Search Results (338)

Background: The Magnetic Resonance Imaging (MRI)-based tumor segmentation remains a challenging problem in medical imaging due to tumor heterogeneity, unpredictable morphological features, and the high complexity of calculations needed to implement it in clinical practice, putting it out of the scope of real-time applications. Although neural networks have significantly improved segmentation performance, they still struggle to capture morphological tumor features while maintaining computational efficiency. This work introduces Hybrid Adaptive Attention U-Net (HAAU-Net) framework, combining context-aware morphologically stable features and spatial channel attention to achieve high-quality tumor segmentation with less computational cost. Methods: The proposed HAAU-Net framework integrates multi-scale Adaptive Attention Blocks (AAB), Context-Aware Morphological Feature Module (CAMFM) and Spatial-Channel Hybrid Attention Mechanism (SCHAM). CAMFM is used to maintain the stability of morphological features by hierarchical aggregation and dynamic normalization of features. SCHAM enhances feature representation by modelling channels and spatial regions where the strongest feature are determined to use in segmentation. On the BRaTS 2022/2023 data, the proposed HAAU-Net is evaluated using four modalities including T1, T1GD, T2 and T2-FLAIR sequences. Results: The proposed model able to obtain 96.8% segmentation accuracy with a Dice coefficient of 0.89 on the entire tumor region, outperforming the alternative U-Net (0.83) and conventional CNN methods of segmentation (0.81). The proposed HAAU-Net architecture cuts the computational complexity of the standard deep learning models by 43% and still achieve real-time inference (28 FPS on a regular GPU). The hybrid model used to predict survival has a C-Index of 0.91 which is higher than the traditional SVM-based methods (0.72).Conclusions: Spatial-channel attention, combined with morphologically stable features, can be combined to allow clinically significant interpretability in attention maps. The proposed framework significantly improves segmentation performance while maintaining computational effeciency. This broad system has a serious potential of AI- enabled clinical decision support system and early prognostic diagnosis in neuro-oncology with practical deployment capability. Full article

Melissopalynological analysis is essential for determining the botanical origin of honey, corbicular pollen and bee bread, as well as detecting adulteration. However, it traditionally relies on labor-intensive and subjective manual pollen identification. As a proof-of-concept preceding full honey analysis, this study evaluates artificial intelligence methods for automated pollen grain recognition under controlled conditions. Hazel (Corylus avellana L.) and dandelion (Taraxacum officinale F.H. Wigg.) were used as model taxa to validate the proposed approach before its application to real varietal honey samples. This study introduces a novel three-stage pipeline that decouples object detection from feature extraction, utilizing YOLOv12m for region-of-interest generation and, for the first time in melissopalynology, DINOv3 ConvNeXt-B for deep feature representation. Microscopic images acquired at 400× magnification yielded 2498 dandelion and 1941 hazel pollen grains. The detector achieved an mAP@0.5 of 0.936 with an F1 score of 0.88, while the classifier reached 98.1% accuracy with good class separability (Silhouette coefficient: 0.407). The primary technical contribution is the systematic optimization of the detection-to-classification interface. Context-aware bounding box expansion (12%) and an optimized IoU-NMS threshold (0.65) significantly improve the stability of morphological feature extraction, as confirmed by ablation studies. Computational cost reporting further supports reproducible, deployment-oriented comparison. The results confirm the feasibility of this AI-based framework as an intermediate step toward automated melissopalynological analysis, with future work focusing on standardized microscopy protocols and expanded pollen databases for varietal honey authentication. Full article

Underwater images often suffer from color distortion and low contrast, severely limiting the reliability of visual perception systems. Existing methods struggle to balance enhancement quality and computational efficiency. To address this issue, we propose PCF-Net (Perception-driven Color Fusion Network), a lightweight dual-branch network for underwater image enhancement based on a stable HSV-CS (Hue-Saturation-Value with sine–cosine transformation) color-space representation. Specifically, a sine–cosine transformation is introduced to construct a stable HSV-CS color space, effectively avoiding hue discontinuities at boundary regions in conventional HSV representations. To compensate for underwater degradation, a Color-Bias-Aware module and a Value-Confidence module are designed to adaptively correct color distortion and luminance degradation. Furthermore, a lightweight Channel-Spatial Adaptive Gated Fusion module dynamically aggregates features from the RGB and HSV-CS branches in a perception-driven manner. The overall architecture incorporates multi-branch re-parameterizable convolutions, significantly reducing computational cost while preserving strong representational capacity. Extensive experiments on underwater image enhancement benchmarks, including UIEB and RUIE, demonstrate that PCF-Net achieves state-of-the-art performance in terms of PSNR, SSIM, and UIQM, along with visually superior color correction and contrast enhancement. With only 0.17 M parameters, the proposed model runs at 118.6 FPS on an RTX 3090 and 35.3 FPS on a Jetson Orin Nano at a resolution of 512 × 512, making it well suited for resource-constrained real-time underwater vision applications. Full article

