Search Results (4,243)

The further development of scalable fabrication for perovskite solar cells has been considerably constrained by strong process variability and the lack of a reliable real-time predictive mechanism during the thin-film formation process. Existing machine learning-based methods are incapable of capturing the inherent multi-stage kinetic characteristics and uncertainties of the perovskite crystallization process, as they rely on deterministic point prediction models and flatten time-series signals into static features, which necessitates more advanced modeling strategies. To address these challenges, an in situ process monitoring and predictive modeling framework based on a lightweight probabilistic Transformer is proposed for the scalable preparation of perovskite thin films. The strategically designed inputs, consisting of time-resolved photoluminescence (PL) and diffuse reflectance imaging signals acquired during the vacuum quenching process, enable the model to directly learn the conditional probability distribution of the final device performance metrics. Rather than producing a single predicted value, this method enables the explicit quantification of prediction uncertainty, providing statistical support for uncertainty-aware process assessment. Leveraging its advantages over feed-forward neural networks and traditional tree-based machine learning methods, the proposed Transformer architecture effectively captures the staged and non-stationary kinetic features of thin-film formation. Consequently, it exhibits higher robustness and superior uncertainty calibration capability during the early-stage prediction phase. The results demonstrate that the probabilistic Transformer-based modeling paradigm provides a viable pathway toward uncertainty-aware, data-driven process evaluation in perovskite manufacturing. This framework extends its application beyond perovskite photovoltaic device fabrication, providing a generalizable modeling strategy for real-time predictive assessment in the preparation of other complex materials governed by irreversible stochastic dynamics. Full article

Existing hyperspectral denoising networks typically rely on large amounts of high-quality paired noisy–clean images for training, which are often unavailable. Moreover, the noise distribution in real hyperspectral images (HSIs) is complex and variable, making it challenging for existing networks to handle noise distributions not present in the training dataset, resulting in poor generalization. To address these issues, this paper proposes an unsupervised Hyperspectral image Denoising approach exploiting the spectral learning preference of neural networks with an adaptive early stopping strategy (termed HyDePre). Inspired by the Deep Image Prior, which reveals that neural networks tend to capture natural image structures before fitting noise, we observe that deep neural networks exhibit a similar learning preference in the spectral domain. Specifically, as training progresses, the network first fits smooth spectral feature curves and only later adapts to Gaussian noise and complex impulse noise. This observation provides an opportunity to use an early stopping strategy, allowing the network to fit only the clean spectral signals and thus achieve denoising. Our method does not require clean images for training, but instead optimizes network parameters to automatically learn prior spectral information from a single noisy image, modeling the intrinsic structure of the input data to uncover its underlying patterns.However, finding the optimal stopping point is challenging without access to clean images as sources of prior information. To tackle this challenge, we introduce an adaptive early stopping strategy based on the average spectral maximum variation of the reconstructed image, effectively preventing overfitting. The experimental results demonstrate that HyDePre outperforms existing methods in terms of both visual quality and quantitative metrics. Full article

The spatial variability of soil physicochemical properties significantly influences irrigation efficiency, nutrient availability, and the long-term sustainability of irrigated agriculture in semi-arid regions. This study aimed to quantify and model the spatial distribution of soil properties in a semi-arid irrigation district in central Mexico (Irrigation District 001 “Pabellón de Arteaga”, Aguascalientes), providing spatially explicit information for differential irrigation and fertilization management. Ninety-seven crop and four natural sampling sites were established under a stratified random design at two soil depths (0–30 and 30–60 cm). Geostatistical and machine learning models (Ordinary Kriging, OK; Generalized Additive Models, GAM; and Random Forest, RF) were applied to predict spatial patterns, and their performance was evaluated using statistical metrics. The findings reveal high spatial and vertical variability, with most properties (such as organic matter, total nitrogen, and texture) showing significant stratification with depth. In contrast, others (pH and electrical conductivity, EC) remained remarkably homogeneous vertically. Correlation patterns were identified, highlighting the negative influence of alkaline pH (≈8.0) on the availability of micronutrients (Fe²⁺ and Mn²⁺) and the positive association between EC and soluble cations (Ca²⁺, K⁺, and Na⁺). Moran’s Index confirmed significant spatial autocorrelation for most properties, reducing the effective sample size by 30–70%. The comparative evaluation of predictive models demonstrated the superiority of RF over OK and GAMs for predicting chemical properties, thanks to its ability to capture nonlinear relationships and complex interactions. However, the overall predictive performance was moderate, reflecting the multifactorial complexity of the edaphic system. This study lays the foundation for the development of an accessible, low-cost Decision Support System by providing a robust methodological framework for spatial soil characterization and contributing to more sustainable, resilient agriculture, where decision-making is based on quantitative data and predictive models. Full article

