Search Results (5,351)

Hyperspectral image super-resolution (HSI SR) focuses on enhancing the spatial resolution of HSIs while preserving their inherent spectral information. Existing single-image HSI SR methods still suffer from blurred spatial edges and spectral distortion. Although numerous spatial–spectral enhancement networks can enhance spatial–spectral feature extraction, they often lead to a cumbersome network architecture. To address these issues, we propose a Spatial–Spectral Distillation Mamba Network, called SSDM-Net, for HSI SR, which contains a main reconstruction branch and two training-only auxiliary branches for spatial and spectral knowledge distillation. Specifically, the spatial and spectral auxiliary branches, which are utilized exclusively during training, provide edge-aware guidance and capture spectral correlations, respectively. During training, the spatial–spectral knowledge is transferred to the main branch. During inference, the auxiliary branches are removed, improving reconstruction quality without extra computational burden. In the main branch, a Mamba-based spatial–spectral global enhancement module processes spatial and latent inter-channel sequences using selective scanning whose cost is linear in the processed sequence lengths when the feature dimensions are fixed. In addition, a dynamic loss weighting strategy is developed to balance reconstruction, distillation, and auxiliary losses during optimization. Comprehensive experiments conducted on the CAVE and Houston datasets with three scale factors demonstrate that SSDM-Net produces more accurate reconstruction results than existing representative HSI SR methods. Cross-dataset experiments on the Harvard dataset further suggest that the method can maintain competitive reconstruction performance under the evaluated cross-dataset settings. Full article

Although asymbiotic orchid seed germination was first achieved in vitro in 1922, the prediction of germination requirements under in vitro conditions still remains complicated. To address this, we developed a machine learning framework to classify the ex situ asymbiotic germination potential of wild orchids into four discrete groups: Low (0–30%), Mid (31–50%), High (51–80%), and Max (81–100%). Models were trained on a dataset of 203 species, utilizing seed morphometrics—specifically, the embryo-to-testa (E:S) length ratio—alongside core ecological traits (subfamily, growth habit, habitat, and climate zone), as well as chemical scarification duration as a proxy of seed permeability. Validation leveraged novel germination and trait data from 26 taxa from Greece (17) and Armenia (9), published here for the first time. To mitigate class imbalance and prevent algorithmic bias toward highly germinating species, we applied inverse frequency weighting during training. Iterative testing of six algorithms revealed that the “Step 4” feature matrix (excluding climate zone and pretreatment duration) yielded the optimal predictive balance. K-Nearest Neighbor (KNN) and Support Vector Machine (SVM) emerged as the superior models, achieving overall accuracies of 44.4% and 61.1%, respectively, with both achieving 100% accuracy for low-germinating species. Finally, we synthesized a novel database compiling new seed morphometrics from Armenia (17 taxa), Greece (52 taxa), and the data from the literature (479 taxa). After filtering previously utilized species, we generated a prediction pool of 361 orchid taxa. Applying our Step 5 KNN and SVM models to forecast their germination behavior revealed distinct variations linked to ecological profiles. This high-accuracy framework, particularly for low-germinability groups, offers a powerful screening tool for ex situ conservation planning. The final trained models are compiled in the publicly available R (v. 4.6.0) package OrchidGermClass. Full article

Seasonal variation in SARS-CoV-2 transmissibility is difficult to estimate directly from year-round epidemic data because interventions, behavior, reporting, immunity, and viral evolution change concurrently. We therefore asked whether cross-country differences observed during the initial exponential-growth phase could be used to infer country-specific seasonal modulation. Early-pandemic basic reproduction numbers (R₀) from 118 countries were linked to 96 harmonized environmental and population-level predictors. Among nine candidate algorithms evaluated across 100 repeated train–test splits, ridge regression using the combined predictor set provided the best balance of predictive accuracy, generalization, and temporal stability. The selected model was then driven by daily climatological covariates to reconstruct annual baseline R₀(t) profiles. Predicted transmissibility generally peaked during winter in the Northern Hemisphere and approximately six months later in the Southern Hemisphere, whereas equatorial countries showed weaker or multimodal patterns. Seasonal forcing amplitude increased strongly with absolute latitude (r = 0.85; mean 0.064 across 77 temperate countries), and predicted R₀ peaks aligned more closely with minimum ultraviolet radiation than with minimum temperature. These ecological associations do not establish causality, but they provide country-specific seasonal-forcing parameters for epidemic models and a baseline environmental context for comparing early-pandemic trajectories. Full article

