sets, thereby reducing information leakage caused by the repeated occurrence of the same drugs in different pairs. Results: Models were developed for 14 clinically relevant ADR types, including cardiovascular, hepatotoxic, nephrotoxic, hemorrhagic, metabolic, and neurological effects. The unweighted class-level macro-average AUC values ranged from 0.830 for the Major category to 0.911 for the Minor category, while balanced accuracy ranged from 0.776 to 0.857. Predictive performance varied significantly between ADR types and severity categories. Higher accuracy was observed for some ADR types that were better captured by the structure-based descriptors used in this study, whereas complex multifactorial reactions, such as hepatotoxicity, were less accurately predicted. Case-based assessment using clinically documented drug combinations showed that the predicted severity profiles were generally consistent with the expected clinical risk patterns. Conclusions: The proposed approach demonstrates that the PoSMNA descriptors of drug pairs can be used for preliminary prediction of DDI-associated ADR severity. The developed models can help to filter out potentially dangerous drug combinations at an early stage and are implemented in the AdverDDIPred web-application. Full article