Search Results (6,551)

In railway operations and safety oversight, vast amounts of accident-related data are recorded in unstructured textual formats, posing challenges for efficient information extraction and analysis. To address this, we constructed a knowledge graph from railway accident profile texts and integrated it with a Graph Retrieval-Augmented Generation (GraphRAG)-enhanced retrieval framework to support both basic and composite queries on railway accident information. First, railway accident profile texts were preprocessed, and Easy Data Augmentation (EDA) was used to expand samples of different types, thereby constructing a railway accident profile text dataset for subsequent knowledge extraction. We then introduced a deep learning model, RoBERTa-CNN-BiLSTM-CRF (RCBC), for automatic extraction of seven categories of entities, including accident IDs and causes. To extract semantic relations, we designed a prompt-based template leveraging large language models (LLMs). To mitigate information loss during entity extraction, a GCN-Attention-LLM (GA-LLM) model was further designed for knowledge graph completion. Experimental results show that RCBC achieves MicroF scores above 85% across entity extraction tasks, while GA-LLM attains an average Hits@3 of 82.84% in knowledge completion. In tests on basic and composite questions, LLM-GraphRAG outperformed both LLM and LLM-RAG in faithfulness, semantic similarity, context precision, and context recall. The resulting knowledge graph contains 1493 entities and 1832 relations. Combined with the retrieval framework, the system enables access to key railway accident information and offers technical support for intelligent railway safety management. Full article

Reliable automated monitoring of photovoltaic modules is essential for improving energy efficiency, operational safety, and predictive maintenance. Infrared thermography is one of the most effective solutions for identifying localised thermal anomalies, such as hotspots, micro-cracks, connection faults, shading effects and other conditions of degradation that can compromise the performance of the photovoltaic system. The interpretation of thermographic images is still frequently reliant on the operator’s experience or on automated procedures based exclusively on image processing techniques or artificial intelligence models often regarded as black-box models, thereby limiting their reliability, robustness and interpretability. This study presents an integrated diagnostic framework combining infrared thermography, computer vision, and artificial intelligence for the automated diagnosis and health monitoring of photovoltaic modules operating under real-world conditions. The proposed methodology extends beyond hotspot detection by integrating thermal image preprocessing, anomaly detection and segmentation, extraction of thermal and geometric descriptors, and intelligent fault classification. The resulting diagnostic information enables automated fault-type classification and quantitative severity assessment, providing interpretable condition indicators for photovoltaic module monitoring. The methodology was validated using a database comprising 1560 thermographic images acquired from photovoltaic modules under representative operating conditions. The experimental evaluation demonstrated an overall classification accuracy of 97.6%, an F1-score of 97.0%, and an area under the ROC curve (AUC) of 0.991 for the fault-type classification task. The proposed framework therefore provides an interpretable and computationally efficient decision-support methodology for photovoltaic condition assessment, while its integration into longitudinal predictive-maintenance systems remains a subject for future investigation. Full article

Despite recent advances in computer vision and optical imaging, the image-based recognition of ophthalmic lens categories characterized by different refractive indices and anti-reflective coating types remains largely unexplored. In this work, we formulate this problem as a non-destructive optical inspection task, where lens-induced visual changes are treated as subtle micro-texture signatures produced by the interaction between the lens and a controlled visual target. Although these signatures are often imperceptible to the human eye, they can be quantified using local texture analysis. To establish the first systematic benchmark for this emerging application, we present a comprehensive evaluation of state-of-the-art local texture descriptors, with particular emphasis on Local Binary Pattern (LBP)-like methods, under unified evaluation protocols. The evaluation is conducted on two dedicated ophthalmic lens image datasets acquired under controlled conditions, using consistent visual targets designed to reveal lens-induced texture and color variations. This setting enables a rigorous assessment of whether handcrafted micro-texture descriptors can capture discriminative image signatures associated with different refractive indices and anti-reflective coating types. The experimental results demonstrate that several modern LBP variants provide excellent recognition performance, confirming the relevance of local micro-texture analysis for image-based ophthalmic lens category recognition. Comparisons with pretrained deep feature extractors further show that although deep representations generally achieve the highest recognition rates, several handcrafted descriptors offer comparable performance while requiring neither network training nor fine-tuning. Beyond recognition accuracy, the benchmark investigates performance stability across datasets, robustness under limited training samples, classifier influence, and the statistical significance of the observed performance differences. Overall, this work establishes the first reproducible image-based benchmark for non-destructive ophthalmic lens inspection and provides practical guidelines for selecting local texture descriptors in ophthalmic lens recognition systems. Full article

