Search Results (88)

Facial emotion recognition (FER) in real-world settings is limited by the semantic mismatch between discrete emotion categories and continuous Valence–Arousal–Dominance (V-A-D) dimensions and the lack of reliable uncertainty estimates under incomplete facial evidence. Existing uncertainty-aware FER studies mainly address annotation ambiguity or training-time reliability, leaving the behavior of predictive uncertainty under progressive input degradation insufficiently examined. This paper proposes BGDC (Bayesian Gaussian-mixture Distributional Consistency), a multi-task FER framework that integrates a GMM-based soft consistency module with a context-conditioned Bayesian regression head and explicitly models aleatoric and epistemic uncertainty. To evaluate predictive reliability, a controlled masking protocol is introduced to remove facial information under different spatial configurations. On FER2013-VAD, BGDC attains the highest classification accuracy of 0.6943 and the highest mean V-A-D CCC of 0.6079 among the compared configurations, and it yields a stronger epistemic uncertainty-error correspondence than MC Dropout in a single-model setting. Controlled masking further shows that the epistemic uncertainty of BGDC tracks task-relevant facial information loss rather than masking ratio alone: it rises with regression error when diagnostically important regions are removed, and it contracts when the masked region is largely task-irrelevant. Combining Bayesian uncertainty with the GMM-based distributional prior thus enables reliability-aware multi-task FER, in which controlled masking serves as a diagnostic intervention rather than as a benchmark of accuracy degradation alone. Full article

Structural health monitoring (SHM) of additively manufactured (AM) small and complex components is investigated using a sensor-based signal processing and machine-learning framework. Guided-wave responses acquired from piezoelectric transducers are analyzed to evaluate the performance of sweep-sine and pulse excitation signals, as well as the influence of infill patterns, part geometry, and defect type on system reliability. Test specimens, including dogbone structures and a simulated rocket-nozzle component, were fabricated using AM, and nonstationary guided-wave signals were recorded and processed. Time–frequency signal representations (scalograms) were generated using the Continuous Wavelet Transform (CWT). Convolutional Neural Networks (CNNs) and Gaussian Mixture Models (GMMs) were employed for supervised classification and unsupervised clustering, respectively. Sweep-sine excitation consistently yielded higher classification accuracy, with CNN analysis achieving near-perfect performance and GMM clustering demonstrating improved group separability. In contrast, pulse excitation revealed transient signal features associated with wave interactions, including reflections, mode conversion, and scattering, highlighting its potential for complementary signal-based diagnostics. Importantly, the proposed hybrid supervised–unsupervised learning framework enables the quantification of previously unseen intermediate load states, demonstrating strong adaptability and generalizability beyond the conditions represented in the training data. Full article

Port State Control (PSC) inspections play a critical role in enforcing international maritime safety and environmental standards, yet little is known about how compliance behaviour interacts with economic cycles. This study examines the relationship between vessel detentions and freight market conditions using monthly data from the Paris and Tokyo Memoranda of Understanding (MoUs) over the period 2010–2021. A system of simultaneous equations is estimated using the Generalized Method of Moments (GMM) and Three-Stage Least Squares (3SLS) to address the bidirectional relationship between detention activity and freight market conditions, proxied by the Baltic Dry Index (BDI) and, for tanker specifications, the Baltic Dirty Tanker Index (BDTI). The results are consistent with a positive and statistically significant bidirectional relationship: vessel detentions increase during periods of strong freight market conditions, while past detentions are positively associated with freight rates, a pattern consistent with a signalling and sentiment channel. Institutional factors, including flag state quality, classification society affiliation, and ISM-related deficiencies, are also found to significantly influence detention risk, though their direction and magnitude vary across MoUs and vessel segments. These findings are consistent with the presence of opportunistic incentives during economic upswings, challenging the conventional expectation that stronger market conditions promote higher compliance. The study contributes to the literature by linking regulatory compliance with economic cycles and highlighting the importance of adaptive, risk-based enforcement strategies. It is important to note, however, that the aggregate nature of the data does not permit direct identification of firm-level behavioural mechanisms, and the findings should be interpreted as associational evidence consistent with these theoretical mechanisms. Full article