Sea-level rise (SLR) and regional precipitation pattern change cause island subsurface freshwater, typically shaped like a thin lens, to be at higher risk of contamination from seawater intrusion (SWI). Installing a cutoff wall is considered a feasible strategy for protecting coastal fresh groundwater from SWI. However, the performance of the cutoff wall in managing freshwater lens (FWL) development and mitigating SWI into island aquifers under SLR and aquifer recharge (RCH) fluctuations remains inadequately quantified. This study investigates how water table elevation (WTE), FWL depth, thickness, and SWI extent, measured by aquifer salt mass and freshwater volume, in an island aquifer equipped with cutoff walls, respond to SLR and RCH fluctuations. It focuses on a two-dimensional, variable-density island groundwater simulation model based on hydrogeological conditions of San Salvador Island, Bahamas. The results demonstrate that RCH critically influences cutoff wall effectiveness for FWL development and SWI mitigation, with higher RCH amplifying gains in WTE, FWL metrics, freshwater storage, and aquifer salt removal, but this influence diminishes with wall depth increasing. SLR elevates WTE in a stable manner associated with its magnitude but negligibly affects the cutoff wall performance in FWL enhancement and SWI mitigation. Under simultaneous SLR and RCH fluctuations, SLR can offset the WTE reduction caused by reduced RCH, but the joint effects of SLR and RCH on FWL metrics, freshwater storage and aquifer salt removal align with their individual impacts. Moreover, cutoff walls are more efficient in low-RCH settings, yielding greater relative improvements in FWL development and SWI mitigation per unit wall depth increase. Full article

Color multi-threshold image segmentation is a non-convex, gradient-free global optimization problem. The number of decision variables increases with the number of thresholds, leading to a rapid expansion of the search space and increased computational complexity. To address this problem, this paper proposes a Multi-Mechanism Artificial Lemming Algorithm (MALA). When applied to color multi-threshold image segmentation, the original Artificial Lemming Algorithm (ALA) suffers from an imbalance between exploration and exploitation, excessive reliance on the current best solution, and rigid boundary handling, which may lead to premature convergence and suboptimal threshold selection. MALA integrates three lightweight yet structurally enhancement mechanisms to enhance the stability of the exploration–exploitation process, population-level guidance, and boundary-handling behavior. To verify its general optimization capability, MALA is evaluated on the CEC2017 benchmark suite, where it shows competitive convergence behavior and improved objective values compared with ALA and representative baseline algorithms. Furthermore, segmentation experiments on six benchmark images using Otsu’s criterion show that MALA attains competitive fitness values and generally higher PSNR, SSIM, and FSIM metrics. These results suggest that MALA can serve as a general optimization method with applicability to color multi-threshold image segmentation. Full article

Membrane fouling remains a major obstacle to the performance of the reverse osmosis (RO) desalination processes. Artificial intelligence (AI) is now a promising approach for the reliable modeling of these complex systems. This study evaluates three modeling techniques—multiple linear regression (MLR), artificial neural networks (ANNs), and support vector regression (SVR)—for predicting transmembrane pressure (TMP) at the Boujdour desalination plant, based on five input parameters: temperature, turbidity, pH, conductivity, and feedflow. The analysis is based on an original dataset of 195 daily measurements, and due to the absence of timestamps, the study focuses on state-to-TMP prediction rather than chronological forecasting, with no temporal generalization claimed. Approximately 2000 augmented training samples generated using a conservative SMOGN approach were used for model development, while performance evaluation relied exclusively on 39 independent real test observations. Two modeling strategies were adopted: (i) a minimalist approach based on significant variables identified by an ordinary least squares (OLS) model (pH and conductivity), and (ii) a multivariate approach integrating all parameters to capture non-linear interactions. A rigorous validation framework was put in place to avoid information leakage and ensure the robustness and generalizability of the models. Performance was evaluated using R², RMSE, and MAE metrics, supplemented by robustness and significance analyses including bootstrap confidence intervals, paired statistical comparisons, and interpretability analyses based on permutation importance, partial dependence plots (PDPs), and individual conditional expectation (ICE) curves. The results indicate that the SVR model achieves the best average predictive accuracy among the tested models, albeit with moderate explanatory power. Full article

