Search Results (1,195)

Cell confluence and number are critical indicators for assessing cellular growth status, contributing to disease diagnosis and the development of targeted therapies. Accurate and efficient cell segmentation is essential for quantifying these indicators. However, current segmentation methodologies still encounter significant challenges in addressing multi-scale heterogeneity, poorly delineated boundaries under limited annotation, and the inherent trade-off between computational efficiency and segmentation accuracy. We propose an innovative network architecture. First, a preprocessing pipeline combining contrast-limited adaptive histogram equalization (CLAHE) and Gaussian blur is introduced to balance noise suppression and local contrast enhancement. Second, a bidirectional feature pyramid network (BiFPN) is incorporated, leveraging cross-scale feature calibration to enhance multi-scale cell recognition. Third, adaptive kernel convolution (AKConv) is developed to capture the heterogeneous spatial distribution of glioma stem cells (GSCs) through dynamic kernel deformation, improving boundary segmentation while reducing model complexity. Finally, a probability density-guided non-maximum suppression (Soft-NMS) algorithm is proposed to alleviate cell under-detection. Experimental results demonstrate that the model achieves 95.7% mAP50 (box) and 95% mAP50 (mask) on the GSCs dataset with an inference speed of 38 frames per second. Moreover, it simultaneously supports dual-modality output for cell confluence assessment and precise counting, providing a reliable automated tool for tumor microenvironment research. Full article

The compound effects of extreme climate change and intensive urban development have led to more frequent urban inundation, highlighting the urgent need for the fine-scale evaluation of stormwater drainage system performance in high-density urban built-up areas. A typical basin, located in Shenzhen, was selected, and a pipe–river coupled SWMM was developed and calibrated via a genetic algorithm to simulate the storm drainage system. Design storm scenario analyses revealed that regional inundation occurred in the central area of the basin and the enclosed culvert sections of the midstream river, even under a 0.5-year recurrence period, while the downstream open river channels maintained a substantial drainage capacity under a 200-year rainfall event. To systematically identify bottleneck zones, two novel metrics, namely, the node cumulative inundation volume and the conduit cumulative inundation length, were proposed to quantify the local inundation severity and spatial interactions across the drainage network. Two critical bottleneck zones were selected, and strategic improvement via the cross-sectional expansion of pipes and river culverts significantly enhanced the drainage efficiency. This study provides a practical case study and transferable technical framework for integrating hydraulic modeling, spatial analytics, and targeted infrastructure upgrades to enhance the resilience of drainage systems in high-density urban environments, offering an actionable framework for sustainable urban stormwater drainage system management. Full article

Educational admission systems, particularly in developing countries, often suffer from opaque decision processes, unstructured data, and limited analytic insight. This study proposes a novel methodology that integrates structural causal models (SCMs), ontological modeling, and machine learning to uncover and apply interpretable knowledge from an admission database. Using a dataset of 12,043 records from Benue State Polytechnic, Nigeria, we demonstrate this approach as a proof of concept by constructing a domain-specific SCM ontology, validate it using conditional independence testing (CIT), and extract features for predictive modeling. Five classifiers, Logistic Regression, Decision Tree, Random Forest, K-Nearest Neighbors (KNN), and Support Vector Machine (SVM) were evaluated using stratified 10-fold cross-validation. SVM and KNN achieved the highest classification accuracy (92%), with precision and recall scores exceeding 95% and 100%, respectively. Feature importance analysis revealed ‘mode of entry’ and ‘current qualification’ as key causal factors influencing admission decisions. This framework provides a reproducible pipeline that combines semantic representation and empirical validation, offering actionable insights for institutional decision-makers. Comparative benchmarking, ethical considerations, and model calibration are integrated to enhance methodological transparency. Limitations, including reliance on single-institution data, are acknowledged, and directions for generalizability and explainable AI are proposed. Full article

