Accurate segmentation of cardiac structures in magnetic resonance imaging (MRI) is essential for reliable diagnosis and quantitative analysis of cardiovascular diseases. However, conventional convolutional neural networks often struggle to maintain both semantic consistency and geometric smoothness, particularly in challenging slices with high anatomical variability. In this work, we propose CASNet, a novel U-Net-based architecture that integrates three key enhancements to address these limitations. First, we introduce a Multi-Scale Context Block (MSCB) at the network bottleneck to enrich encoder features with diverse receptive fields, enabling robust representation of cardiac structures across varying spatial scales. Second, we replace standard skip connections with Cross-Attentive Skip Connections (CASC), allowing the decoder to selectively aggregate spatial features from encoder layers via attention-weighted fusion. This mitigates semantic mismatch and promotes more effective feature reuse. Third, we incorporate a Curvature-Aware Loss that penalizes second-order spatial discontinuities in the predicted segmentation, thereby improving the smoothness and anatomical plausibility of the boundaries. Extensive experiments on the ACDC dataset demonstrate that CASNet outperforms baseline U-Net models and recent attention-based architectures, achieving superior performance in both region overlap and boundary accuracy metrics. The proposed approach provides a robust and generalizable solution for high-precision cardiac MRI segmentation, which may serve as a foundation for future downstream clinical applications in AI-assisted cardiac analysis.

Text-to-audio (TTA) generation often suffers from inconsistencies between audio output and textual input due to insufficient representation of textual details. In this paper, we present a lightweight keyword-guided generation framework for soundscape synthesis that explicitly maps spatiotemporal cues in text to the corresponding sound effects. Based on this framework, we developed two generative systems and validated their flexibility and effectiveness. System I leverages the large language model (LLM) to enhance semantic representation by introducing a structured mapping mechanism. We constructed a small-scale dataset based on this system and fine-tuned a state-of-the-art TTA model. System II employs traditional semantic analysis and generates soundscape using a predefined base dataset and an effect-mapping set. A subjective evaluation involving 24 participants demonstrated that the soundscape generated using the proposed approach yielded higher semantic consistency compared with traditional TTA generation. In addition, the reading experiments showed that the generated soundscape significantly improved the immersion experience during both silent and aloud reading. These results highlight the importance of fine-grained textual cues in cross-modal generation and support the use of structured rule-based mapping to improve the semantic alignment of TTA systems.

Safety testing serves as the fundamental pillar for the development of autonomous driving systems (ADSs), and decision-making plays a key role in ADSs. To ensure the safety of ADSs, it is paramount to generate a range of critical test scenarios to test the safety of decision-making in ADSs. While existing researches primarily focus on reproducing real-world traffic accidents in simulation environments to create test scenarios, it is essential to highlight that many of these accidents do not result in safety violations of decision-making in ADSs due to the differences between human driving and autonomous driving. More importantly, we observe that some accident-free real-world scenarios can lead to misbehaviors of ADSs. Therefore, orthogonally to existing work, it is equally important to discover safety violations of ADSs from routine traffic scenarios (i.e., accident-free scenarios) to ensure the safety of Autonomous Vehicles (AVs). We introduce CRISER , a novel methodology to achieve the above goal. It automatically generates abstract and concrete scenarios from real-traffic videos where human-driving worked safely. Based on them, CRISER discovers safety violations of the ADS's decision-making in semantic equivalent scenarios (i.e., the test scenarios with the same semantics as the original accident-free traffic videos). Specifically, CRISER enhances the ability of Large Multimodal Models (LMMs) to accurately extract scenario semantics from accident-free traffic videos and generate test scenarios by multi-modal few-shot Chain of Thought (CoT). Based on them, CRISER explores the behavior differences between the ego vehicle (i.e., the vehicle connected to the ADS under test) and human-driving in semantic equivalent scenarios. During the exploration search, CRISER keeps the semantic consistency of test scenarios with accident-free traffic videos, and explores the universality of discovered safety violations of the ADS. We implement and evaluate CRISER on the industrial-grade Level-4 ADS, Apollo. The experimental results demonstrate that CRISER can accurately extract scenario semantics and generate test scenarios from traffic videos, and effectively discover distinct types of safety violations of Apollo’s decision-making in accident-free traffic scenarios.