Knowledge-Tracing (KT) models typically rely on ID-based representations, such as question IDs and skill IDs, which limit their generalizability to unseen questions (cold-question scenarios). In this study, we improve the generalization of cold-question scenarios by embedding a MathBERT-based semantic question in a BERT-based KT model. The proposed model enhances the input representations with domain-specific textual information to enable the model to understand relationships beyond ID-based features. The proposed method was evaluated on the XES3G5M dataset under both the standard setting, where training and testing share the same question set, and the cold-question setting, where test questions are unseen during training. The results show small improvements in the standard setting (Area Under the Curve (AUC): 0.8844 → 0.8854) and larger improvements in the cold-question setting (AUC: 0.8123 → 0.8443). To further evaluate the consistency of the cold-question results, five balanced cold-question selections were generated using different random seeds. Across these selections, the proposed model achieved a mean AUC of 0.7448 ± 0.0428 compared with 0.7183 ± 0.0452 for MLFBK, with a statistically significant mean paired improvement of 0.0265 (p = 0.0054). These results provide evidence that incorporating semantic question embeddings can improve generalization to unseen questions when their associated skills are represented in the training data. Full article

Manual fruit thinning is labor-intensive and inefficient, making the development of intelligent visual detection systems a crucial approach for improving the quality and production efficiency of the pear industry. However, in natural orchard environments, young pear fruits are small in size with slender fruit stalks, and their texture and color characteristics are highly similar to those of tender branches. Furthermore, variations in illumination and occlusions caused by branches and leaves make it difficult for existing detection models to simultaneously and accurately identify fruits and fruit stalks, limiting their application in automated thinning equipment. In this study, Yuluxiang pear was selected as the research object, and image data were collected under diverse field conditions, including forward-lighting, backlighting, close-range shooting, long-range shooting and fruit overlapping. A dedicated dataset containing 3057 images was established. Based on the RT-DETR-R18 network, a lightweight and high-precision fruit–stalk synchronous detection model was proposed. Specifically, the backbone network was reconstructed by integrating GCConv with C2f modules to enhance global feature extraction for slender fruit stalks. The bottleneck structure was optimized using GCConvC3 to reduce feature degradation under occlusion conditions, and an additional 4× down-sampling P2 detection head was introduced to improve the detection capability for small targets. To fully validate the model performance and stability, three types of experiments were conducted in this study: ablation experiments, repeated experiments with different random seeds, and comparative experiments. Ablation experiments verified the cumulative performance improvements brought by the introduced modules. Repeated experiments with different random seeds were performed to explore training randomness-induced performance fluctuations, and the results demonstrated that the proposed model maintains stable overall detection accuracy with minor metric fluctuations. Comparative experiments demonstrated that the proposed model achieved a compact parameter size of only 15.97 M, with a precision of 95.0% for young pear fruit detection and an mAP50 of 83.0% for fruit stalk detection, outperforming all comparative models in overall mAP50. The training convergence curves and Grad-CAM++ visualization results further confirmed the stable optimization process and enhanced feature attention capability of the proposed model. By achieving a favorable balance between detection accuracy and model lightweightness, this approach provides effective technical support for the development of intelligent fruit thinning equipment and vision-based systems for smart pear orchards. Full article

