Search Results (2,196)

Change detection in high-resolution optical remote sensing images is essential for dynamic surface monitoring and land resource management. However, existing methods still face challenges in representing multiscale change information, accurately locating change boundaries, and handling the imbalance between changed and unchanged samples. This study proposes SAM3TGNet, which transforms the general visual features of Segment Anything Model 3 (SAM3) into change-oriented representations through lightweight channel adaptation, multiscale cross-temporal interaction, and spatiotemporal feature fusion, while the Global Context Spatial Attention (GCSA) module and dynamic weighted loss enhance boundary representation and alleviate class imbalance, respectively. First, a SAM3-based bi-temporal remote sensing image feature encoder with channel adapters is developed, where intermediate multiscale feature layers are exploited to enhance semantic representation and generalization for diverse change patterns. Second, the GCSA module is introduced after spatiotemporal feature fusion to model global dependencies and enhance local details, improving boundary localization accuracy. Third, a dynamic weighted change loss function is designed to adaptively adjust the contribution of changed pixels according to their proportion in each batch, reducing background bias caused by sample imbalance and improving change localization and completeness. Experiments on the Wuhan University Change Detection (WHU-CD) and Sun Yat-sen University Change Detection (SYSU-CD) datasets demonstrate that the proposed method achieves F1-scores of 94.08 ± 0.12% and 82.91 ± 0.20%, respectively, outperforming existing approaches in multiscale change detection and boundary delineation. Full article

Intelligent monitoring systems in domains such as renewable energy, electrical grid management, and environmental sensing continuously generate high-dimensional multivariate time-series data characterized by non-stationarity and multi-scale temporal dependencies. Accurate long-term forecasting of system parameters is essential for proactive maintenance, operational scheduling, and cost reduction, yet many existing models rely solely on numerical sequences and lack mechanisms to incorporate higher-level contextual information. This study proposes a multimodal long-term forecasting framework that integrates frequency-aware signal decomposition with semantic-enhanced representation learning. The framework comprises four components: (1) a wavelet-based feature extraction module that captures multi-scale periodic patterns through energy-guided frequency selection; (2) a semantic feature extraction module that encodes statistical summaries of the input into language embeddings via a pretrained language model; (3) a cross-attention fusion module that dynamically aligns temporal and semantic representations; and (4) a multi-scale MLP ensemble for robust prediction. Experiments on three benchmark datasets—Solar Power, ETTh1, and Weather—show that the framework achieves competitive accuracy against strong baselines, including PatchTST and iTransformer, with its strongest results on data exhibiting complex multi-scale seasonal structure, where it attains the second-best mean squared error on the Weather dataset. A controlled ablation isolating the pretrained embedding from a direct numerical encoding of the same statistics indicates a small, dataset-specific benefit that is comparable in magnitude to seed-to-seed variation. Overall, the proposed framework provides a modular and interpretable architecture combining frequency-aware and semantic-aware processing for intelligent system management. Full article

Semantic segmentation assigns a semantic label to every pixel in an image, which is a fundamental task in computer vision with applications such as autonomous driving and medical imaging. Existing segmentation methods, whether CNN-based or transformer-based, both have limitations: the former are constrained by fixed dilation rates that hinder scale adaptation, while the latter suffer from quadratic computational complexity that prevents efficient deployment. To address these issues, this paper proposes SA-DeepLab, a CNN framework that combines a switchable atrous spatial pyramid pooling (SA-ASPP) module and a channel shuffle operation in the decoder. SA-ASPP employs a learnable spatial switch map to dynamically fuse two atrous convolutions with different dilation rates, enabling adaptive receptive field selection at each pixel, while a depthwise separable atrous convolution replaces the most dilated branch to reduce computational overhead. The channel shuffle operation rearranges feature channels across groups to encourage cross-group information exchange without extra parameters. Experiments on Pascal VOC 2012 and Cityscapes demonstrate that SA-DeepLab achieves competitive mIoU (86.30% and 81.78%) with a lightweight framework, outperforming both CNN-based and transformer-based competitors in the accuracy-efficiency tradeoff. Full article

