Search Results (2,528)

For high-intensity focused ultrasound (HIFU) thermal ablation to be safe and effective, real-time, high-precision monitoring of tissue coagulative necrosis is essential. However, decoding the ultra-long, non-stationary radio frequency (RF) echoes produced during tissue phase transitions usually results in severe feature aliasing and high computational costs for conventional deep learning models. This paper suggests a highly interpretable, asymmetric classification framework that combines a lightweight Vision Transformer (ViT) with multifractal detrended fluctuation analysis (MFDFA) in order to overcome this obstacle. In terms of methodology, ViT may independently capture cross-scale thermodynamic dependencies without local inductive biases by using MFDFA as a physical prior to compress 1D RF sequences into dense 2D fractal tensors. This method greatly improved the algorithmic recognition of the extremely elusive “partially degenerated” transient state, achieving a strong 96.5% classification accuracy when validated on an ex vivo pig liver dataset. Furthermore, by firmly attaching its classifications to the macroscopic statistical correlates of the acoustic scattering process, the model achieves great decision transparency instead of functioning as an opaque black box. Importantly, this MFDFA-ViT architecture provides an ideal accuracy–latency trade-off with only 3.45M parameters and an end-to-end inference latency of 20.6 ms. This offers a real-time, intelligent monitoring paradigm that is highly deployable and specifically designed for upcoming clinical HIFU applications. Full article

Hyperspectral (HS) sensors offer high spectral discrimination but generally limited spatial coverage, whereas multispectral (MS) sensors provide broad coverage with lower spectral detail. HS-MS complementary mapping aims to reduce coverage gaps in HS sensors by training classifiers within HS-MS overlap regions and applying them to surrounding MS-only areas. However, comprehensive comparisons across classifiers remain scarce in this complementary mapping setting. Furthermore, lightweight pixel-wise classifiers often produce spatially inconsistent predictions, whereas spatial–spectral deep learning models demand substantial computational resources. In this study, we propose a flexible HS-MS complementary mapping framework by systematically evaluating 13 classifiers across mineral and land-use/land-cover (LULC) mapping tasks and introducing class-adaptive uncertainty revocation (CAUR), a lightweight, classifier-independent post-processing module. Performance was evaluated via spatial holdout cross-validation within the HS-MS overlap area. When computational resources are sufficient, 3D convolutional neural networks (3D-CNN) achieve the highest accuracy. Conversely, lightweight models such as random forest (RF) and k-nearest neighbors (kNN) provide computationally efficient and robust baselines. Applying CAUR consistently improves spatial consistency and overall classification accuracy without model retraining, with the largest improvements observed for coarser-resolution HS reference data. These findings provide practical design guidelines for constructing efficient HS-MS complementary mapping pipelines tailored to application demands and sensor characteristics. Full article

To address the challenges of strong spectral mixing, pronounced feature redundancy, and limited generalization capability of traditional methods in complex heterogeneous river water body extraction, this study uses Sentinel-2 Level-2A multispectral imagery obtained from the Copernicus Data Space Ecosystem to investigate water body extraction in the middle reaches of the Heilongjiang River and proposes an optimized Extreme Gradient Boosting (XGBoost) model based on generalized features and balanced sampling, implemented using XGBoost version 2.1.4. The proposed method is compared with the Normalized Difference Water Index (NDWI)-based threshold segmentation method, U-Net deep learning segmentation model and the Random Forest (RF) classifier. The results show that the proposed XGBoost model achieves the best classification performance, with a Precision of 0.9821 and an F1-score of 0.9695 for water body extraction, significantly outperforming the benchmark methods and demonstrating stronger robustness and adaptability. Both ablation experiments and Shapley Additive exPlanations (SHAP)-based interpretability analysis consistently identified the Sentinel-2 red-edge band (B5) as the most influential feature. Its removal resulted in a substantial decline in classification performance, while B5 exhibited the highest SHAP importance. The water-vapour band (B9) and the Automated Water Extraction Index with Shadow (AWEIsh) were the next most influential features, while water indices, band-ratio features, and atmospheric-related bands mainly played auxiliary roles. This study provides an effective technical framework for high-precision water body extraction and river dynamic monitoring in complex fluvial environments. Full article