Nowadays, educational software across all learning levels is increasingly enhanced with Artificial Intelligence (AI), primarily through content generation or post-session learning analytics. However, most existing systems remain weakly connected to learners’ real-time affective states and rarely exploit emotional information as a direct control signal for instructional adaptation. In this work, we propose a proof-of-concept closed-loop affect-aware educational adaptation framework that integrates real-time facial emotion recognition into a dynamic learning control system. The proposed approach is built upon a dual-model ensemble architecture, combining a transformer-based model (CAGE) and a CNN-based model (DDAMFN++) trained on large-scale in-the-wild datasets. To bridge heterogeneous emotion representations, we introduce a probabilistic fusion strategy that aligns continuous valence–arousal predictions with discrete emotion classification via a Gaussian Mixture Model (GMM), enabling unified emotion inference in real time. Based on the fused emotional state, a temporal aggregation mechanism is applied to capture sustained affective trends rather than transient expressions. These aggregated signals are then mapped to instructional decisions through an emotion-driven adaptive control policy, which adjusts activity difficulty using an Average Emotion Score (AES). This establishes a fully automated closed-loop adaptation cycle, where detected learner affect directly influences the learning environment without requiring explicit user input or post-session questionnaires. The framework is integrated into an open-source educational platform (eduActiv8) to demonstrate feasibility and system-level behavior. Results from alpha-level validation show that the system can continuously monitor learner affect, generate interpretable emotional analytics, and dynamically adjust task difficulty in real time, while reducing user interaction overhead. This study contributes a modular architecture for affect-aware educational systems by combining real-time ensemble emotion recognition, probabilistic fusion of heterogeneous outputs, and closed-loop instructional adaptation. The proposed framework provides a foundation for future research in scalable, emotion-driven intelligent tutoring and adaptive learning environments. Full article

Automated detection of suspicious human activities in complex and crowded environments remains a critical challenge in modern surveillance systems due to high false-positive rates, poor contrast and generalization across diverse scenes. We propose a GM_CNN3D Model for the classification of suspicious activity based on a Deep Fused Feature Block (DFFB) framework that integrates handcrafted spatial descriptors (PCA-HOG and Motion-HOG) with deep spatiotemporal features extracted from 3D Convolution Neural Network (3D-CNN). Motion regions are first localized using a Gaussian Mixture Model (GMM), after which handcrafted and deep features are concatenated in a dimensionality-normalized fusion stage, followed by a fully connected layer and softmax classification. The system is evaluated on five diverse and publicly available datasets: Violent Crowd, Hockey Fight, Kaggle Fight, Movies Fight, and Custom Annotated YouTube Clips, achieving up to 99.12% accuracy, 98.7% F1-score, and a ROC-AUC of 0.992, outperforming state-of-the-art CNN, LSTM, and SlowFast models. All datasets include real world scenarios with varying lighting, crowd density, and camera viewpoints, with annotations created manually where unavailable. The proposed method demonstrates robust cross-scene performance, enabling automated alarming and reduced false positives in real-time security operations. Full article

Therapist shortages make home-based rehabilitation an essential component of post-stroke care, yet patients often exhibit reduced adherence when functional gains are difficult to quantify and interpret. This study presents a cloud-enabled assessment framework centered on a dynamic reaching task for upper-limb rehabilitation in individuals with mild stroke. The proposed system combines wearable sensing and Internet of Things (IoT) connectivity to stream kinematic data to the cloud for near real-time analysis, and integrates a force-feedback rehabilitation robot to deliver motion guidance during training. The pipeline proceeds in three stages. First, smoothness-related kinematic descriptors are extracted and fed into a deep multi-class classifier to discriminate the affected side (left, right, or healthy). Second, movement quality is modeled using a Gaussian Mixture Model (GMM) trained on IoT-acquired trajectories to quantify performance via probabilistic similarity. Third, a calibrated scoring function transforms GMM log-likelihood into a normalized 0–1 quality index, producing visual reports that support interpretable feedback for patients and therapists. The framework is validated using motion data collected from stroke patients at Taipei Veterans General Hospital. Experimental results demonstrate that the neural network multi-classifier achieved an F1-score of 0.95. Incorporating robot-derived interaction signals further improved classification performance by approximately 5%. For movement quality assessment, the derived scores showed a significant positive correlation (Pearson correlation = 0.632, p = 0.02) with therapist-defined gold reference standards for right-affected patients. Additionally, integrating robot force-feedback signals and AIoT-enabled dynamic streams improved score accuracy by 8% and score responsiveness by 10%. These quantitative outcomes substantiate the efficacy of combining IoT-driven sensing and robot-assisted training for objective, interpretable, and remotely deployable motor assessment. Full article