Due to its multi-factor mechanism, variable opioid response, and high-risk adverse reactions, cancer pain remains a major challenge in oncology. To overcome these obstacles, we have developed a collaboration framework based on large language models (LLMs): OncoPainBot. This framework can simulate the reasoning and decision-making of multiple clinical experts to conduct comprehensive cancer pain assessment and management. Our OncoPainBot integrates four specialized agents: Pain-Extraction, Pain-Mechanism Reasoning, Treatment-Planning, and Safety-Check, each corresponding to a unique clinical role. In this paper, we compare seven LLMs and three Retrieval-Augmented Generation(RAG) strategies to determine the optimal model configuration. The final framework was verified on 516 real-world electronic medical records of cancer pain collected. We tested our solution through multiple dimensions. Ultimately, Claude-4 combined with RAG achieved the best overall performance, demonstrating outstanding semantic consistency and evidence-based reasoning in multiple metrics. In clinical validation, OncoPainBot achieved a high degree of consistency between the generated reports and actual clinical documents, while maintaining a high decision-making accuracy (0.841) in the analgesic recommendation task. At the same time, our error analysis shows that most of the differences are caused by patient-specific factors and monitoring recommendations rather than incorrect drug selection, which demonstrates the reliability of our framework. OncoPainBot has demonstrated the feasibility of a cancer pain management system based on LLMs, providing a transparent, evidence-based, and clinical-based framework for personalized analgesic care.

Large Language Models (LLMs) have demonstrated impressive generative and reasoning abilities, yet their tendency to produce factually incorrect or fabricated information—so-called hallucinations—remains a key limitation. This study systematically examines how temperature and system instruction strategies affect hallucination behavior in open-source LLMs executed through the Ollama framework. Three representative models—Gemma 2B, Mistral 7B Instruct, and Phi-3 Mini—were evaluated on the TruthfulQA benchmark using zero-shot, few-shot, and “say-I-don’t-know” prompting paradigms. Performance was measured through exact match, token-level F1, semantic similarity, and embedding-based similarity metrics. Two-way ANOVA and3 Tukey post-hoc analyses revealed that system instruction significantly influenced factual accuracy across all models, while temperature effects were comparatively minor. Few-shot prompting achieved the highest mean F1 score (0.1889), indicating that example conditioning effectively constrained hallucinations. Conversely, “say-I-don’t-know” prompts increased semantic alignment but reduced precision, suggesting a conservative refusal bias. Embedding-based similarity analyses confirmed higher semantic consistency for zero-shot responses. The results highlight that prompt design exerts a stronger and more interpretable influence on hallucination than sampling stochasticity, offering practical guidance for improving the factual reliability of open-source LLMs.

Recent advancements in large language models (LLMs) have significantly improved text-to-image (T2I) generation, enabling systems to produce visually compelling and semantically meaningful images. However, preserving fine-grained semantic consistency in generated images, particularly in response to complex and region-specific textual prompts, remains a key challenge. In this work, we propose a context-aware hierarchical agent mechanism that integrates a semantic condensation strategy to enhance attention efficiency and maintain critical visual-textual alignment. By dynamically fusing contextual information, the method effectively balances computational efficiency and ensures semantic alignment with textual descriptions. Experimental results demonstrate improved visual coherence and semantic consistency across diverse prompts, validated through quantitative metrics and qualitative analysis. Our contributions include: (i) introducing a novel semantic condensation strategy that enhances attention efficiency while preserving critical feature information; (ii) developing a new hierarchical agent attention mechanism to enhance computation efficiency; (iii) designing an iterative feedback method based on CLIP Score to improve image diversity and overall quality.

Visual object tracking using Siamese networks has proven effective by matching a reference target with candidate regions. However, their performance is limited by static templates, insufficient context modeling, and weak multi-level feature integration, especially under occlusion, background clutter, and appearance variation. To address these limitations, we propose TSDTrack, a transformer-augmented Siamese tracker designed for quality-aware and robust tracking. Our framework employs a ResNet backbone to extract multi-scale hierarchical features, which are fused using a transformer-based module that applies global attention to enhance semantic and spatial consistency. The prediction head consists of two branches: a confidence aware branch (CAB) that assesses the confidence of classification responses, and a regression distribution learning (RDL) branch that models bounding box localization as discrete probability distributions, improving precision under uncertainty. Furthermore, we introduce a confidence-gated template update strategy that selectively refreshes the target representation based on the CAB score, enabling adaptive appearance modeling while avoiding drift. Experiments on LaSOT, GOT-10k, OTB100, and UAV123 demonstrate that TSDTrack achieves state-of-the-art performance in both accuracy and robustness, achieving 55.5% success on LaSOT, 67.5% AO on GOT-10k, 71.6% AUC on OTB100, and 66.4% success on UAV123, outperforming recent transformer-based and Siamese trackers.

