Search Results (6,485)

Unsupervised change detection (UCD) from heterogeneous bitemporal optical–SAR imagery is challenging due to modality discrepancy, speckle/illumination variations, and the absence of change annotations. We propose MV-S2CD, a vision foundation model (VFM)-based framework that learns a modality-bridged latent space and produces dense change maps in a fully unsupervised manner. To robustly adapt pretrained VFM priors to heterogeneous inputs with minimal task-specific parameters, MV-S2CD incorporates lightweight modality-specific adapters and parameter-efficient low-rank adaptation (LoRA) in high-level layers. A shared projector embeds the two observations into a common geometry, enabling consistent cross-modal comparison and reducing sensor-induced domain shift. Building on the bridged representation, we design a dual-branch change reasoning module that decouples structure-sensitive cues from semantic-consistency cues: a structure pathway preserves fine boundaries and local variations, while a semantic-consistency pathway employs reliability gating and multi-scale context aggregation to suppress pseudo-changes caused by modality-specific nuisances and residual misregistration. For label-free optimization, we develop a difference-centric self-supervision scheme with two perturbation views and reliability-guided pseudo-partitioning, jointly enforcing pseudo-unchanged invariance, pseudo-changed/unchanged separability, and sparsity and edge-preserving regularization. Experiments on three heterogeneous optical–SAR benchmarks demonstrate that MV-S2CD consistently improves the Precision–Recall trade-off and achieves state-of-the-art performance among unsupervised baselines, while remaining backbone-flexible and efficient. Full article

Bird’s-eye-view (BEV) perception has emerged as a key representation for unified scene understanding in autonomous driving. However, current BEV methods relying solely on monocular cameras suffer from severe degradation under adverse weather and dynamic scenes due to limited depth cues and illumination dependency. To address these challenges, we propose a robust multi-modal BEV perception framework that integrates dual-source 4D millimeter-wave radar and multi-view camera images. The proposed architecture systematically exploits Doppler velocity and temporal information from 4D radar to model dynamic object motion, while introducing a deformable fusion strategy in the BEV space for accurate semantic alignment across modalities. Our design includes four key modules: a Doppler-Aware Radar Encoder (DARE) that enhances motion-sensitive features via velocity-guided attention; a Fog-Aware Feature Denoising Module (FADM) that suppresses modality inconsistency in low-visibility conditions through cross-modal attention and residual enhancement; a Multi-Modal Temporal Fusion Module (TFM) that encodes radar temporal sequences using a Transformer encoder for motion continuity modeling; and a confidence-aware multi-task loss that jointly supervises semantic segmentation, motion estimation, and object detection. Extensive experiments on the DualRadar dataset and adverse-weather simulations demonstrate that our method achieves significant gains over state-of-the-art baselines in BEV segmentation accuracy, detection robustness, and motion stability. The proposed framework offers a scalable and resilient solution for real-world autonomous perception, especially under challenging environmental conditions. Full article

This study investigates how financial technology (FinTech) contributes to economic growth, focusing on whether it acts primarily as a mediator or a moderator within the finance–growth nexus. A composite FinTech index is constructed using Principal Component Analysis based on cross-country data for 2021, and the analysis distinguishes between High-Income and Non-High-Income economies following the World Bank classification. The results show that in developing and emerging economies, FinTech mainly serves as a mediator, helping to close structural gaps in financial intermediation and expanding access to financial services. In High-Income countries, by contrast, FinTech acts as a moderator, enhancing innovation and efficiency in mature financial systems. When financial inclusion is disaggregated by gender, the findings reveal additional nuances. FinTech fosters growth through inclusion for both men and women, but its effects are stronger for male account ownership in developing economies and more balanced in High-Income contexts. In general, the study contributes to the literature by developing a multidimensional FinTech index, clarifying its dual mediating and moderating functions, and introducing a gender-sensitive perspective that highlights the uneven distribution of FinTech’s growth benefits between income levels and genders. Full article

