Search Results (3,920)

Reliable object detection in marine radar imagery is critical for maritime situational awareness, collision avoidance, and autonomous navigation. However, it remains challenging due to sea clutter, small targets, and interference from fixed navigational aids. This study proposes a high-precision detection pipeline that integrates tiled training, Sliced Aided Hyper Inference (SAHI), and an oriented bounding box (OBB) variant of the lightweight YOLOv11 architecture. The proposed approach effectively addresses scale variability in Plan Position Indicator (PPI) radar images. Experiments were conducted on the real-world DAAN dataset provided by the German Aerospace Center (DLR). The dataset consists of 760 full-resolution radar frames containing multiple moving vessels, dynamic own-ship, and clutter sources. A semi-automatic contour-based annotation pipeline was developed to generate multi-format labels, including axis-aligned bounding boxes, oriented bounding boxes (OBBs), and instance segmentation masks, directly from radar echo characteristics. The results demonstrate that the tiled YOLOv11n-OBB model with SAHI achieves an mAP@0.5 exceeding 0.95, with a mean center localization error below 10 pixels. The proposed method shows better performance on small targets compared to standard full-image baselines and other YOLOv11 variants. Moreover, the lightweight models enable near real-time inference at 4–6 FPS on edge hardware. These findings indicate that OBBs and scale-aware strategies enhance detection precision in complex marine radar environments, providing practical advantages for tracking and navigation tasks. Full article

The main characteristics of the large number of coffee species are differences in aroma and caffeine content. Labeled blends of Coffea arabica (C. arabica) and Coffea canephora (C. canephora) are common to broaden the flavor profile or enhance the stimulating effect of the beverage. New emerging species such as Coffea liberica (C. liberica) further increase the variability in blends. However, significant price differences between coffee species increase the risk of unlabeled blends and thus influence food quality and safety for consumers. In this study, a prototypic hyphenation of trapped headspace-gas chromatography-ion mobility spectrometry-quadrupole mass spectrometry (THS-GC-IMS-QMS) was used for the detection of characteristic compounds of C. arabica, C. canephora, and C. liberica in green and roasted coffee samples. For the discrimination of coffee species with IMS data, multivariate resolution with multivariate curve resolution–alternating least squares (MCR-ALS) prior to partial least squares–discriminant analysis (PLS-DA) was evaluated. With this approach, the classification accuracy, as well as sensitivity and specificity, of the PLS-DA model was significantly improved from an overall accuracy of 87% without prior feature selection to 92%. As MCR-ALS preserves the physical and chemical properties of the original data, characteristic features were determined for subsequent substance identification. The simultaneously generated QMS data allowed for partial annotation of the characteristic volatile organic compounds (VOC) of roasted coffee. Full article

The increasing ability of Large Language Models (LLMs) to generate fluent and coherent text has heightened the need for resources to analyze and detect synthetic content, particularly in Spanish, where the scarcity of datasets hinders the development of reliable detection systems. This work presents a Spanish-language dataset of 18,236 synthetic news descriptions generated from real journalistic headlines using a fully reproducible, open-source pipeline. The methodology used to produce the dataset includes both a Retrieval Augmented Generation (RAG) approach, which incorporates contextual information from recent news descriptions, and a NO-RAG approach, which relies solely on the headline. Texts were generated with the instruction-tuned Mistral 7B Instruct model, systematically varying temperature to explore the effect of generation parameters. The dataset includes detailed metadata linking each synthetic description to its source headline, generation settings, and, when applicable, retrieved contextual content. By combining contextual grounding, controlled parameter variation, and source-level traceability, this dataset provides a reproducible and richly annotated resource that supports research in Spanish synthetic text and evaluation of LLM-based generation. Full article

Modern Standard Arabic (MSA) and the many regional dialects differ substantially in vocabulary, morphology, and pragmatic usage. Most available annotated resources are in MSA, and zero-shot transfer from MSA to dialectal tasks suffers a large performance drop. This paper addresses generalised cross-dialectal Arabic question answering (QA), where the context and the question are written in different Arabic varieties. We propose a training-free augmentation framework that generates code-mixed questions to bridge lexical gaps across Arabic varieties. The method produces semantically faithful, balanced code-mixed questions through the following two-stage procedure: lexicon-based partial substitution with semantic similarity and substitution-rate constraints, followed by fallback neural machine translation with word-level alignment when needed. We also introduce automated multidialectal lexicon construction using machine translation, embedding-based alignment, and semantic checks. We carry out our evaluation in a zero-shot setting, where the model is fine-tuned only on MSA and then tested on dialectal inputs using ArDQA, covering five Arabic varieties and three domains (SQuAD, Vlogs, and Narratives). Experiments show consistent improvements under context-question dialect mismatch as follows: +1.09 F1/+0.87 EM on SQuAD, +1.54/+1.25 on Vlogs, and +2.75/+2.27 on Narratives, with the largest gains for Maghrebi questions in Narratives (+12.13 F1/+8.45 EM). These results show that our method improves zero-shot cross-dialectal transfer without fine-tuning or retraining. Full article

