Search Results (314)

Graph Neural Networks (GNNs) have emerged as the standard for learning representations from graph-structured data. While traditional architectures relying on message-passing mechanisms excel in homophilic settings, they essentially function as fixed low-pass filters. However, this smoothing operation limits their ability to generalize to heterophilic graphs, where connected nodes often exhibit dissimilar labels and high-frequency signals are crucial for discrimination. Furthermore, existing Mixture-of-Experts (MoE) methods for graphs often suffer from local-view routing, failing to capture global structural context during expert selection. To address these challenges, this paper proposes SS-AdaMoE, a novel Spatio-Spectral Adaptive Mixture of Experts framework designed for robust node classification across diverse graph patterns. Specifically, a Dual-Domain Expert System is constructed, integrating heterogeneous spatial aggregators with learnable spectral filters based on Bernstein polynomials. This allows the model to adaptively capture arbitrary frequency responses—including high-pass and band-pass signals—which are overlooked by standard GNNs. To resolve the locality bias, a Hierarchical Global-Prior Gating Network augmented by a Linear Graph Transformer is introduced, ensuring that expert selection is guided by both local node features and global topological awareness. Extensive experiments are conducted on five benchmark datasets spanning both homophilic and heterophilic networks. The results demonstrate that SS-AdaMoE consistently outperforms baselines, achieving accuracy improvements of up to 2.65% on Chameleon and 1.41% on Roman-empire over the strongest MoE baseline, while surpassing traditional GCN architectures by margins exceeding 28% on heterophilic datasets such as Texas. These findings validate that the synergy of learnable spectral priors and global gating effectively bridges the gap between spatial aggregation and spectral filtering. Full article

Few-shot semantic segmentation aims to learn accurate pixel-level classification from limited annotated samples, a critical capability for real-world applications where data acquisition is expensive or impractical. However, existing methods often struggle with fine-grained texture details and complex boundaries under data-scarce conditions, particularly when applied to domains with intricate visual patterns (such as batik patterns). To address this few-shot learning challenge, we constructed a few-shot batik pattern dataset and proposed a novel network architecture centered on attention weighting and hierarchical decoding. Our method leverages a pre-trained ResNet101 backbone for transfer learning to establish a strong feature foundation. It incorporates a dual-attention module that combines spatial and channel attention to dynamically highlight semantically rich regions and intricate texture boundaries specific to batik. For multi-scale context aggregation, a lightweight module utilizing parallel dilated convolutions is introduced to efficiently capture features from varying receptive fields. Finally, a hierarchical decoder progressively integrates these enhanced, multi-scale features with high-resolution shallow features to reconstruct precise segmentation maps. Comprehensive evaluations on a dedicated batik dataset show that our model achieves state-of-the-art performance, with a mean Intersection over Union (mIoU) of 79.22% and a pixel accuracy (PA) of 92.47%. It notably improves over the strong DeepLabV3+ baseline by 3.3% in mIoU and 0.95% in PA, demonstrating its effectiveness for the task of batik pattern segmentation under data-scarce conditions. Full article

The accurate and efficient extraction of individual tree phenotypic traits for seabuckthorn (Hippophae rhamnoides L.) in natural forests is crucial for germplasm exploration, precision silviculture, and ecological restoration. This study extracted structural and biochemical traits of seabuckthorn in Tibet’s Lhasa valley using Unmanned aerial vehicle (UAV) LiDAR, multispectral imagery, and the N-PROSAIL model. Firstly, building on a classification conducted through multi-scale spatial analysis and hierarchical clustering with dynamic thresholds, shrub interference was effectively reduced, thereby improving the accuracy of individual tree segmentation. Tree height and crown width were derived from the segmentation results, and a DBH estimation model was developed using handheld LiDAR data. Finally, leaf nitrogen content was mapped within canopies using random forest combined with the N-PROSAIL model and nitrogen reference data. The results demonstrated that the optimized segmentation method successfully extracted structural traits (F1 = 84.21%). Tree height was accurately estimated (R² = 0.814, RMSE = 0.580 m), and the DBH prediction model performed satisfactorily (R² = 0.779, RMSE = 1.725 cm). The random forest model also effectively estimated leaf nitrogen content (R² = 0.680, RMSE = 2.074 mg/g). Full article