Mapping debris flow susceptibility is essential for disaster risk reduction in mountainous regions. This study proposes a spatially enhanced modeling framework to evaluate these mass-wasting hazards. The framework integrates conventional topographic parameters with localized micro-geomorphic indicators to assess susceptibility in Bomi County, Tibet. Extracted geomorphological variables include the Topographic Position Index (TPI), surface roughness, local relief, and flow accumulation. We applied multi-scale moving window operations to explicitly quantify spatial heterogeneity and sediment connectivity. This approach systematically evaluates the influence of localized landscape variations on debris flow kinematics. The Random Forest (RF) algorithm was utilized to construct the spatial susceptibility model, and the area under the receiver operating characteristic curve (AUC) quantified its predictive capability. The proposed framework achieves a high predictive accuracy with an AUC of 0.9434. Feature importance analysis demonstrates that micro-geomorphic variables contribute significantly to the predictions; specifically, TPI and local relief primarily drive the overall classification performance. Furthermore, the multi-scale spatial enrichment enables the model to effectively capture complex physical interactions across the terrain. Ultimately, this methodology provides an objective, spatially explicit tool for mass-wasting susceptibility mapping, offering reliable data to support disaster prevention and risk management in the Tibetan Plateau and similar highly incised alpine ecosystems. Full article

Accurate mapping of commercial durian plantations is essential for agricultural inventory, precision agriculture, and sustainable land management but remains challenging because of spectral similarity with other evergreen vegetation. This study developed a multisource remote sensing approach integrating multi-temporal Sentinel-2 imagery, Sentinel-1 Synthetic Aperture Radar (SAR), DEM-derived terrain variables, recursive feature selection, and machine learning for binary durian plantation classification in Sisaket Province, Thailand. A total of 2430 field reference samples were used to evaluate Random Forest (RF), Support Vector Machine (SVM), and Classification and Regression Tree (CART). RF achieved the strongest overall performance using the integrated Sentinel-2, Sentinel-1, and DEM dataset, with an Overall Accuracy of 90.38%, an F1-score of 90.59%, a Kappa coefficient of 0.81, and an AUC of 0.991. Recursive feature selection reduced the predictor set from 53 to 12 variables (77.4%) while retaining high classification performance. The retained 12-predictor subset comprised nine Sentinel-2-derived variables (five seasonal EVI and four NDBI variables), two Sentinel-1 VV/VH ratio variables, and one DEM-derived elevation variable, corresponding to 75.0%, 16.7%, and 8.3% of the retained predictors, respectively. The proposed approach provides an accurate and computationally efficient method for durian plantation mapping within the study area, with broader applicability requiring further validation. Full article

Phytophthora root and crown rot causes irreversible canopy decline in citrus before ground-level symptoms appear, yet UAV-based detection still relies almost exclusively on RGB imagery and, in citrus, on leaf-level classification rather than field-scale localisation. This work addresses that gap with a crown-level object-detection pipeline for Phytophthora in orange orchards, in three contributions. First, a mean-initialised patch embedding expansion that adapts a pretrained detection transformer to N-channel input while preserving its DINOv2 representations and activation magnitude, applicable to any ViT-based detector. Second, a two-stage protocol that screens seven vegetation indices (GNDVI, SAVI, EVI, GRVI, ExG, CARI, MCARI) as fourth channels over three seeds; the screening does not resolve them, and GNDVI is retained because both bands of its ratio respond to root dysfunction-induced chlorophyll degradation and both come from a single sensor. Third, a matched-modality comparison isolating the contribution of the architecture from that of the spectral channel. On 1147 georeferenced RGB–multispectral pairs with 5560 expert-annotated instances, RF-DETR + GNDVI attains a test mAP_50:95 of $0.590\pm 0.008$ and mAP₅₀ of $0.873\pm 0.005$ over three seeds, exceeding a YOLO26n baseline on identical four-channel input by 9.6 and 8.4 percentage points at half the resolution, the architecture proving the decisive component and supporting georeferenced crown-level alerts. Full article

