Search Results (201)

EEG-based classification of mental disorders has increasingly relied on deep learning models, which are computationally intensive and difficult to interpret, limiting reproducibility and clinical deployment in resource-constrained or cross-site settings. We propose a configurable and modular machine learning framework for EEG-based classification that emphasizes interpretability, flexibility, and rigorous evaluation using schizophrenia detection as a representative use case. Our framework integrates standardized preprocessing, multiscale feature extraction, minimum redundancy–maximum relevance feature selection, and configurable ensemble learning. It also supports multiple validation strategies, including random splits, k-fold cross-validation, and leave-one-subject-out (LOSO), enabling systematic assessment of subject-level generalization. We evaluated the framework on two open EEG datasets: Warsaw IPN (Institute of Psychiatry and Neurology, 19 channels, 250 Hz; 28 subjects) and a Moscow adolescent cohort (16 channels, 128 Hz; 84 subjects). Results show that validation strategy strongly affects model performance. While K-fold validation yielded epoch-level accuracies of 98.06% and 91.47%, LOSO results were much lower: 76.12% (epoch-level) and 82.14% (subject-level) for Dataset 1, and 70.71% (epoch-level) and 77.38% (subject-level) for Dataset 2. These findings demonstrate the risk of overestimated performance due to data leakage and underscore the importance of subject-independent evaluation. Our proposed framework provides a low-complexity, interpretable, and extensible benchmark for reproducible EEG-based machine learning, with interpretable feature representations linked to EEG dynamics and potential applicability to broader neuroengineering and clinical decision-support systems. Full article

Oil spills represent a severe environmental hazard, threatening marine and coastal ecosystems, biodiversity, and socio-economic stability. Timely and accurate detection of such incidents is critical for mitigating their ecological and economic consequences. Conventional detection techniques, including manual inspection and satellite-based observation, remain limited by high operational costs, temporal delays, and restricted spatial coverage. To overcome these limitations, this study introduces a comprehensive computer vision framework that addresses two core challenges: (i) the construction of a large-scale, high-quality synthetic oil spill dataset through mask extraction and seamless blending of oil spill regions with diverse oceanic backgrounds, and (ii) the development of a fine-tuned YOLOv11m-seg detection model trained on this enriched dataset. To further enhance the realism and spatial distinctiveness of oil spill textures, the Line Integral Convolution (LIC) is applied to estimate and visualize ocean surface flow patterns, generating coherent streamline textures that simulate the natural diffusion and transport of oil in water. The model exhibited strong generalization and precision, achieving a training accuracy exceeding IoU@0.50-0.95 to 85% over 50 epochs. Evaluation metrics confirmed its reliability, with an F1 score of 94%, precision of 94%, and recall (mAP@0.50) of 94%. These results demonstrate that the developed approach not only enhances dataset diversity but also substantially improves the accuracy and representativeness of real-time oil spill detection in marine environments. Full article

Classical Be stars are key laboratories for investigating how rapid rotation, pulsations, and mass loss couple to the formation and evolution of circumstellar decretion disks. However, few studies have combined Kepler/K2 photometry with multi-epoch H$\alpha$ monitoring. Here we present four previously unclassified Be-type variable stars observed by K2 (three in Campaign 11 and one in Campaign 15) and followed up with ground-based spectroscopy. We analyzed public PDC light curves and extracted variability frequencies using Lomb–Scargle periodograms and iterative prewhitening with a conservative detection threshold of S/N ≥ 5. Optical spectra obtained at the Observatório Pico dos Dias (Brazil) over a multi-year baseline (2017–2025) include repeated H$\alpha$ observations and blue-region spectra for photospheric characterization. All targets show detectable K2 variability on timescales from hours to days, with frequency spectra ranging from close multi-periodic components producing beating patterns to power dominated by low frequencies. Each star exhibits H$\alpha$ emission at multiple epochs, with long-term changes in line-profile morphology and equivalent width, indicating disk variability on year-long timescales. These results demonstrate that disk evolution can occur without conspicuous photometric outbursts over the time span of space-based observations, highlighting the diagnostic value of combining high-precision space photometry with long-term spectroscopy to characterize multiscale variability in Galactic Be stars. Full article