Classic machine learning and regime identification methods applied to financial time series lack theoretical guarantees and exhibit systematic failure modes: heavy-tails invalidate moment-based geometry, rendering distances and centroids dominated by extremes or unstable; jumps violate smoothness, destabilizing local regressions, kernel methods, and gradient-based learning; and non-stationarity disrupts neighborhood relations, so distances in classical feature spaces no longer reflect meaningful proximity. To address these challenges, we propose a topology-based machine-learning framework grounded on probabilistic reconstruction of state-space geometry, which replaces moment- and smoothness-dependent representations with deformation-stable summaries of state-space geometry, preserving neighborhoods, adjacency, and topology. The finite-sample validity of homeomorphic state-space reconstruction, required for topology-based machine learning, is assessed through numerical studies on synthetic data with heavy tails, jumps, and known ground-truth regimes. Further diagnostics of local invertibility and bounded geometric distortion quantify when embedding windows are consistent with local diffeomorphic behavior, enabling metric-sensitive, geometry-aware learning. Clustering of Hilbert-space summaries accurately recovers underlying market tail-risk regimes with robust results across selected filtrations. Temporal, feature-space, and cluster-label null tests confirm that topology-based clustering captures genuine topological structure rather than noise or artifacts, and encodes temporal dependencies at local, mesoscopic, and network levels associated with market regimes. Full article

Multi-view 3D reconstruction is essential for smart city, supporting applications such as smart city planning and autonomous navigation. While traditional reconstruction pipelines and recent neural implicit methods, such as NeRF, achieve high visual fidelity, they often struggle with geometric accuracy and sparse-view scenarios. To address this challenge, we present DiffLiGS, a novel multi-modal 3D reconstruction framework that integrates LiDAR point clouds and LiDAR-guided diffusion-based priors into the 3D Gaussian Splatting (3DGS) pipeline, enabling high-fidelity and geometrically accurate models. Our method first densifies sparse LiDAR depths using a diffusion model and refines them through multi-view geometric constraints, producing dense LiDAR depth maps that provide robust supervision for 3DGS optimization. Leveraging these dense depth maps, we guide a Stable Video Diffusion model to synthesize novel view images, which are incorporated into training to enhance reconstruction completeness and visual realism. By jointly fusing rich appearance cues from multi-view images with precise LiDAR-derived geometry and diffusion priors, DiffLiGS achieves unified, geometry-aware 3D scene representations. Our extensive experiments demonstrate that our approach significantly improves both geometric accuracy and rendering quality compared to existing 3D reconstruction methods, enabling real-time, high-precision modeling of complex urban environments. Full article

Heavy-load unmanned aerial vehicles (UAVs) are increasingly being applied in logistics, infrastructure installation, and emergency response missions, where complex payload dynamics and unstructured environments pose significant challenges to safe and efficient operation. Conventional manual teleoperation interfaces, such as dual-joystick control, impose a high cognitive workload and provide limited support for expressing high-level operator intent, while fully autonomous solutions remain difficult to deploy reliably under real-world uncertainty. To address these limitations, this paper proposes the Multimodal Fusion Cooperation Network (MFCN), an end-to-end shared autonomy framework that integrates speech commands, visual gestures, and haptic cues through cross-modal feature fusion to infer operator intent in real time. The fused intent representation is translated into dynamically feasible control commands by a cooperative control policy with embedded physics-aware constraints to suppress payload oscillations and ensure flight stability. Extensive semi-physical simulations and real-world experiments demonstrate that the MFCN significantly improves the task success rate, positioning accuracy, and payload stability while reducing the task completion time and operator cognitive workload compared with manual, unimodal, and heuristic multimodal baselines. Full article