Implantable Cardioverter-Defibrillators (ICDs) are life-sustaining devices used in the prevention of sudden death in patients suffering from advanced cardiac diseases. Although improvements in ICD technology and monitoring capabilities have been made, existing techniques are still not effective in predicting accelerated battery drain, thereby causing unplanned generator replacement and clinically significant device-related events. In this study, machine learning techniques were employed in the assessment of the early detection of ICD battery depletion risk using the collected device interrogation reports. The dataset used consisted of 32 devices in the training set and nine in the testing set. In order to mitigate the problem of a small sample size, a GMM-based data augmentation technique was applied solely to the training data, and actual devices were used for the testing data. Five supervised models, including Logistic Regression, Random Forest, SVM, CatBoost, and a Neural Network (MLP), have been utilized using a repeated stratified cross-validation and a separate held-out test data set. All the models have been tested for their performance using classification metrics. All models demonstrated variable performance with wide confidence intervals due to limited sample size. Support vector machines showed the highest cross-validation discrimination 0.889 ± 0.203, though uncertainty remains substantial given the small datasets (n = 41). From the feature analysis, it was found that atrial sensing amplitude, RV/LV capture threshold, output settings, and implant duration were the most important features for the prediction of high battery depletion risk. These findings suggest that changes in device parameters and implant age are associated with elevated battery depletion risk, supporting the feasibility of telemetry-driven risk stratification for device management. Full article

This study proposes a hybrid deep learning framework that integrates supervised classification and unsupervised profiling for electricity consumption analysis. In the supervised phase, a one-dimensional Convolutional Neural Network combined with Long Short-Term Memory (1D CNN–LSTM) architecture is developed to classify daily load patterns. The performance of the proposed model is compared with traditional machine learning and deep learning approaches, including Support Vector Machine (SVM), k-Nearest Neighbors (KNN), a standalone Long Short-Term Memory (LSTM) model, a Transformer-based model, and a standalone 1D CNN model. Experimental results on the Precon house dataset and the CU-BEMS dataset demonstrate that the proposed hybrid architecture outperforms the benchmark models, achieving classification accuracies of 87.59% and 86.40%, respectively. In the unsupervised phase, the trained CNN–LSTM encoder is utilized as a deep feature extractor. The resulting 32-dimensional latent embeddings are clustered using K-Means, Gaussian Mixture Model (GMM), Agglomerative, Spectral, and Ensemble methods. Clustering robustness is evaluated through bootstrap-based stability analysis using the Adjusted Rand Index (ARI) and the Normalized Mutual Information (NMI). The results demonstrate stable and interpretable electricity consumption profiles, particularly in the residential dataset, where near-perfect clustering stability is observed for K-Means. The proposed framework provides both improved classification performance and robust consumption profiling based on deep embedding, offering a practical tool for energy management. Full article

Accurate engine fault classification with limited labeled data is critical for the safety and reliability of rotating machinery. This task is challenging because operating regimes are time-varying, and key variables must satisfy physical constraints, under which traditional feature classifier pipelines degrade and deep networks tend to overfit. We propose PhysGMM-MoE, a physics-aware Gaussian Mixture Model (GMM)-Mixture-of-Experts (MoE) framework for small-sample engine fault classification. At the data level, PhysGMM-MoE fits class-conditional, regime-aware GMMs and performs physically constrained, distance-based quality control to selectively augment minority classes while preserving engine operating semantics. At the model level, a heterogeneous pool of lightweight statistical experts and a lightweight Transformer-based deep expert (ECFT-Transformer) capture complementary neighborhood cues and high order multi-sensor correlations, and an L2-regularized logistic regression meta-learner fuses expert outputs via stacking. We evaluate fault classification on the 3500-DEFault diesel-engine dataset using the adopted eight-class cylinder-fault labeling (H, F1–F7) built from in-cylinder pressure statistics and torsional-vibration harmonics; although severity levels exist in the dataset, this study focuses on classification rather than severity estimation. With 40 training samples per class, PhysGMM-MoE achieves a mean accuracy of 0.9875, exceeding SMOTE+XGBoost by 0.0086, and attains the best macro precision/recall/F1 of 0.9878/0.9826/0.9889, demonstrating strong performance under the adopted small-sample setting. Full article