Multimodal sentiment analysis (MSA) has emerged as a prominent research frontier, enabling a comprehensive understanding of complex human emotions through the synergistic integration of heterogeneous multimodal signals. However, most existing approaches rely on idealized signal distribution assumptions, overlooking the detrimental impact of demographic bias on representation fairness and fusion robustness. This paper proposes a Label-Guided Contrastive Decoupling Fusion (LGCDF) framework that enhances model robustness to demographic bias by learning and fusing multimodal representations invariant to Sensitive Attributes (SAs). Specifically, the proposed LGCDF framework employs gender-sensitive attribute information as modality-level constraints to achieve language-centric cross-modal sentiment alignment, which is accomplished by computing contrastive losses between text–audio and text–visual feature pairs. Moreover, it introduces a SA-guided contrastive decoupling mechanism that decomposes multimodal representations into SA-related and -independent components. The SA-independent components are subsequently fused through a cross-modal attention fusion strategy, thereby facilitating fair sentiment representation and enabling efficient and robust multimodal information fusion. Extensive experimental results demonstrate that the proposed LGCDF framework achieves superior performance in fair representation learning and cross-modal information fusion while maintaining strong robustness under varying gender distribution biases. Full article

Background/Objectives: Cross-linguistic studies show that the production of morphosyntactic elements (e.g., clitics) is problematic and often omitted in nonfluent agrammatic aphasia (NFA), with the degree of impairment varying across languages. Serbian, with its rich clitic system, provides a sensitive window into grammatical impairment. This study is the first to examine the production of proclitics and enclitics in Serbian speakers with aphasia and their relationship to short-term and working memory. Methods: Forty-six individuals with stroke-induced aphasia (25 NFA and 21 fluent aphasia [FA]) and 54 healthy controls completed an experimental Serbian clitic production test. Participants were prompted to produce clitic sentences (12 proclitics, such as prepositions or conjunctions, and 18 clitics, such as pronouns or auxiliary verbs) in response to various scenarios. Performances were correlated with sentence repetition and digit span (forward/backward). Results: Both aphasia groups produced significantly fewer clitics than controls (p < 0.001). Participants with NFA produced fewer overall clitics and showed no clitic type effects (p = 0.329), whereas participants with FA produced proclitics more accurately than enclitics (p = 0.028). Clitic production correlated with performance on sentence repetition and digit span tasks, but patterns differed by aphasia group. In NFA, both enclitics and proclitics were associated with sentence repetition and digit span (p < 0.05), whereas in FA, these measures were primarily associated with enclitic production (p < 0.05). Conclusions: Clitics production distinguishes nonfluent from fluent aphasia in Serbian and is differentially supported by working and verbal memory resources. The Serbian clitic production test reveals a selective proclitic advantage that is observed only in fluent aphasia, serving as a sensitive clinical marker in this population. Full article

Urban playgrounds are vital public spaces that support children’s play, social interaction, and well-being. However, many playgrounds in socially disadvantaged or aging urban areas experience physical deterioration, limited play diversity, and declining use. Although corporate social responsibility (CSR) initiatives have increasingly supported playground regeneration, many projects continue to emphasize short-term physical improvements rather than participatory processes and social value creation. This study conceptualizes CSR-supported, child-participatory playground regeneration as a social value creation process and examines how CSR enables process continuity through a structured six-stage participatory approach spanning planning, design, construction, and post-opening use. Two cases were selected from the “Save the Playground” program in Seoul, Korea: Saerok Children’s Park in a stable residential neighborhood and Mukjeong Children’s Park in a high-mobility, multicultural commercial district. Using a qualitative multiple-case study design, the study triangulates workshop outputs, observational records, facilitator field notes, and official program documents through thematic and cross-case analyses. The findings indicate that CSR support primarily ensured process continuity and facilitated multi-actor coordination across project stages. By securing implementation continuity and stabilizing governance arrangements, CSR support allowed participatory outputs to be iteratively translated into design development and post-opening evaluation. Post-opening outcomes differed by urban context; nevertheless, both cases showed social value creation through strengthened place attachment, responsibility-oriented use, and inclusive mixed-group play. This study advances a cross-case analytical framework linking urban context, participatory mechanisms, and post-opening social value outcomes, contributing to a more context-sensitive understanding of CSR-supported participatory design processes and their implications for sustainable urban public space development. Full article