With the diversification of information retrieval methods, news retrieval tasks have gradually evolved towards multimodal retrieval. Existing methods often encounter issues such as inaccurate alignment and unstable feature matching when handling cross-modal data like text and images, limiting retrieval performance. To address this, this paper proposes an innovative multimodal news retrieval method by introducing the Learnable Alignment Module (LAM), which establishes a learnable alignment relationship between text and images to improve the accuracy and stability of cross-modal retrieval. Specifically, the LAM, through trainable label embeddings (TLEs), enables the text encoder to dynamically adjust category information during training, thereby enhancing the alignment capability of text and images in the shared embedding space. Additionally, we propose three key alignment strategies: logits calibration, parameter consistency, and semantic feature matching, to further optimize the model’s multimodal learning ability. Extensive experiments conducted on four public datasets—Visual News, MMED, N24News, and EDIS—demonstrate that the proposed method outperforms existing state-of-the-art approaches in both text and image retrieval tasks. Notably, the method achieves significant improvements in low-recall scenarios (R@1): for text retrieval, R@1 reaches 47.34, 44.94, 16.47, and 19.23, respectively; for image retrieval, R@1 achieves 40.30, 38.49, 9.86, and 17.95, validating the effectiveness and robustness of the proposed method in multimodal news retrieval. Full article

Accurate melt pool geometry prediction is essential for ensuring quality and reliability in Laser Powder Bed Fusion (L-PBF). However, small experimental datasets and limited physical interpretability often restrict the effectiveness of traditional machine learning (ML) models. This study proposes a hybrid framework that integrates an explicit thermal model with ML algorithms to improve prediction under sparse data conditions. The explicit model—calibrated for variable penetration depth and absorptivity—generates synthetic melt pool data, augmenting 36 experimental samples across conduction, transition, and keyhole regimes for 316 L stainless steel. Three ML methods—Multilayer Perceptron (MLP), Random Forest, and XGBoost—are trained using fivefold cross-validation. The hybrid approach significantly improves prediction accuracy, especially in unstable transition regions (D/W ≈ 0.5–1.2), where morphological fluctuations hinder experimental sampling. The best-performing model (MLP) achieves R² > 0.98, with notable reductions in MAE and RMSE. The results highlight the benefit of incorporating physically consistent, nonlinearly distributed synthetic data to enhance generalization and robustness. This physics-augmented learning strategy not only demonstrates scientific novelty by integrating mechanistic modeling into data-driven learning, but also provides a scalable solution for intelligent process optimization, in situ monitoring, and digital twin development in metal additive manufacturing. Full article

Background/Objectives: Educational background is an aspect of socio-economic status, that may be associated with higher caries risk. This study aimed to investigate differences in caries prevalence between different school types for older adolescents in Greifswald, Germany. Methods: Cross-sectional data were collected as part of compulsory dental school examinations between 2020 and 2023. Oral health status was assessed according to WHO criteria by six calibrated examiners and reported as mean D₃MFT (D₃: dentin caries, M: missing, F: filled, SD/±: standard deviation). To compare educational backgrounds, the adolescents were divided into two groups according to their age and type of school (11–15 and 16–18 years old). Results: The study included 5816 adolescents (48.7% females) with a mean D₃MFT of 0.65 (Q1–Q3: 0–1); 73.8% were clinically caries-free, having D₃MFT = 0, confirming the polarization in caries experience with 2.5 ± 2.13 SaC index. The logistic regression model showed a significantly increased Odds Ratio for having caries in relation to age, being male, having plaque or gingivitis (p < 0.005). There were significant differences in caries experience and prevalence between school types, where high schools had the lowest D₃MFT values in both age groups (0.39 ± 1.17 and 0.64 ± 1.49, respectively). The highest D₃MFT values were in schools for special educational needs in younger adolescents (1.12 ± 1.9) and in vocational schools in older adolescents (1.63 ± 2.55). Conclusions: In a low-caries-risk population, there were significant differences in caries experience and prevalence among adolescents in different school types. Prevention strategies should aim to reduce the polarization in caries across different educational backgrounds in late adolescence. Full article