Accurate quantification of vegetation vertical structure is crucial for analyzing the ecological functions of urban green spaces. However, constrained by the complexity of vegetation structure and spatial heterogeneity, current approaches for extracting vegetation vertical structure by airborne LiDAR have limitations in terms of layer boundary identification stability, threshold dependency, and ecological plausibility. This study developed two integrated UAV LiDAR-based stratification frameworks for identifying urban riparian vegetation vertical structure by combining established statistical modeling and signal processing techniques: (1) a Gaussian Mixture Model with Bayesian Information Criterion (GMM-BIC)-based probabilistic stratification framework; (2) a Savitzky–Golay filtering and Pruned Exact Linear Time (SG-PELT)-based change-point detection framework. Furthermore, the ecological height constraint was incorporated into the model to achieve biological adjustments. Two models were applied in the study area and compared using reference data. The results showed that the GMM-BIC method achieved an overall classification accuracy of 91.06%, with a macro-averaged F1-score of 87.77%, while the SG-PELT method attained an overall accuracy of 84.57%, with a macro-averaged F1-score of 79.20%. These results demonstrate that both models can effectively identify the vertical structure of urban vegetation. In particular, the two models exhibited distinct characteristics across different scenarios. The GMM-BIC model showed superior stratification accuracy in regions where vegetation height distribution displayed pronounced multi-peak characteristics and distinct differences among height segments. In comparison, the SG-PELT model demonstrated greater sensitivity in areas with significant height variation and clearly defined abrupt transitions between layers. These models could provide new methodologies for monitoring vegetation vertical structure and offer data support for biodiversity monitoring and ecological function assessment within urban ecosystems. Full article

Unsafe events such as slips and trips occur regularly on construction sites. Efficient identification of these events can help protect workers from accidents and improve site safety. However, current detection methods rely on subjective reporting, which has several limitations. To address these limitations, this study presents a sound-based slip and trip classification method using wearable sound sensors and machine learning. Audio signals were recorded using a smartwatch during simulated slip and trip events. Various 1D and 2D features were extracted from the processed audio signals and used to train several classifiers. Three key findings are as follows: (1) The hybrid CNN-LSTM network achieved the highest classification accuracy of 0.966 with 2D MFCC features, while GMM-HMM achieved the highest accuracy of 0.918 with 1D sound features. (2) 1D MFCC features achieved an accuracy of 0.867, outperforming time- and frequency-domain 1D features. (3) MFCC images were the best 2D features for slip and trip classification. This study presents an objective method for detecting slip and trip events, thereby providing a complementary approach to manual assessments. Practically, the findings serve as a foundation for developing automated near-miss detection systems, identification of workers constantly vulnerable to unsafe events, and detection of unsafe and hazardous areas on construction sites. Full article

Crop health assessment is essential for the early detection of nutrient deficiencies, diseases, and pests, allowing for timely interventions that optimize yield, reduce losses, and support sustainable agricultural practices. While traditional methods and satellite-based remote sensing offer broad scale monitoring, they often suffer from coarse spatial resolution, and insufficient precision at the plant level. These limitations hinder accurate and dynamic assessment of crop health, particularly for high-resolution applications such as nutrient diagnosis during different crop growth stages. This study addresses these gaps by leveraging high-resolution UAV (Unmanned Aerial Vehicle) imagery to monitor the health of paddy crops across multiple temporal stages. A novel methodology was implemented to assess the crop health condition from the predicted Above-Ground Biomass (AGB) and essential macro-nutrients (N, P, K) using vegetation indices derived from UAV imagery. Four machine learning models were used to predict these parameters based on field observed data, with Random Forest (RF) and XGBoost outperforming other algorithms, achieving high regression scores (AGB > 0.92, N > 0.96, P > 0.92, K > 0.97) and low prediction errors (AGB < 80 gm/m², N < 0.11%, P < 0.007%, K < 0.08%). A significant contribution of this study lies in the development of decision-making rules based on threshold values of AGB and specific nutrient critical, optimum, and toxic levels for the paddy crop. These rules were used to derive crop health maps from the predicted AGB and NPK values. The resulting spatial health maps, generated using RF and XGBoost models with high classification accuracy (Kappa coefficient > 0.64), visualize intra-field variability, allowing for site-specific interventions. This research contributes significantly to precision agriculture by offering a robust, plant-level monitoring approach that supports timely, site-specific nutrient management and enhances sustainable crop production practices. Full article