First-year computer-science students judged their grade on two assessments differing in judgment type: a concurrent self-assessment embedded in an early quiz (Assessment 1, $N=101$) and a postdiction after a mid-course written-and-laboratory exam (Assessment 2, $N=117$), with 94 students linked across both. The concurrent judgment was optimistic (mean signed error $+9.10$ pp; $70.3\%$ overestimators; Spearman $\rho =0.533$). The postdictions reversed the bias: students underestimated the written component by $10.08$ pp and the laboratory by $4.21$ pp, rank-order calibration being markedly better for the laboratory task ($\rho =0.804$). In the linked subsample, the reversal was large (paired $d_z=1.00$) but absolute error did not improve ($p=0.582$): the error changed direction, not magnitude. A tertile split shows broad-based quiz optimism and, on the postdictions, a gradient compatible with regression to the mean. An exploratory post hoc re-banding of the laboratory scores, excluding the administrative 0–1 segment, yields a Dunning–Kruger-compatible between-band difference of $1.80$ raw points (95% CI $[0.94,2.66]$, $p=0.0001$), robust to alternative bandings, though the low-band overestimation is not. The Assessment 1 overestimation replicates in two further cohorts. Calibration therefore differed markedly across two contexts that differ simultaneously in judgment type, timing, format, content and diagnostic cues, which these data cannot disentangle. Full article

This systematic review examines how artificial intelligence (AI) technologies relevant to human-factor risk reduction can be integrated into international and Spanish maritime safety frameworks. The formal PRISMA corpus comprises 23 core sources (13 peer-reviewed studies and 10 regulatory, institutional or technical documents), while a separate supplementary search provides recent independent technical and regulatory evidence up to 31 August 2026. The analysis covers computer vision, thermal and near-infrared sensing, multimodal fusion, behavioral and fatigue analysis, and onboard edge processing, with particular attention to precision, recall, false alarms, latency, computational requirements and operational robustness. The evidence shows that high detection performance can be achieved in specific maritime datasets, but the reported values depend strongly on the task, sensor, dataset and hardware and cannot be treated as a universal accuracy threshold. A system architecture is therefore proposed in which heterogeneous sensors feed synchronized edge processing, event verification, alarm management, VDR-compatible event logging, and human confirmation with defined fail-safe behavior. On the regulatory side, the study proposes staged adaptations of SOLAS, the ISM Code, STCW, MLC and Spanish inspection frameworks. The 2026 IMO MASS Code, considered as supplementary regulatory evidence, provides a relevant precedent for goal-based approval, risk assessment and progressive operational experience. Fixed tonnage and implementation-date thresholds are consequently treated as illustrative parameters rather than validated requirements; any mandatory carriage provision should be supported by formal safety assessment, type approval and operational evidence. The resulting framework links electronics implementation with a short-, medium- and long-term regulatory roadmap for safer merchant-ship operations. Full article

National school-leaving examinations do more than certify attainment: through task design, they define which forms of literacy, knowledge, and judgement receive institutional recognition. This study examines how Italy’s upper-secondary school-leaving examination is administered in Slovene, a protected minority language, and how task design gives institutional form to minority-language rights. A comparative document analysis covered six regular-session papers from 2019 and 2022–2026, comprising 42 options. Each option was analysed for source material, question sequence, final writing demand, identity positioning, and functions assigned to Slovene. The examination retains a three-type architecture: literary interpretation in Type A, source-based reasoning and argumentation in Type B, and autonomous critical-explanatory writing in Type C. Eight options address border- or minority-related content, six of them in Type B, while nine have a documented Italian-language counterpart. Minority-related content is framed for interpretation, contextualisation, and reasoned judgement without requiring identity affirmation. Candidates select one of seven options, so engagement with Slovene literature and minority-specific content is available but not compulsory. At the level of task design, the examination requires Slovene for literary interpretation, historical reasoning, source-based argumentation, and public-discourse writing, making these functions part of state certification in the protected language. Full article