Physical-domain adversarial attacks have been extensively studied in face recognition and object detection, yet the field of palmprint recognition remains largely unexplored. Existing methods generate grayscale patches constrained by the single-channel input of most palmprint models. When deployed on skin, these patches contrast sharply with the surrounding tissue and are readily noticeable to human observers, undermining the covertness required in practical attacks. To address this limitation, we propose a Color Adversarial Patch (CAP) generation algorithm that leverages style transfer principles to produce visually natural color patches while maintaining high attack success rates. The method initiates the patch with a style prior using a pre-trained Contrastive Arbitrary Style Transfer (CAST) model and jointly optimizes adversarial loss, style loss, and smoothness loss within a unified framework. A three-channel averaging strategy is adopted to ensure compatibility with single-channel recognition models during gradient backpropagation. Experiments on the Tongji palmprint dataset show that the generated color patches achieve average cosine similarity values above the decision threshold in physical-domain tests, with peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) values significantly higher than those for their grayscale counterparts. Ablation studies validate the indispensable role of each loss component. CAP offers a practical balance between attack effectiveness and visual camouflage, demonstrating the feasibility of concealed physical-domain attacks on palmprint recognition systems. Full article

With the rapid development of Industrial Internet of Things (IIoT), large amounts of sensor data generated by industrial devices and edge nodes have become the basis of intelligent manufacturing applications. Collaborative modeling on these distributed data is important for tasks such as anomaly detection, equipment monitoring, and predictive maintenance. Federated learning offers a practical way to train models without exposing raw sensor data, but it still faces privacy leakage and malicious poisoning attacks. To address these issues, this paper proposes a hierarchical privacy protection and poisoning-robust defense framework for industrial federated learning. Starting from the sensitivity differences among parameters at different model layers, the proposed method designs a hierarchical privacy-budget allocation strategy that enhances protection for sensitive information while minimizing the performance impact of perturbation. Meanwhile, a multi-layer, multi-feature anomaly-detection mechanism is adopted to identify malicious updates by jointly exploiting directional consistency, scale stability, and inter-layer similarity, and majority voting together with update clipping is used to further improve system robustness. Experiments on Fashion-MNIST, MVTec AD, and C-MAPSS demonstrate that the proposed method can effectively suppress global-model degradation under multiple poisoning attacks and achieves a favorable balance among privacy protection strength, robustness, and training efficiency. Full article

Metabolic dysfunction-associated steatotic liver disease (MASLD) is the most prevalent chronic liver disease, characterized by hepatic steatosis, mitochondrial dysfunction, and oxidative stress. Food-derived polyphenols are promising ingredients that modulate cellular metabolism and redox balance. Açaí (Euterpe oleracea Mart.) seed, a polyphenol-rich agro-industrial by-product, has shown hepatoprotective potential; however, no previous study has comprehensively evaluated its effects on key mitochondrial functions in experimental MASLD. Therefore, this study investigated whether açaí seed extract (ASE), administered alone or in combination with physical training, could improve hepatic mitochondrial respiration, biogenesis, ultrastructural integrity, and redox homeostasis in HF diet-fed Sprague–Dawley rats. Male rats were fed a control or HF diet for 16 weeks and treated with ASE (200 mg/kg/day), aerobic training, or both during the final six weeks. ASE reduced hepatic steatosis by approximately 52% compared with the HF group, improved mitochondrial ultrastructure, increased the expression of the mitochondrial biogenesis regulators PGC-1α and NRF-1, with a 3-fold increase in PGC-1α mRNA expression, increased TFAM immunoreactivity, increased PPARα expression, reduced oxidative stress, and restored lipid-driven mitochondrial respiration. Physical training improved glycemic control and mitochondrial structure, with limited effects on mitochondrial function and redox balance. Combined treatment reduced plasma triglycerides by 41%, increased CPT1α expression, and enhanced carbohydrate-supported mitochondrial respiration. These findings provide mechanistic support for further investigation of açaí seed-derived polyphenols as candidates for nutritional strategies targeting MASLD. Full article