Recent image diffusion models have enabled automatic 3D object creation from text or image guidance, but their 2D generative priors often bake illumination and shadow into textures, making relighting and physically based rendering (PBR) difficult. To address this issue, we propose MaterialSeg3D, a framework that predicts surface materials for 3D assets by leveraging 2D material semantics. Given a mesh and its albedo UV map, MaterialSeg3D renders multi-view images, performs material segmentation using a 2D prior model, projects the predictions back to UV space, and fuses them through weighted voting and region unification to obtain coherent material maps. To train the prior model, we construct MIO++, a large-scale single-object material segmentation dataset containing 115,542 images, 12 object themes, and 30 fine-grained material categories, substantially extending the previous MIO dataset. Each MIO++ material category is associated with an independent pair of roughness and metallic values for PBR assignment. We further observe that poor topology in AI-generated assets can degrade PBR quality even when plausible materials are assigned, and introduce a plane-simplification strategy as an auxiliary preprocessing step for such meshes. Experiments show that MIO++ improves material segmentation and that the resulting material maps support more consistent relightable renderings for both human-crafted and AI-generated 3D assets. Full article

Fine-grained visual classification (FGVC) aims to distinguish highly similar subcategories, and its performance relies heavily on the accurate modeling of discriminative local parts and their structural relationships. However, existing Vision Transformer-based methods are susceptible to background noise interference, and the relationship modeling approach relying on explicit spatial coordinates struggles to maintain stable structural representations when targets undergo pose variations and non-rigid deformations. To address these issues, this paper proposes a Semantic Topology Part Network (STP-Net). First, a Prior-Guided Part Aggregator (PGA) is designed, which leverages the foreground prior provided by foundation models to guide discriminative part discovery, enhancing target region responses while suppressing background interference. Second, a Topology-Informed Semantic Graph Convolutional Network (TIS-GCN) is designed to dynamically construct topological relationships among parts in an implicit semantic space, achieving robust modeling against complex structural variations. Furthermore, a Semantic–Spatial Cross-Attention (SSCA) mechanism is introduced to establish bidirectional interaction between semantic relationships and spatial features, and combined with a Global-Context Adaptive Gating mechanism to accomplish multi-scale feature fusion. On four mainstream fine-grained visual classification benchmarks, namely CUB-200-2011, Stanford Cars, Stanford Dogs, and NABirds, the proposed model achieves Top-1 accuracies of 92.7%, 94.9%, 95.2%, and 92.3%, respectively. Comprehensive ablation studies and visualization analyses further validate the effectiveness of the proposed method in background suppression, structural relationship modeling, and discriminative feature learning. Full article

Early detection and timely management of mango malformation disease (MMD) are essential for stabilizing yield and improving orchard management efficiency. However, existing mango disease-recognition studies have mainly focused on leaf diseases, fruit defects, or image-level classification, whereas object detection of MMD-related mango inflorescences under natural orchard conditions remains insufficiently investigated. This task is challenging because malformed inflorescences show large morphological variation, occur at different scales, and are frequently affected by branch occlusion, illumination variation, and complex orchard backgrounds. To address these challenges, this study proposes an improved YOLOv11n model, termed LRA-YOLOv11n, for field detection of MMD. YOLOv11n was used as the baseline. First, LSKNet was introduced to reconstruct the backbone network and provide a dynamic large-receptive-field mechanism for representing large-span and irregular malformed inflorescence contours. Second, receptive-field attention was incorporated into the neck network by constructing C3k2_RFAConv and RFAConv modules, thereby refining multiscale semantic features and suppressing background noise caused by overlapping branches, leaves, and shadows. Finally, an auxiliary detection branch was added during training to strengthen the learning of small or visually weak inflorescence features and blurred target boundaries through deep supervision. Tests on a self-built field dataset showed that the improved model achieved mAP@0.5 and mAP@0.5:0.95 values of 82.4% and 49.4%, respectively, representing improvements of 5.0 and 3.4 percentage points over the baseline YOLOv11n. The model contained 5.57 M parameters and achieved competitive overall performance among models of comparable scale. These results indicate that LRA-YOLOv11n alleviates feature confusion, missed detection, and false detection in the tested orchard scenes while maintaining a moderate computational cost, providing preliminary algorithmic support for further evaluation of MMD detection in field images. Full article