Infrared–visible multimodal object detection has attracted increasing attention for its robustness under challenging conditions such as low illumination, occlusion, and complex backgrounds. However, existing fusion methods often suffer from coarse illumination modeling and insufficient cross-modal semantic alignment, leading to performance degradation in scenes with strong illumination variations or modality imbalance. To address these issues, this paper proposes IAF-RTDETR (Illumination-Aware Fusion RT-DETR), an illumination-aware fusion real-time detection network built upon the RT-DETR framework. The proposed method introduces a progressive fusion pipeline composed of four key modules: (1) a Modality-Specific Feature Enhancer to recalibrate modality-dependent representations and suppress low-quality feature interference; (2) a lightweight Global Light Estimator that learns a continuous illumination score via self-supervised proxy supervision derived from RGB image statistics; (3) a Light-Aware Fusion module that dynamically adjusts multi-scale fusion weights of infrared and visible features according to the estimated illumination; and (4) a Cross-Layer Dual-Branch Interaction Module that alleviates cross-modal semantic shift through bidirectional attention-guided interaction and channel reweighting. Extensive experiments on the M3FD dataset demonstrate that the proposed method achieves consistent performance improvements under diverse lighting conditions, outperforming RGB-only and IR-only baselines by 7.4% and 16.1% in mAP@50, respectively, while maintaining real-time inference speed (≈17.3 ms). Further evaluations on the LLVIP dataset validate the robustness and generalization ability of IAF-RTDETR in real low-illumination scenarios. Moreover, compared with representative multimodal fusion methods such as TFDet and TarDAL, the proposed method achieves superior detection accuracy. Visualization and quantitative semantic consistency analyses further confirm the effectiveness of the proposed illumination-aware fusion and cross-layer interaction mechanisms. These results indicate that IAF-RTDETR provides an effective and practical solution for real-time infrared–visible object detection under complex lighting environments. Full article

Adaptive learning at scale requires explicit handling of uncertainty and information flow across diverse educational technologies. This paper proposes a TOGAF-conformant enterprise architecture for the University of Tabuk (UT) that embeds entropy- and uncertainty-aware requirements from the outset and aligns them with institutional goals in teaching, research, and administration. Using the Architecture Development Method (ADM), we map information-theoretic requirements to architectural artifacts across the architecture vision, business, information systems, and technology domains; formally specify core entropy-informed observables, including predictive entropy, expected information gain, workflow variability entropy, and uncertainty hot-spot severity; and define semantic and metadata standards for their near-real-time computation. These indicators are positioned explicitly across the TOGAF domains: business architecture identifies where uncertainty matters, information systems architecture defines the computable data and application representations, technology architecture operationalizes secure and scalable computation, and later ADM phases use the resulting metrics for prioritization and governance. The architecture also establishes governance that ranks initiatives by their expected uncertainty reduction through Architecture Review Board (ARB) decision gates. We address three research questions: (R.Q.1) how to design a TOGAF-conformant architecture for UT that natively encodes uncertainty-aware requirements and aligns with institutional needs; (R.Q.2) how to integrate dispersed data, achieve semantic harmonization, and deliver analytics-ready streams that support information-theoretic indicators for personalization without delay; and (R.Q.3) how to embed IT demand planning in opportunities and solutions and migration planning using uncertainty reduction and expected information gain as prioritization criteria. The resulting architecture offers a university-wide foundation for adaptive learning: it unifies learner and system interaction data under governed schemas, supports low-latency analytics, and formalizes decision processes that treat uncertainty as a primary metric. Though learner-level operational validation is future work, the design establishes the technical and organizational foundations for responsible, large-scale deployment of entropy-driven learner modeling, content sequencing, and feedback optimization. Full article