Neonatal seizures are a critical early indicator of neurological injury, yet effective automated detection is challenged by significant inter-subject variability in electroencephalogram (EEG) signals. To address this generalization gap, this study introduces the Domain-Adversarial Spatiotemporal Network (DA-STNet) for robust cross-subject seizure detection. Utilizing Short-Time Fourier Transform (STFT) spectrograms, the architecture employs a hierarchical backbone comprising a Channel-Independent CNN (CI-CNN) for local texture extraction, a Spatial Bidirectional Long Short-Term Memory (Bi-LSTM) for modeling topological dependencies, and Attention Pooling to dynamically prioritize pathological channels while suppressing noise. Crucially, a Gradient Reversal Layer (GRL) is integrated to enforce domain-adversarial training, decoupling pathological features from subject-specific identity to ensure domain invariance. Under rigorous 5-fold cross-validation, the model achieves State-of-the-Art performance with an average Area Under the Curve (AUC) of 0.9998 and an F1-score of 0.9952. Data scaling experiments further reveal that optimal generalization is attainable using only 80% of source data, highlighting the model’s superior data efficiency. These findings demonstrate the proposed method’s capability to reduce reliance on extensive clinical annotations while maintaining high diagnostic precision in complex clinical scenarios. Full article

Today, foundation models simulate humans’ skills in translation, literature review, fact checking, fake-news detection, novel and poetry production. However, generative AI can also be applied to discourse analysis. This study instructed the Gemini 2.5 model to analyze multimodal political discourse. We selected some fragments from the Trump–Zelensky debate held at the White House on 28 February 2025 and annotated each sentence, gesture, intonation, gaze, and facial expression in terms of LEP (Logos, Ethos, Pathos) analysis to assess when speakers, in words or body communication, rely on rational argumentation, stress their own merits or the opponents’ demerits, or express and try to induce emotions in the audience. Through detailed prompts, we asked the Gemini 2.5 model to run the LEP analysis on the same fragments. Then, considering the human’s and model’s annotations in parallel, we proposed a metric to compare their respective analyses and measure discrepancies, finally tuning an optimized prompt for the model’s best performance, which in some cases outperformed the human’s analysis: an interesting application, since the LEP analysis highlights deep aspects of multimodal discourse but is highly time-consuming, while its automatic version allows us to interpret large chunks of speech in a fast but reliable way. Full article

Underwater structural surface defect detection—such as identifying cavities and spalling—faces significant challenges due to complex environments, scarce annotated data, and the reliance of modern detectors on large-scale datasets. While current approaches often combine large-data training with fine-tuning or image enhancement, they still require extensive underwater samples and are typically too computationally heavy for resource-constrained robotic platforms. To address these issues, we introduce a defect detection model based on feature distillation, which achieves high detection accuracy with limited samples. We tackle three key challenges: enhancing sample diversity under data scarcity, selecting and training a baseline model that balances accuracy and efficiency, and improving lightweight model performance using augmented samples under computational constraints. By integrating a feature distillation mechanism with a sample augmentation strategy, we develop a compact detection strategy and framework that delivers notable performance gains in limited data, offering a practical and efficient solution for real-world underwater inspection. Full article

The exponential growth and complexity of legal agreements pose significant Big Data challenges and strategic risks for modern organizations, often overwhelming traditional, manual contract management workflows. While AI has enhanced legal research, most current applications require extensive domain-specific fine-tuning or substantial annotated data, and Large Language Models (LLMs) remain susceptible to hallucination risk. This paper presents an AI-based Agreement Management System that addresses this methodological gap and scale. The system integrates a Python 3.1.2/MySQL 9.4.0-backed centralized repository for multi-format document ingestion, a role-based Collaboration and Access Control module, and a core AI Functions module. The core contribution lies in the AI module, which leverages zero-shot learning with OpenAI’s GPT-4o and structured prompt chaining to perform advanced contractual analysis without domain-specific fine-tuning. Key functions include automated metadata extraction, executive summarization, red-flag clause detection, and a novel feature for natural-language contract modification. This approach overcomes the cost and complexity of training proprietary models, democratizing legal insight and significantly reducing operational overhead. The system was validated through real-world testing at a leading industry partner, demonstrating its effectiveness as a scalable and secure foundation for managing the high volume of legal data. This work establishes a robust proof-of-concept for future enterprise-grade enhancements, including workflow automation and predictive analytics. Full article