Fault diagnosis in modern power systems increasingly depends on unstructured operation and maintenance (O&M) logs, yet real-world logs are often small in scale and highly imbalanced across fault types, which degrades the generalizability of standard neural models. This paper proposes a graph-augmented diagnostic framework that integrates imbalance-aware knowledge extraction with interpretable reasoning. The framework consists of three stages: (1) domain adaptation of a BERT–BiLSTM–CRF NER model and a BERT–MLP RE model using an imbalance-aware training recipe that combines Low-Rank Adaptation (LoRA), a mixed focal–range loss, and undersampling; (2) construction of a power-system knowledge graph that organizes extracted entities and relations (e.g., fault devices, abnormal phenomena, causes, and handling measures); and (3) a graph-augmented assistant agent that reuses the NER model as a graph-aware retriever within a retrieval-augmented generation (RAG) architecture to support contextualized and interpretable diagnostic reasoning. Experiments on 3921 real-world fault-processing logs show consistent gains: NER reaches 92.0% accuracy and 71.3% Macro-F1 (vs. 80.3% and 63.2%), and RE achieves 88.0% accuracy and 70.1% F1 (vs. 82.1% and 60.4%), while reducing average training time per epoch by about 18%. These results demonstrate an efficient and practical path toward robust log-based fault diagnosis under scarce and imbalanced data. Full article

This study focuses on enhancing wind speed prediction for wildfire spread simulation by proposing an integrated forecasting approach. The original wind speed series is first processed via variational mode decomposition (VMD), with its parameters [K, $\alpha$] optimized via particle swarm optimization (PSO). Every intrinsic mode function (IMF) resulting from this decomposition is predicted using a bidirectional long short-term memory model incorporating an attention mechanism (AM-BiLSTM), and the final wind series is reconstructed from these predictions. Model training and validation were conducted using data from controlled burning experiments in the Mao’er Mountain area of Heilongjiang Province, China. Predictive performance is evaluated through multiple statistical metrics, error distribution analysis, and Taylor diagrams. To assess practical utility, the predicted wind field is further applied in FARSITE to drive wildfire spread simulations. Results demonstrate that the PSO-VMD-AM-BiLSTM model provides reliable wind forecasts and contributes to improved fire spread prediction accuracy, indicating its potential for decision support in wildfire management. To achieve accurate forest fire spread prediction, we construct the MCNN model, which is based on early perception of understory wind fields using predicted wind speed data and adopts a multi-branch convolutional neural network architecture to extract fire spread features. FARSITE is employed to simulate forest fire spread in the Mao’er Mountain region, generating a dataset for model training and testing. After 50 training epochs, the loss value of the MCNN model converges, achieving optimal prediction performance when the combustion threshold is set to 0.7. Compared to models such as CNN, DCIGN, and DNN, MCNN shows improvements in evaluation metrics including precision, recall, Sørensen coefficient, and Kappa coefficient. To validate the model’s predictive performance in real fire scenarios, four field ignition experiments were conducted at the Liutiao Village test site: homogeneous fuel combustion, long fire line combustion, alternating fuel combustion, and multiple ignition source merging combustion. Comprehensive evaluation across the four experiments indicates that the model achieves precision, recall, Sørensen coefficient, and Kappa coefficient values of 0.940, 0.965, 0.953, and 0.940, respectively, with stable prediction errors below 6%. These results represent improvements over the comparative models DCIGN and DNN. The proposed MCNN model can adapt to forest fire spread prediction under different scenarios, offering a novel approach for accurate forest fire prediction and prevention. Full article