Fine-grain heterogeneity in agricultural landscapes is often obscured by coarse land-use classification schemes. This study provides a structural characterization of agricultural land use in selected sites within the Dukagjini and Kosova Plains of Kosovo using fine-grain, field-mapped data. Agricultural land-use structure was analyzed across three hierarchical classification levels, from broad categories to specific crop types, focusing on patterns of composition and configuration. Descriptive analyses and non-metric multidimensional scaling (NMDS) were used to examine structural patterns across thematic resolution and spatial grouping, with topographic and geographic variables included as contextual variables. Landscape metrics derived from field mapping were also compared with the ESA WorldCover dataset to evaluate how global land-cover products represent agricultural landscape structure. The results show that coarse classifications limit detectable structural differentiation. While broad land-use categories showed limited compositional variation and low diversity, finer classification levels revealed stronger contrasts in composition, configuration, and diversity. At the finest classification level, significant differentiation was detected among villages and municipalities, while contrasts between plains were weak. Topographic and geographic variables showed limited but detectable associations with structural patterns. Overall, this study provides a descriptive baseline of agricultural land-use structure in a data-scarce region. Full article

The objective of this research is to determine whether levels of road accessibility in urban, peri-urban, and sub-urban localities within the municipalities of Mexicali and San Felipe, in Baja California, Mexico, can be associated with processes of territorial expansion, population growth, and changes in urban marginalization levels. This is assessed through a methodology that combines ex-ante and ex-post analysis, the use of the Urban Marginalization Index (UMI) at the AGEB scale, and a hierarchical accessibility classification (Levels A, B, and C), thereby contributing a replicable tool for analyzing socio-spatial impacts derived from road infrastructure. To this end, modernization, maintenance, and reconstruction works, as well as the construction of an interchange carried out between 2006 and 2017 along Federal Highway No. 5—specifically the Mexicali–San Felipe section—were examined in relation to the accessibility they provide to ten nearby localities. UMI values were estimated for 134 AGEB using data from 2000, 2010, and 2020, which enabled the assessment of changes in quality of life before, during, and after the execution of these works. The results show significant population growth in six localities, accompanied by territorial expansion processes. Localities with direct connection to the study corridor tended to exhibit middle to low marginalization levels, while those with indirect accessibility or direct access through another federal highway section tended toward middle to high levels, with some shifting to middle to low. It is concluded that road accessibility constitutes a relevant factor in the progressive improvement in socioeconomic conditions and quality of life in urban, peri-urban, and sub-urban areas. Full article

Timely and accurate detection of sow oestrus is crucial for enhancing reproductive efficiency and reducing non-productive days (NPDs) in large-scale pig farms. However, traditional manual observation is labour-intensive and subjective, while cloud-based deep learning solutions face challenges such as high latency and privacy risks when applied in intensive housing environments. This study developed an edge-intelligent monitoring system that integrates deep temporal modelling with sound source localisation technology. A three-stage hierarchical screening strategy was utilised to select and deploy a lightweight Stacked-LSTM model on the resource-constrained ESP32-S3 hardware platform. This model was trained and calibrated using a high-quality acoustic dataset validated against serum reproductive hormones, specifically follicle-stimulating hormone (FSH), luteinising hormone (LH), and progesterone ($P_4$). Experimental results demonstrate that the optimised model achieved a classification accuracy of 96.17%, with an inference latency of only 41 ms, thereby fully satisfying the stringent real-time monitoring requirements while maintaining a minimal memory footprint. Furthermore, the system integrates a localisation algorithm based on Generalised Cross-Correlation with Phase Transform (GCC-PHAT). Through spatial geometric modelling, the system successfully implements the functional mapping of vocalisation events to individual gestation stalls (Stall IDs). Laboratory pressure tests validated the robustness and low-cost deployment advantages of the “edge recognition–cloud synchronization” architecture, providing a reliable technical framework for the precision management of smart livestock farming. Full article