Introduction: Acute ischemic stroke demands interventions that restore perfusion and protect neurons within a narrow therapeutic window. We propose a unified theranostic platform that couples adaptive imaging, topology-aware decision-making, and immediate neuroprotective and micro-dosimetric intervention. Methods: The platform integrates three components. First, a topology-preserving MR–PET engine employs adaptive Poisson-disc sampling, partial Fourier constraints, and structured Hankel low-rank priors in a closed loop. Persistent-homology metrics quantify vascular graph uncertainty and guide subsequent k-space and PET projections, reducing acquisition time while preserving collateral topology. Second, immediate post-reperfusion delivery of glycolic acid attenuates glutamate-driven calcium influx and stabilizes mitochondrial function. Third, trace doses of sol–gel-derived, neutron-activated ⁹⁰Y₂O₃ microspheres provide sharply confined beta irradiation for micro-scale metabolic modulation. Results: In a porcine stroke model replicating the human recanalization workflow, the imaging engine maintained vascular Betti-number invariants within three percent of fully sampled reference scans while reducing acquisition time by nearly half. Glycolic acid reduced glutamate-induced intracellular calcium rise by approximately sixty percent in vitro and decreased infarct volume by thirty-eight percent in vivo. Micro-dosimetry confirmed a mean perivascular beta dose of twenty-eight grays, and histology demonstrated a forty-two percent increase in NeuN-positive neuronal survival compared with standard recanalization. Conclusions: These results demonstrate that intelligent compressed-sensing MR–PET, targeted micro-radioembolization, and glycolic acid neuroprotection can act synergistically to bridge diagnostic imaging and immediate intervention. By coupling imaging, decision-making, and therapy in a closed-loop manner and elevating topological fidelity from a reconstruction byproduct to a control variable, the proposed platform reframes MR–PET from passive diagnostics into an active, decision-driven theranostic system and establishes a foundation for future human trials. Full article

Background/Objectives: Aggressive variants of prostate cancer pose significant diagnostic and prognostic challenges due to atypical imaging appearances, variable prostate-specific antigen behavior, and distinct molecular features. Conventional imaging may underestimate their biological aggressiveness. This review aimed to synthesize current evidence on imaging characteristics, biomarker dynamics, tumor localization, histology, and radiomic features of aggressive prostate cancer variants, and to evaluate the potential role of radiomics in early recognition and risk stratification. Methods: A structured narrative review was performed of studies reporting imaging, clinical, and molecular features of aggressive prostate cancer variants. Imaging modalities included multiparametric magnetic resonance imaging, positron emission tomography with prostate-specific membrane antigen or fluorodeoxyglucose, bone scintigraphy, and transrectal ultrasound. Data on prostate-specific antigen levels and kinetics, intraprostatic tumor location, tumor size, metastatic patterns, and molecular alterations were extracted. Evidence for rare entities such as basaloid and primary squamous carcinomas was derived from published case reports and series, while selected variants were complemented by institutional imaging and histopathologic observations. Results: Neuroendocrine and small cell carcinomas frequently showed low prostate-specific antigen levels, high fluorodeoxyglucose uptake, low prostate-specific membrane antigen expression, and central or transitional zone involvement with large tumor size at diagnosis. Ductal adenocarcinoma demonstrated marked diffusion restriction and elevated prostate-specific antigen, whereas basal cell carcinoma often appeared inconspicuous on conventional imaging. Radiomic analysis consistently captured tumor heterogeneity and spatial complexity beyond standard qualitative metrics. Conclusions: Aggressive prostate cancer variants represent a diagnostic blind spot in routine imaging. Radiomics offers complementary quantitative information that may improve early detection, subtype differentiation, and risk stratification when integrated into multimodal imaging workflows. Further prospective and radiogenomic studies are warranted to validate these findings. Full article

This study examined weekly internal load and neuromuscular performance in elite junior female boxers over 10 weeks. Internal load was quantified using session rating of perceived exertion (sRPE), from which weekly monotony and strain were derived. Neuromuscular performance was assessed weekly using wall-sit endurance and a repetitive jump test. Twenty elite junior female boxers (Mean ± SD: 18.9 ± 1.2) were monitored during regular training without experimental manipulation. Weekly sRPE-derived training load, monotony, and strain showed statistically significant week-to-week fluctuations (p < 0.001). Neuromuscular performance improved in week 2, declined during weeks 3–5, and partially recovered in week 6. The findings demonstrated consistent temporal alignment between internal-load indices and week-to-week neuromuscular performance changes within an observational monitoring framework. Inter-individual variability was observed across athletes. Overall, sRPE-derived indices reflected training stress patterns and were aligned with neuromuscular performance changes in elite female boxers, supporting their use for contextual monitoring of weekly training responses. These findings support the practical integration of internal-load and performance monitoring in elite female combat-sport settings. Future research incorporating boxing-specific external-load metrics, physiological markers, and longer monitoring periods may further refine individualized load-management strategies. Full article