Human–computer interaction (HCI) plays a crucial role across various fields, with eye-tracking technology emerging as a key enabler for intuitive and dynamic control in assistive systems like Assistive Robotic Arms (ARAs). By precisely tracking eye movements, this technology allows for more natural user interaction. However, current systems primarily rely on the single gaze-dependent interaction method, which leads to the “Midas Touch” problem. This highlights the need for real-time eye movement classification in dynamic interactions to ensure accurate and efficient control. This paper proposes a novel Gaussian Mixture Model–Hidden Markov Model (GMM-HMM) classification algorithm aimed at overcoming the limitations of traditional methods in dynamic human–robot interactions. By incorporating sum of squared error (SSE)-based feature extraction and hierarchical training, the proposed algorithm achieves a classification accuracy of 94.39%, significantly outperforming existing approaches. Furthermore, it is integrated with a robotic arm system, enabling gaze trajectory-based dynamic path planning, which reduces the average path planning time to 2.97 milliseconds. The experimental results demonstrate the effectiveness of this approach, offering an efficient and intuitive solution for human–robot interaction in dynamic environments. This work provides a robust framework for future assistive robotic systems, improving interaction intuitiveness and efficiency in complex real-world scenarios. Full article

Acoustic emission (AE) signals exhibit a strong correlation with concrete damage. However, the relationship between column damage and AE signals under eccentric loading conditions, combined with the application of traditional RA-AF classification methods for crack characterization, demonstrates limitations. These approaches provide insufficient resolution to accurately identify damage types throughout the entire structural failure process. This study employed K-means clustering algorithm and Gaussian mixture models (GMMs) to analyze AE signal features from reinforced concrete (RC) columns undergoing failure under the eccentric compression loading of different eccentricity. Subsequently, a random forest model was used for automated damage stage classification. Experimental results demonstrate that the damage progression in eccentrically compressed columns comprises four distinct stages, each exhibiting unique AE signal characteristics. The integrated approach of clustering and random forest modeling demonstrates robust feasibility in identifying AE signal patterns associated with specific damage stages, achieving an 85% recognition rate for damage stage classification. These findings provide quantitatively validated evidence supporting the efficacy of machine learning-based methodologies for enabling stage-specific damage characterization in structural health monitoring applications. Full article

Regarding the issue of constructing Fourier transform infrared (FTIR) spectral characteristics of hot jet of aero-engines, this paper presented a construction algorithm for the FTIR spectral characteristics of an aero-engine hot jet, which integrated staged refined processing and statistical feature fusion. First, a remote-sensing Fourier transform infrared spectrometer was employed to collect data on the hot jets of two distinct types of aero-engines, thereby establishing a measured spectral dataset. Subsequently, a multi-dimensional feature extraction vector construction algorithm was proposed, encompassing a peak feature extraction algorithm based on staged refined processing and a high-order statistical feature extraction algorithm. The peak feature extraction algorithm based on staged refined processing consisted of four steps: “coarse detection—local optimization—dynamic screening—intelligent merging”. It adopted an adaptive threshold for the initial coarse detection of peaks, enhanced the positioning accuracy through local gradient optimization, dynamically screened the local strongest peak according to intensity information, and resolved the problem of overlapping peak resolution via an intelligent merging strategy based on the physical characteristics of spectral lines, achieving high-precision and high-robustness peak feature extraction. The high-order statistical feature extraction algorithm realized the extraction of the intensity distribution information and waveform symmetry information of the spectral signal by fusing the kurtosis and skewness statistics. Compared with the traditional feature construction algorithms, the multi-dimensional feature vector construction algorithm proposed in this paper possessed a higher-dimensional comprehensive representation capability. In the experiment, we selected the GMM classifier of the unsupervised clustering algorithm. The classification accuracy of the features extracted by the algorithm in this paper on this classifier reached 82.42%, thereby validating the effectiveness of the algorithm presented in this paper. Full article