Search Results (5,228)

Cotton is a globally strategic crop that plays a crucial role in sustaining national economies and livelihoods. To address the challenges of accurate cotton field extraction in the complex planting environments of Xinjiang’s Alaer reclamation area, a cotton field identification model was developed that integrates multi-source satellite remote sensing data with machine learning methods. Using imagery from Sentinel-2, GF-1, and Landsat 8, we performed feature fusion using principal component, Gram–Schmidt (GS), and neural network techniques. Analyses of spectral, vegetation, and texture features revealed that the GS-fused blue bands of Sentinel-2 and Landsat 8 exhibited optimal performance, with a mean value of 16,725, a standard deviation of 2290, and an information entropy of 8.55. These metrics improved by 10,529, 168, and 0.28, respectively, compared with the original Landsat 8 data. In comparative classification experiments, the endmember-based random forest classifier (RFC) achieved the best traditional classification performance, with a kappa value of 0.963 and an overall accuracy (OA) of 97.22% based on 250 samples, resulting in a cotton-field extraction error of 38.58 km². By enhancing the deep learning model, we proposed a U-Net architecture that incorporated a Convolutional Block Attention Module and Atrous Spatial Pyramid Pooling. Using the GS-fused blue band data, the model achieved significantly improved accuracy, with a kappa coefficient of 0.988 and an OA of 98.56%. This advancement reduced the area estimation error to 25.42 km², representing a 34.1% decrease compared with that of the RFC. Based on the optimal model, we constructed a digital map of continuous cotton cropping from 2021 to 2023, which revealed a consistent decline in cotton acreage within the reclaimed areas. This finding underscores the effectiveness of crop rotation policies in mitigating the adverse effects of large-scale monoculture practices. This study confirms that the synergistic integration of multi-source satellite feature fusion and deep learning significantly improves crop identification accuracy, providing reliable technical support for agricultural policy formulation and sustainable farmland management. Full article

Predicting driver injury severity is critical for enhancing road safety, but it is complicated because fatal accidents inherently create class imbalance within datasets. This study conducts a comparative analysis of machine-learning (ML) and deep-learning (DL) models for multi-class driver injury severity prediction using a comprehensive dataset of 107,195 traffic accidents from the Adana, Mersin, and Antalya provinces in Turkey (2018–2023). To address the significant imbalance between fatal, injury, and non-injury classes, the hybrid SMOTE-ENN algorithm was employed for data balancing. Subsequently, feature selection techniques, including Relief-F, Extra Trees, and Recursive Feature Elimination (RFE), were utilized to identify the most influential predictors. Various ML models (K-Nearest Neighbors (KNN), XGBoost, Random Forest) and DL architectures (Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Recurrent Neural Network (RNN)) were developed and rigorously evaluated. The findings demonstrate that traditional ML models, particularly KNN (0.95 accuracy, 0.95 F1-macro) and XGBoost (0.92 accuracy, 0.92 F1-macro), significantly outperformed DL models. The SMOTE-ENN technique proved effective in managing class imbalance, and RFE identified a critical 25-feature subset including driver fault, speed limit, and road conditions. This research highlights the efficacy of well-preprocessed ML approaches for tabular crash data, offering valuable insights for developing robust predictive tools to improve traffic safety outcomes. Full article

This study investigates the certification outcomes of intermediate-level ICs in The National Fire Service Academy in South Korea through a comprehensive quantitative analysis of their evaluated competencies. Using assessment data from 141 candidates collected from 2022 to 2024, we examine how scores on six higher-order competencies—comprising 35 sub-competencies—influence pass or fail results. Descriptive statistics, correlation analysis, logistic regression (a statistical model for binary outcomes), random forest modeling (an ensemble decision-tree machine-learning method), and principal component analysis (PCA; a dimensionality reduction technique) were applied to identify significant predictors of certification success. Visualization techniques, including heatmaps, box plots, and importance bar charts, were used to illustrate performance gaps between successful and unsuccessful candidates. Results indicate that competencies related to decision-making under pressure and crisis leadership most strongly correlate with positive outcomes. Furthermore, unsupervised clustering analysis (a data-driven grouping method) revealed distinctive performance patterns among candidates. These findings suggest that current evaluation frameworks effectively differentiate command readiness but also highlight specific skill domains that may require enhanced instructional focus. The study offers practical implications for fire training academies, policymakers, and certification bodies, particularly in refining curriculum design, competency benchmarks, and evaluation criteria to improve fireground leadership training and assessment standards. Full article