Most existing driver behavior classification technologies have focused on highways and other primary road infrastructures, despite secondary roads accounting for a disproportionately large number of traffic fatalities worldwide. Compared with highways, secondary roads present greater variability in road geometry, infrastructure quality, and traffic interactions, making driver behavior recognition considerably more challenging. This paper investigates the classification of driver behavior on secondary roads using machine learning techniques. Naturalistic driving data obtained from the publicly available UAH-DriveSet dataset were analyzed using two complementary feature groups describing lane detection and traffic status. Four supervised machine learning algorithms, namely, Logistic Regression (LR), gradient boosting (GB), Random Forest (RF), and Artificial Neural Networks (ANNs), were evaluated to classify driving behavior into three categories: Normal, Aggressive, and Drowsy. The extracted features were first analyzed through statistical profiling and exploratory feature analysis before training and evaluating the classification models. The experimental results show that gradient boosting consistently achieved the highest performance for both feature groups, attaining an overall classification accuracy of approximately 67% while providing balanced precision, recall, and F1-scores across all behavioral classes. Logistic regression and random forest produced competitive but lower performance, whereas the Artificial Neural Network yielded the lowest classification accuracy. The obtained results demonstrate the effectiveness of ensemble learning methods for driver behavior recognition under secondary-road conditions and highlight their potential for integration into intelligent driver monitoring and Advanced Driver Assistance Systems (ADASs). By enabling earlier identification of aggressive and drowsy driving behaviors on secondary roads, the proposed approach could support timely driver warnings and safety interventions, potentially reducing accident risk. Furthermore, the findings provide a benchmark for future machine learning models designed for real-world secondary-road environments, where driving conditions are more variable and challenging than on highways. Full article

Development of new agricultural fungicides requires reliable computational tools to screen candidate molecules before investing time, money, and effort in experimental assays. In this study, curated datasets of fungicidal and non-fungicidal compounds were employed in the construction of possible predictive models using diverse machine-learning algorithms and molecular descriptors. The best performance was obtained using balanced datasets and topological descriptors. Two classification models based on different kinds of negative class instances were selected, achieving results of accuracy, sensitivity, and specificity for the test set of 0.806, 0.778, and 0.828 for the first model (RF_BF_14), and 0.937, 0.943, and 0.933 for the second model (RF_BF_17). Differences in performance were consistent with the chemical nature of the negative class. Both models showed excellent applicability domain coverage (>99.6%). Predictions of fungicidal likeness were applied to a curated database of about 1.2 million molecules from the ChEMBL database by using an in-house developed Python3 tool: FLiP-Z (Fungicide Likeness Predictor by Zimeck). Roughly 22% of molecules were predicted by positive consensus as fungicidal candidates. A subsequent screening based on Acute Oral Toxicity reduced the set to 295 molecules with low-toxicity and positive fungicide-likeness predictions. Proprietary rights for FLiP-Z are held by Zimeck C.L.; the tool is available by permission. Full article