Robotic timber joinery demands integrated, adaptive methods to compensate for the inherent dimensional variability of wood. We introduce a seamless robotic workflow to enhance the measurement accuracy of the Workpiece Coordinate System (WCS). The approach leverages a Zivid 3D camera mounted in an eye-in-hand configuration on a KUKA industrial robot. The proposed algorithm applies a geometric method that strategically crops the point cloud and fits planes to the workpiece surfaces to define a reference frame, calculate the corresponding transformation between coordinate systems, and measure the cross-section of the workpiece. This enables reliable toolpath generation by dynamically updating WCS and effectively accommodating real-world geometric deviations in timber components. The workflow includes camera-to-robot calibration, point cloud acquisition, robust detection of workpiece features, and precise alignment of the WCS. Experimental validation confirms that the proposed method is efficient and improves milling accuracy. By dynamically identifying the workpiece geometry, the system successfully addresses challenges posed by irregular timber shapes, resulting in higher accuracy for timber joints. This method contributes to advanced manufacturing strategies in robotic timber construction and supports the processing of diverse workpiece geometries, with potential applications in civil engineering for building construction through the precise fabrication of structural timber components. Full article

Focusing on the problem of insufficient detection accuracy caused by blurred target boundaries, variable scales, and severe noise interference in side-scan sonar images, this paper proposes a high-precision detection network named ESL-YOLO, which integrates edge perception and adaptive quality assessment. Firstly, an Edge Fusion Module (EFM) is designed, which integrates the Sobel operator into depthwise separable convolution. Through a dual-branch structure, it realizes effective fusion of edge features and spatial features, significantly enhancing the ability to recognize targets with blurred boundaries. Secondly, a Self-Calibrated Dual Attention (SCDA) Module is constructed. By means of feature cross-calibration and multi-scale channel attention fusion mechanisms, it achieves adaptive fusion of shallow details and deep-rooted semantic content, improving the detection accuracy for small-sized targets and targets with elaborate shapes. Finally, a Location Quality Estimator (LQE) is introduced, which quantifies localization quality using the statistical characteristics of bounding box distribution, effectively reducing false detections and missed detections. Experiments on the SIMD dataset show that the mAP@0.5 of ESL-YOLO reaches 84.65%. The precision and recall rate reach 87.67% and 75.63%, respectively. Generalization experiments on additional sonar datasets further validate the effectiveness of the proposed method across different data distributions and target types, providing an effective technical solution for side-scan sonar image target detection. Full article

Accurate assessments of aboveground biomass (AGB) stocks and their changes in extensive Miombo forests are challenging due to the lack of site-specific allometric equations (AEs). Terrestrial Laser Scanning (TLS) is a non-destructive method that enables the calibration of AEs and has recently been validated by the IPCC guidelines for carbon accounting within the REDD+ framework. TLS surveys were carried out in five non-contiguous 1-ha plots in two study sites in the wet Miombo forest of Katanga, in the Democratic Republic Congo. Local wood densities (WD) were determined from wood cores taken from 619 trees on the sites. After a careful checking of Quantitative Structure Models (QSMs) output, the individual volumes of 213 trees derived from TLS data processing were converted to AGB using WD. Four AEs were calibrated using different predictors, and all presented strong performance metrics (e.g., R² ranging from 90 to 93%), low relative bias and relative individual mean error (11.73 to 16.34%). Multivariate analyses performed on plot floristic and structural data showed a strong contrast in terms of composition and structure between sites and between plots within sites. Even though the whole variability of the biome has not been sampled, we were thus able to confirm the transposability of results within the wet Miombo forests through two cross-validation approaches. The AGB predictions obtained with our best AE were also compared with AEs found in the literature. Overall, an underestimation of tree AGB varying from −35.04 to −19.97% was observed when AEs from the literature were used for predicting AGB in the Miombo of Katanga. Full article