Given the complexity of near-surface soil moisture retrieval, a single machine learning algorithm often struggles to capture the intricate relationships among multiple features, resulting in limited generalization and robustness. To address this issue, this study proposes a multi-satellite fusion GNSS-IR soil moisture retrieval method based on heterogeneous ensemble machine learning models. Specifically, two heterogeneous ensemble learning strategies (Bagging and Stacking) are combined with three base learners, Back Propagation Neural Network (BPNN), Random Forest (RF), and Support Vector Machine (SVM), to construct eight ensemble GNSS-IR soil moisture retrieval models. The models are validated using data from GNSS stations P039, P041, and P043 within the Plate Boundary Observatory (PBO) network. Their retrieval performance is compared against that of individual machine learning models and a deep learning model (Multilayer Perceptron, MLP), enabling an optimized selection of algorithms and model architectures. Results show that the Stacking-based models significantly outperform those based on Bagging in terms of retrieval accuracy. Among them, the Stacking (BPNN-RF-SVM) model achieves the highest performance across all three stations, with R of 0.903, 0.904, and 0.917, respectively. These represent improvements of at least 2.2%, 2.8%, and 2.1% over the best-performing base models. Therefore, the Stacking (BPNN-RF-SVM) model is identified as the optimal retrieval model. This work aims to contribute to the development of high-accuracy, real-time monitoring methods for near-surface soil moisture. Full article

Floods are among the most devastating natural disasters; predicting their depth and extent remains a global challenge. Machine Learning (ML) models have demonstrated improved accuracy over traditional probabilistic flood mapping approaches. While previous studies have developed ML-based models for specific local regions, this study aims to establish a methodology for estimating flood depth on a global scale using ML algorithms and freely available datasets—a challenging yet critical task. To support model generalization, 45 catchments from diverse geographic regions were selected based on elevation, land use, land cover, and soil type variations. The datasets were meticulously preprocessed, ensuring normality, eliminating outliers, and scaling. These preprocessed data were then split into subgroups: 75% for training and 25% for testing, with six additional unseen catchments from the USA reserved for validation. A sensitivity analysis was performed across several ML models (ANN, CNN, RNN, LSTM, Random Forest, XGBoost), leading to the selection of the Random Forest (RF) algorithm for both flood inundation classification and flood depth regression models. Three regression models were assessed for flood depth prediction. The pixel-based regression model achieved an R² of 91% for training and 69% for testing. Introducing a pixel clustering regression model improved the testing R² to 75%, with an overall validation (for unseen catchments) R² of 64%. The catchment-based clustering regression model yielded the most robust performance, with an R² of 83% for testing and 82% for validation. The developed ML model demonstrates breakthrough computational efficiency, generating complete flood depth predictions in just 6 min—a 225× speed improvement (90–95% time reduction) over conventional HEC-RAS 6.3 simulations. This rapid processing enables the practical implementation of flood early warning systems. Despite the dramatic speed gains, the solution maintains high predictive accuracy, evidenced by statistically robust 95% confidence intervals and strong spatial agreement with HEC-RAS benchmark maps. These findings highlight the critical role of the spatial variability of dependencies in enhancing model accuracy, representing a meaningful approach forward in scalable modeling frameworks with potential for global generalization of flood depth. Full article

