Search Results (7,441)

Deep learning models for power quality (PQ) disturbance classification remain critically vulnerable to adversarial perturbations, with classification performance degrading severely under white-box attacks. Existing defenses address individual models in isolation and provide no mechanism for operators to assess whether the system is under attack or which classifier remains trustworthy. This paper proposes a two-stage framework that combines adversarial training with large language model (LLM) reasoning to improve both robustness and interpretability. In the first stage, four architecturally diverse classifiers, including a proposed Multi-Scale Temporal Attention Network (MSTAN), are evaluated under four adversarial attacks (FGSM, PGD, C&W, and UAP), and their failure patterns are recorded as structured vulnerability fingerprints. The ensemble is then retrained via adversarial training on mixed clean and perturbed signals. In the second stage, an LLM analyzes the ensemble predictions alongside the fingerprint knowledge base to perform attack detection, fingerprint-guided meta-classification, and operator-facing threat report generation. On a 17-class, 255,000-signal synthetic benchmark, adversarial training recovers FGSM and PGD accuracy from below 25% to the 53–78% range, with MSTAN achieving the highest post-training robustness (78.26% under FGSM, 65.41% under PGD). The LLM reasoning layer provides an additional 3.5–6.2 percentage point improvement over majority voting by selecting the most reliable ensemble member based on the inferred attack condition, and detects adversarial attacks with 87.6% overall accuracy. To our knowledge, this is the first integration of LLM-based ensemble reasoning into the PQ adversarial robustness pipeline and the first application of the C&W optimization attack to power quality signals. Full article

This review analyses and logically structures modern intelligent sensor technologies in the context of animal husbandry, feed production, and veterinary medicine. The main research discussed in the article focuses on machine learning based on modern neural network models, computer vision, and sensor systems that are transforming the methods for assessing the health, behavior, and nutrition of farm animals. The first part examines modern approaches to quality control and optimization of mineral and vitamin premixes, including visual inspection using visual sensors and neural networks. Key roles are played by precise dosing, component stability (minerals, vitamins), and the transition to more bioefficient organic forms of micronutrients to reduce environmental impact. Improvements in feed and premix production are analyzed, including automation, energy management, and the use of machine learning for non-destructive quality control, defect detection, mixing homogeneity assessment, and vitamin stability prediction. The second part analyzes methods for animal location and behavior detection. This article presents computer vision-based systems, including modifications of YOLO, for automatically tracking and classifying key behavioral patterns (lying down, standing, feeding, and aggression) in cattle and pigs, even in crowded conditions. It also discusses the use of ultra-wideband (UWB) systems and accelerometers combined with machine learning for high-precision positioning and detection of specific behavioral anomalies, such as lameness and playfulness. The third section focuses on the application of machine learning in veterinary diagnostics, including the automated interpretation of medical images (X-ray, ultrasound, and MRI) as sensor data streams for the diagnosis of cardiovascular, oncological, and orthopedic diseases in farm and small animals. Furthermore, the article examines the use of machine learning models for proactive disease diagnosis in farm animals and poultry based on multimodal data and image analysis. Considerable attention is given to methods and tools for radiometric diagnosis of animal diseases at an early stage using microwave sensors, as well as laser therapy and surgery in veterinary medicine. The review concludes that the integration of intelligent systems enables a transition to data-driven livestock management, significantly improving animal welfare and, consequently, the efficiency and sustainability of agricultural production. Full article

