Search Results (4,091)

Pine Wilt Disease (PWD) is a global destructive forest disease. It poses a serious threat to ecological security and forestry economy, and early detection of PWD is crucial for its prevention and control. Most current studies on identifying infected pine trees based on multispectral data only rely on Vegetation Indices (VIs). They fail to fully explore the role of Texture Features (TFs) in disease identification. Furthermore, existing models generally lack interpretability. To address these issues, this study proposes a machine learning classification framework integrating VIs and TFs. It also introduces the SHAP algorithm to clarify the contribution of key features to classification decisions. The results show that the method using fused VIs and TFs as input features performs significantly better than using single features. Among the four models evaluated, LGBM achieved the best performance (OA: 0.897, Macro-F1: 0.895), followed by LR (OA: 0.818, Macro-F1: 0.830), RF (OA: 0.790, Macro-F1: 0.786), and SVM (OA: 0.770, Macro-F1: 0.787) when using fused VIs-TFs. SHAP analysis further reveals that VIs such as Vegetation Atmospherically Resistant Index (VARI), Plant Senescence Reflectance Index (PSRI), Difference Vegetation Index (DVI), Anthocyanin Reflectance Index (ARI), and Normalized Difference Red Edge Index (NDRE), as well as TFs like NIR-Mean (NIR-M), play a dominant role in identifying disease stages. Among the VIs, VARI demonstrated the highest contribution, while NIR-M showed the most significant contribution among TFs. Specifically, VIs are more advantageous in distinguishing the pre-visual, early, middle, and late stages. In contrast, TFs contributed more to identifying healthy and dead trees. This study confirms that fusing VIs and TFs can effectively complement the physiological and structural information of pine canopies. Combined with the interpretable LGBM model, it provides a new technical path for the accurate monitoring of PWD. Full article

In Internet of Things (IoT) systems, data collected by geographically distributed sensors is often incomplete due to device failures, harsh deployment conditions, energy constraints, and unreliable communication. Such data gaps can significantly degrade downstream data processing and decision-making, particularly when failures result in the loss of all locally redundant sensors. Conventional imputation approaches typically rely on historical trends or multi-sensor fusion within the same target environment; however, historical methods struggle to capture emerging patterns, while same-location fusion remains vulnerable to single-point failures when local redundancy is unavailable. This article proposes a correlation-aware, cross-location data fusion framework for data imputation in IoT networks that explicitly addresses single-point failure scenarios. Instead of relying on co-located sensors, the framework selectively fuses semantically similar features from independent and geographically distributed gateways using summary statistics-based and correlation screening to minimize communication overhead. The resulting fused dataset is then processed using a lightweight KNN with an Iterative PCA imputation method, which combines local neighborhood similarity with global covariance structure to generate synthetic data for missing values. The proposed framework is evaluated using real-world weather station data collected from eight geographically diverse locations across the United States. The experimental results show that the proposed approach achieves improved or comparable imputation accuracy relative to conventional same-location fusion methods when sufficient cross-location feature correlation exists and degrades gracefully when correlation is weak. By enabling data recovery without requiring redundant local sensors, the proposed approach provides a resource-efficient and failure-resilient solution for handling missing data in IoT systems. Full article

Out-of-distribution (OOD) detection is essential for ensuring the reliability of machine learning models deployed in safety-critical applications. Existing methods often rely solely on statistical properties of feature distributions while ignoring the geometric structure of learned representations. We propose TopoAD, a topology-aware OOD detection framework that leverages Euler Characteristic Curves (ECCs) extracted from intermediate convolutional activation maps and fuses them with standardized energy scores. Specifically, we employ a computationally efficient superlevel-set filtration with a local estimator to capture topological invariants, avoiding the high cost of persistent homology. Furthermore, we introduce task-adaptive aggregation strategies to effectively integrate multi-scale topological features based on the complexity of distribution shifts. We evaluate our method on CIFAR-10 against four diverse OOD benchmarks spanning far-OOD (Textures), near-OOD (SVHN), and semantic shift scenarios. Our results demonstrate that TopoAD-Gated achieves superior performance on far-OOD data with 89.98% AUROC on Textures, while the ultra-lightweight TopoAD-Linear provides an efficient alternative for near-OOD detection. Comprehensive ablation studies reveal that cross-layer gating effectively captures multi-scale topological shifts, while threshold-wise attention provides limited benefit and can degrade far-OOD performance. Our analysis demonstrates that topological features are particularly effective for detecting OOD samples with distinct structural characteristics, highlighting TopoAD’s potential as a sustainable solution for resource-constrained applications in texture analysis, medical imaging, and remote sensing. Full article