Blood glucose monitoring is crucial for the daily management of diabetic patients. In this study, we developed a differential absorbance and photoplethysmography (PPG)-based non-invasive blood glucose measurement system (NIBGMS) using visible–near-infrared (Vis-NIR) light. Three light-emitting diodes (LEDs) (625 nm, 850 nm, and 940 nm) and three photodetectors (PDs) with different source–detector separation distances were used to detect the differential absorbance of tissues at different depths and PPG signals of the index finger. A spatiotemporal multimodal fused long short-term memory (STMF-LSTM) model was developed to improve the prediction accuracy of blood glucose levels by multimodal fusion of optical spatial information (differential absorbance and PPG signals) and glucose temporal information. The validity of the NIBGMS was preliminarily verified using multilayer perceptron (MLP), support vector regression (SVR), random forest regression (RFR), and extreme gradient boosting (XG Boost) models on datasets collected from 15 non-diabetic subjects and 3 type-2 diabetic subjects, with a total of 805 samples. Additionally, a continuous dataset consisting 272 samples from four non-diabetic subjects was used to validate the developed STMF-LSTM model. The results demonstrate that the STMF-LSTM model indicated improved prediction performance with a root mean square error (RMSE) of 0.811 mmol/L and a percentage of 100% for Parkes error grid analysis (EGA) Zone A and B in 8-fold cross validation. Therefore, the developed NIBGMS and STMF-LSTM model show potential in practical non-invasive blood glucose monitoring. Full article

Aboveground biomass (AGB) in grasslands is a vital metric for assessing ecosystem functioning and health. Accurate and efficient AGB estimation is essential for the scientific management and sustainable use of grassland resources. However, achieving low-cost, high-efficiency AGB estimation via remote sensing remains a key challenge. This study integrates Sentinel-1 and Sentinel-2 imagery to derive 38 multi-source feature variables, including backscatter coefficients, texture, spectral reflectance, vegetation indices, and topographic factors. These features are combined with AGB data from 112 field plots in the Three Parallel Rivers area. Feature selection was performed using Pearson correlation, Random Forest (RF), and SHAP values to identify optimal variable sets. Genetic Programming (GP) was then applied for nonlinear optimization of the selected features. Three machine learning models—RF, GBRT, and KNN—were used to estimate AGB and generate spatial distribution maps. The results revealed notable differences in model accuracy, with RF performing best overall, outperforming GBRT and KNN. After GP optimization, all models showed improved performance, with the RF model based on RF-selected features achieving the highest accuracy (R² = 0.90, RMSE = 0.31 t/ha, MAE = 0.23 t/ha), improving R² by 0.03 and reducing RMSE and MAE by 0.05 and 0.03 t/ha, respectively. Spatial mapping showed the AGB ranged from 0.41 to 3.59 t/ha, with a mean of 1.39 t/ha, closely aligned with the actual distribution characteristics. This study demonstrates that the RF model, combined with multi-source features and GP optimization, provides an effective approach to grassland AGB estimation and supports ecological monitoring in complex areas. Full article

Assessing ecological quality in mining areas is critical for environmental protection and sustainable resource management. However, most previous studies concentrate on large-scale analysis, overlooking fine-scale assessment in mining areas. To address this issue, this study proposed a novel analysis framework for mining areas by integrating high-resolution Landsat data, the Remote Sensing Ecological Index (RSEI), and the Random Forest regression method. Based on the framework, four decades of spatiotemporal dynamics and drivers of ecological quality were revealed in Youjiang River Valley. Results showed that from 1986 to 2024, ecological quality in Youjiang River Valley exhibited a fluctuating upward trend (slope = 0.004/year), with notable improvement concentrated in the most recent decade. Spatially, areas with a significant increasing trend in RSEI (48.71%) were mainly located in natural vegetation regions, whereas areas with a significant decreasing trend (9.11%) were concentrated in impervious surfaces and croplands in northern and central regions. Driver analysis indicates that anthropogenic factors played a crucial role in ecological quality changes. Specifically, land use intensity, precipitation, and sunshine duration were main determinants. These findings offer a comprehensive understanding of ecological quality evolution in subtropical karst mining areas and provide crucial insights for conservation and restoration efforts in Youjiang River Valley. Full article