Background: Spinocerebellar ataxia type 2 (SCA2) is an inherited neurodegenerative disorder characterized by progressive cerebellar degeneration. One difficulty in treating this disease lies in identifying preclinical carriers: individuals who carry the pathogenic ATXN2 mutation but remain asymptomatic with respect to motor manifestations. Though magnetic resonance imaging (MRI) has proven valuable in supporting the diagnosis of ataxia, traditional univariate approaches using linear measurements have shown limited ability to capture the complex anatomical changes that occur across the disease spectrum, particularly during the preclinical phase. Methods: This study employed a comprehensive multivariate approach to improve the classification of individuals across the SCA2 spectrum. We developed a multinomial logistic regression model incorporating multiple linear measurements derived from magnetic resonance imaging to discriminate between healthy controls (n = 72), preclinical carriers (n = 17), and patients with manifest SCA2 (n = 61). To mitigate inherent class imbalance, particularly in the smaller preclinical subgroup, we implemented the Synthetic Minority Over-sampling Technique (SMOTE), generating a balanced dataset that enhances the model’s ability to discern the distinctive anatomical features. This was compared to the model applied to the unbalanced data. An improvement was observed when applying SMOTE. Results: The multivariate model demonstrated discriminatory performance, achieving an overall accuracy of 80.7%. The ability to identify healthy controls (AUC: 0.96), preclinical individuals (AUC: 0.75), and clinical individuals (AUC: 95%). This represents an advance over previous univariate approaches, which have had difficulty capturing the neurodegenerative changes characteristic of the preclinical stage. Conclusions: By integrating multiple neuroimaging biomarkers into a multivariable model, this study provides a tool for early identification of preclinical SCA2 carriers. The ability to accurately classify these individuals opens an opportunity for early therapeutic intervention before irreversible neurological deterioration occurs. This approach shows promise for optimizing clinical trial design and personalized care in SCA2. Full article

Film posters serve as front-end visual communication media that shape viewers’ initial judgments of film genre, emotional tone, and viewing appeal. However, whether the optimal overall color configuration follows a universal rule or varies across poster types remains insufficiently examined. This study investigated how overall lightness and chroma influence the communication effects of film posters and identified type-specific optimal color intervals. Based on a cross-type poster sample library, film posters were classified into four visual grammar types: affable-entertaining, relational-emotional, spectacle-dynamic, and threat-suspenseful. Type-specific quantile thresholds for lightness and chroma were established within each category. Eye-tracking data, subjective ratings, mixed-effects response surface modeling, and constrained desirability optimization were combined to identify optimal regions of overall color configuration. The results show that no single optimal lightness–chroma interval applies across all poster types. The dominant optimal interval was low lightness–high chroma for affable-entertaining and relational-emotional posters, high lightness–low chroma for spectacle-dynamic posters, and medium lightness–high chroma for threat-suspenseful posters. These findings indicate that overall color optimization varies across poster types within the present experimental context and provide practical support for evidence-based, type-specific poster color design. Full article

Background: Advances in monitoring systems featuring wearable sensors, computer vision, and artificial intelligence (AI) have been increasingly used in sports science and rehabilitation practices as a means of movement pattern analysis, injury prevention, and training optimization. These technologies are becoming essential components of athlete-performance analysis and rehabilitation-monitoring systems designed to support biomechanical assessment, athlete development, and movement-quality evaluation. Athlete-performance analysis and rehabilitation monitoring increasingly rely on intelligent multimodal sensing systems capable of continuously evaluating movement quality, biomechanical patterns, training execution, and recovery progress. Human activity recognition (HAR) serves as a key enabling technology for these applications by providing automated assessment of human movement using wearable and vision-based sensing modalities. Therefore, the purpose of this study was to develop and evaluate an attention-based multimodal framework that integrates wearable inertial sensing and RGB video analysis for robust athlete-performance assessment and rehabilitation monitoring through accurate recognition of human movement patterns. Methods: Athlete-performance analysis and rehabilitation monitoring combining inertial sensor data and RGB-based visual information was introduced. Inertial signals were segmented with adaptive windowing, whereas silhouette refinement was performed to analyze motion structures from visual inputs in support of athlete-performance analysis and rehabilitation monitoring. Temporal, spatial, and motion features such as trajectory, orientation, and skeleton-based space-time representations were calculated from multimodal inputs. The proposed framework was designed to capture complex movement dynamics associated with rehabilitation exercises and sports-related motion patterns across heterogeneous sensing environments. Extracted features were then combined and optimized with a multimodal feature fusion approach, while the Ranger optimization algorithm was utilized during the process. An attention-based deep learning classifier was implemented to classify movement activities. Results: The results showed that the proposed framework reached accuracy scores of 88.40% and 87.96% on the VIDIMU dataset and the UTD-MHAD dataset respectively. Recognition performance across both inertial and vision-based modalities provided greater robustness than single-modality solutions. The integration of wearable sensing and computer vision modalities further improved the ability of the framework to analyze complex movement behaviors under varying execution conditions and environmental variations. Conclusion: The proposed multimodal framework provides a foundation for intelligent athlete-performance and rehabilitation-monitoring systems by integrating wearable sensing, computer vision, and attention-based artificial intelligence for robust movement analysis. The findings highlight its potential to support biomechanical assessment, movement-quality evaluation, training-performance monitoring, rehabilitation tracking, and injury-risk management in modern sports and healthcare environments. Full article