Background/Objectives: Pharmacokinetic and pharmacodynamic drug–drug interactions are major determinants of drug safety in polypharmacy, with potential consequences including reduced therapeutic efficacy, altered drug exposure, and increased adverse effects. This study presents DDI-HierPred, an artificial intelligence-based hierarchical framework for drug–drug interaction prediction using Morgan fingerprint-based drug-pair representations. Methods: At Level 1, Logistic Regression, Random Forest, Linear SVM, and XGBoost were compared for pharmacokinetic/pharmacodynamic (PK/PD) classification. At Level 2, an XGBoost multiclass classifier was used to predict 61 interaction subtype classes. End-to-end performance was evaluated using predicted Level 1 routing, and additional drug-identity-disjoint evaluations were conducted to assess generalization when one or both drugs were unseen during training. Y-randomization analyses were performed for both classification levels. Results: Linear SVM achieved the best Level 1 performance, yielding an accuracy of 0.8661, a ROC-AUC of 0.9380, and an MCC of 0.7303 on the independent test set. At Level 2, the XGBoost classifier achieved an accuracy of 0.8712, a balanced accuracy of 0.9087, a macro F1-score of 0.9115, and a Top-3 accuracy of 0.9868. Because the Level 2 test partition was also used for algorithm comparison and model selection, these results should be regarded as exploratory and potentially optimistic rather than as an independent final evaluation. When evaluated end-to-end using predicted Level 1 routing, performance decreased to an accuracy of 0.5767, balanced accuracy of 0.3573, and macro F1-score of 0.4379. Additional drug-identity-disjoint evaluations showed further performance reductions when one or both drugs were unseen during training, highlighting the greater difficulty of generalization to previously unseen drug identities. Y-randomization analyses supported the robustness of both classification levels. Conclusions: The framework was deployed as a publicly accessible web platform integrating documented interaction lookup, hierarchical PK/PD classification, Level 1-constrained subtype prediction, confidence scoring, single-pair and batch analysis, ranked Top-3 predictions, and downloadable reports. These findings indicate that upstream routing and unseen-drug generalization remain important limitations of the current framework. DDI-HierPred therefore provides a computational platform for research-oriented screening, interpretation, and prioritization of potential drug–drug interactions. Full article

Cattle producers must balance animal welfare with productivity, but regular observation of animals in extensively managed operations is often impractical. Artificial intelligence (AI) integrated with computer vision offers an automated alternative, but few studies compare object detection architectures within the same dataset or focus on pastured cattle. Groups (n = 2–7) of heterogeneous beef cattle were recorded on pasture with nine solar trail cameras. Footage (~132 h) was curated in VideoLAN; annotated in Computer Vision Annotation Tool (CVAT) with bounding boxes and behavioral classes; and split 62/21/17% into training (24,508 frames), validation (8231 frames), and testing (6625 frames) sets. Four architectures were trained: Faster R-CNN (ResNet-50 FPN), Single Shot MultiBox Detector (SSD300, VGG-16), RetinaNet (ResNet-50 FPN with focal loss), and YOLOv8 nano (Ultralytics). With validation at 0.50 confidence and 0.50 IoU, Faster R-CNN achieved the highest overall F1 (0.79) and best per-class balance; RetinaNet was intermediate (peak F1 = 0.72); SSD300 saturated at F1 = 0.40; and YOLOv8 nano achieved some minority class recall at lower confidence. Each model detected the classes “grazing” and “hay feeding” accurately but confused cattle with the visually similar “normal” class. Datasets, checkpoints, and analysis scripts are provided to support further refinement of AI-enabled monitoring of extensive cattle systems. Full article