Financial institutions face an escalating volume of cybersecurity threats, yet existing decision frameworks rarely link predictive analytics to operational security priorities. Drawing on Task-Technology Fit theory, this study develops a machine learning-based decision-support framework to classify cybersecurity breach types in financial institutions and to identify the organizational risk factors that determine them. Analyzing 935 publicly disclosed incidents from the VERIS Community Database (VCDB, NAICS 52), we compare XGBoost against Random Forest, Logistic Regression, and Decision Tree. XGBoost achieves the most balanced performance (accuracy: 95.19%; weighted F1: 0.9519; 5-fold CV: 96.68% ± 0.21%). Feature importance analysis reveals ATM/kiosk infrastructure and breach pattern as the strongest predictors, translating into concrete SOC monitoring priorities. This framework supports UN/SDG 9 (Industry, Innovation and Infrastructure) and UN/SDG 16 (Peace, Justice and Strong Institutions) by strengthening the cyber resilience of financial institutions through open, replicable, data-driven methods. The open-data framework is replicable without commercial threat intelligence licenses. Full article

Landslides induced by slope deformation in open-pit coal mines pose significant risks to personnel safety and production continuity. Intelligent slope recognition is a prerequisite for early deformation warning, yet no publicly available segmentation dataset specifically targeting open-pit mine slopes currently exists, hindering progress in vision-based monitoring. To address this gap, we present OPMS-Seg, a benchmark dataset for semantic segmentation of open-pit mine slopes. The dataset contains 2814 UAV-captured RGB images and 15,334 polygon instances, covering four geometric slope types (Backlight, Stepped, Rubble, and Bottom types). Annotation files are provided in COCO (JSON), YOLO (TXT), and binary mask (PNG) formats to support diverse segmentation models. All annotations were validated by mining engineering experts. The dataset supports two segmentation tasks: (1) binary segmentation (slope vs. background) and (2) four-class fine-grained segmentation (distinguishing the four slope geometry types). In this paper, we present baseline results for the binary segmentation task, while the four-class task is provided as a benchmark for future research. We evaluated four classic models—U-Net, DeepLabV3+, YOLOv5-Seg, and YOLOv11-Seg—on OPMS-Seg, achieving strong performance and confirming annotation accuracy and scene representativeness. This open-access dataset is intended to accelerate intelligent monitoring, algorithm benchmarking, and low-altitude remote sensing tasks in open-pit environments. Full article

Background: Tumor heterogeneity arises from complex interactions among diverse biological factors, posing a major challenge for the development of robust multi-omics data integration methods. While the existing Tumor Multi-Omics pre-trained Network (TMO-Net) enables the fusion of multi-omics features into unified representations, its practical utility is constrained by issues such as missing modalities, incomplete within-omics data, and high-dimensional noise. To overcome these limitations, we propose TMO-Net+, an enhanced architecture specifically designed to improve the robustness and reliability of multi-omics modeling. Methods: TMO-Net+ introduces several coordinated architectural enhancements. First, a feature attention encoder is applied to each omics data type to reduce the influence of modality-dependent input variation. Second, we combine a gated Mixture-of-Experts (MoE) module with a Product-of Experts (PoE) mechanism to capture sample-specific contributions and enable robust inference even when partial omics data are available. Additionally, a supervised deep classification head with a tailored loss function is incorporated to enhance the separability of learned embeddings in the latent space. Results: Extensive experiments on pan-cancer datasets demonstrate that TMO-Net+ consistently outperforms the original TMO-Net, as measured by LogME scores. Furthermore, in various downstream tasks (e.g., pan-cancer classification, primary/metastatic site prediction, and prognostic modeling), TMO-Net+ achieves superior performance under partial-omics settings, which proves that it enhances the robustness and cross-cancer transferability of the multi-omics representations. Conclusions: The proposed TMO-Net+ improves the robustness and cross-cancer transferability of multi-omics representations within the evaluated TCGA cohorts. Biological interpretability analyses further show that TMO-Net+ prioritizes established cancer-driver genes, preserves cancer-dependent molecular-state information, and adaptively redistributes relative modality contributions across molecular states. By addressing modality-level missingness and modality-dependent input variation, it offers a reliable framework for integrative tumor analysis within the evaluated TCGA cohorts. Full article