The use of fiber-reinforced polymers (FRPs) for strengthening existing reinforced concrete (RC) structures has significantly improved structural rehabilitation processes, providing efficient, durable, and non-invasive solutions. This study presents an advanced deep learning-based predictive model specifically developed to estimate the shear strength of concrete beams strengthened externally with carbon fiber-reinforced polymer (CFRP) composites. Using a comprehensive dataset of 216 experimentally tested CFRP-wrapped concrete beams drawn from existing research, a deep neural network model was rigorously optimized with the Optuna hyperparameter tuning framework and k-fold cross-validation to ensure robustness and generalizability. Model validation involved a thorough comparative analysis against established international design codes (ACI PRC-440.2-17, CSA-S806-12, JSCE) and a parametric study examining the sensitivity of shear strength predictions to key influencing factors, including concrete compressive strength, beam depth, and CFRP wrap thickness. Results demonstrated superior prediction accuracy and reliability of the deep learning approach compared to traditional empirical design models. Consequently, this research significantly enhances the precision of shear strength predictions for CFRP-strengthened concrete beams, supporting the development of more efficient and accurate structural rehabilitation and design guidelines. Full article

Reinforcement corrosion is one of the primary deterioration mechanisms affecting the long-term seismic performance of reinforced concrete (RC) structures. Although the effects of corrosion on individual RC members have been widely investigated, its influence on the cyclic behavior of RC frame systems has received limited attention. This study numerically investigates the seismic response of a single-bay reinforced concrete frame subjected to cyclic lateral loading under various corrosion scenarios. A three-dimensional nonlinear finite element model was developed in ABAQUS, incorporating corrosion-induced effects such as reinforcement cross-sectional loss, degradation of mechanical properties, bond strength deterioration, and concrete softening. The corrosion propagation rate and exposure duration were considered as key parameters, and different corrosion scenarios were comparatively evaluated. The numerical model was validated using an experimentally tested non-corroded reinforced concrete frame subjected to cyclic loading. The results demonstrate that reinforcement corrosion leads to significant degradation in the seismic performance of RC frames. Depending on corrosion severity, reductions of up to approximately 25% in lateral load capacity and up to 27% in both initial stiffness and energy dissipation capacity were observed. The findings further indicate that stiffness- and energy-based performance indicators are more sensitive to corrosion damage than strength-based indicators. The study highlights the importance of explicitly accounting for corrosion effects in the seismic performance assessment of reinforced concrete frame systems and provides a practical numerical framework for evaluating corrosion-induced performance degradation. Full article

High-precision detection of hazardous gases with low refractive indices ranging from 1.000 to 1.100, specifically including methane, carbon monoxide, and sulfur dioxide, is critical for industrial safety, yet conventional sensors often suffer from limited sensitivity and severe thermal cross-sensitivity. This work presents a Magneto-Optical Differential Photonic Crystals Sensor (MO-DPCS) utilizing indium antimonide (InSb) to address these constraints. Employing the Multi-Objective Dragonfly Algorithm (MODA), the system was inversely optimized to maximize magneto-optical polarization splitting while rigorously maintaining an ultra-high transmission efficiency. Crucially, an angular interrogation architecture operating under oblique incidence was established to maximize the magneto-optical non-reciprocity, where the detection was realized by fixing the terahertz source frequency and monitoring the precise angular displacements of the steep spectral edges. A differential detection technique was employed to utilize the non-reciprocal phase changes wherein Transverse Electric (TE) and Transverse Magnetic (TM) modes display contrasting kinematic characteristics in the presence of an external magnetic field. The findings indicate that with an adjusted magnetic field of 0.033 T, the MO-DPCS attains an exceptional differential sensitivity of 30.8°/RIU, much above the 0.8°/RIU seen in the unmagnetized condition. The differential approach efficiently eliminates common-mode thermal noise, minimizing temperature-induced drift to below 0.35° across a 1 K range. The suggested MO-DPCS offers a robust, self-referencing solution for stable and high-sensitivity gas sensing applications with a detection limit of 4.18 × 10⁻⁴ RIU. Full article