A behavioral and neuropsychological disorder that develops in young children during their early school years is called attention-deficit hyperactivity disorder (ADHD). When young children are diagnosed with ADHD, they have a tendency not to concentrate on academic and extracurricular activities. Moreover, children affected with ADHD suffer from mood swings, so it becomes quite difficult for them to establish good connections with teachers and friends. In the field of clinical research, deploying Electroencephalography (EEG) signals, a rapid and accurate diagnosis of ADHD is essential so that an effective treatment can be given to the children affected with ADHD. In this work, a unified framework is proposed for the detection of ADHD using EEG signals and some coherent models. The framework initially employs the concept of normalization of EEG signals, followed by the usage of dimensionality reduction techniques such as Local Linear Embedding (LLE), Sammon Mapping (SM) and Locally Linear Coordination (LLC). The dimensionally reduced EEG values are further clustered using four techniques such as spectral clustering, K-means clustering, Fuzzy C-means (FCM) clustering, Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN), and finally, silhouette coefficient analysis is used to analyze the clustering effectiveness. The features are then extracted from the clustered values using an Improved Wavelet Transform (IWT) and then the features are selected with four efficient techniques such as the chi-squared test, Mutual Information (MI), Mahalanobis analysis and Binary Horse Herd Optimization (BHHO) techniques. Finally, the selected values are fed into classifiers for classification with the help of ten traditional machine learning classifiers. The work is tested on a publicly available ADHD dataset and the analysis shows that the best results are obtained when the LLC dimensionality reduction is utilized with FCM clustering and IWT feature extraction, BHHO feature selection, and classified with LGBA classifier reporting a high classification accuracy of 98.12%. Full article

Autism classification is challenging due to the subtle, heterogeneous, and overlapping neural activation profiles that occur in individuals with autism. Novel deep learning approaches, such as Convolutional Neural Networks (CNNs) and their variants, as well as Transformers, have shown moderate performance in discriminating between autism and normal cohorts; yet, they often struggle to jointly capture the spatial–structural and temporal–functional variations present in autistic brains. To overcome these shortcomings, we propose a novel hierarchical deep learning framework that extracts the inherent spatial dependencies from the dual-modal MRI scans. For sMRI, we develop a 3D Hierarchical Convolutional Neural Network to capture both fine and coarse anatomical structures via multi-view projections along the axial, sagittal, and coronal planes. For the fMRI case, we introduced a bidirectional LSTM-based temporal encoder to examine regional brain dynamics and functional connectivity. The sequential embeddings and correlations are combined into a unified spatiotemporal representation of functional imaging, which is then classified using a multilayer perceptron to ensure continuity in diagnostic predictions across the examined modalities. Finally, a cross-modality fusion scheme was employed to integrate feature representations of both modalities. Extensive evaluations on the ABIDE I dataset (NYU repository) demonstrate that our proposed framework outperforms existing baselines, including Vision/Swin Transformers and various newly developed CNN variants. For the sMRI branch, we achieved 90.19 ± 0.12% accuracy (precision: 90.85 ± 0.16%, recall: 89.27 ± 0.19%, F1-score: 90.05 ± 0.14%, and focal loss: 0.3982). For the fMRI branch, we achieved an accuracy of 88.93 ± 0.15% (precision: 89.78 ± 0.18%, recall: 88.29 ± 0.20%, F1-score: 89.03 ± 0.17%, and focal loss of 0.4437). These outcomes affirm the superior generalization and robustness of the proposed framework for integrating structural and functional brain representations to achieve accurate autism classification. Full article