Accurate brain tumor magnetic resonance imaging (MRI) segmentation is essential for objective lesion assessment and treatment planning, yet it remains challenging because tumors exhibit large-scale variations, irregular morphologies, heterogeneous internal textures, and indistinct boundaries. Conventional U-Net variants commonly improve either spatial feature extraction or skip connection fusion, but they do not jointly address the adaptive representation of pathological texture changes and the semantic filtering of noisy shallow features. To bridge this gap, a texture-span and frequency-gated U-Net (TSF U-Net) is proposed, establishing a collaborative spatial and frequency-domain feature-learning framework. Within this framework, a texture-span gated multi-scale (TSGM) block is developed by integrating parallel multi-scale context extraction with a pathological texture-span gating mechanism derived from local maximum and minimum feature responses. This design enables the encoder to emphasize heterogeneous tumor regions and boundary-related variations without requiring additional boundary supervision. Moreover, a discrete cosine frequency-gated fusion (DCT-FGB) block is introduced into the skip connections. It employs deep semantic features to generate top-down frequency-domain gating masks for shallow skip features, allowing useful structural information to be retained while reducing background and noise interference before decoder fusion. On the Brain Tumor Dataset, TSF U-Net achieves an mIoU of 86.471%, a Recall of 87.424%, an F1-score of 84.693%, and a Dice coefficient of 0.819, improving the U-Net baseline by 1.243 percentage points in mIoU and 1.645 percentage points in F1-score. It also achieves the highest mIoU of 88.882% on the BRICS 2025 Dataset. Ablation results confirm the complementary contributions of TSGM and DCT-FGB. The bidirectional cross-dataset experiments yield mIoU values of 77.369% and 70.881%, providing initial evidence of transferability under dataset distribution shifts. These findings indicate that jointly modeling pathological texture variations in the spatial domain and semantically selecting skip features in the frequency domain provides an effective strategy for improving boundary sensitivity, noise resistance, and cross-dataset robustness in brain tumor MRI segmentation. Full article

Agricultural operations such as pepper harvesting, fruit counting, and field phenotyping rely on accurate visual recognition and instance segmentation algorithms. However, pepper fruits in complex field environments often exhibit slender and curved shapes, partial occlusion, ambiguous boundaries, and adhesion between adjacent instances. Existing object detection and instance segmentation methods therefore struggle to obtain complete fruit masks, which adversely affects subsequent fruit counting, contour measurement, and picking-point localization. To improve the instance segmentation accuracy of occluded peppers in complex field scenes, this study proposes OccPepSeg-YOLO, an improved model based on YOLO11n-seg. First, a P2FreqFusion module is introduced to fuse shallow, high-resolution detail features with deep semantic features, thereby enhancing the representation of fruit edges and tip regions. Second, an ASC module is designed to model the directional and scale-related morphological characteristics of pepper fruits, while a BoundaryGate module strengthens responses at occlusion interfaces and boundaries between adjacent instances. Finally, an OccPepSegment multi-scale prototype segmentation head is constructed, and a BDoU loss function is introduced to improve the boundary consistency of instance masks. Experiments on a self-constructed field-pepper instance segmentation dataset showed that OccPepSeg-YOLO achieved M-P, M-R, M-mAP50, and M-mAP50–95 values of 93.87%, 92.09%, 97.17%, and 82.31%, respectively, representing improvements of 5.59, 3.18, 3.83, and 9.52 percentage points over YOLO11n-seg. Further comparisons with representative YOLO-based instance segmentation models, including YOLOv8n-seg, YOLOv9c-seg, YOLO12n-seg, and YOLOv26n-seg, demonstrated that OccPepSeg-YOLO achieved the best overall segmentation performance. In particular, its M-mAP50–95 exceeded the best competing result obtained by YOLOv9c-seg by 8.35 percentage points. Under a unified repeated-inference protocol on an RTX 3090 GPU using FP32 precision, a batch size of 1, and 640 × 640 inputs, OccPepSeg-YOLO achieved a mean inference latency of 15.801 ± 1.238 ms, a P95 latency of 17.323 ms, and a throughput of 63.29 FPS. These results demonstrate that the proposed model can produce more complete pepper instance masks under leaf occlusion, fruit overlap, and complex background conditions, providing technical support for field-pepper recognition, fruit counting, and visual perception by agricultural robots. Full article