The growing problem of air pollution and its direct impact on human health have created an urgent need for reliable, intelligent, and machine learning (ML)-enabled gas-sensing technologies. Among various sensing platforms, metal oxide gas sensors (MO-GSs) have emerged as promising candidates owing to their low cost, high sensitivity, and scalability. However, their practical application is limited by poor selectivity, cross-sensitivity, sensor drift, and high operating temperatures. Recent advances in ML have provided effective strategies to overcome these limitations through data-driven optimization of sensing performance. This review summarizes recent progress in ML-assisted MO-GSs, covering sensor array design, feature engineering, and classification algorithms, including support vector machines (SVMs), random forests (RFs), and deep neural networks (DNNs). In addition, key data-processing techniques such as preprocessing, dimensionality reduction, and hybrid learning approaches are critically discussed. The application of ML-enabled MO-GSs in medical diagnostics, environmental monitoring, industrial safety, and food quality assessment is also reviewed. Despite significant progress, challenges including limited dataset availability, sensor drift, and poor model generalization remain. Future research should focus on developing adaptive, energy-efficient, and IoT-enabled smart sensing systems. The integration of machine learning with metal oxide gas sensors represents a significant step toward intelligent, next-generation, high-performance gas-sensing technologies. Full article

The dynamics of aboveground biomass (AGB) in rubber plantations (RPs) provide an important basis for evaluating carbon stocks and environmental adaptability in tropical plantations. However, continuous monitoring of AGB of RPs at the regional scale is lacking, and its nonlinear responses to hydrothermal conditions remain insufficiently understood. This study focused on RPs in western Hainan Island (WHI), including Danzhou, Baisha, Lingao, and Chengmai, and integrated field plot data with multi-source remote sensing datasets. A framework for mapping RPs combining rule-based constraints and phenology-based random forest (RF) classification was developed. After key variable screening, extreme gradient boosting (XGBoost), Shapley additive explanations (SHAP), and generalized additive model (GAM) were used for AGB estimation and identification of climatic responses. The results showed that mapping of RPs achieved an overall accuracy of 92.89% and a Kappa coefficient of 0.854. The XGBoost-derived estimates showed that AGB of RPs in the study area increased by approximately 1.43 × 10⁶ Mg from 2017 to 2025, with growth areas mainly concentrated in the Danzhou–Baisha and western Chengmai. AGB exhibited significant nonlinear responses to climatic factors. Specifically, the effect of precipitation (PRE) shifted to negative after approximately 1945 mm yr⁻¹, whereas annual mean maximum temperature (TMAX) shifted to a positive effect after about 29.72 °C, although this effect gradually weakened as temperature continued to rise. Combinations such as PRE × annual mean temperature (PRE × TMP), PRE × TMAX, and PRE × potential evapotranspiration (PRE × PET) exhibited significant nonlinear interactions, indicating that the direction and magnitude of the effect of PRE shifted with changes in temperature and PET levels. These findings link the spatiotemporal changes in AGB of RPs in WHI with hydrothermal thresholds and their interacting effects, deepening our understanding of the climatic response characteristics of AGB in RPs in this region. They also provide a scientific basis for RP monitoring, carbon stock assessment, and climate-adaptive management in WHI. Full article

Certified seed purity is a prerequisite for sustaining the agronomic performance and commercial value of Khao Dawk Mali 105 (KDML-105), Thailand’s premium aromatic rice; however, conventional inspection methods are destructive, labour-intensive, and poorly suited to high-throughput operations. This study developed a non-destructive purity inspection system integrating five-band MSI acquired using a MicaSense RedEdge-MX sensor with three machine learning classifiers: Extra Trees (ET), Random Forest (RF), and Support Vector Machine (SVM). The system was evaluated systematically across four illumination levels (360.90–13,188.46 lx) to discriminate KDML-105 from three morphologically similar contaminating varieties: Chainat-1, RD-6, and RD-15. Two-way ANOVA confirmed that classifier type was the dominant performance determinant (η² = 0.847). Under optimal illumination (L4, 13,188.46 lx), ET achieved the highest accuracy (88.8%), recall (86.6%), and F1-score (88.4%) with a training time of 0.098 s. External validation confirmed model generalisability: KDML-105 seeds were identified with 90.6% accuracy and 95.6% recall, while Chainat-1 and RD-6 yielded accuracies of 85.1% and 90.6%, respectively; RD-15 remained challenging (62.9%; 67.4%) owing to spectral proximity to KDML-105. These findings establish MSI combined with ET classification as a viable, cost-effective approach for automated seed purity screening in certified rice production. Full article