This study examines the application of ensemble machine learning methods for identifying and flagging potentially risky transactions in military public procurement in Ukraine, a sector characterized by elevated financial and security sensitivity and limited capacity for comprehensive ex post control. Using an integrated dataset of procurement procedures conducted between 2021 and 2025, enriched with 56 financial, economic, and behavioral indicators of suppliers, the study develops and compares standard logistic and LASSO-penalized regression as econometric benchmarks, Random Forest, XGBoost, XGBoost with SMOTE balancing, and CatBoost classification models. The target variable is defined on the basis of officially detected violations identified through state monitoring. Model performance is evaluated using standard binary classification metrics, with particular emphasis on recall. Model uncertainty and predictive robustness are addressed through partial dependence analysis, temporal stability assessment, and out-of-sample residual diagnostics. The results indicate that the CatBoost model demonstrates the most balanced performance across evaluation measures. Feature importance analysis identifies expected contract value, procurement method, CPV code, and suppliers’ financial capacity as significant determinants of procurement-related risk. The findings provide empirical evidence on the usefulness of risk-oriented machine learning tools in supporting earlier detection and monitoring of irregularities in military procurement. Full article

Background and Purpose: We present a pilot benchmark dataset of 378 preclinical histological samples for evaluating large language model (LLM) performance on multi-dimensional classification tasks. This dataset addresses the lack of standardized benchmarks for assessing LLMs in preclinical histopathology, encompassing species identification (mouse, rabbit, rat), organ recognition, staining methods, and preparation techniques. Methods: We evaluated the LLMs GPT-4.1, GPT-4o-mini, and Llama 3.2 on 378 histological samples across four classification dimensions: species identification (mouse, rabbit, rat), organ recognition (kidney, liver, prostate, spleen), staining method classification (H&E, Elastica van Gieson, collagen, iron, IHC-elastin, MOVAT’s pentachrome), and preparation type determination (frozen vs. paraffin-embedded). Performance was assessed using sensitivity and specificity metrics with confusion matrix analysis. Results: Model performance varied substantially across tasks and exhibited strong sensitivity to class imbalance. For preparation type classification, GPT-4.1 achieved the most balanced performance (50% frozen sensitivity, 85.7% paraffin sensitivity), while Llama 3.2 failed to recognize paraffin samples (0% sensitivity). In species classification, Llama 3.2 was the only model capable of identifying all three species (rabbit: 75% sensitivity, rat: 85.7% sensitivity) despite poor mouse recognition (0.3% sensitivity). GPT-4.1 achieved higher mouse sensitivity within this dataset (70.4% sensitivity) but failed with minority species. For staining classification, Llama 3.2 demonstrated highest overall performance, achieving >88% sensitivity for most staining types, while GPT-4o-mini showed perfect H&E recognition (100% sensitivity). Conclusions: Current LLMs demonstrate variable performance for histological classification with substantial sensitivity to class imbalance. While not suitable for standalone diagnostic use, they may serve as useful screening tools in research settings with appropriate human oversight. Full article

Effective management zone delineation is key to implementing site-specific strategies that address spatiotemporal heterogeneity in agriculture. Although time-series remote sensing offers a dynamic perspective, most current methods lack the framework to integrate it with soil properties, thereby hindering accurate characterization of crop growth variability. This study bridges the gap by developing a spatiotemporal framework that synthesizes remote sensing-derived phenology and soil attributes for daylily management zoning. Through a sequential approach—phenological metric extraction, SNIC-based segmentation, and STSF classification—we produce refined phenological time-series stacks. These outputs are designed to elucidate the drivers of field heterogeneity and directly inform precision management strategies. Compared to pixel-based and SNIC-based random forest, the STSF–SNIC framework increased spatial overlap rates by 5.4–8.0% (reaching 88.6%), despite comparable overall accuracy and kappa coefficients (OA/kappa: 92–94%). Geographical detector analysis identified village boundaries, soil type, total nitrogen, and organic carbon as key drivers of spatial patterns. A spatial generalized fuzzy c-means model, incorporating crop growth dynamics and soil gradients, reduced management zone fragmentation by 27.8% compared to conventional methods, with spatial autocorrelation analysis confirming enhanced spatial consistency (Moran’s I = 0.600 vs. 0.433, p < 0.001). In conclusion, by integrating time-series remote sensing phenology with soil attribute analysis within a spatially constrained clustering scheme, this study (1) provides a novel method for delineating coherent management zones, (2) reveals key drivers of crop growth heterogeneity, and (3) demonstrates a transferable pathway for translating satellite data into precision management actions. It thereby exemplifies the value of applied remote sensing in addressing practical challenges in sustainable agriculture. Full article