Medical image segmentation is a critical task in computer-assisted diagnosis, but accurate delineation remains challenging in sensor-captured dermoscopic and endoscopic images because of low contrast, blurred boundaries, acquisition artifacts, and large appearance variations. Although CNN-based methods are effective in capturing local texture and boundary cues, they often struggle to explicitly model long-range dependencies and global structural relationships. Transformer-based architectures can capture global context, but their self-attention mechanism may become computationally costly when processing high-resolution feature maps. To address these challenges, we propose MGA-UNet, a frequency-aware multi-scale encoder–decoder segmentation framework that integrates wavelet-based frequency decomposition with Mamba-based long-range dependency modelling. Specifically, the Wavelet-Mamba feature extraction backbone (WMB) decomposes features into low- and high-frequency components to enhance boundary-aware representation, the Gated Multi-scale Aggregation Module (GMAM) aggregates parallel multi-scale encoder features and applies a content-dependent gate to the fused response, and the Adaptive Sparse Attention Module (ASAM) refines bottleneck representations with sparse attention for global semantic modelling. Across three independent runs with random seeds 42, 123, and 2026, MGA-UNet achieves mean Dice Similarity Coefficients of $88.92\pm 0.04\%$, $88.01\pm 0.07\%$, and $85.91\pm 0.04\%$ on ISIC2018, ISIC2017, and Kvasir-SEG, respectively. These results demonstrate competitive segmentation performance among the compared representative CNN-based, Transformer-based, and Mamba-based methods, including the recent H-VMUNet baseline. These results indicate that frequency-domain decomposition and state-space modelling can complement each other for accurate medical image segmentation, particularly in images with ambiguous boundaries and complex background interference. Full article

Real-time event detection from social media has become increasingly important for emergency response, public safety, and situational awareness. However, accurately identifying emerging events from Thai social media remains challenging because Thai is a low-resource language without explicit word boundaries, while social media text is often characterized by informal writing, spelling variations, and noisy user-generated content. This paper proposes an entity-centric real-time event detection framework for Thai social media that employs a Multi-Granularity Thai Character Cluster (TCC)-Aware Named Entity Recognition (NER) model as its core information extraction component. The proposed NER architecture integrates contextual word embeddings, character-level representations, Thai Character Cluster features, and Part-of-Speech embeddings through an attention-based feature fusion mechanism to improve entity recognition under noisy conditions. Recognized entities are subsequently used as semantic anchors for event construction, clustering, temporal trend analysis, event ranking, and alert generation within a unified streaming framework. Event discovery combines Density-Based Spatial Clustering of Applications with Noise (DBSCAN) and Exponential Moving Average (EMA)-based temporal analysis to identify emerging events in real time. Experiments conducted on a large-scale Thai social media corpus demonstrate that the proposed model achieves an F1-score of 94.14% for named entity recognition and 92.2% for downstream event detection, outperforming representative baseline methods. Additional ablation studies, qualitative error analysis, and statistical significance tests confirm the effectiveness of the proposed multi-granularity representation. These results demonstrate that the proposed framework provides an effective solution for real-time event monitoring in low-resource language environments. Full article