The V/G criterion is a critical indicator for monitoring dynamic changes during Czochralski silicon single crystal (Cz-SSC) growth. However, the inability to measure it in real time forces reliance on offline feedback for process regulation, leading to imprecise control and compromised crystal quality. To overcome this limitation, this paper proposes a novel soft sensor modeling framework that integrates both mechanism-based knowledge and data-driven learning for the real-time prediction of the crystal quality parameter, specifically the V/G value (the ratio of growth rate to axial temperature gradient). The proposed approach constructs a hybrid prediction model by combining a data-driven sub-model with a physics-informed mechanism sub-model. The data-driven component is developed using an attention-based dynamic stacked enhanced autoencoder (AD-SEAE) network, where the SEAE structure introduces layer-wise reconstruction operations to mitigate information loss during hierarchical feature extraction. Furthermore, an attention mechanism is incorporated to dynamically weigh historical and current samples, thereby enhancing the temporal representation of process dynamics. In addition, a robust ensemble approach is achieved by fusing the outputs of two subsidiary models using an adaptive weighting strategy based on prediction accuracy, thereby enabling more reliable V/G predictions under varying operational conditions. Experimental validation using actual industrial Cz-SSC production data demonstrates that the proposed method achieves high-prediction accuracy and effectively supports real-time process optimization and quality monitoring. Full article

In many land cover classification tasks, the limited precision of individual sensors hinders the accurate separation of certain classes, largely due to the complexity of the Earth’s surface morphology. To mitigate these issues, decision fusion methodologies are employed, allowing data from multiple sensors to be synthesized into robust and more conclusive classification outcomes. This study employs fully polarimetric Synthetic Aperture Radar (PolSAR) imagery and leverages the strengths of three decomposition methods, namely Pauli’s, Krogager’s, and Cloude’s, by extracting their respective components for improved detection. From each decomposition method, three scattering components are derived, enabling the extraction of informative features that describe the scattering behavior associated with various land cover types. The extracted scattering features, treated as independent sensors, were used to train three neural network classifiers. The resulting outputs were then considered as local decisions for each land cover type and subsequently fused through a decision fusion rule to generate more complete and accurate classification results. Experimental results demonstrate that the proposed Multi-Class Bahadur–Lazarsfeld Expansion (MC-BLE) fusion significantly enhances classification performance, achieving an overall accuracy (OA) of 95.78% and a Kappa coefficient of 0.94. Compared to individual classification methods, the fusion notably improved per-class accuracy, particularly for complex land cover boundaries. The core innovation of this work is the transformation of the Bahadur–Lazarsfeld Expansion (BLE), originally designed for binary decision fusion into a multi-class framework capable of addressing multiple land cover types, resulting in a more effective and reliable decision fusion strategy. Full article

Traffic state estimation (TSE) is a core task in intelligent transportation systems (ITSs) that seeks to infer key operational parameters—such as speed, flow, and density—from limited observational data. Existing methods often face challenges in practical deployment, including limited estimation accuracy, insufficient physical consistency, and weak generalization capability. To address these issues, this paper proposes a hybrid estimation framework that integrates multi-agent reinforcement learning (MARL) with the Lighthill–Whitham–Richards (LWR) traffic flow model. In this framework, each roadside detector is modeled as an agent that adaptively learns fundamental diagram (FD) parameters—the free-flow speed and jam density—by fusing local detector measurements with global CAV trajectory sequences via an interactive attention mechanism. The learned parameters are then passed to an LWR solver to perform sequential (rolling) prediction of traffic states across the entire road segment. We design a reward function that jointly penalizes estimation error and violations of physical constraints, enabling the agents to learn accurate and physically consistent dynamic traffic state estimates through interaction with the physics-based LWR environment. Experiments on simulated and real-world datasets demonstrate that the proposed method outperforms existing models in estimation accuracy, real-time performance, and cross-scenario generalization. It faithfully reproduces dynamic traffic phenomena, such as shockwaves and queue waves, demonstrating robustness and practical potential for deployment in complex traffic environments. Full article