This article addresses the problem of finding the shortest route for Automated Guided Vehicles (AGVs) in a production environment with constrained battery state-of-charge (SoC) and time-dependent operating conditions. The route map is divided into a uniform grid containing stationary obstacles and two types of dynamic obstacles: human, for which AGV transportation is prohibited, and inanimate (moving objects), which impose a penalty function. A key contribution of the proposed methodology is the introduction of a battery residual charge matrix, which embeds cell-level energy feasibility directly into the grid-based environment representation by determining minimum admissible SoC constraints and accounting for transition-dependent energy costs. This matrix restricts the set of traversable cells under low-energy conditions, enabling energy-aware route feasibility evaluation during both initial planning and adaptive replanning. The proposed approach is based on the A* and D* Lite algorithms, providing shortest-path construction that explicitly integrates battery SoC into the spatio-temporal cost function. To avoid collisions in a multi-agent environment during routing, a simplified hybrid scheme with M* elements performs local coordination and adaptive trajectory replanning. The effectiveness of the proposed methodology was assessed using travel time, temporal complexity, and spatial complexity metrics. Simulation results on a $10\times 10$ grid showed that agents with sufficient battery completed routes of 8 and 11 cells with travel times of 7.2 to 10.7 conventional units. A critically low-energy agent was initially unable to move, but after adjusting the minimum SoC constraint, all agents completed their routes with travel times up to 11.4 conventional units, demonstrating the direct impact of energy constraints on system performance. Additional experiments with varying agent counts and SoC thresholds confirmed reliable balancing of route feasibility and energy constraints across configurations. Full article

Accurate tree species mapping is critical for forest inventory, biodiversity assessment, and ecosystem management. In mountainous regions, terrain-induced radiometric non-stationarity and limited field access often produce scarce, clustered, and environmentally biased samples, limiting model generalization. To address this issue, this study proposes a terrain-aware self-supervised representation learning framework for tree species classification under small-sample conditions. The framework integrates terrain information into representation learning and adopts a hybrid contrastive–generative self-supervised strategy to learn discriminative and terrain-robust features from large volumes of unlabeled multi-source remote sensing data. These learned representations are subsequently combined with limited field samples to produce regional-scale tree species maps. Experiments conducted across Yunnan Province, China, using Sentinel-1, Sentinel-2 and Landsat time-series data show that the proposed framework substantially improvesa class separability and classification robustness in complex mountainous environments. The framework achieves an overall accuracy of 75.8%, significantly outperforming conventional feature engineering (38.3–40.6%) and supervised deep learning models (37.3–47.8%). Species with relatively homogeneous structure and strong ecological niche dependence can be accurately mapped with limited training samples, whereas structurally complex forest communities require broader environmental sample coverage. Overall, the results highlight the potential of terrain-aware self-supervised representation learning as a scalable and data-efficient paradigm for forest mapping in mountainous and environmentally heterogeneous regions. Full article

As cyberattacks grow increasingly diverse and sophisticated, achieving accurate yet efficient network traffic anomaly detection has become a fundamental challenge in modern cybersecurity. While existing machine learning methods enable effective feature extraction, they remain limited in jointly modeling multi-scale temporal dynamics and frequency-domain characteristics of anomalous network behaviors, and typically incur substantial computational overhead when processing long traffic sequences. These limitations hinder their effectiveness in real large-scale deployments. To overcome these challenges, this paper proposes a Multi-scale time-frequency Representation learning and Lightweight Attention (MRLA)-based model, which unifies hierarchical time and frequency feature learning with efficient long-range dependency modeling. Extensive experiments on the CIC-IDS2018, CIC-DDoS2019, and UNSW-NB15 datasets with session-aware data splits demonstrate that MRLA achieves F1-scores of 99.94%, 99.78%, and 93.74%, respectively. These results indicate that MRLA consistently delivers high detection accuracy with improved computational efficiency, offering a robust and scalable solution for network traffic anomaly detection across diverse attacks. Full article

Accurate forecasting of equity returns remains fundamentally constrained by weak short-horizon predictability, pronounced noise, and structural non-stationarity. While deep learning models have been widely applied to financial time series, most studies prioritize point prediction and provide limited guidance on reliable uncertainty quantification, particularly in emerging markets. This study developed an uncertainty-aware forecasting framework for the South African equity market by integrating variational mode decomposition (VMD), gated recurrent units (GRUs), and temporal conformal prediction (TCP) to construct distribution-free prediction intervals with finite-sample coverage guarantees. Using daily returns from the FTSE/JSE All Share Index, we first confirmed that baseline recurrent models applied directly to raw returns exhibited negligible out-of-sample explanatory power, consistent with weak-form market efficiency. Incorporating VMD enhanced representation learning and improved point forecast accuracy by isolating latent frequency components. However, model-based predictive variance alone proved insufficient for reliable calibration. Embedding the models within a rolling conformal prediction framework restored near-nominal coverage across multiple confidence levels while allowing interval widths to adapt dynamically to changing volatility regimes. Robustness analyses, including walk-forward validation, stress-regime evaluation, and block permutation negative control experiments, indicated that the observed performance was not driven by temporal leakage or alignment artifacts. The results further highlight a trade-off between interval sharpness and tail-risk protection, particularly during extreme market events. Overall, the findings support a shift from return-level prediction toward calibrated uncertainty estimation as a more stable and economically meaningful objective in non-stationary financial environments. Full article