Prefabricated mechanical, electrical, and plumbing (MEP) systems have been increasingly adopted in energy station projects; however, systematic evaluation frameworks capable of integrating construction performance, cost constraints, and uncertain multi-indicator assessments remain limited. To address this gap, this study constructs an Analytic Hierarchy Process (AHP)–Entropy–Fuzzy evaluation framework to assess the comprehensive benefits of BIM-enabled prefabricated MEP construction in energy stations. A hierarchical evaluation system was established based on five dimensions: schedule, quality, cost, safety, and environmental performance, and ten secondary indicators were defined. The Analytic Hierarchy Process was used to determine expert-based subjective weights, the entropy method was applied to capture objective data variability, and multiplicative normalization was employed to obtain combined weights. A fuzzy comprehensive evaluation model was then introduced to transform heterogeneous construction records into comparable benefit levels and scores. The prefabricated method scored 87.80 and was classified as “high”, whereas the conventional method scored 60.85 and was classified as “low”. A Technique for Order Preference by Similarity to Ideal Solution (TOPSIS)-based sensitivity analysis further showed that, under 10%, 20%, and 50% criterion-weight perturbations, the prefabricated group consistently achieved higher closeness coefficients than the conventional group. The smallest margin occurred when the schedule weight was reduced by 50%, but the prefabricated group retained a positive advantage. The results demonstrate that Building Information Modeling (BIM)-enabled prefabricated MEP construction can achieve superior overall project performance through the coordinated optimization of schedule, cost, safety, quality, and environmental objectives, offering a practical evaluation framework and decision-support tool for the industrialized delivery of future energy infrastructure projects. Full article

Diagnosing bearing faults remains a crucial challenge, particularly in effectively extracting fault information and achieving high diagnostic accuracy. To address this issue, this study presents a model for diagnosing bearing faults, which comprises three primary stages: feature extraction, feature selection, and classification. In the feature extraction stage, features are extracted from raw motor signals using empirical mode decomposition (EMD) and fast Fourier transform (FFT). In the feature selection stage, an effective method based on binary grey wolf optimization (BGWO) and the equilibrium optimizer (EO) is developed to remove redundant and irrelevant features. Finally, k-nearest neighbours (KNNs) and support vector machine (SVM) classifiers are used to identify bearing fault conditions. The proposed model is evaluated using four datasets: the University of California, Irvine (UCI) benchmark datasets, a motor bearing fault current-signal dataset, the Case Western Reserve University (CWRU) benchmark dataset, and the Machinery Failure Prevention Technology (MFPT) benchmark dataset. The experimental results show that the proposed method improves bearing fault diagnosis accuracy and demonstrates strong robustness compared with conventional methods. Full article