With the rapid growth of online presentations, there has been an increasing need for efficient review of recorded materials. In typical presentations, speakers verbally elaborate on each slide, providing details not captured in the slides themselves. Automatically extracting and embedding these verbal explanations at their corresponding slide locations can greatly enhance the review process for audiences. This paper presents a Slide Annotation System that employs a robust hybrid two-stage detector to identify slide boundaries, extracts slide text through Optical Character Recognition (OCR), transcribes narration, and employs a multimodal Large Language Model (LLM) to generate concise, context-aware annotations that are added to their corresponding slide locations. For evaluations, the technical performance was validated on five recorded presentations, while the user experience was assessed by 37 participants. The results showed that the system achieved a macro-average $F_1$ score of 0.879 ($SD=0.024$, 95% $CI[0.849,0.909]$) for slide segmentation and 90.0% accuracy (95% $CI[74.4\%,96.5\%]$) for annotation alignment. Subjective evaluations revealed high annotation validity and usefulness as rated by presenters, and a high System Usability Scale (SUS) score of 80.5 ($SD=6.7$, 95% $CI[78.3,82.7]$). Qualitative feedback further confirmed that the system effectively streamlined the review process, enabling users to locate key information more efficiently than standard video playback. These findings demonstrate the strong potential of the proposed system as an effective automated annotation system. Full article

Background/Objectives: Accurate and reliable automated dermoscopic lesion classification remains challenging. This is due to pronounced dataset bias, limited expert-annotated data, and poor cross-dataset generalization of conventional supervised deep learning models. In clinical dermatology, these limitations restrict the deployment of data-driven diagnostic systems across diverse acquisition settings and patient populations. Methods: Motivated by these challenges, this study proposes a transformer-based, dermatology-specific foundation model. The model learns transferable visual representations from large collections of unlabeled dermoscopic images via self-supervised pretraining. It integrates large-scale dermatology-oriented self-supervised learning with a hierarchical vision transformer backbone. This enables effective capture of both fine-grained lesion textures and global morphological patterns. The evaluation is conducted across three publicly available dermoscopic datasets: ISIC 2018, HAM10000, and PH2. The study assesses in-dataset, cross-dataset, limited-label, ablation, and computational-efficiency settings. Results: The proposed approach achieves in-dataset classification accuracies of 94.87%, 97.32%, and 98.17% on ISIC 2018, HAM10000, and PH2, respectively. It outperforms strong transformer and hybrid baselines. Cross-dataset transfer experiments show consistent performance gains of 3.5–5.8% over supervised counterparts. This indicates improved robustness to domain shift. Furthermore, when fine-tuned with only 10% of the labeled training data, the model achieves performance comparable to fully supervised baselines. Conclusions: This highlights strong data efficiency. These results demonstrate that dermatology-specific foundation learning offers a principled and practical solution for robust dermoscopic lesion classification under realistic clinical constraints. Full article

Multi-object tracking (MOT) in vineyard environments remains challenging due to frequent and long-term occlusions caused by dense foliage, overlapping grape clusters, and complex plant structures. These characteristics often result in identity switches and fragmented trajectories when using conventional tracking methods. This paper proposes OATSAM-Track, an occlusion-aware multi-object tracking framework designed for vineyard fruit monitoring. The framework integrates lightweight MobileSAM-assisted instance segmentation to estimate target visibility and occlusion severity. Occlusion-state reasoning is further incorporated into temporal association, appearance memory updating, and identity recovery. An adaptive temporal memory mechanism selectively updates appearance features according to predicted occlusion states, reducing identity drift under partial and severe occlusions. To facilitate occlusion-aware evaluation, an extended vineyard multi-object tracking dataset (GrapeOcclusionMOTS) with SAM-refined instance masks and fine-grained occlusion annotations is constructed. The experimental results demonstrate that OATSAM-Track improves identity consistency and tracking robustness compared to representative baseline trackers, particularly under medium and severe occlusion scenarios. These results indicate that explicit occlusion modeling is beneficial for reliable fruit monitoring in precision agriculture. Full article