To investigate whether spectral information from the backside of soybean leaves can help to improve the accuracy of disease identification models, based on spectral data from the front and back surfaces of leaves, as well as fused spectral data derived from them, a classification model was established using machine learning algorithms in this study. The study first used a spectral acquisition system to obtain spectral information from the front and back surfaces of the leaves, respectively, and calculated the averages of the two types of data to generate fused spectral data from both surfaces. For the three types of spectral data mentioned above, the following four preprocessing methods were applied: Savitaky–Golay smoothing (SG), multiplicative scatter correction (MSC), standard normal variate (SNV), and second-order derivative (2nd Der). At the same time, five modeling methods—support vector machines (SVM), partial least squares discriminant analysis (PLS-DA), convolutional neural network (CNN), random forest (RF), and back propagation neural network (BPNN)—were introduced to establish classification models of soybean leaf diseases, with the aim of selecting the optimal model that achieves the highest identification accuracy in each type of data. The results of the study indicate the following: Among the classification models based on spectral data from the front surface of the leaves, the BPNN model constructed after SG smoothing preprocessing (SG-BPNN) performed the best, achieving recognition accuracy of 88.89% on the testing set. Among the models based on spectral data from the back surface of the leaves, the MSC-PLS-DA model was identified as the optimal model, achieving an accuracy of 98.61% on the testing set. Among the models based on fused spectral data from both front and back surfaces, the MSC-PLS-DA model also demonstrated optimal performance, achieving a classification accuracy of 100% on the testing set. Its accuracy was 11.11% higher than that of the best model using only front-surface data and 1.39% higher than that of the best model using only back-surface data, which verified the effectiveness of fused spectral information from the front and back surfaces of the leaves in improving the accuracy of the soybean disease classification models. Therefore, this study provides a new approach and theoretical basis for the non-invasive, efficient, and precise detection of soybean diseases, and offers valuable reference for promoting the practical application of spectroscopy in the diagnosis of agricultural diseases. Full article

Accurately assessing the fatigue levels of air traffic controllers is crucial for reducing human errors in ATC and ensuring the safe and orderly operation of the civil aviation transportation system. In remote tower scenarios, air traffic controllers’ work environments and task interaction modes differ significantly from those in traditional towers, and traditional fatigue detection approaches relying on physiological monitoring can cause intrusive disruptions to ATC operations. To overcome these limitations, this study proposes a scenario-based, non-invasive assessment framework for accurate and low-interference fatigue recognition. Taking three key scenario elements (traffic load, main operation screen brightness, and core work area illuminance) as the basis for measuring fatigue, the framework bridges the mapping from scenario elements to fatigue status, thereby enabling the transition of assessment inputs from physiological metrics to scenario features. In this mapping, fatigue labels are determined using a fusion strategy. Specifically, objective fatigue labels are derived from optimal wave features extracted from electroencephalogram data using one-way analysis of variance (OW-ANOVA), which are then fused with subjective labels based on the Karolinska Sleepiness Scale (KSS) self-reports through fuzzy C-means (FCM) clustering. Ultimately, a hybrid intelligent classification model integrating the Gannet optimization algorithm (GOA) and random forest (RF) is constructed to perform the primary assessment task. The experimental results indicate that the proposed framework achieves a recognition accuracy of 95.00%, outperforming six other commonly used classification or combination models. Ablation experiments and robustness tests validate the effectiveness of the fused labeling strategy and GOA modules, as well as the method’s excellent stability in resisting data noise. Furthermore, feature interpretability analysis reveals the quantitative influence of the three core fatigue drivers used. The research findings confirm the feasibility of non-invasive fatigue assessment for remote tower controllers leveraging scenario-based elements, which can offer intelligent decision support for controller shift scheduling, visual environment optimization, and targeted safety interventions. Full article