Background/Objectives: The “small-n-large-p” dilemma in mammography artificial intelligence (AI)—where the number of candidate imaging features far exceeds the number of labeled cases—commonly results in model overfitting, unstable feature selection, and poor generalization across clinical settings. This study aims to reassess the internal performance of the HiTopology-GOA-CSA (Grasshopper Optimization Algorithm–Crow Search Algorithm) feature-selection framework for mammography using a larger real Curated Breast Imaging Subset of Digital Database for Screening Mammography (CBIS-DDSM) cohort and a stricter leakage-aware evaluation strategy. Methods: In this retrospective computational study using public anonymized datasets, an expanded internal cohort of 98 CBIS-DDSM mass cases (49 benign, 49 malignant) was assembled from digital imaging and communications in medicine (DICOM) region of interest (ROI) series. A total of 1074 features were extracted per case, including 88 handcrafted radiomic descriptors and 986 EfficientNet-B5 deep features. HiTopology-GOA-CSA selected 102 features, corresponding to 91% feature reduction. Two internal evaluation modes were compared: Mode A, which matched the original pilot methodology by performing feature selection once on the full cohort before cross-validation, and Mode B, which used strict nested feature selection within training folds. Performance was assessed with 5-fold stratified cross-validation using a multilayer perceptron (MLP) classifier. Results: On the expanded cohort, Mode A achieved an area under the receiver operating characteristic curve (AUC) of 0.726 (95% CI: 0.594–0.858), sensitivity of 0.658, specificity of 0.651, and F1-score of 0.644. Under the stricter nested evaluation, Mode B achieved AUC of 0.683 (95% CI: 0.549–0.817), sensitivity of 0.598, specificity of 0.631, and F1-score of 0.595. Mean pairwise Jaccard similarity across nested folds was 0.604, indicating moderate feature stability. Benchmark comparisons showed that the proposed method was competitive but did not outperform standard baselines; LASSO logistic regression achieved the highest AUC of 0.739, while the proposed HiTopology-GOA-CSA + MLP achieved an AUC of 0.683. Real external validation on the locked VinDr-Mammo subset (n = 25) remained near-random (AUC of 0.500 [95% CI: 0.304–0.696]), with complete prediction collapse (sensitivity of 1.000, specificity of 0.000). Conclusions: The framework demonstrated feasibility for structured feature selection and stress testing in a small-cohort mammography AI setting; however, external validation revealed near-random discrimination and prediction collapse, indicating limited generalizability. These findings emphasize the need for benchmark comparisons, transparent uncertainty reporting, patient-level validation, and larger multicenter datasets before clinical translation. Full article

Reliable age information is needed for fisheries assessment, but conventional otolith reading requires trained readers and considerable time. This study evaluated whether convolutional neural networks could classify reader-assigned age classes of Japanese jack mackerel Trachurus japonicus directly from sagittal otolith images. Otolith images from fish aged 0 to 4 years were used to compare three image-only backbones: Inception v3, Xception, and EfficientNet B4. The models were trained under the same data split, preprocessing, augmentation, and evaluation framework. In Stage 1, Inception v3 showed the highest validation macro F1 score (0.933) and was selected as the image-only baseline. After additional optimization, the selected model reached a validation macro F1 score of 0.944, validation exact accuracy of 0.935, and validation agreement within one age class of 1.000. On the independent test set, the optimized image-only model achieved exact accuracy of 0.866, macro F1 score of 0.873, and agreement within one year of 1.000. These results indicate that otolith images contain useful age-related visual information. Convolutional neural networks may support age class screening in T. japonicus, although they should complement rather than replace expert otolith reading. These findings apply to the initial screening of T. japonicus within the 0 to 4 age range represented in commercial purse seine catches, and performance for ages older than 4 was not evaluated. Full article