Early and accurate date palm disease detection is the key to successful smart farming ecosystem sustainability. In this paper, we introduce DoST-DPD, a new Dual-Stream Transformer architecture for multimodal disease diagnosis utilizing RGB, thermal and NIR imaging. In contrast with standard deep learning approaches, our model receives ontology-based semantic supervision (via per-dataset OWL ontologies), enabling knowledge injection via SPARQL-driven reasoning during training. This structured knowledge layer not only improves multimodal feature correspondence but also restricts label consistency for improving generalization performance, particularly in early disease diagnosis. We tested our proposed method on a comprehensive set of five benchmarks (PlantVillage, PlantDoc, Figshare, Mendeley, and Kaggle Date Palm) together with domain-specific ontologies. An ablation study validates the effectiveness of incorporating ontology supervision, consistently improving the performance across Accuracy, Precision, Recall, F1-Score and AUC. We achieve state-of-the-art performance over five widely recognized baselines (PlantXViT, Multi-ViT, ERCP-Net, andResNet), with our model DoST-DPD achieving the highest Accuracy of 99.3% and AUC of 98.2% on the PlantVillage dataset. In addition, ontology-driven attention maps and semantic consistency contributed to high interpretability and robustness in multiple crop and imaging modalities. Results: This work presents a scalable roadmap for ontology-integrated AI systems in agriculture and illustrates how structured semantic reasoning can directly benefit multimodal plant disease detection systems. The proposed model demonstrates competitive performance across multiple datasets and highlights the unique advantage of integrating ontology-guided supervision in multimodal crop disease detection.

The rapid development of intelligent transportation systems and autonomous driving technologies has made visual perception a key component in ensuring safety and improving efficiency in complex traffic environments. As a core task in visual perception, object detection directly affects the reliability of downstream modules such as path planning and decision control. However, adverse weather conditions (e.g., fog, rain, and snow) significantly degrade image quality—causing texture blurring, reduced contrast, and increased noise—which in turn weakens the robustness of traditional detection models and raises potential traffic safety risks. To address this challenge, this paper proposes an enhanced object detection framework, ECL-YOLOv11 (Edge-enhanced, Context-guided, and Lightweight YOLOv11), designed to improve detection accuracy and real-time performance under adverse weather conditions, thereby providing a reliable solution for in-vehicle perception systems. The ECL-YOLOv11 architecture integrates three key modules: (1) a Convolutional Edge-enhancement (CE) module that fuses edge features extracted by Sobel operators with convolutional features to explicitly retain boundary and contour information, thereby alleviating feature degradation and improving localization accuracy under low-visibility conditions; (2) a Context-guided Multi-scale Fusion Network (AENet) that enhances perception of small and distant objects through multi-scale feature integration and context modeling, improving semantic consistency and detection stability in complex scenes; and (3) a Lightweight Shared Convolutional Detection Head (LDHead) that adopts shared convolutions and GroupNorm normalization to optimize computational efficiency, reduce inference latency, and satisfy the real-time requirements of on-board systems. Experimental results show that ECL-YOLOv11 achieves mAP@50 and mAP@50–95 values of 62.7% and 40.5%, respectively, representing improvements of 1.3% and 0.8% over the baseline YOLOv11, while the Precision reaches 73.1%. The model achieves a balanced trade-off between accuracy and inference speed, operating at 237.8 FPS on standard hardware. Ablation studies confirm the independent effectiveness of each proposed module in feature enhancement, multi-scale fusion, and lightweight detection, while their integration further improves overall performance. Qualitative visualizations demonstrate that ECL-YOLOv11 maintains high-confidence detections across varying motion states and adverse weather conditions, avoiding category confusion and missed detections. These results indicate that the proposed framework provides a reliable and adaptable foundation for all-weather perception in autonomous driving systems, ensuring both operational safety and real-time responsiveness.