Background/Objectives: Paget’s disease of bone (PDB) and osteoporosis are chronic metabolic bone disorders characterized by disrupted bone remodeling and increased skeletal fragility; however, the underlying mechanism of PDB remains poorly understood. Artificial intelligence (AI) has emerged as a transformative tool in medical imaging, enabling automated feature extraction and improved diagnostic classification of skeletal disorders. This study aimed to investigate whether AI could distinguish subtle variations in bone morphology between PDB and osteoporotic bone. Methods: C57BL/6 mice femurs were scanned by µCT: 16 optineurin-knockout mice with a PDB phenotype (20–26 months), 25 genetically matched wild-type Aging mice (20–26 months), and 15 ovariectomized (OVX) mice with osteoporotic bone phenotype (4.5 months). Two AI algorithms were investigated: a machine learning (ML) model using 22 µCT-derived features trained with a Random Forest (RF) classifier, and a deep learning (DL) model using a 3D convolutional neural network (3D-CNN) trained on raw µCT images. Leave-one-out cross-validation was applied to evaluate model robustness. Results: Significant differences in volumetric, density, and morphological parameters of cortical and trabecular bone were observed between PDB and osteoporosis (p < 0.05). The RF algorithm achieved 90% accuracy in distinguishing PDB from both aging- and OVX-induced osteoporosis and provided feature importance rankings that improved model interpretability. The 3D-CNN achieved classification accuracies of 70% for PDB vs. OVX and 68% for PDB vs. aging, demonstrating the feasibility of an image-based DL approach. Conclusions: AI-based RF and 3D-CNN models demonstrated promising performance in differentiating PDB from osteoporosis using µCT-derived bone features. These findings suggest potential for using AI to assist with analyzing skeletal images in the diagnosis of metabolic bone disorders. Full article

Objectives: Sleep staging from polysomnographic (PSG) recordings is clinically critical for diagnosing sleep-related disorders, yet manual scoring by certified technologists remains time-consuming, costly, and subject to inter-rater variability. Methods: This study presents an automated, explainable, and multimodal framework for five-class sleep stage classification using simultaneously acquired electroencephalography (EEG), electrooculography (EOG), and electromyography (EMG) signals. A total of 1946 annotated 30 s epochs from 30 healthy adult recording sessions (Sleep-EDF Expanded and Sleep Cassette subset) were processed through a 37-dimensional multimodal feature extraction pipeline encompassing temporal amplitude statistics, frequency-domain spectral band powers, nonlinear entropy and complexity measures, and Daubechies-4 discrete wavelet transform (DWT) energy coefficients. Four classical machine learning classifiers -Random Forest (RF), Support Vector Machine with radial basis function kernel (SVM-RBF), Gradient Boosting (GB), and K-Nearest Neighbours (KNN, k = 7) were benchmarked under stratified five-fold cross-validation. Results: SVM-RBF achieved the highest macro-averaged F1-score of 0.7322 (Cohen’s kappa 0.6784, overall accuracy 75.18%). N3 deep slow-wave sleep achieved the highest per-class F1 of 0.879, while N1 light sleep was the most challenging (F1 = 0.668). SHapley Additive exPlanations (SHAP) and RF mean decrease in Gini impurity (MDGI) analysis jointly identified EMG root mean square amplitude (MDGI = 0.0805), gamma band power (0.0784), and permutation entropy (0.0434) as the three most discriminative features. As a novel methodological contribution, sixteen categories of signal sculpting visualisations were developed, translating abstract multivariate features into clinically interpretable graphical representations. Conclusions: The proposed framework achieves substantial kappa agreement approaching the lower bound of expert inter-rater reliability (0.76–0.82) while providing full model transparency, with direct implications for wearable sleep monitoring device design. Full article