The hyperspectral image (HSI) is rich in spectral information and has important applications in the field of ground objects classification. However, HSI data have high dimensions and variable spatial–spectral features, which make it difficult for some models to adequately extract the effective features. Recent studies have shown that fusing spatial and spectral features can significantly improve accuracy by exploiting multi-dimensional correlations. Based on this, this article proposes a spectral integration and focused attention network (SIFANet) with a two-branch structure. SIFANet captures the local spatial features and global spectral dependencies through the parallel-designed spatial feature extractor (SFE) and spectral sequence Transformer (SST), respectively. A cross-module attention fusion (CMAF) mechanism dynamically integrates features from both branches before final classification. Experiments on the Salinas dataset and Xiong’an hyperspectral dataset show that the overall accuracy on these two datasets is 99.89% and 99.79%, which is higher than the other models compared. The proposed method also had the lowest standard deviation of category accuracy and optimal computational efficiency metrics, demonstrating robust spatial–spectral feature integration for improved classification. Full article

Ophiocordyceps sinensis is a precious medicinal material with significant pharmacological and economic value. However, the visual similarity between its wild and cultivated forms poses a challenge for authentication. This study investigates the influence of specimen orientation on the accuracy of hyperspectral identification. Hyperspectral data were systematically acquired from four standard specimen orientations (left lateral, right lateral, dorsal, and ventral) for each sample. Random Forest (RF), Support Vector Machine (SVM), Logistic Regression (LR), and Fully Connected Neural Network (FCNN) models were trained and evaluated using both single-orientation and multi-orientation fused data. Results indicate that the LR model achieved superior and stable performance, with an average identification accuracy exceeding 98%. Crucially, for all tested models, no statistically significant difference in identification accuracy was observed across the different specimen orientations. This finding demonstrates that specimen orientation does not significantly influence identification accuracy. The conclusion was further corroborated in experiments using randomly orientation-fused datasets, in which model performance remained consistent and reliable. It is therefore concluded that precise specimen orientation control is unnecessary for the hyperspectral identification of Ophiocordyceps sinensis. This insight substantially simplifies the hardware design of dedicated identification devices by eliminating the need for complex orientation-fixing mechanisms and facilitating the standardization of operational protocols. The study provides a practical theoretical foundation for developing cost-effective, user-friendly, and widely applicable identification instruments for Ophiocordyceps sinensis and offers a reference for similar non-destructive testing applications involving anisotropic medicinal materials. Full article

In view of the depletion of high-grade magnesite resources in China, this study presents a comparative analysis of two industrial fused magnesia products produced via a flotation–fusion route. A low-grade magnesite (DSQLM-3, MgO 41.48 wt.%) was upgraded by reverse flotation to a concentrate (FDSQLM-3, MgO 47.55 wt.%) with >97% SiO₂ removal. Two fused magnesia samples (FM-1 from natural high-grade ore DSQLM-1; FFM-3 from concentrate FDSQLM-3) were produced under identical arc-furnace melting (2800 °C, 4 h), followed by natural cooling. Although FFM-3 showed higher MgO (97.61 vs. 97.25 wt.%), its bulk density was comparable to FM-1 (3.45 vs. 3.46 g/cm³). XRD/Rietveld refinement and SEM-EDS indicated that CMS dominated the Ca–silicate assemblage in FM-1, whereas β/γ-C₂S was observed in FFM-3, coinciding with a higher CaO/SiO₂ (C/S) ratio (2.85 vs. 0.68). Image analysis further showed higher grain boundary microcrack metrics in FFM-3. These observations are consistent with reports in the literature stating that the β → γ transformation of C₂S during cooling involves ~12% volume expansion that can contribute to cracking; however, cooling history and composition were not independently controlled in this industrial comparison, so the relationships are interpreted as data-supported associations rather than isolated causality. The results suggest that beneficiation strategies may benefit from managing residual oxide balance (especially C/S ratio) in addition to reducing total impurities. Mechanical and thermomechanical properties were not measured and should be evaluated in future work. Full article