In public safety scenarios, the precise detection and positioning of prohibited weapons such as firearms and knives along with the involved personnel are the core pre-requisite technologies for violent risk warning and emergency response. However, in security surveillance scenarios, there are common problems such as object occlusion, difficulty in capturing small-sized weapons, and complex background interference, which lead to the shortcomings of existing general object detection models in the tasks of detecting and locating security-related objects, including poor adaptability, low detection accuracy, and insufficient robustness in complex scenarios. Therefore, this paper proposes a threat object detection framework for security scenarios (PDGV-DETR) based on adaptive dynamic convolution and cross-scale semantic fusion, specifically optimized for the detection and positioning tasks of weapons and personnel objects in static security surveillance images. This research focuses on category recognition at the object level and pixel-level spatial positioning, and does not involve the classification and identification of violent behaviors based on temporal information. There are clear technical boundaries and scene limitations between the two. This framework is optimized through three core modules: designing a dynamic hierarchical channel interaction convolution module to reduce computational complexity while enhancing the ability to detect occluded and incomplete objects; constructing an improved bidirectional hybrid feature pyramid network, combining the cross-scale fusion module to strengthen multi-scale feature expression, and adapting to the simultaneous detection requirements of small weapon objects and large personnel objects; and introducing a global semantic weaving and elastic feature alignment network to solve the problem of low discrimination between objects and complex backgrounds. Under the same experimental configuration, the proposed model is verified against current mainstream models on typical datasets: on a dataset of 2421 conflict scene personnel violent images, the peak average precision mAP50 of PDGV-DETR reached 85.9%. Through statistical verification, compared with the baseline model RT-DETR with an average value ± standard deviation of 0.840 ± 0.007, the average value ± standard deviation of PDGV-DETR reached 0.858 ± 0.004, demonstrating statistically significant performance improvement, with a p-value less than 0.01. This model can accurately complete the task of locating the object area of personnel, and compared with the deformable DETR, the accuracy improvement rate reached 15.1%.; on the weapon-specific dataset OD-WeaponDetection, the mAP for gun and knife detection reached 93.0%, improving by 2.2% compared to RT-DETR. Compared to the performance fluctuations of other general object detection models in complex security scenarios, PDGV-DETR not only has better detection and positioning accuracy for security-related objects, but also significantly improves the generalization and stability of the model. The results show that PDGV-DETR effectively balances the accuracy of positioning, detection, and computational efficiency, accurately completing end-to-end detection and positioning of weapon and personnel objects in static security surveillance images, demonstrating highly competitive performance in the detection and positioning of security-related objects in security scenes, providing core object-level pre-processing technology support for scenarios such as public area monitoring, intelligent video monitoring, and early warning of violent risks, and providing basic data for subsequent violent behavior recognition based on temporal data. Full article

The taste and quality of kiwifruit are key factors affecting consumers’ purchase intention and satisfaction. As an important indicator for measuring kiwifruit quality, sugar content is crucial for quality grading. Accurate and rapid kiwifruit grading based on sugar content is of great significance for ensuring product quality and enhancing market competitiveness. Traditional grading methods mostly adopt destructive sampling, which are cumbersome, low in efficiency, and difficult to meet the needs of modern large-scale production. Therefore, this paper proposes a kiwifruit classification method based on the Hierarchical 3D Convolution and Attention Mechanism Network (H3DAMNet). This method performs 3D convolution operations on multiple dimensions of hyperspectral data blocks simultaneously to deeply extract spatial–spectral features. It assigns weights to each channel through the channel attention mechanism to weaken attention to irrelevant information, and introduces the bottleneck self-attention mechanism to capture the positional dependence in input features, thereby effectively modeling global information. Referring to industry standards, kiwifruit are classified into three grades based on sugar content: first-grade (≥14.5 °Brix), second-grade (13.5–14.5 °Brix), and third-grade (≤13.5 °Brix). On the test set containing 280 kiwifruit samples, the overall accuracy (OA) of this method reaches 97.5% and the average accuracy (AA) is 97.3%, successfully realizing the accurate classification of kiwifruit according to sugar content and setting a reference example for the classification of other similar fruits. Full article