Proton therapy has emerged as a highly precise and tissue-sparing radiotherapy technique, capitalizing on the unique energy deposition pattern of protons characterized by the Bragg peak. Ensuring treatment accuracy relies on calibration phantoms, often composed of tissue-equivalent polymeric materials. This study investigates the dosimetric behavior of four commonly used polymers—Parylene, Epoxy, Lexan, and Mylar—by analyzing their linear energy transfer (LET) values and Bragg curve characteristics across various proton energies. Experimental LET data were collected and used to train and evaluate the predictive power for Bragg peak of multiple artificial intelligence models, including kNN, SVR, MLP, RF, LWRF, XGBoost, 1D-CNN, LSTM, and BiLSTM. These algorithms were optimized using 10-fold cross-validation and assessed through statistical error and performance metrics including MAE, RAE, RMSE, RRSE, CC, and R². Results demonstrate that certain AI models, particularly RF and LWRF, accurately (in terms of all evaluation metrics) predict Bragg peaks in Epoxy polymers, reducing the reliance on costly and time-consuming simulations. In terms of CC and R² metrics, the LWRF model demonstrated superior performance, achieving scores of 0.9969 and 0.9938, respectively. However, when evaluated against MAE, RMSE, RAE, and RRSE metrics, the RF model emerged as the top performer, yielding values of 12.3161, 15.8223, 10.3536, and 11.4389, in the same order. Additionally, the SVR model achieved the highest number of statistically significant differences when compared pairwise with the other eight models, showing significance against six of them. The findings support the use of AI as a robust tool for designing reliable calibration phantoms and optimizing proton therapy planning. This integrative approach enhances the synergy between materials science, medical physics, and data-driven modeling in advanced radiotherapy systems. Full article

The global atmospheric concentration of carbon dioxide (CO₂) has exceeded 420 ppm. Adsorption-based carbon capture technologies, offer energy-efficient, sustainable solutions. Relying on classical adsorption models like Langmuir to predict CO₂ uptake presents limitations due to the need for case-specific parameter fitting. To address this, the present study introduces a universal machine learning (ML) framework using multiple algorithms—Generalized Linear Model (GLM), Feed-forward Multilayer Perceptron (DL), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), and Gradient Boosted Trees (GBT)—to reliably predict CO₂ adsorption capacities across diverse zeolite structures and conditions. By compiling over 5700 experimentally measured adsorption data points from 71 independent studies, this approach systematically incorporates critical factors including pore size, Si/Al ratio, cation type, temperature, and pressure. Rigorous Cross-Validation confirmed superior performance of the GBT model (R² = 0.936, RMSE = 0.806 mmol/g), outperforming other ML models and providing comparable performance with classical Langmuir model predictions without separate parameter calibration. Feature importance analysis identified pressure, Si/Al ratio, and cation type as dominant influences on adsorption performance. Overall, this ML-driven methodology demonstrates substantial promise for accelerating material discovery, optimization, and practical deployment of zeolite-based CO₂ capture technologies. Full article

To address issues commonly observed during the inference phase of large language models—such as inconsistent labels, formatting errors, or semantic deviations—a series of targeted strategies has been proposed. First, a relation label refinement strategy based on semantic similarity and syntactic structure has been designed to calibrate the model’s outputs, thereby improving the accuracy and consistency of label prediction. Second, to meet the contextual modeling needs of different types of instance bags, a multi-level contextual augmentation strategy has been constructed. For multi-sentence instance bags, a graph-based retrieval enhancement mechanism is introduced, which integrates intra-bag entity co-occurrence networks with document-level sentence association graphs to strengthen the model’s understanding of cross-sentence semantic relations. For single-sentence instance bags, a semantic expansion strategy based on term frequency-inverse document frequency is employed to retrieve similar sentences. This enriches the training context under the premise of semantic consistency, alleviating the problem of insufficient contextual information. Notably, the proposed multi-granularity framework captures semantic symmetry between entities and relations across different levels of context, which is crucial for accurate and balanced relation understanding. The proposed methodology offers practical advancements for semantic analysis applications, particularly in knowledge graph development. Full article