Fact checking is a key task in the field of natural language processing, and aims to judge the authenticity of a given claim by using external evidence. Existing multi-agent-based fact-checking systems achieve significant improvements on multiple benchmark datasets, but still have two core limitations: firstly, they lack a full-process faithfulness-verification mechanism to test the faithfulness of intermediate products at each stage, resulting in the amplification of hallucination errors cascading through the pipeline; secondly, when both supportive and refuting evidence exist, the system often relies on the implicit preferences of the model to make judgments. To address these issues, this paper proposes a multi-agent fact-checking framework called FRAME (Faithful, Reproducible, Agent-based, Multi-stage, and Evidence perception). FRAME addresses these limitations through two core designs: firstly, it designs a three-stage faithfulness-gating mechanism, embedding a faithfulness-verifier agent into the pipeline for faithfulness verification, and triggering targeted repairs when unfaithfulness is detected; secondly, it constructs a Bayesian-inspired structured perception framework for evidence-conflict perception, classifying the retrieved evidence, performing multi-attribute reweighting and weighted synthesis, and, finally, outputting a structured decision with a complete reasoning chain. FRAME is systematically evaluated on multiple benchmark datasets, and the experimental results demonstrate the advantages, generalization ability, and fault tolerance of FRAME when dealing with different types of datasets. Full article

The growth of big data ecosystems has shifted the classical paradigm of selective storage toward an approach that preserves large volumes of heterogeneous data for subsequent exploitation, thereby strengthening the adoption of repositories such as data warehouses and, particularly, data lakes. In this context, data ingestion from multiple sources, formats, and structures is a critical activity in implementing and operating these environments. However, it is often carried out in a highly ad hoc manner, with low levels of standardization and with variability managed informally. Beyond the operational complexity of ingestion itself, the variability in features such as source types, ingestion frequencies, transformation needs, and loading strategies constitutes an additional engineering problem that must be addressed systematically. This work tackles both issues in the context of a consulting firm involved in data migration projects to data lakes under governance constraints defined by clients in the BFSI sector. The goal is to formalize the data ingestion process through an architecture that provides technical, documentation, and training support for engineering teams, while also incorporating a variability management tool to model, analyze, and guide the configuration of ingestion solutions according to project-specific needs. In this way, the proposal seeks to reduce uncertainty, improve development quality, optimize resource utilization, and provide a more systematic treatment of variability in data ingestion projects. The proposal was evaluated through a structured survey answered by two cohorts totaling 29 respondents: an enterprise cohort of 15 practitioners (60% of the firm’s staff) and a prospective cohort of 14 engineering interns. For the enterprise cohort, the survey obtained average scores of 76 (individual) and 83.1 (role-averaged) for usability and 85.71 (individual) and 91.56 (role-averaged) for perceived quality; for the intern cohort, the corresponding individual averages were 70.18 for usability and 74.74 for perceived quality. In addition, a before/after comparison against the previous ad hoc workflow, covering objective engineering indicators, was conducted with both cohorts, providing task-based evidence that complements the perception-based results. Full article

Images from high-resolution cameras are mapped onto a sparse pattern of low spatial resolution and intensity in the retina, which limits visual perception in retinal prosthetic vision. When the entire scene is converted into phosphenes, it may allow unnecessary background information to be retained and may cause visual clutter, which may make it hard for prosthetic vision users to interpret the scene. In order to tackle this issue, this paper presents a context-, depth-, and user-preference-aware method for selecting the objects of interest in the generation of phosphene images. The proposed method does not show all the objects equally but learns to sort the objects according to their relevance to prosthetic vision. Manual annotation of a subset of COCO images was conducted where the most salient object was selected based on environment type, scene type, user mode, safety, navigation relevance, task importance, and distance. All of the candidate objects are described by full-scene visual features, object-crop features, handcrafted priority features, context embeddings, and monocular depth features. To predict object-level importance scores and identify the Top-1 and Top-4 important objects in unseen scenes, a hybrid deep learning model combining twin ResNet-18 backbones for scene and object feature extraction with embedding-based context encoding was trained. Priority maps and phosphene images were then created using the selected object masks and were depth-weighted. Two types of phosphene representations were also produced: Canny-edge-based and direct full images. The proposed framework is designed to suppress irrelevant background areas and improve important and closer objects in order to obtain a simplified and informative prosthetic-vision representation of the scene. The experimental evaluation, including Top-1 accuracy, Top-3 accuracy, mean reciprocal rank (MRR), and visual comparison, demonstrates the effectiveness of the proposed framework, achieving a Top-1 accuracy of 90.12%, a Top-3 accuracy of 97.45%, and an MRR of 0.9368. Furthermore, the proposed Canny-priority phosphene representation achieved an average human-participant recognition accuracy of approximately 86%. The proposed method offers a user-adaptive strategy for selecting and visualizing the information of a scene under the severe constraint of the bandwidth of retinal prosthetic vision. Full article