This paper addresses the issue of low detection accuracy in underwater optical images for unmanned underwater vehicles (UUVs) during practical operations, caused by factors such as uneven lighting, blur, complex backgrounds, and target occlusion. To enhance the autonomous perception and control capabilities of UUVs, a high-precision algorithm named MBACA-YOLO is proposed based on the YOLOv13n model. Firstly, the convolutional layers in the backbone network of YOLOv13n are optimized by replacing stride-2 convolutions with stride-1 and embedding SPD layers to enable richer feature extraction. Secondly, the newly proposed MBACA attention mechanism is integrated into the final layer of the backbone network, enhancing effective features and suppressing background noise interference. Thirdly, traditional upsampling in the neck network is replaced with CARAFE upsampling to mitigate noise pollution. Finally, an Alpha-Focal-CIoU loss function is designed to improve the accuracy of bounding box regression for underwater targets. To validate the algorithm’s effectiveness, experiments were conducted on the URPC dataset with the following evaluation protocol: 640 × 640 input resolution, batch size 1, FP32 precision, and standard NMS. All results are from a single random seed with 300 epochs of training. The proposed MBACA-YOLO algorithm outperforms the baseline YOLOv13n model, improving mAP@0.5 and mAP@0.5:0.95 by 3.1% and 2.8% respectively, while adding only 0.49M parameters and 1.0 GFLOPs, with an FPS drop of just 2 frames. This makes it an efficient, deployable perception solution for automated Unmanned Underwater Vehicles (UUVs), significantly advancing intelligent underwater systems. Full article

Medical image segmentation, particularly for pathological diagnosis, faces challenges in leveraging patient clinical metadata that could enhance diagnostic accuracy. This study presents MMY-Net (Multimodal Y-shaped Network), a novel deep learning framework that effectively fuses patient metadata with pathological images for improved tumor segmentation performance. The proposed architecture incorporates a Text Processing Block (TPB) utilizing BERT for metadata feature extraction and a Text Encoding Block (TEB) for multi-scale fusion of textual and visual information. The network employs an Interlaced Sparse Self-Attention (ISSA) mechanism to capture both local and global dependencies while maintaining computational efficiency. Experiments were conducted on two open/public eyelid tumor datasets (Dataset 1: 112 WSIs for training/validation; Dataset 2: 107 WSIs as an independent test set) and the public Dataset 3 gland segmentation benchmark. For Dataset 1, 7989 H&E-stained patches (1024 × 1024, resized to 224 × 224) were extracted and split 7:2:1 (train:val:test); Dataset 2 was used exclusively for external validation. All images underwent Vahadane stain normalization. Training employed SGD (lr = 0.001), 1000 epochs, and a hybrid loss (cross-entropy + MS-SSIM + Lovász). Results show that integrating metadata—such as age and gender—significantly improves segmentation accuracy, even when metadata does not directly describe tumor characteristics. Ablation studies confirm the superiority of the proposed text feature extraction and fusion strategy. MMY-Net achieves state-of-the-art performance across all datasets, establishing a generalizable framework for multimodal medical image analysis. Full article

Electroencephalography (EEG) is an effective tool for monitoring the psycho-emotional state induced by open-skill sport activities characterized by dynamic environments and unpredictable situations, offering objective insights into mental engagement. This study aims to characterize the psycho-emotional state induced by table tennis sport activity by exploiting EEG-derived biomarkers. The ‘Real World Table Tennis’ database was analyzed, which includes EEG signals of 25 subjects acquired before, during and after playing table tennis. After preprocessing, 30-s EEG epochs were recursively extracted every 5 s. For each epoch, EEG rhythms were extracted and combined to obtain 37 engagement indexes, defined as ratios of two or more EEG rhythm powers. Median trends of each index were obtained for five cortical regions, and the Wilcoxon signed-rank test was applied to assess significant temporal changes. Results show that engagement indexes can effectively characterize psycho-emotional dynamics during table tennis, capturing the transition from resting to game phase in all cortical regions and exhibiting an increasing trend when having beta/alpha in their definition, and a decreasing trend when having high-frequency rhythms in the denominator. Our findings demonstrate the feasibility of using engagement indexes to monitor psycho-emotional states induced by open-skill sports and provide a framework for investigating mental engagement over time. Full article