Multistep-ahead prediction of critical states in the iron ore sintering process is essential for maintaining production stability, enhancing energy efficiency, and reducing industrial emissions. However, large time delays, strong coupling, and condition drifts challenge existing spatiotemporal graph neural networks (STGNNs). This paper proposes Process-aware Mixed Dynamic Graph Network (ProMix-DGNet), which integrates a Decoupled Two-Stream Topology Learning mechanism—fusing Adaptive Static Graph with a Radial Basis Function (RBF)-driven Dynamic Graph Constructor—to ensure robust spatial modeling under high-noise conditions. Furthermore, Process-View Global Mixer explicitly captures long-range process coupling across the entire sintering strand, overcoming the receptive field limitations of traditional graph convolutions. In the decoding phase, a future control-informed module utilizes a bidirectional Long Short-Term Memory (BiLSTM) and a global mixer to align known future control setpoints with the system’s spatial topology. These features are integrated via a gated residual mechanism that dynamically modulates the interaction between control intents and historical representations. Extensive experiments conducted on two real-world industrial datasets, Sinter-A and Sinter-B, demonstrate that ProMix-DGNet consistently outperforms mainstream baselines across multiple metrics, including Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). The results verify the model’s higher accuracy and robustness in complex large-time-delay systems, offering a reliable framework for the intelligent monitoring and closed-loop optimization of sintering process. Full article

Autonomous systems are expanding rapidly, driving a demand for robust perception technologies capable of navigating challenging, unstructured environments. While urban autonomy has made significant progress, off-road environments pose unique challenges, including dynamic terrain and limited communication infrastructure. This research addresses these challenges by introducing a novel three-part solution for off-road autonomous vehicles. First, we present a large-scale off-road dataset curated to capture the visual complexity and variability of unstructured environments, providing a realistic training ground that supports improved model generalization. Second, we propose a Confusion-Aware Loss (CAL) that dynamically penalizes systematic misclassifications based on class-level confusion statistics. When combined with cross-entropy, CAL improves segmentation mean Intersection over Union (mIoU) on the off-road test set from 68.66% to 70.06% and achieves cross-domain gains of up to ~0.49% mIoU on the Cityscapes dataset. Third, leveraging semantic segmentation as an intermediate representation, we introduce a spatial overlay video encoding scheme that preserves high-fidelity RGB information in semantically critical regions while compressing non-essential background regions. Experimental results demonstrate Peak Signal-to-Noise Ratio (PSNR) improvements of up to +5 dB and Video Multi-Method Assessment Fusion (VMAF) gains of up to +40 points under lossy compression, enabling efficient and reliable off-road autonomous operation. This integrated approach provides a robust framework for real-time remote operation in bandwidth-constrained environments. Full article

Battery-powered electric unmanned surface vehicles (e-USVs) and electrified surface-cleaning platforms require reliable onboard vision under strict compute and power constraints. In reflective water environments, tiny floating debris is often obscured by specular highlights, reflection bands, ripples, motion blur, and camera jitter, while label noise further degrades training stability. To improve robustness without increasing onboard inference burden, this paper proposes YOLOv11-IMP, a lightweight detector for reflective water-surface scenes and embedded edge inference. The method integrates a transformer-enhanced backbone stage, a Global Channel–Spatial Attention module in the neck, and a median-enhanced channel–spatial module in the neck to improve global-context modeling, cross-scale interaction, and weak-boundary representation. WIoU-v3 is adopted to improve localization, and a train-time-only noise-aware screening strategy based on the small-loss principle is introduced to suppress unreliable labels without extra inference cost. Experiments on the CAS dataset and a self-built debris dataset show gains of 3.3% in AP@0.75 and 6.5% in AP for small objects over YOLOv11, while maintaining 7.3 GFLOPs and real-time inference on Jetson Nano, demonstrating practical potential for energy-constrained onboard missions. Full article