Despite the ongoing debate surrounding oral corrective feedback (CF), recent studies emphasize CF’s pedagogical value in foreign/second language teaching, as it informs learners about the accuracy of their utterances and helps increase their awareness of language forms. Viewing CF as a classroom-based instructional strategy for supporting language development, this study examined the effects of different types of oral CF (recasts, explicit correction, and metalinguistic feedback) on young learners’ grammar acquisition within Focus-on-Form (FonF) instruction in an EFL context. The study involved 88 fifth-grade students, aged 11–12, who were assigned to three experimental groups, each receiving one type of oral CF during FonF instruction, and a control group that received FonF instruction without feedback. Data were collected through oral production and scenario description tasks administered as pre-tests, immediate post-tests, and delayed post-tests. The instructional phase included approximately 10.7 h of treatment, and the assessment phase included approximately 15 h of testing. All classroom interactions and test recordings were transcribed and quantitatively analyzed. The results showed that oral CF statistically significantly enhanced young learners’ grammar acquisition compared to no oral feedback. Among the feedback types, explicit correction showed a significant advantage over the other feedback conditions in both immediate and delayed-test performance; however, this advantage was not consistent across all assessment tasks. These findings also highlight the pedagogical value of integrating oral CF into FonF instruction as an evidence-based classroom practice for promoting grammar acquisition and language development in young learner EFL classrooms. Full article

The inspection of conventional door locks in public buildings is a repetitive security task commonly performed manually at predefined times. This paper presents the development and experimental validation of a non-invasive, low-complexity robotic system designed for autonomous door lock-state verification. The core contribution is a novel physical-interaction method integrated into an indoor omnidirectional mobile robot that infers the lock state of a lever-type door handle without infrastructure modifications. The system executes a three-stage operational workflow: 2D LiDAR-based global positioning in front of target doors, depth-camera-based local realignment of the robot and the door handle, and physical actuation coupled with state inference via kinematic feedback. By depressing the handle during a controlled forward motion, forward displacement identifies an unlocked door, whereas motion resistance signals a locked state. The system was evaluated in a real facility across 18 target doors during eight complete inspection missions. Out of 144 verification attempts, the system achieved a 97.9% success rate; LiDAR global positioning enabled immediate handle actuation in 96 cases (66.7%), while depth-camera realignment successfully corrected 45 handle misalignments. These results validate the reliability and low-complexity of physical-feedback inference for routine autonomous facility security. Full article

Background: Markerless motion capture systems are increasingly used in clinical and educational settings to assess functional movement. However, differences in system design and data processing may influence measurement outputs, raising questions about the systematic differences in these technologies. Methods: A within-subject, repeated-measures design was used to compare range of motion (ROM) measurements obtained from two markerless motion capture systems during a sit-to-stand task. Fifty healthy adults (18–50 years) performed the task under three surface conditions (firm, compliant, and commode) while both systems recorded movement simultaneously. Joint ROM at the trunk, hips, knees, and ankles was analyzed using linear mixed-effects modeling to examine device, surface, and interaction effects. ROM was standardized to account for differences in absolute ROM magnitude and variability across biomechanically distinct joint measures. Results: A linear mixed-effects model showed significant main effects of both Device and Surface (all p’s < 0.001). There were also significant interactions between Device × Surface (p = 0.032), Device × Joint (p < 0.001), and Surface × Joint (p < 0.001), indicating that differences between devices and surfaces varied across biomechanical measures. The device-related discrepancies in standardized ROM (zROM) were greatest for hip adduction and trunk forward lean, whereas smaller between system differences were observed for ankle dorsiflexion and knee flexion. The firm surface demonstrated relatively larger zROM measurements for ankle dorsiflexion, knee flexion, and hip flexion compared to the compliant and commode surfaces. Conclusions: Markerless motion capture systems produced systematically different measurements during a functional task, and these differences were influenced by movement type and task conditions. The findings suggest that kinematic data derived from different markerless motion capture systems should be interpreted cautiously when comparing movement performance. Consistent use of a single system may be important for clinical tracking, and educational programs should emphasize critical interpretation of technology-derived movement data. Full article