Pediatric gait abnormalities are closely intertwined with musculoskeletal dysfunctions and heightened injury risk, underscoring the urgency of early and accessible screening tools. Here, we develop and validate a video-based semi-supervised Abnormal Gait Recognition Module (AGRM) to address unmet needs in pediatric gait assessment, with a focus on diagnostic performance and clinical interpretability. The AGRM is built on a 3D ResNet backbone, synergistically integrated with a Mean Teacher Module (MTM) to mitigate the limitations of limited labeled clinical data, and a Spatial Hierarchical Pooling Module (SHPM) for robust multiscale spatiotemporal feature extraction—two core innovations tailored to gait dynamics. We trained and validated the model on a hybrid dataset combining self-collected pediatric gait videos and the public CASIA-B dataset, evaluating its performance in binary (normal vs. abnormal) and three-class (normal, genu varum, genu valgum) classification tasks using accuracy, macro-precision, macro-recall, and macro-F1 score. Ablation studies quantified the incremental contributions of MTM and SHPM, while Grad-CAM visualization was employed to enhance model interpretability. In the three-class classification task, the AGRM achieved a 70.5% accuracy, 72.1% macro-precision, 71.5% macro-recall, and a macro-F1 score of 0.718; in the binary task, it yielded a 80.3% precision and 79.2% recall. SHPM significantly augmented spatiotemporal feature aggregation, capturing fine-grained gait dynamics, whereas MTM improved model generalization under constrained labeled data scenarios—findings corroborated by ablation experiments. Grad-CAM visualization confirmed the model’s targeted attention to lower extremity regions, particularly the knee joints, aligning with the pathological loci of gait abnormalities. Collectively, our AGRM demonstrates robust performance and generalization in identifying pediatric gait abnormalities, while effectively capturing key pathological gait characteristics. This video-based intelligent approach offers a promising tool for early gait screening in both clinical and community settings, addressing barriers to accessible pediatric musculoskeletal assessment. Full article

Unauthorized farmland excavation is a prominent manifestation of farmland non-agriculturalization, and its effective monitoring depends on structured representations of objects and their spatial interactions in complex scenes. However, the existing computer vision research mainly focuses on object-level recognition or scene-level classification, while lacking datasets that explicitly model topological relationships in farmland excavation scenarios. To address this limitation, this paper presents TopoFarm, a topology-annotated panoptic dataset for unauthorized farmland excavation scenes. TopoFarm provides fine-grained panoptic segmentation annotations together with pairwise object contact relationship labels, enabling joint object–relation modeling and topology-aware scene representation. To improve annotation reliability under complex conditions, a human-in-the-loop hybrid intelligence framework, termed HITPA, is introduced to integrate automatic panoptic segmentation, depth-aware topological reasoning, and expert-guided refinement, achieving high annotation quality with controlled manual effort. Based on TopoFarm, systematic benchmark experiments are conducted for panoptic segmentation and topological relationship reasoning, along with a hierarchical evaluation protocol to analyze the impact of object-level representation quality on relational inference. The results demonstrate that TopoFarm poses substantial challenges for both tasks and highlight the strong dependence of topological reasoning on object accuracy and global scene context. Overall, TopoFarm provides a new data foundation and evaluation benchmark for topology-aware perception in farmland monitoring applications. Full article

Wildfire events are increasing in frequency and severity, creating an urgent need for early, accurate, and energy-efficient forest fire prediction systems that can operate at a large scale. A fundamental challenge in edge-level forest fire prediction lies in jointly achieving high detection accuracy while minimizing wireless transmissions and communication-related energy consumption. This paper proposes a communication-aware hierarchical wireless sensor network (WSN) framework that performs fire versus normal environmental state classification directly at the network edge. Multi-modal physical and constrained virtual sensor readings are fused into short-term temporal supervectors and processed locally using lightweight random forest classifiers deployed on sensor nodes and cluster heads. A temporal 2-of-3 voting mechanism is applied at the edge to suppress transient noise and improve prediction reliability before triggering communication. The proposed design enables selective, event-driven transmission, where only temporally validated abnormal states are forwarded through the hierarchy, thereby decoupling detection accuracy from continuous data reporting. Extensive experiments using real multi-modal environmental sensor data and statistically rigorous 5-fold GroupKFold cross-validation—ensuring strict node-level separation between training and testing—demonstrate the effectiveness of the approach. The proposed framework achieves a node-level accuracy of 98.82 ± 1.75% and a scenario-level detection accuracy of 96.52 ± 0.89%. Compared to periodic reporting and the LEACH protocol, the system reduces wireless transmissions by over 66% and communication-related energy consumption by more than 66% across network sizes ranging from 100 to 1000 nodes. The main contributions of this work are summarized as follows: (1) a communication-aware hierarchical Edge-AI framework for early forest fire prediction that performs local inference and temporal validation directly at sensor nodes; (2) a constrained virtual sensing strategy integrated with temporal supervector modeling to enhance spatial coverage while preserving reliability; and (3) a statistically rigorous large-scale evaluation demonstrating joint optimization of prediction accuracy, transmission reduction, and communication energy efficiency across network sizes ranging from 100 to 1000 nodes. These results show that accurate early forest fire prediction can be achieved through edge-level inference and selective communication, substantially extending network lifetime while maintaining statistically reliable detection performance. Full article