Accurate wave forecasting under typhoon conditions is essential for coastal safety in the Pearl River Estuary. This study explores the use of Random Forest (RF) and Long Short-Term Memory (LSTM) models to predict significant wave heights, using SWAN-simulated data from 87 historical typhoon events. Ten representative typhoons were reserved for independent testing. Results show that the LSTM model outperforms RF in 3 h forecasts, achieving a lower mean RMSE and higher R², particularly in capturing wave peaks under highly dynamic conditions. For 6 h forecasts, both models exhibit decreased accuracy, with RF performing slightly better in stable scenarios, while LSTM remains more responsive in complex wave evolution. Generalization tests at three nearby stations demonstrate that both models, especially LSTM, retain strong predictive skill beyond the training location. These findings highlight the potential of combining numerical wave models with machine learning for short-term, data-driven wave forecasting in typhoon-prone and observation-sparse regions. The study also points to future improvements through integration of wind field predictors, model updating strategies, and ensemble meteorological data. Full article

Annealing of steel coils in bell-type furnaces is a critical process in steel production, requiring precise temperature control to ensure desired mechanical properties and microstructure. However, direct measurement of inner coil temperatures is impractical in industrial conditions, necessitating model-based estimation. This study presents a comparative analysis of physics-based and machine learning (ML) approaches for predicting internal temperatures during annealing. A finite difference method (FDM) was developed as a physics-based model and validated against experimental data from both laboratory and industrial annealing cycles. Furthermore, several ML models, including support vector regression (SVR), neural networks (NN), multivariate adaptive regression splines (MARS), k-nearest neighbors (k-NN), and random forests (RFs), were trained on surface temperature measurements to predict inner temperatures. The results demonstrate that the MARS, k-NN, and RF models achieved high prediction accuracy with performance index (PI) values below 1.0 on unseen data, demonstrating excellent generalization capabilities. In contrast, SVR with polynomial kernels and NN exhibited poor performance in specific configurations, highlighting their sensitivity to overfitting and data variability. The findings suggest that combining physics-based models with data-driven techniques offers a robust framework for soft-sensing in annealing operations, enabling improved process monitoring and control. Full article

Early Autism Spectrum Disorder (ASD) detection is important for early intervention. This study investigates the potential of eye-tracking (ET) data combined with machine learning (ML) models to classify ASD and Typically Developed (TD) children. Using a publicly available dataset, five ML models were evaluated: Support Vector Machine (SVM), Random Forest, Convolutional Neural Network (CNN), Artificial Neural Network (ANN), and Random Forest improved with Convolutional Filters (ConvRF). The models were trained and tested using a set of evaluation metrics, including accuracy, precision, recall, F1-score, and ROC Area Under the Curve (AUC). Among these, the ConvRF model attained superior performance, achieving a recall of 90% and an AUC of 88%, indicating its robustness in identifying ASD children. These results highlight the model’s effectiveness in ensuring high sensitivity, which is critical for early ASD detection. This study shows the promise of combining ML and eye-tracking technology as accessible non-invasive tools for enhancing early ASD detection, resulting in timely and personalized interventions. Limitations and recommendations for future research are also included. Full article

Purpose: This study aimed to develop and validate a new machine learning (ML)-based screening tool for a two-step prediction of the need for and type of nutritional therapy (enteral, parenteral, or combined) using Nutrition Risk Screening 2002 (NRS-2002) and other demographic parameters from the Optimal Nutrition Care for All (ONCA) national cohort data. Methods: This multicenter retrospective cohort study included 191,028 patients, with data on age, gender, body mass index (BMI), NRS-2002 score, presence of cancer, and hospital unit type. In the first step, classification models estimated whether patients required nutritional therapy, while the second step predicted the type of therapy. The dataset was divided into 60% training, 20% validation, and 20% test sets. Random Forest (RF), Artificial Neural Network (ANN), deep learning (DL), Elastic Net (EN), and Naive Bayes (NB) algorithms were used for classification. Performance was evaluated using AUC, accuracy, balanced accuracy, MCC, sensitivity, specificity, PPV, NPV, and F₁-score. Results: Of the patients, 54.6% were male, 9.2% had cancer, and 49.9% were hospitalized in internal medicine units. According to NRS-2002, 11.6% were at risk of malnutrition (≥3 points). The DL algorithm performed best in both classification steps. The top three variables for determining the need for nutritional therapy were severe illness, reduced dietary intake in the last week, and mild impaired nutritional status (AUC = 0.933). For determining the type of nutritional therapy, the most important variables were severe illness, severely impaired nutritional status, and ICU admission (AUC = 0.741). Adding gender, cancer status, and ward type to NRS-2002 improved AUC by 0.6% and 3.27% for steps 1 and 2, respectively. Conclusions: Incorporating gender, cancer status, and ward type into the widely used and validated NRS-2002 led to the development of a new scale that accurately classifies nutritional therapy type. This ML-enhanced model has the potential to be integrated into clinical workflows as a decision support system to guide nutritional therapy, although further external validation with larger multinational cohorts is needed. Full article