Residual creases generated during the leather spreading process exhibit highly variable morphologies and irregular feature distributions, causing significant challenges for feature extraction and leading to low localization accuracy. To address these issues, this paper proposes the Dynamic, Cross-Calibrated, and Edge-Enhanced YOLOv8n-Pose (DCE-YOLOv8n-Pose) algorithm. Instead of providing regional approximations, this framework outputs precise spatial coordinates for robotic grasping by integrating three synergistic components in a progressive network flow. First, dynamic snake convolution adaptively perceives the continuous geometric features of elongated creases; subsequently, an efficient multi-scale attention mechanism provides cross-dimensional weight calibration to suppress highly homochromatic background interference and correct spatial misalignments; finally, an edge-enhanced content-aware reassembly of features module preserves high-frequency gradients and prevents feature fracturing during multi-scale fusion. For comprehensive evaluation, a dataset comprising 700 original laboratory images was constructed. To prevent data leakage, the dataset was partitioned into training and validation sets based on individual leather specimens, ensuring that images of the same leather piece do not appear in both sets. Additionally, an independent test set of 500 images collected from an actual processing plant was designed for industrial validation. Experimental results indicate that, at an Intersection over Union (IoU) threshold of 0.5, the DCE-YOLOv8n-Pose model achieves a bounding box mean average precision (mAP@0.5) of 91.8% and a keypoint mAP@0.5 of 85.1%, with a keypoint precision of 87.9%. The computational load is maintained at 9.2 GFLOPs, alongside an inference speed of 114.3 FPS. Furthermore, consistent convergence across four independent training runs substantiates the model’s reliability in reducing missed detection rates and localization deviations. In conclusion, the proposed algorithm demonstrates practical applicability for the visual guidance of automated leather spreading equipment by balancing detection precision and inference speed, thereby offering an effective coordinate reference for subsequent robotic stretching operations. Full article

Rapid leakage prediction during pressure-reducing valve manoeuvres requires models that reproduce inertial hydraulic effects at low computational cost. This study proposes a reproducible surrogate-modelling framework in which transient leakage responses are generated with an extended rigid water column model incorporating time-dependent valve resistance and then used to train machine learning regressors. Twenty-eight regression models were evaluated using four SCADA-oriented predictors: time, inlet flow, upstream pressure head, and valve position. Model selection followed an accuracy–cost–predictive-stability assessment that considered predictive error, training time, inference speed, model size, and behaviour under near-domain, boundary-unseen, and extreme extrapolation scenarios. Gaussian process regression achieved the lowest in-domain errors, with test root mean square error values of 0.0017–0.0019 L/s, but required training times above 21,000 s and inference speeds below 700 observations/s. Bagged Trees provided the most balanced option for PRV-operation screening within the represented hydraulic domain, combining low prediction error, high inference capacity, and stable behaviour within the evaluated near-domain and boundary-unseen range. The P3 extrapolation test showed that prediction beyond the represented hydraulic envelope requires scenario-library expansion and model reassessment. The framework supports rapid valve-operation screening and numerical assessment of RWCM-generated transient leakage responses, while field or SCADA-supported use requires local calibration and validation. Full article

Classification is a fundamental technique in data mining that predicts categorical labels by analyzing input features. However, training Artificial Neural Networks (ANNs) using traditional methods often encounters challenges, such as getting stuck in local minima and slow convergence. To address these issues, this study proposes a novel hybrid model, IPA-ANN, which integrates the Immune Plasma Algorithm (IPA) to optimize the ANN’s connection weights and biases. The IPA, inspired by the immune plasma treatment process, utilizes a unique donor-receiver mechanism to balance exploration and exploitation in the search space. The proposed model was evaluated on nine benchmark datasets from the UCI repository and compared with 18 state-of-the-art metaheuristic algorithms, including Grey Wolf Optimization (GWO), Differential Evolution (DE), and Particle Swarm Optimization (PSO). Experimental results were analyzed using accuracy, F1-score, confusion matrices, and convergence graphs. The findings indicate that IPA-ANN achieves competitive and stable classification performance across different datasets while demonstrating favorable convergence characteristics in several cases. Furthermore, the study investigates the influence of donor–receiver parameters on the optimization process, highlighting the adaptability of the proposed framework. The reliability of these findings was further examined through repeated stratified 5-fold cross-validation and paired Wilcoxon signed-rank tests with Holm–Bonferroni correction on representative datasets, confirming that a subset of the observed performance differences are statistically significant, and through a computational cost analysis showing that IPA-ANN incurs no additional overhead relative to the majority of the compared algorithms. This study contributes to the literature by presenting the first documented application of IPA in ANN training and by providing a modular infrastructure for future metaheuristic-based ANN optimization studies. Full article