Student dropout remains an important challenge in higher education because it affects degree completion, institutional resource efficiency, workforce preparation, and students’ long-term socioeconomic opportunities. This requires not only accurate predictions but also decision support that is both reliable and aware of uncertainty. This study posits that the amalgamation of probabilistic modeling, uncertainty quantification, and graph-based refinement can augment both predictive reliability and decision support for the early detection of dropouts. A reliability-centered predictive framework is presented, integrating Educational Competition Optimization (ECO)-based feature selection, probabilistic Support Vector Classification (SVC), isotonic regression for probability calibration, and split conformal prediction for distribution-free uncertainty quantification. In addition, a similarity-driven Graph-based Fuzzy Cellular Automata (Graph-FCA) refinement mechanism is developed, where student relationships are modeled using a k-nearest neighbor graph with radial basis function similarity. Entropy-based confidence weighting is used to control uncertainty-aware propagation. An Explainable Artificial Intelligence layer based on SHAP provides both global and local interpretability, and fairness-aware evaluation assesses consistency across demographic groups. The suggested framework maintains predictive performance while improving probabilistic reliability. The Graph-FCA refinement achieves an accuracy of 0.7503, which is close to the calibrated ECO–SVC baseline (Accuracy = 0.7537; Macro-F1 = 0.6704) and also reduces the Brier score. The conformal prediction layer achieves empirical coverage close to the desired confidence level, ensuring reliable uncertainty estimates. The ECO–SVC–Conformal–GraphFCA framework transforms traditional classification into a reliable, understandable, and uncertainty-aware early warning system, enhancing its usefulness for ethical and informed decision-making in engineering education. Full article

Pose estimation is an important task in the field of machine vision, being widely used in robot grasping, augmented reality, and other applications. Weakly textured objects pose severe challenges due to scarce texture and low-density features, becoming a bottleneck in robot grasping. This paper systematically reviews recent progress in weakly textured object pose estimation, classifying methods into traditional and deep learning categories, and further dividing deep learning methods into instance-level, category-level, and unseen object-level. This review further summarizes the core issues of generalization limitations, real-time contradictions, and data bottlenecks in existing research. Combined with the practical needs of weakly textured scenes, the review points out that multimodal fusion optimization, lightweight model design, and low-cost annotation technology development are the future core research directions. The research results can provide a reference for algorithm design, experimental verification, and engineering applications in the field of weakly textured object pose estimation. Full article

Underground mine production scheduling under uncertainty is a complex and multi-field coupling system project. In this study, underground mine production scheduling seeks to determine the optimal start time of extraction-related projects, with the objectives of maximizing net present value, minimizing makespan, and maximizing resource utilization rate. The Copula function is adopted to formulate the correlation between uncertain project duration and cost and generate a set of stochastic scenarios. Then, the K-means algorithm classifies the scenarios into multiple scenario families, and the SBR algorithm is adopted to perform scenario reduction. Moreover, a rank choice function-based hyper-heuristic algorithm is extended to solve the multi-objective optimization model, which makes an excellent balance among the three objective functions. For determining the optimal scheduling plan, the cross-efficiency DEA algorithm is used to evaluate the archive set, sort the optimal solution, and guide the next iteration. The computational case verifies the effectiveness and efficiency of the multi-objective underground mine scheduling model, stochastic scenario and technical and hyper-heuristic algorithm. Full article