Background/Objectives: Mitochondrial DNA (mtDNA) analysis is an essential tool for human identification in contexts such as disaster victim identification (DVI) and missing persons cases, where the remains may be highly degraded or even skeletonised. Traditionally, capillary electrophoresis (CE)-based Sanger sequencing has been the standard method for analysing the mtDNA control region. With the development of massively parallel sequencing (MPS) technologies, mtDNA sequencing using MPS offers advantages over traditional Sanger sequencing, such as increased sensitivity, higher throughput, and less sample consumption. The Ion Chef™ and Ion S5™ XL system from Thermo Fisher Scientific represents one such MPS system. Methods: We conducted an internal validation study evaluating key parameters including (a) concordance, repeatability and reproducibility; (b) potential cross-contamination; (c) sensitivity; (d) effects of library pooling on read depth; and (e) mixture sample analysis. Additionally, to mimic samples typically encountered during forensic investigations, case type samples were also used to evaluate the performance of this workflow. While the entire mitochondrial genome was sequenced in this validation study, considering that the international guidelines for full mtDNA genome analysis and interpretation have yet to be fully updated, our analysis, interpretation and subsequent implementation are limited to the control region only. Results: The results obtained demonstrated the reliability, sensitivity and reproducibility of this MPS workflow. Conclusions: This internal validation study supports the implementation of this workflow in our laboratory for the analysis of forensic casework samples. Full article

Active Channel Width (ACW) provides a robust indicator for tracking river corridor dynamics, yet automated extraction from multisensory imagery remains limited by spatial and temporal variability in spectral conditions. We developed and validated a workflow in Google Earth Engine (GEE) to delineate the active channel using multispectral indices derived from annual composite Landsat and Sentinel-2 imagery. The indices include the Modified Normalized Difference Water Index (MNDWI), Normalized Difference Vegetation Index (NDVI), and Enhanced Vegetation Index (EVI). The 34 km study segment of the Lircay River (Chile) served as a demonstration site undergoing substantial geomorphic change over a 20-year period (2003–2023) that spanned a decade-long mega drought (2010–2023) and two major floods (2006, 2023). Multispectral index thresholds were calibrated using manually digitized active channel polygons for a reference year and validated for five different years within the study period to assess their spatial transferability across reaches and temporal stability under varying hydrologic regimes. Sentinel-2 annual composites with the MNDWI-EVI pairing achieved the highest overall accuracy in estimating ACW (mean Kling-Gupta Efficiency = 0.72; Percent Bias = 12.69 across study reaches). Threshold values were tested at the cross-sectional and reach scales. Using cross-section-specific thresholds enhanced the accuracy of ACW estimation, indicating that threshold performance is strongly conditioned by the local characteristics present in the immediate surroundings of each cross section. These results suggest that spectral threshold selection is sensitive to small scale factors that vary across the river corridor, underscoring the need to explicitly consider local geomorphic and ecological conditions when defining thresholds. This reproducible, open-source workflow links automated channel delineation with cross-section-based morphology and explicitly quantifies uncertainty from spatiotemporal spectral variability. It enables high-resolution, repeatable measurements of river corridor change and underscores the need to consider evolving spectral and vegetation conditions when interpreting remotely sensed geomorphic indicators. Full article

Background: Trigeminal neuralgia (TN) is a severe neuropathic pain disorder primarily diagnosed on clinical grounds, and objective molecular biomarkers that could support diagnosis remain limited. Increasing evidence suggests that inflammation–apoptosis interactions contribute to TN pathophysiology. Methods: In this exploratory prospective case–control study, circulating apoptosis-related microRNAs (miRNAs) were analyzed in 30 patients with idiopathic TN and 20 healthy controls. Plasma miRNA expression levels were quantified using quantitative real-time polymerase chain reaction. Diagnostic performance of individual miRNAs was assessed using receiver operating characteristic (ROC) curve analysis. A multivariable logistic regression model integrating multiple miRNAs was constructed to evaluate combined diagnostic performance, with internal validation performed using five-fold cross-validation. Results: Circulating miRNA expression profiles differed between TN patients and controls. Among individual markers, hsa-miR-183-5p demonstrated the highest diagnostic accuracy (AUC = 0.72), followed by hsa-miR-23a-3p (AUC = 0.65). hsa-miR-223-3p showed reversed directionality (AUC = 0.28), consistent with lower expression in TN and high specificity but low sensitivity at the optimal threshold. The combined miRNA panel achieved an apparent AUC of 0.86, with a mean cross-validated AUC of 0.84 ± 0.12, suggesting improved discrimination over single miRNAs but with variability consistent with the limited sample size. Conclusions: Apoptosis-related circulating miRNAs exhibit distinct expression patterns in idiopathic TN. While individual miRNAs show modest diagnostic performance, integration into a multi-miRNA panel improved discrimination between TN patients and healthy controls in this pilot dataset. These findings support the potential of apoptosis-based miRNA signatures as candidate minimally invasive biomarkers for TN, warranting further validation in larger, independent cohorts, ideally including clinically relevant disease-control facial pain conditions. Full article