Background/Objectives: Regular participation in physical activity is essential for maintaining health and well-being and plays a key role in disease prevention and quality of life, both in the general population and among people with disabilities, who represent approximately 16% of the global population. Martial arts-based physical activity programs incorporating safe-falling techniques may contribute to improving physical fitness, reducing fall-related injuries, and supporting adherence to the World Health Organization physical activity recommendations. This systematic review primarily aimed to identify martial arts-based fall-training programs, particularly judo-based interventions, and examine their effects on health and well-being in individuals with intellectual disability. As a secondary objective, evidence from other populations and martial arts disciplines was considered to identify relevant fall-training protocols and transferability. Methods: The research question was formulated using the PEO framework. Eligibility criteria included: people with intellectual and developmental disabilities or special educational needs; exposure to exercise programs involving martial arts, combat sports, or judo; and outcomes related to safe falling, falls, balance, motor skills, and physical condition. A systematic literature search was conducted between December 2025 and April 2026 on the Web of Science Core Collection, Scopus, PubMed, and SPORTDiscus in accordance with PRISMA guidelines. Results: From a total of 658 records identified, 27 studies met the inclusion criteria, encompassing 1892 participants with various disabilities, including intellectual disability, autism spectrum disorder, visual impairment, and deafness. The interventions involved different martial arts modalities, such as judo, taekwondo, karate, tai chi, and capoeira. Overall, the available evidence suggests beneficial effects on physical fitness, motor abilities, balance, psychosocial outcomes, and safe-falling skills across different populations; however, substantial clinical and methodological heterogeneity was observed among the included studies. Conclusions: No studies investigating structured martial arts-based fall-training programs in individuals with intellectual disability were identified. Although evidence from other populations is promising, it remains heterogeneous. Further high-quality studies are needed to evaluate these interventions in individuals with intellectual disabilities. Full article

Although deep learning methods have shown high performance in chest X-ray classification, high accuracy alone does not guarantee reliable reasoning. A model may still exhibit pathological behavior, such as unstable evidence usage under harmless input changes, inconsistent reasoning across augmented views, excessive dependence on surrounding frame information, and appearance bias dominance, where prediction relies too heavily on intensity while neglecting texture and shape. In this paper, we propose a bio-inspired pathology-aware, factor-aware framework for explainable and reliable chest X-ray classification, inspired by biological vision principles such as figure–ground separation, selective attention, and balanced use of complementary visual cues. During training, the method regularizes appearance bias dominance through evidence-guided counterfactual perturbations that mimic cue-suppression analysis in biological perception, thereby revealing and penalizing excessive factor dependence. During testing, it evaluates model behavior using four criteria: reasoning stability, augmentation inconsistency, appearance bias dominance, and framing sensitivity. This combination enables the framework to go beyond conventional inference-time explanation by both correcting pathological behavior during training and exposing it during evaluation. From a biomimetic perspective, the framework encourages the model to separate relevant foreground anatomy from surrounding background and to avoid over-reliance on a single dominant cue. The proposed approach improves interpretability and reliability without modifying the backbone architecture or increasing model size or inference-time cost. The proposed training process improved the F1-score of DenseNet-121 from 0.899 to 0.931, while also producing more stable and balanced reasoning. Full article