In order to improve the poor prediction effect of current filling body strength design, a support vector machine (SVM) and Lib Toolbox were used to build an optimal match model or strength index of unclassified tailings filling body. Eight main factors were analyzed and screened as condition attributes, and backfill strength as a decision attribute. Next, we selected 72 groups of training samples and 6 groups of calibration samples. Our model adopts a radial basis function (RBF) as the kernel function and uses a grid search method to optimize parameters; it then tests the combination of optimal parameters by cross-validation. Results show that the mean error of regression prediction and verified predictions made by the SVM match model were 1.01%, which were more accurate than the BP neural network model’s predictions. On the premise that stope stability is ensured, the SVM match model may decrease cement consumption and the cost of backfill more effectively, and improve economic efficiency. Full article

Renewable fuels produced from animal- and plant-based edible oils have emerged as an alternative to oil and natural gas. Burgeoning interest in renewables can be attributed to the rapid depletion of fossil fuels caused by the global energy demand and the environmental advantages of renewables, specifically reduced emissions of greenhouse gases. An important property of the feedstock that is crucial for the conversion of edible oils to renewable fuels is the total acid number (TAN), as even a small increase in TAN for the feedstock can lead to corrosion of the catalyst in the refining process. Currently, the TAN is determined by potentiometric titration, which is time-consuming, expensive, and requires the preparation of reagents. As part of an effort to promote the use of renewable fuels, a partial least squares regression method with orthogonal signal correction to remove spectral information related to the sample background was developed to determine the TAN from the mid-infrared (IR) spectra of the feedstock. Digitally blended mid-IR spectral data were generated to fill in regions of the PLS calibration where there were very few samples. By combining experimental and digitally blended mid-IR spectral data to ensure adequate sample representation in all regions of the spectra–property calibration and better understand the spectra–property relationship through the identification of sample outliers in the original data that can be difficult to detect because of swamping, a PLS regression model for TAN (R² = 0.992, cross-validated root mean square error = 0.468, and bias = 0.0036) has been developed from 118 experimental and digitally blended mid-IR spectra of commercial feedstock. Thus, feedstock whose TAN value is too high for refining can be flagged using the proposed mid-IR method, which is faster and easier to use than the current titrimetric method. Full article

This study addresses the challenges of manual measurement bias and low robustness in detecting small, occluded targets in complex marine environments during real-time onboard sorting of Portunus trituberculatus. We propose YOLO-DFAM, an enhanced YOLOv11n-based model that replaces the global average pooling in the Focal Modulation module with a spatial–channel dual-attention mechanism and incorporates the ASF-YOLO cross-scale fusion strategy to improve feature representation across varying target sizes. These enhancements significantly boost detection, achieving an mAP@50 of 98.0% and precision of 94.6%, outperforming RetinaNet-CSL and Rotated Faster R-CNN by up to 6.3% while maintaining real-time inference at 180.3 FPS with only 7.2 GFLOPs. Unlike prior static-scene approaches, our unified framework integrates attention-guided detection, scale-adaptive tracking, and lightweight weight estimation for dynamic marine conditions. A ByteTrack-based tracking module with dynamic scale calibration, EMA filtering, and optical flow compensation ensures stable multi-frame tracking. Additionally, a region-specific allometric weight estimation model (R² = 0.9856) reduces dimensional errors by 85.7% and maintains prediction errors below 4.7% using only 12 spline-interpolated calibration sets. YOLO-DFAM provides an accurate, efficient solution for intelligent onboard fishery monitoring. Full article