This research presents an advanced YOLOv8-MMD framework specifically designed for intelligent white radish harvesting systems, addressing the critical need for simultaneous species recognition and quality evaluation. The proposed architecture is built upon a dual-branch detection system (YOLOv8-Dual) with a shared Backbone network, and is further enhanced by two novel components: the Multi-Scale Attention Aggregation (MSAA) module that strategically combines channel-wise and spatial attention mechanisms to refine feature representation, and the Multi-scale Feature Enhancement (MAFE) module that facilitates effective information fusion across different hierarchical levels of the network. Extensive experimental validation reveals that the YOLOv8-MMD model achieves remarkable performance metrics, including a species detection precision of 0.945 and a quality assessment precision of 0.812, representing improvements of 1.4% and 4%, respectively, over the baseline YOLOv8-Dual model. Under the comprehensive mAP@50 evaluation standard, the model reaches 0.949 for species identification and 0.859 for quality classification, while maintaining impressive recall rates of 0.924 and 0.836 for the respective tasks. The system demonstrates exceptional robustness when deployed in challenging field conditions, consistently performing well under varying lighting intensities, different growth stages, and partial occlusion scenarios. Computational analysis confirms the model’s practical viability, achieving a processing throughput of 112 frames per second with 8.1 GFLOPs of computational overhead, thereby meeting stringent real-time operational requirements for agricultural robotic applications. Comparative studies with existing methods further substantiate the superiority of the proposed approach in balancing detection accuracy with computational efficiency. The integration of multi-scale attention mechanisms and hierarchical feature enhancement strategies provides a comprehensive solution for automated agricultural harvesting in complex, unstructured environments, offering significant potential for practical implementation in precision agriculture systems. Full article

The accurate upscaling of peatland carbon stocks is fundamentally limited by fine-scale microrelief (hummocks/depressions), which has not yet been resolved by conventional satellite or field methods. We demonstrate the critical advantage of using Uncrewed Aerial System LiDAR (UAS-LiDAR) for mapping the hierarchical microrelief of a Western Siberian ombrotrophic bog to enhance ground-layer phytomass estimation. The rule-based classification of a normalized digital terrain model generated a high-resolution microform map (overall accuracy = 79%, Kappa = 0.72). This map was used to upscale field-measured phytomass and compared against estimates generated through satellite imagery (SuperView-2) and traditional field-visual extrapolation. While total landscape-level phytomass stocks were similar across methods (~93–97 t ha⁻¹), their spatial allocation differed fundamentally. The satellite-based method exhibited a predictable, landscape-dependent systematic bias (overestimation by 7–25% in some units) and a substantially lower microtopography accuracy (OA = 77%, Kappa = 0.53) compared to the aggregated LiDAR map (OA = 95%, Kappa = 0.89). Crucially, only the LiDAR-based approach accurately resolved the biomasses of key microforms (e.g., hummocks within hollows contributing up to 6.2 ± 1.4 tonnes per unit), which were missed or misaggregated when using traditional techniques. We conclude that objective, high-resolution microrelief mapping via UAS-LiDAR is essential for spatially explicit and ecologically coherent phytomass upscaling, providing an indispensable structural template for credible carbon accounting in heterogeneous peatlands. Full article