3D semantic image synthesis with geometric and semantic consistency

The fashion industry increasingly relies on artificial intelligence technologies to enhance creative workflows and accelerate design innovation. This research presents a comprehensive framework that employs Generative Adversarial Net- works and advanced diffusion models to generate high-quality fashion imagery from textual descriptions. The proposed system integrates Stable Diffusion architecture with specialized text preprocessing pipelines to create diverse, photo realistic fashion designs that align with textual specifications while maintaining aesthetic coherence and commercial viability. The framework was evaluated using a dataset of 10,000 high-resolution fashion images, with systematic assessment conducted across multiple performance dimensions including creativity, aesthetic appeal, design diversity, and semantic consistency. Experimental results demonstrate exceptional performance in creative design generation, achieving average scores of 4.7 for originality and 4.5 for aesthetic quality based on comprehensive evaluation by thirty participants. The system successfully produces varied design alternatives from similar prompts, indicating robust exploration of design possibilities rather than repetitive pattern generation. While text prompt accuracy achieved a moderate score of 3.8, highlighting opportunities for enhanced semantic interpretation, the overall results validate the framework’s capability to support professional fashion design workflows. The research contributes to the growing body of knowledge in AI-assisted creative applications and demonstrates significant potential for transforming traditional fashion design processes through intelligent automation and creative augmentation technologies.

The redundancy of SMILES notation, where multiple strings can describe the same molecule, remains a challenge in computational chemistry and cheminformatics. To mitigate this issue, we introduce TokenSMILES, a grammatical framework that standardizes SMILES into structured sentences composed of context-free words. By applying five syntactic constraints (including branch limitations, balanced parentheses, and aromaticity exclusion), TokenSMILES minimizes redundant SMILES enumerations for alkanes while maintaining valence and octet compliance through semantic parsing rules. TokenSMILES does not replace SMILES but rather formalizes its syntax into a standardized, machine-interpretable form. This grammatical structure enables controlled generation and manipulation of valid SMILES strings, ensuring syntactic and semantic consistency while substantially reducing redundancy. Implemented into SmilX, an open-source tool, TokenSMILES generates valid SMILES with accuracy comparable to existing computational implementations for molecules with low hydrogen deficiency (HDI ≤ 4). Its applicability extends beyond alkanes through stoichiometric modifications such as bond insertion, cyclization, and heteroatom substitution. Nevertheless, challenges remain for highly unsaturated systems, where canonicalization artifacts highlight the need for dynamic feasibility checks. By integrating linguistic principles with cheminformatics, TokenSMILES establishes a scalable framework for systematic chemical space exploration, supporting applications in drug discovery, materials design, and machine learning-driven molecular innovation.

DSSFusion: A dynamic expert-guided infrared and visible image fusion network with semantic and structural consistency constraints

GAGM: Geometry-aware graph matching framework for weakly supervised gyral hinge correspondence.

This study presents SemanDICT, a modular semantic dictionary system designed to enhance the semantic richness, traceability, and interoperability of IFC-based Building Information Modeling (BIM) data. The system supports RDF triple generation, SPARQL querying, and multi-user data editing processes through interface panels structured according to ontology engineering principles. Users can select IFC classes to define data types and unit-defined properties; semantic consistency can be verified through bSDD integration. The AI Feature Assistant Panel suggests missing features based on contextual inference, while the Material Properties Database supports domain-focused semantic modeling. RDF export is performed in OWL-compliant formats such as Turtle, RDF/XML, and JSON-LD. The panel-based workflow covers eight core processes: IFC class selection, property suggestion, semantic definition, ontological validation, RDF export, SPARQL querying, statistical analysis, and project collaboration. Each process is paired with a corresponding interface panel, offering modular interaction and extensible development capabilities. The SPARQL Query Panel facilitates semantic queries, while the Dictionary Statistics Panel visualizes data type distribution and semantic density. The Project Collaboration Panel supports multi-user development with simultaneous editing and version control. SemanDICT contributes to academia and industry in the areas of ontology-focused design, data management, and open standards by bringing semantic web technologies together with BIM production environments.

Semantic segmentation of crowdsourced street-level imagery plays a critical role in urban analytics by enabling pixel-wise understanding of urban scenes for applications such as walkability scoring, environmental comfort evaluation, and urban planning, where robustness to geometric transformations and projection-induced symmetry variations is essential. This study presents a comparative evaluation of two primary families of semantic segmentation models: transformer-based models (SegFormer and Mask2Former) and prompt-based models (CLIPSeg, LangSAM, and SAM+CLIP). The evaluation is conducted on images with varying geometric properties, including normal perspective, fisheye distortion, and panoramic format, representing different forms of projection symmetry and symmetry-breaking transformations, using data from Google Street View and Mapillary. Each model is evaluated on a unified benchmark with pixel-level annotations for key urban classes, including road, building, sky, vegetation, and additional elements grouped under the “Other” class. Segmentation performance is assessed through metric-based, statistical, and visual evaluations, with mean Intersection over Union (mIoU) and pixel accuracy serving as the primary metrics. Results show that LangSAM demonstrates strong robustness across different image formats, with mIoU scores of 64.48% on fisheye images, 85.78% on normal perspective images, and 96.07% on panoramic images, indicating strong semantic consistency under projection-induced symmetry variations. Among transformer-based models, SegFormer proves to be the most reliable, attains higher accuracy on fisheye and normal perspective images among all models, with mean IoU scores of 72.21%, 94.92%, and 75.13% on fisheye, normal, and panoramic imagery, respectively. LangSAM not only demonstrates robustness across different projection geometries but also delivers the lowest segmentation error, consistently identifying the correct class for corresponding objects. In contrast, CLIPSeg remains the weakest prompt-based model, with mIoU scores of 77.60% on normal images, 59.33% on panoramic images, and a substantial drop to 59.33% on fisheye imagery, reflecting sensitivity to projection-related symmetry distortions.