To reduce manual operation and enhance the intelligence of the high-altitude maintenance wall-climbing robot during its operation, path planning and autonomous navigation need to be implemented. Due to non-uniform magnetic adhesion between the wall-climbing robot and the steel plate, often caused by variations in steel thickness or surface pitting, the wall-climbing robot may experience motion deviations and deviate from its planned trajectory. In order to obtain the actual deviation from the expected trajectory, it is necessary to accurately locate the wall-climbing robot. This allows for the generation of precise control signals, enabling trajectory correction and ensuring high-precision autonomous navigation. Therefore, this paper proposes an external visual localization system based on a pan–tilt laser tracker unit. The system utilizes a zoom camera to track an AprilTag marker and drives the pan–tilt platform, while a laser rangefinder provides high-accuracy distance measurement. The robot's three-dimensional (3D) pose is ultimately calculated by fusing the visual and ranging data. However, due to the limited tracking speed of the pan–tilt mechanism relative to the robot’s movement, we introduce an Extended Kalman Filter (EKF) to robustly predict the robot's true spatial coordinates. The robot's three-dimensional coordinates are periodically compared with the predefined Full article

Fused deposition modeling (FDM) suffers from unpredictable mechanical properties in nominally identical prints. Current quality assurance relies on destructive testing or expensive post-process inspection, while existing machine learning approaches focus primarily on printing parameters rather than real-time thermal environments. The objective of this proof-of-concept study is to develop a low-cost, non-destructive framework for predicting tensile strength during FDM printing by directly measuring convective thermal gradients surrounding the print. To accomplish this, we introduce thermal fingerprinting: a novel non-destructive technique that combines Background-Oriented Schlieren (BOS) imaging with machine learning to predict tensile strength during printing. We captured thermal gradient fields surrounding PLA specimens (n = 30) under six controlled cooling conditions using consumer-grade equipment (Nikon D750 camera, household hairdryers) to demonstrate low-cost implementation feasibility. BOS imaging was performed at nine critical layers during printing, generating thermal gradient data that was processed into features for analysis. Our initial dual-model ensemble system successfully classified cooling conditions (100%) and showed promising correlations with tensile strength (initial 80/20 train–test validation: R² = 0.808, MAE = 0.279 MPa). However, more rigorous cross-validation revealed the need for larger datasets to achieve robust generalization (five-fold cross-validation R² = 0.301, MAE = 0.509 MPa), highlighting typical challenges in small-sample machine learning applications. This work represents the first successful application of Schlieren imaging to polymer additive manufacturing and establishes a methodological framework for real-time quality prediction. The demonstrated framework is directly applicable to real-time, non-contact quality assurance in FDM systems, enabling on-the-fly identification of mechanically unreliable prints in laboratory, industrial, and distributed manufacturing environments without interrupting production. Full article

Deepfake detection in video data typically relies on mining deep embedded representations across multiple modalities to obtain discriminative fused features and thereby improve detection accuracy. However, existing approaches predominantly focus on how to exploit complementary information across modalities to ensure effective fusion, while often overlooking the impact of noise and interference present in the data. For instance, issues such as small objects, blurring, and occlusions in the visual modality can disrupt the semantic consistency of the fused features. To address this, we propose a Multimodality Semantic Consistency Fusion model for video forgery detection. The model introduces a semantic consistency gating mechanism to enhance the embedding of semantically aligned information across modalities, thereby improving the discriminability of the fused representations. Furthermore, we incorporate an event-level weakly supervised loss to strengthen the global semantic discrimination of the video data. Extensive experiments on standard video forgery detection benchmarks demonstrate the effectiveness of the proposed method, achieving superior performance in both forgery event detection and localization compared to state-of-the-art approaches. Full article

Accurately identifying drug off-targets is essential for reducing toxicity and improving the success rate of pharmaceutical discovery pipelines. However, current deep learning approaches often struggle to fuse chemical structure, protein biology, and multi-target context. Here, we introduce HDPC-LGT (Hybrid Dual-Prompt Cross-Attention Ligand–Protein Graph Transformer), a framework designed to predict ligand binding across sixteen human translation-related proteins clinically associated with antibiotic toxicity. HDPC-LGT combines graph-based chemical reasoning with protein language model embeddings and structural priors to capture biologically meaningful ligand–protein interactions. The model was trained on 216,482 experimentally validated ligand–protein pairs from the Chemical Database of Bioactive Molecules (ChEMBL) and the Protein–Ligand Binding Database (BindingDB) and evaluated using scaffold-level, protein-level, and combined holdout strategies. HDPC-LGT achieves a macro receiver operating characteristic–area under the curve (macro ROC–AUC) of 0.996 and a micro F1-score (micro F1) of 0.989, outperforming Deep Drug–Target Affinity Model (DeepDTA), Graph-based Drug–Target Affinity Model (GraphDTA), Molecule–Protein Interaction Transformer (MolTrans), Cross-Attention Transformer for Drug–Target Interaction (CAT–DTI), and Heterogeneous Graph Transformer for Drug–Target Affinity (HGT–DTA) by 3–7%. External validation using the Papyrus universal bioactivity resource (Papyrus), the Protein Data Bank binding subset (PDBbind), and the benchmark Yamanishi dataset confirms strong generalisation to unseen chemotypes and proteins. HDPC-LGT also provides biologically interpretable outputs: cross-attention maps, Integrated Gradients (IG), and Gradient-weighted Class Activation Mapping (Grad-CAM) highlight catalytic residues in aminoacyl-tRNA synthetases (aaRSs), ribosomal tunnel regions, and pharmacophoric interaction patterns, aligning with known biochemical mechanisms. By integrating multimodal biochemical information with deep learning, HDPC-LGT offers a practical tool for off-target toxicity prediction, structure-based lead optimisation, and polypharmacology research, with potential applications in antibiotic development, safety profiling, and rational compound redesign. Full article