Load forecasting has become a key tool, especially for distribution system operators, to ensure optimal grid management and control. In recent years, attention has shifted toward probabilistic load forecasting (PLF), as it can model forecast uncertainty. Because electricity demand is strongly influenced by time-dependent factors such as seasonal patterns and daily habits, non-parametric PLF methods are particularly suitable because they make no assumptions about the distribution of variables. This study focuses on quantile regression (QR), a widely studied non-parametric PLF technique that models forecast uncertainty by only assuming a linear dependency among variables. It is applied every hour to forecast the daily consumption of three large public buildings—an elderly healthcare center, a biomedical research facility, and a polyclinic—with different demand variability profiles. Forecasts are carried out using real-world consumption data and evaluated considering both univariate and multivariate approaches. The performance of both QR approaches is rigorously evaluated against that of two persistence-based methods through standard evaluation metrics. For the univariate case, two aggregation levels are considered: single buildings and aggregation of buildings. The results confirm the effectiveness of both uQR and mQR, which consistently outperform persistence-based benchmarks. In terms of the pinball loss (PL) function, the QR approaches exhibit values ranging from 1% to 1.8% across all case studies. Both approaches demonstrate reliable and sharp prediction intervals (PIs); for example, for the PI(10–90) using the uQR, the PI coverage probability (PICP) ranges from 0.78 to 0.89 and the PI normalized average width (PINAW) from 0.09 to 0.26. Overall, uQR achieves lower PL, whereas mQR yields slightly better PICP and PINAW results for the building characterized by an irregular and unpredictable consumption profile. Full article

To address the limited detection accuracy of casting defects—including pores, inclusions, and looseness—in digital radiography (DR) images, which stems from their small scale, high morphological variability, and interference from complex background textures, we propose MTS-YOLOv11: an edge–discrepancy collaborative defect detection framework tailored for casting DR imagery. Built upon YOLOv11, MTS-YOLOv11 incorporates three key innovations: (1) a Multi-Scale Edge Information Enhancement System (MSEES), integrated into the C3K2 module of the backbone network, to strengthen discriminative feature extraction for minute defects; (2) a TripletAttention mechanism embedded in high-level backbone stages to jointly calibrate channel–spatial dependencies and suppress texture-induced spurious responses under complex backgrounds; (3) a Scale-Discrepancy-Aware Gated Fusion (SDAGFusion) module positioned immediately before the detection head, enabling scale-discrepancy-aware gated fusion of multi-scale features, emphasizing defect regions while suppressing background interference. Experimental results show that on the casting DR dataset, MTS-YOLOv11 achieves mAP@0.5 = 96.5% and mAP@0.5:0.95 = 68.5%—improvements of 1.3 and 1.2 percentage points over the baseline YOLOv11—across all three defect categories. Moreover, on the same platform, MTS-YOLOv11 achieves an inference speed of 359.07 FPS, compared with 346.86 FPS for the baseline. Meanwhile, the model has 2.72M parameters and 7.8G FLOPs. These results indicate a consistent improvement in detection accuracy while maintaining a practical balance between precision and computational efficiency. Moreover, cross-dataset generalization tests on newly acquired industrial DR data show that MTS-YOLOv11 consistently outperforms the baseline across evaluation metrics, suggesting improved robustness to unseen imaging conditions and supporting its potential for real-world foundry inspection. Full article

Precipitation data is a primary influencing factor in hydrological modeling. However, the sparse distribution of surface hydrological stations and the lack of available data constrain the development of watershed models and the management and allocation of water resources. This study employs statistical metrics to evaluate discrepancies between observed precipitation data and multi-source precipitation products (CMADS, ERA5, GPM IMERG, and TRMM). It identifies highly sensitive parameters in the SWAT model established using observed hydrological data and quantitatively assesses runoff simulation performance in the Manas River Basin using the coefficient of determination and Nash index. Results indicate the following: (1) CMADS and TRMM exhibit good overall trends within a year. For multi-year monthly precipitation averages, CMADS performs best at monthly and seasonal scales (CC > 0.7), while TRMM performs best at the annual scale (CC > 0.75). (2) At spatial scales, IMERG shows the poorest performance compared to observed stations, and ERA5 exhibits anomalous points. (3) TRMM achieved the best monthly runoff simulation performance in the Manas River Basin, with an average NSE value of 0.73, average R² of 0.80, and average KGE of 0.80. This study provides valuable scientific support for hydrological forecasting in data-scarce regions with complex topography and similar climate variability. Full article