A single user-invariant tensor, kinematically impossible for any human finger to produce, bypasses bidirectional long short-term memory (BiLSTM) continuous-authentication defenders with numerically identical structure across four independently trained generators. We arrive at this finding by training generative adversarial networks against BiLSTM defenders on 51 users across three independent random seeds, with the data partition held fixed, to test the prevailing assumption that successful generative attacks must reproduce the victim’s kinematic behavior. Aggregate attack success rate varies from 31.4% to 45.1% across seeds, a 13.7 percentage-point spread arising purely from optimization stochasticity, demonstrating how unreliable single-seed reporting is as an estimator of the true attack surface. A four-group descriptive stratification shows that 8% of users are attacked across all three seeds, 31% are consistently safe, and 61% exhibit seed-dependent outcomes. Classifier accuracy on zero-effort impostors does not predict adversarial vulnerability (Spearman $\rho =-0.058$, permutation $p=0.688$), whereas intra-user behavioral variance does ($\rho =+0.351$, permutation $p=0.012$, Bonferroni-corrected). The mechanism is not behavioral emulation but convergence to an Adversarial Skeleton Key, a tensor located in an unregularized region of the BiLSTM’s decision surface that the network reliably maps to acceptance, despite lying many standard deviations outside any genuine human distribution. The mimicry-centric evaluation paradigm underestimates the real threat surface. Input-space plausibility must be treated as a defensive layer rather than a preprocessing concern. Full article

Objectives: Multimodal large language models (MLLMs) have shown potential for medical image classification. We evaluated four optimization strategies in two MLLMs—GPT-4o (gpt-4o-2024-08-06) and Gemini 2.5 Flash-Lite—for ultrasound-based thyroid nodule malignancy classification using two public datasets and a clinical cohort of nodules with atypia of undetermined significance (AUS) cytology. Methods: Text prompting, few-shot learning, fine-tuning, and a hybrid strategy combining fine-tuning with few-shot learning were evaluated for each model. Performance was assessed using the Digital Database of Thyroid Images (DDTI; n = 80), a 1000-image test subset of TN5000, and an institutional AUS cohort with surgical pathology (n = 84). In the AUS cohort, the best-performing strategy was compared with the consensus classification of three endocrinologists and the American Thyroid Association (ATA) ultrasound risk stratification. Results: For GPT-4o, the hybrid strategy achieved the highest area under the receiver operating characteristic curve (AUC) in DDTI (0.866), TN5000 (0.689), and the AUS cohort (0.836). In the AUS cohort, its specificity was higher than that of endocrinologist consensus and ATA risk stratification when only high-suspicion nodules were classified as malignant (95.1% vs. 70.7% and 70.7%; p = 0.002 and p = 0.001, respectively), while sensitivity did not differ significantly (72.1% vs. 74.4% and 79.1%, respectively; both p > 0.05). However, the hybrid model misclassified 12 of 43 malignant nodules, corresponding to a false-negative rate of 27.9%. When high- and intermediate-suspicion ATA categories were classified as malignant, ATA sensitivity increased to 83.7% and specificity decreased to 56.1%; the hybrid model had a higher AUC than ATA risk stratification (0.836 vs. 0.749; p = 0.017). For Gemini 2.5 Flash-Lite, few-shot learning, fine-tuning, and the hybrid strategy did not improve AUC relative to text prompting in any dataset. Conclusions: The hybrid strategy produced the most consistent performance gains for GPT-4o across the three datasets but did not improve Gemini 2.5 Flash-Lite. The optimized GPT-4o model achieved high specificity in the diagnostically challenging AUS cohort, although its false-negative rate limits its use as a stand-alone diagnostic tool. Further validation in larger, prospective multicenter cohorts is required before clinical use. Full article

Autonomous vehicle open-road testing is a crucial component in the development of intelligent and connected vehicle (ICV) industries. The classification of road complexity plays a key role in ensuring the safety and efficiency of such tests. This study, based on the practices of the High-Level Autonomous Driving Demonstration Zone in Beijing, proposes a scientific and systematic framework for classifying road complexity. The framework integrates static road features, dynamic traffic flow indicators, and safety event metrics, employing the Analytic Hierarchy Process (AHP) to quantify road complexity and categorize roads into five distinct levels. The findings provide significant guidance for the phased opening of test roads, optimization of autonomous driving algorithms, construction of accident scenario databases, and deployment of infrastructure. This paper further explores the practical applications and future development directions of road complexity classification, aiming to offer theoretical and practical support for the testing and demonstration of intelligent and connected vehicles. Full article