A miniaturised bionic electronic nose system was developed to solve the problems of expensive equipment and long response time for soil pesticide residue detection. The structure of the bionic electronic nasal cavity is designed based on the spatial structure and olfactory principle of the sturgeon nasal cavity. Through experimental study, the structure of the nasal cavity of the sturgeon was extracted and analyzed. The 3D model of the bionic electronic nasal cavity was constructed and verified by Computational Fluid Dynamics (CFD) simulation. The results show that the gas flow distribution in the bionic chamber is more uniform than that in the ordinary chamber. The airflow velocity near the sensor in the bionic chamber is lower than in the ordinary chamber. The eddy current intensity near the bionic chamber sensor is 2.29 times that of the ordinary chamber, further increasing the contact intensity between odor molecules and the sensor surface and shortening the response time. The 10-fold cross-validation method of K-Nearest Neighbor (K-NN), Random Forest (RF) and Support Vector Machine (SVM) was used to compare the recognition performance of the bionic electronic nasal cavity with that of the ordinary electronic nasal cavity. The results showed that, when the bionic electronic nose detection system identified the concentration of pesticide residues in soil, the recognition rate of the above three recognition algorithms reached 97.3%, significantly higher than that of the comparison chamber. The bionic chamber electronic nose system can improve the detection performance of electronic noses and has a good application prospect in soil pesticide residue detection. Full article

This study investigates the impact of countries’ fundamental risk on the speed of adjustment (SOA) towards firms’ target capital structures. Using a dataset comprising 17,747 non-financial firms from 44 countries, this study finds that a reduction in country-specific fundamental risk significantly increases a firm’s rate of leverage adjustment. More specifically, we observe that a one standard deviation reduction in fundamental risk results in a substantial 12.79% increase in SOA for book leverage and a 4.81% increase for market leverage. The study also finds evidence of the influence of individual dimensions of fundamental risk on SOA. It implies that improved operational efficiency, high foreign accessibility, enhanced corporate transparency, and increased political stability expedite the pace of leverage adjustment within firms. Robustness checks using a machine learning random forest estimator predicted leverage targets to corroborate these findings. The results highlight the critical role of institutional quality in reducing financing frictions and promoting more efficient corporate capital adjustments. These insights have profound implications for policymakers, emphasising the need to strengthen institutional and regulatory frameworks to enhance capital market integrity and reduce friction, which could ultimately create value for the firm stakeholders. Full article

Bathymetry, the measurement of water depth and underwater terrain, is vital for scientific, commercial, and environmental applications. Traditional methods like shipborne echosounders are costly and inefficient in shallow waters due to limited spatial coverage and accessibility. Emerging technologies such as satellite imagery, drones, and spaceborne LiDAR offer cost-effective and efficient alternatives. This research explores integrating multi-sensor datasets to enhance bathymetric mapping in coastal and inland waters by leveraging each sensor’s strengths. The goal is to improve spatial coverage, resolution, and accuracy over traditional methods using data fusion and machine learning. Gülbahçe Bay in İzmir, Turkey, serves as the study area. Bathymetric modeling uses Sentinel-2, Göktürk-1, and aerial imagery with varying resolutions and sensor characteristics. Model calibration evaluates independent and integrated use of single-beam echosounder (SBE) and satellite-based LiDAR (ICESat-2) during training. After preprocessing, Random Forest and Extreme Gradient Boosting algorithms are applied for bathymetric inference. Results are assessed using accuracy metrics and IHO CATZOC standards, achieving A1 level for 0–10 m, A2/B for 0–15 m, and C level for 0–20 m depth intervals. Full article