Background: First-trimester prenatal screening is a fundamental component of modern obstetric care, offering early insights into fetal health and development. A key focus of this screening is the detection of chromosomal abnormalities, such as Trisomy 21 (Down syndrome), which can have significant implications for pregnancy management and parental counseling. Over the years, various non-invasive methods have been developed, with ultrasound-based assessments becoming a cornerstone of early evaluation. Among these, the measurement of Nuchal Translucency (NT) has emerged as a critical marker. This sonographic measurement, typically performed between 11- and 13-weeks 6+ days of gestation, quantifies the fluid-filled space at the back of the fetal neck. An increased NT measurement is a well-established indicator of a higher risk for aneuploidies and other congenital conditions, including heart defects. The Fetal Medicine Foundation has established standardized criteria for this measurement to ensure its reliability and widespread adoption in clinical practice. Methods: We utilized two datasets comprising 2425 ultrasound images from Shenzhen People’s Hospital China and the National Hospital of Obstetrics and Gynecology Vietnam. The methodology employs a two-stage Deep Learning framework: first, a DenseNet121 model assesses image quality to filter non-standard planes; second, a novel DenseNet-based segmentation delineates the NT region for automated measurement. Results: The quality assessment module achieved 94% accuracy in distinguishing standard from non-standard planes. For segmentation, the proposed model achieved a Dice coefficient of 0.897 and an overall accuracy of 98.9%, outperforming the standard U-Net architecture. Clinically, 55.47% of automated measurements deviated by less than 1 mm from expert annotations, and the system demonstrated > 90% sensitivity and specificity for identifying high-risk cases (NT ≥ 2.5 mm). Conclusions: The proposed framework successfully integrates quality assurance with automated measurement, offering a robust decision-support tool to reduce variability and improve screening accuracy in prenatal care. Full article

Large-scale collection and annotation of sensitive facial data in real-world traffic scenarios face significant hurdles regarding privacy protection, temporal consistency, and high costs. To address these issues, this work proposes an integrated method specifically designed for sensitive information anonymization and semi-automatic image annotation (AIA). Specifically, the Nullface anonymization model is applied to remove identity information from facial data while preserving non-identity attributes including pose, expression, and background that are relevant to downstream vision tasks. Secondly, the Qwen3-VL multimodal foundation model is combined with the Grounding DINO detection model to build an end-to-end annotation platform using the Dify workflow, covering data cleaning and automated labeling. A traffic-sensitive information dataset with diverse and complex backgrounds is then constructed. Subsequently, the systematic experiments on the WIDER FACE subset show that Nullface significantly outperforms baseline methods including FAMS and Ciagan in head pose preservation and image quality. Finally, evaluation on object detection further confirms the effectiveness of the proposed approach. The accuracy achieved by the proposed method reaches 91.05%, outperforming AWS, and is almost identical to the accuracy of manual annotation. This demonstrates that the anonymization process maintains critical semantic details required for effective object detection. Full article

Fetal growth restriction (FGR) is a major contributor to perinatal morbidity and mortality, most commonly arising from placental dysfunction, with increasing evidence implicating aberrant DNA methylation in its pathogenesis. To identify robust epigenetic alterations associated with FGR, we analyzed placental chorionic villi from an in-house early-onset FGR cohort and compared them with a publicly available dataset (GSE100197). DNA methylation profiling was performed using Illumina EPIC (in-house) and 450K (public) arrays, processed with identical normalization and quality-control pipelines, including adjustment for gestational age and estimation of placental cell-type composition. Differentially methylated positions (DMPs) were identified using linear regression models, revealing 10,427 DMPs in the in-house cohort and 7467 in the public dataset, with 108 shared DMPs showing consistent direction of change across both cohorts. Promoter-associated DMPs were mapped to genes involved in angiogenesis, morphogenesis, immune regulation, and transcriptional control, including EPHA1, ANGPTL6, ITGAX, BCL11B, and CYP19A1, while additional novel candidates such as SLC39A12, YEATS4, and MIR515 family members were also identified. Functional annotation suggests that these methylation changes may influence pathways essential for placental vascular development and structural organization. Overall, this cross-cohort comparison highlights reproducible epigenetic signatures of FGR and underscores the need for standardized approaches to clarify the molecular mechanisms underlying placental insufficiency. Full article

Bridges are a crucial part of infrastructure, but many are in urgent need of maintenance. Digital methods like bim and Digital Twinning can support this process but depend on digital models that are often missing for existing structures. Automating the reconstruction of these models from existing documentation, such as construction drawings, is essential to accelerate digital adoption. Addressing a key step in the reconstruction process, this paper presents an end-to-end pipeline for extracting bridge cross-sections from drawings. First, the YOLOv8 network locates and classifies the cross-sections within the drawing. The results are then processed by the segmentation model sam, which generates pixel-wise masks without requiring task-specific training data. This eliminates the need for manual mask annotation and enables straightforward adaptation to different cross-section types, making the approach broadly applicable in practice. Finally, a global optimization algorithm fits parametric templates to the masks, minimizing a custom loss function to extract geometric parameters. The pipeline is evaluated on 33 real-world drawings and achieves a median parameter deviation of −2.2 cm and 2.4 cm, with an average standard deviation of 35.4 cm. Full article