This paper presents a hybrid deep learning framework for real-time sign language recognition (SLR) tailored to Internet of Things (IoT)-enabled environments, enhancing accessibility for Deaf communities. The proposed system integrates a Long Short-Term Memory (LSTM) network for static gesture recognition and a 3D Convolutional Neural Network (3D CNN) for dynamic gesture recognition. Implemented on a Raspberry Pi device using MediaPipe for landmark extraction, the system supports low-latency, on-device inference suitable for resource-constrained edge computing. Experimental results demonstrate that the LSTM model achieves its highest stability and performance for static signs at 1000 training epochs, yielding an average F1-score of 0.938 and an accuracy of 86.67%. In contrast, at 2000 epochs, the model exhibits a catastrophic performance collapse (F1-score of 0.088) due to overfitting and weight instability, highlighting the necessity of careful training regulation. Despite this, the overall system achieves consistently high classification performance under controlled conditions. In contrast, the 3D CNN component maintains robust and consistent performance across all evaluated training phases (500–2000 epochs), achieving up to 99.6% accuracy on dynamic signs. When deployed on a Raspberry Pi platform, the system achieves real-time performance with a frame rate of 12–15 FPS and an average inference latency of approximately 65 ms per frame. The hybrid architecture effectively balances recognition accuracy with computational efficiency by routing static gestures to the LSTM and dynamic gestures to the 3D CNN. This work presents a detailed epoch-wise comparative analysis of model stability and computational feasibility, contributing a practical and scalable IoT-enabled solution for inclusive, real-time sign-to-text communication in intelligent environments. Full article

Background and Objective: The accurate detection of sleep stages and disorders in older adults is essential for the effective diagnosis and treatment of sleep disorders affecting millions worldwide. Although Polysomnography (PSG) remains the primary method for monitoring sleep in medical settings, it is costly and time-consuming. Recent automated models have not fully explored and effectively fused the sleep features that are essential to identify sleep stages and disorders. This study proposes a novel automated model for detecting sleep stages and disorders in older adults by analyzing PSG recordings. PSG data include multiple channels, and the use of our proposed advanced methods reveals the potential correlations and complementary features across EEG, EOG, and EMG signals. Methods: In this study, we employed three novel advanced architectures, (1) CNNs, (2) CNNs with Bi-LSTM, and (3) CNNs with a transformer encoder, for the automatic classification of sleep stages and disorders using multichannel PSG data. The CNN extracts local features from RGB spectrogram images of EEG, EOG, and EMG signals individually, followed by an appropriate column-wise feature fusion block. The Bi-LSTM and transformer encoder are then used to learn and capture intra-epoch feature transition rules and dependencies. A residual connection is also applied to preserve the characteristics of the original joint feature maps and prevent gradient vanishing. Results: The experimental results in the CAP sleep database demonstrated that our proposed CNN with transformer encoder method outperformed standalone CNN, CNN with Bi-LSTM, and other advanced state-of-the-art methods in sleep stages and disorders classification. It achieves an accuracy of 95.2%, Cohen’s kappa of 93.6%, MF1 of 91.3%, and MGm of 95% for sleep staging, and an accuracy of 99.3%, Cohen’s kappa of 99.1%, MF1 of 99.2%, and MGm of 99.6% for disorder detection. Our model also achieves superior performance to other state-of-the-art approaches in the classification of N1, a stage known for its classification difficulty. Conclusions: To the best of our knowledge, we are the first group going beyond the standard to investigate and innovate a model architecture which is accurate and robust for classifying sleep stages and disorders in the elderly for both patient and non-patient subjects. Given its high performance, our method has the potential to be integrated and deployed into clinical routine care settings. Full article