Fine-grained visual classification requires recognizing subtle inter-class differences under substantial intra-class variation. Ancient mural recognition poses additional challenges: severe degradation and complex backgrounds introduce noise that obscures discriminative features, limited annotated data restricts model training, and dynasty-specific artistic styles manifest as periodic brushwork patterns and compositional structures that are difficult to capture. Existing spatial-domain methods fail to model the frequency characteristics of textures and the cross-region semantic relationships inherent in mural imagery. To address these limitations, we propose a Vision Transformer (ViT) framework which integrates frequency-domain enhancement, explicit token-relation modeling, adaptive multi-focus inference, and discriminative metric supervision. A Frequency Channel Attention (FreqCA) module applies 2D FFT-based channel gating to emphasize discriminative periodic patterns and textures. A Cross-Token Relation Attention (CTRA) module employs joint global and local gates to strengthen semantically related token interactions across distant regions. An Adaptive Omni-Focus (AOF) block partitions tokens into importance groups for multi-head classification, while Complementary Tokens Integration (CTI) fuses class tokens from multiple transformer layers. Finally, Adaptive Margin Contrastive Center Loss (AMCCL) improves intra-class compactness and inter-class separability with margins adapted to class-center similarities. Experiments on CUB-200-2011, Stanford Dogs, and a Dunhuang mural dataset show accuracies of 91.15%, 94.57%, and 94.27%, outperforming the ACC-ViT baseline by 1.35%, 1.63%, and 2.20%, respectively. Full article

Background and Objectives: Academic point-of-care (POC) manufacturing enables the in-hospital design and production of patient-specific medical devices within certified environments, integrating clinical practice, engineering, and translational research. This model represents a new academic ecosystem that accelerates innovation while maintaining compliance with medical device regulations. Gregorio Marañón University Hospital has established one of the first ISO 13485-certified academic manufacturing facilities in Spain, providing on-site production of anatomical models, surgical guides, and custom implants for oral and maxillofacial surgery. This study presents a retrospective review of all devices produced between April 2017 and September 2025, analyzing their typology, materials, production parameters, and clinical applications. Materials and Methods: A descriptive, retrospective study was conducted on 442 3D-printed medical devices fabricated for oral and maxillofacial surgical cases. Recorded variables included device classification, indication, printing technology, material type, sterilization method, working and printing times, and clinical utility. Image segmentation and design were performed using 3D Slicer and Meshmixer. Manufacturing used fused deposition modeling (FDM) and stereolithography (SLA) technologies with PLA and biocompatible resin (Biomed Clear V1). Data were analyzed descriptively. Results: During the eight-year period, 442 devices were manufactured. Biomodels constituted the majority (approximately 68%), followed by surgical guides (20%) and patient-specific implants (7%). Trauma and oncology were the leading clinical indications, representing 45% and 33% of all devices, respectively. The orbital region was the most frequent anatomical site. FDM accounted for 63% of the printing technologies used, and PLA was the predominant material. The mean working time per device was 3.4 h and mean printing time 12.6 h. Most devices were applied to preoperative planning (59%) or intraoperative use (35%). Conclusions: Academic POC manufacturing offers a sustainable, clinically integrated model for translating digital workflows and additive manufacturing into daily surgical practice. The eight-year experience of Gregorio Marañón University Hospital demonstrates how academic production units can enhance surgical precision, accelerate innovation, and ensure regulatory compliance while promoting education and translational research in healthcare. Full article