Cross-condition fault diagnosis of cotton harvester picking-head drivetrains remains challenging due to significant distribution discrepancies in vibration signals under different operating conditions. Existing transfer learning approaches predominantly focus on domain-invariant features while failing to sufficiently exploit domain-specific information and the structural constraints embedded in target-domain normal samples, which often leads to unstable diagnostic performance across conditions. To address this issue, this paper proposes a prototype-guided dual-feature transfer learning method termed Proto-DISFNet (Prototype-guided Domain-Invariant and Domain-Specific Feature Network). The proposed method explicitly disentangles domain-invariant and domain-specific features to alleviate the impact of operating condition variations. High-confidence pseudo-labeled samples, selected through a filtering strategy, are utilized to construct class prototypes in the target domain, thereby enhancing semantic consistency and structural awareness in the feature space. In addition, a stage-wise training strategy is introduced to coordinate multi-task optimization, which improves training stability and overall adaptability under representative complex engineering operating conditions. Experiments conducted on three vibration datasets, JNU, THU, and CHPH-FETB, demonstrate that Proto-DISFNet achieves stable and competitive cross-condition diagnostic performance under varying degrees of domain shift and operating conditions. These results indicate the engineering relevance and potential applicability of the proposed method for fault diagnosis of cotton harvester picking-head drivetrains.

The development of autonomous driving systems requires seamless integration between high-level system models and simulation environments to enable early validation and verification. This paper presents a novel methodology that bridges the gap between Model-Based Systems Engineering (MBSE) and 3D simulation for autonomous driving scenarios. We in-troduce RoadRunnerSysMLProfile, a specialized SysML profile that extends standard modeling capabilities with domain-specific constructs for autonomous driving environments and behaviors. This profile customizes SysML diagrams to create RoadRunner Scene Integration Diagrams for static road elements and RoadRunner Scenario Integration Diagrams for dy-namic vehicle behaviors. Additionally, we present AutoSim Transfer, an AI-enhanced transformation tool that leverages machine learning techniques to automatically detect complex junctions, preserve semantic consistency, and optimize the conversion of SysML XMI files into standardized OpenDRIVE and OpenSCENARIO formats compatible with the Road-Runner simulation environment. Our approach addresses key challenges in current MBSE-to-simulation workflows, includ-ing semantic preservation, automated junction detection, and bidirectional traceability. In a case study on complex urban driving scenarios, the proposed methodology demonstrated a 78% reduction in manual modeling effort and achieved 92% accuracy in detecting multi-lane intersections compared to conventional approaches. This methodology enables automotive engineers to maintain consistency between system specifications and simulation environments throughout the development lifecycle, facilitating more comprehensive validation of autonomous driving functions.

Object detection in adverse weather remains challenging due to the simultaneous degradation of visibility, structural boundaries, and semantic consistency. Existing restoration-driven or multi-branch detection approaches often fail to recover task-relevant features or introduce substantial computational overhead. To address this problem, DLC-SSD, a lightweight degradation-aware framework for detecting robust objects in adverse weather environments, is proposed. The framework integrates image enhancement and feature refinement into a single detection pipeline and adopts a hierarchical strategy in which global and local degradations are corrected at the image level, structural cues are reinforced in shallow high-resolution features, and semantic representations are refined in deep layers to suppress weather-induced noise. These components are jointly optimized end-to-end with the single-shot multibox detection (SSD) backbone. In rain, fog, and low-light conditions, DLC-SSD demonstrated more stable performance than conventional detectors and maintained a quasi-real-time inference speed, confirming its practicality in intelligent monitoring and autonomous driving environments.

Visualizing poetry with deep semantic understanding and consistency evaluation