Accurate estimation of battery state of health is essential for ensuring safety, supporting fault diagnosis, and optimizing the lifetime of electric vehicles. This study proposes a compact dual-path architecture that combines Convolutional Neural Networks with Convolutional Long Short-Term Memory (ConvLSTM) units to jointly extract spatial and temporal degradation features from charge-cycle voltage and current measurements. Residual and inter-path connections enhance gradient flow and feature fusion, while a three-channel preprocessing strategy aligns cycle lengths and isolates padded regions, improving learning stability. Operating end-to-end, the model eliminates the need for handcrafted features and does not rely on discharge data or temperature measurements, enabling practical deployment in minimally instrumented environments. The model is evaluated on the NASA battery aging dataset under two scenarios: Same-Battery Evaluation and Leave-One-Battery-Out Cross-Battery Generalization. It achieves average RMSE values of 1.26% and 2.14%, converging within 816 and 395 epochs, respectively. An ablation study demonstrates that the dual-path design, ConvLSTM units, residual shortcuts, inter-path exchange, and preprocessing pipeline each contribute to accuracy, stability, and reduced training cost. With only 4913 parameters, the architecture remains robust to variations in initial capacity, cutoff voltage, and degradation behavior. Edge deployment on an NVIDIA Jetson AGX Orin confirms real-time feasibility, achieving 2.24 ms latency, 8.24 MB memory usage, and 12.9 W active power, supporting use in resource-constrained battery management systems. Full article

Predictive maintenance is increasingly important in rotating machinery to prevent unexpected failures, reduce downtime, and improve operational efficiency. This study compares the efficacy of traditional machine learning (ML) and deep learning (DL) techniques in diagnosing bearing faults under varying load and speed conditions. Two classification tasks were conducted: a simpler three-class task that distinguishes healthy bearings, inner race faults, and outer race faults, and a more complex nine-class task that includes faults of varying severity in the inner and outer races. In this study, the machine learning algorithm ensemble bagged trees, achieved maximum accuracies of 93.04% for the three-class and 87.13% for the nine-class classifications, followed by neural network, SVM, KNN, decision tree, and other algorithms. For deep learning, the CNN model, trained on scalograms (time–frequency images generated by continuous wavelet transform), demonstrated superior performance, reaching up to 100% accuracy in both classification tasks after six training epochs for the nine-class classifications. While CNNs take longer training time, their superior accuracy and capability to automatically extract complex features make the investment worthwhile. Consequently, the results demonstrate that the CNN model trained on CWT-based scalogram images achieved remarkably high classification accuracy, confirming that deep learning methods can outperform traditional ML algorithms in handling complex, non-linear, and dynamic diagnostic scenarios. Full article

Stress, workload, and uncertainty characterize occupational tasks across sports, healthcare, military, and transportation domains. Emerging theory and empirical research suggest that coordinated whole-body movements may reflect these transient mental states. Wearable sensors and optical motion capture offer opportunities to quantify such movement dynamics and classify mental states that influence occupational performance and human–machine interaction. We tested this possibility in a small pilot study (N = 10) designed to test feasibility and identify preliminary movement features linked to mental states. Participants performed a perceptual decision-making task involving facial emotion recognition (i.e., deciding whether depicted faces were happy versus angry) with variable levels of stress (via a risk of electric shock), workload (via time pressure), and uncertainty (via visual degradation of task stimuli). The time series of movement trajectories was analyzed both holistically (full trajectory) and by phase: lowered (early), raising (middle), aiming (late), and face-to-face (sequential). For each epoch, up to 3844 linear and non-linear features were extracted across temporal, spectral, probability, divergence, and fractal domains. Features were entered into a repeated 10-fold cross-validation procedure using 80/20 train/test splits. Feature selection was conducted with the T-Rex Selector, and selected features were used to train a scikit-learn pipeline with a Robust Scaler and a Logistic Regression classifier. Models achieved mean ROC AUC scores as high as 0.76 for stress classification, with the highest sensitivity during the full movement trajectory and middle (raise) phases. Classification of workload and uncertainty states was less successful. These findings demonstrate the potential of movement-based sensing to infer stress states in applied settings and inform future human–machine interface development. Full article