Accurate weed mapping in cereal fields requires pixel-level segmentation from unmanned aerial vehicle (UAV) imagery that remains reliable across fields, seasons, and illumination. Existing multispectral pipelines often depend on thresholded vegetation indices, which are brittle under radiometric drift and mixed crop–weed pixels, or on single-stream convolutional neural network (CNN) and Transformer backbones that ingest stacked bands and indices, where radiance cues and normalized index cues interfere and reduce sensitivity to small weed clusters embedded in crop canopy. We propose VISA (Vegetation Index and Spectral Attention), a two-stream segmentation network that decouples these cues and fuses them at native resolution. The radiance stream learns from calibrated five-band reflectance using local residual convolutions, channel recalibration, spatial gating, and skip-connected decoding, which preserve fine textures, row boundaries, and small weed structures that are often weakened after ratio-based index compression. The index stream operates on vegetation-index maps with windowed self-attention to model local structure efficiently, state-space layers to propagate field-scale context without quadratic attention cost, and Slot Attention to form stable region descriptors that improve discrimination of sparse weeds under canopy mixing. To support supervised training and deployment-oriented evaluation, we introduce BAWSeg, a four-year UAV multispectral dataset collected over commercial barley paddocks in Western Australia, providing radiometrically calibrated blue, green, red, red edge, and near-infrared orthomosaics, derived vegetation indices, and dense crop, weed, and other labels with leakage-free block splits. On BAWSeg, VISA achieves 75.6% mean Intersection over Union (mIoU) and 63.5% weed Intersection over Union (IoU) with 22.8 M parameters, outperforming a multispectral SegFormer-B1 baseline by 1.2 mIoU and 1.9 weed IoU. Under cross-plot and cross-year protocols, VISA maintains 71.2% and 69.2% mIoU, respectively. The full BAWSeg benchmark dataset, VISA code, trained model weights, and protocol files will be released upon publication. Full article

As wearable electronic products have been integrated into daily life, flexible pressure sensors, which convert pressure into electrical signals, have become a research focus because of their cross-industry application potential. Despite an increasing number of related studies, the systematic integration of discussions on sensing mechanisms, performance regulation, and multiscenario adaptability remains to be explored. In this paper, core sensing mechanisms such as piezoresistive, capacitive, piezoelectric, and triboelectric mechanisms are systematically reviewed; key performance indicators, including sensitivity, response time, and linearity, are analyzed; construction strategies for diverse substrates and conductive functional materials are explored; and applications in healthcare, human–computer interaction, and electronic skin are elaborated on. The aim of these analyses is to provide practical insights into the development and design of flexible pressure sensors, thus providing a useful reference for advancing these technologies and expanding their cross-domain use. Full article

This study proposes an explainable machine learning framework for estimating the total project cost (TPC) of AI training-data construction, where cost information is difficult to structure due to heterogeneous workflows and quality requirements. Using 386 public AI training-data projects conducted between 2020 and 2022, we derive 24 numerical predictors from standardized final reports and construct three input tracks: a baseline feature set, a principal component analysis (PCA)-enhanced set, and a factor analysis (FA)–enhanced set capturing latent cost structures. Four regression models (Ridge, Random Forest, XGBoost, and LightGBM) are evaluated using nested cross-validation. XGBoost achieves the best overall performance across all three tracks (Baseline, PCA-enhanced, and FA-enhanced). Among them, PCA-enhanced XGBoost attains the highest predictive accuracy (R² = 0.868; RMSE = 1084.9; MAE = 746.9; MAPE = 0.358; pooled out-of-fold), while Baseline XGBoost yields the lowest MAE (731.4; R² = 0.863). To support transparent decision-making, Shapley Additive exPlanations (SHAP)-based attribution and scenario-based sensitivity analyses are conducted. Results show that project scale and process-level unit costs are dominant cost-drivers, while cloud usage, expert participation, and de-identification requirements exhibit secondary effects. The proposed framework provides an interpretable, data-driven approach to cost information management and decision support for data-intensive AI projects. Full article