Background/Objectives: Artificial intelligence (AI) has been increasingly used to support radiological assessment in dentistry. We benchmarked machine learning (ML), deep learning (DL), and a hybrid optimization-assisted approach for the automatic five-class image-level classification of dental restorations (filling, implant, root canal treatment, fixed partial denture/bridge, crown) on panoramic radiographs. Methods: We analyzed 353 anonymized panoramic images comprising 2137 labeled restorations, acquired on the same device. Images were cropped and enhanced (histogram equalization and CLAHE), and texture features were extracted with GLCM. A three-stage pipeline was evaluated: (i) GLCM-based features classified by conventional ML and a baseline DL model; (ii) Hybrid Grey Wolf–Particle Swarm Optimization (HGWO-PSO) for feature selection followed by SVM; and (iii) a CNN trained end-to-end on raw images. Performance was assessed with an 80/20 per-patient split and 5-fold cross-validation on the training set. While each panoramic radiograph may contain multiple restorations, in this study we modeled the task as single-label, image-level classification (dominant restoration type) due to pipeline constraints; this choice is discussed as a limitation and motivates multi-label, localization-based approaches in future work. The CNN baseline was implemented in TensorFlow 2.12 (CUDA 11.8/cuDNN 8.9) and trained with Adam (learning rate 1 × 10⁻⁴), with a batch size 32 and up to 50 epochs with early stopping (patience 5); data augmentation included horizontal flips, ±10° rotations, and ±15% brightness variation. A post hoc power analysis (G*Power 3.1; α = 0.05, β = 0.2) confirmed sufficient sample size (n = 353, power > 0.84). Results: The HGWO-PSO + SVM configuration achieved the highest accuracy (73.15%), with macro-precision/recall/F1 = 0.728, outperforming the CNN (68.52% accuracy) and traditional ML models (SVM 67.89%; DT 59.09%; RF 58.33%; K-NN 53.70%). Conclusions: On this single-center dataset, the hybrid optimization-assisted classifier moderately improved detection performance over the baseline CNN and conventional ML. Given the dataset size and class imbalance, the proposed system should be interpreted as a decision-supportive tool to assist dentists rather than a stand-alone diagnostic system. Future work will target larger, multi-center datasets and stronger DL baselines to enhance generalizability and clinical utility. Full article

Unsupervised Hebbian learning is a biologically inspired algorithm designed to extract representations from input images, which can subsequently support supervised learning. It presents a promising alternative to traditional artificial neural networks (ANNs). Many attempts have focused on enhancing Hebbian learning by incorporating more biologically plausible components. Contrarily, we draw inspiration from recent advances in ANNs to rethink and further improve Hebbian learning in three interconnected aspects. First, we investigate the issue of overfitting in Hebbian learning and emphasize the importance of selecting an optimal number of training epochs, even in unsupervised settings. In addition, we discuss the risks and benefits of anti-Hebbian learning in model performance, and our visualizations reveal that synapses resembling the input images sometimes do not necessarily reflect effective learning. Then, we explore the impact of different activation functions on Hebbian representations, highlighting the benefits of properly utilizing negative values. Furthermore, motivated by the success of large pre-trained language models, we propose a novel approach for leveraging unlabeled data from other datasets. Unlike conventional pre-training in ANNs, experimental results demonstrate that merging trained synapses from different datasets leads to improved performance. Overall, our findings offer fresh perspectives on enhancing the future design of Hebbian learning algorithms. Full article