Remote sensing semantic segmentation is critical for extracting fine-grained spatial information in applications such as urban planning and environmental monitoring. However, existing methods face significant challenges in modeling heterogeneous spatial relationships within complex urban scenes, where semantically related regions are spatially dispersed yet functionally interdependent. Conventional convolutional neural networks exhibit limited receptive fields that fail to capture long-range dependencies, while Transformer-based approaches capture global dependencies but do not explicitly model regional heterogeneity, leading to blurred boundaries and category confusion. To address these limitations, this paper proposes a novel multi-relational-aware segmentation framework that leverages hypergraph theory to dynamically model higher-order semantic groupings across non-adjacent regions. The core innovation lies in a hypergraph structure learning unit that employs fuzzy clustering to partition multi-scale features into adaptive hyperedge sets, enabling joint representation of topological associations and functional dependencies among spatially distributed entities. Additionally, a multi-scale co-modeling strategy integrates stochastic feature masking with weighted fusion to bridge semantic abstraction and spatial localization. Experiments demonstrate that the proposed method achieves state-of-the-art mIoU performance on the LoveDA, Vaihingen, and Potsdam datasets, obtaining mIoU scores of 54.70%, 85.01%, and 87.64%, with improvements of 0.30%, 0.91%, and 0.08%, respectively, over the best existing methods. Full article

Addressing the challenges posed by the small scale, diverse morphology, and severe occlusion of pest and disease targets on lotus leaves in complex aquatic environments—as well as the difficulty of existing methods in automatically quantifying disease severity—this paper proposes an approach for the identification of lotus leaf pests and diseases and quantitative grading of leaf spot disease severity. First, by integrating high-altitude canopy imagery captured by unmanned aerial vehicles (UAVs) with high-definition ground-level data, a multi-perspective dataset comprising object detection bounding box annotations and pixel-level segmentation annotations is constructed. Second, the YOLOv12n-DFFN object detection model is proposed; this model enhances interaction between deep and shallow features through a dynamic feature feedback mechanism, thereby improving the ability to localize disease targets against complex aquatic backgrounds. Finally, taking typical leaf spot disease as the subject, YOLOv12n-DFFN is used to detect and extract diseased leaf regions, while a VGG-UNet semantic segmentation model is employed to achieve pixel-level fine-grained segmentation of leaf and lesion areas. By calculating the ratio of lesion area to total leaf area, automatic quantitative grading of disease severity is realized. Experimental results show that YOLOv12n-DFFN achieved a precision of 93.58%, representing an improvement of 4.37 percentage points over the YOLOv12n baseline, with an mAP50 of 83.87%; the segmentation model attained an average intersection-over-union of 90.47%, and the overall accuracy of the two-stage framework for leaf spot disease severity grading reached 96.0%. Through a “detection first, segmentation second” two-stage strategy, this framework enables the intelligent identification and severity quantification of lotus leaf diseases in complex aquatic environments, providing an effective approach for the intelligent monitoring and precision management of aquatic crop diseases. Full article

Background/Objectives: Chest radiographs contain hierarchical relationships between global thoracic configuration and organ-specific findings, together with scale-dependent anatomical boundary geometry and heterogeneous texture distributions. Coarse image–report alignment may overlook regional evidence, whereas isolated local alignment can fragment complete-examination semantics. We propose Anatomy-Guided Image–Report Alignment (AGIRA) for fine-grained chest X-ray (CXR) representation learning and investigate its behavior through post hoc finite-scale fractal and multifractal analysis. Methods: AGIRA constructs whole-image, left-lung, right-lung, and heart image–text pairs, adapts frozen vision and language backbones using lightweight Anatomy Sensors, and retains masked image and language reconstruction objectives to preserve global–local semantic continuity. Temporal–multiview fusion and hybrid soft labels reduce false-negative supervision. Post hoc analysis estimates effective box-counting dimensions, generalized dimensions, multifractal spectrum width, and a finite-scale attention concentration index. We additionally compare these descriptors with conventional intensity, texture, and uncertainty features using nested cross-validated incremental-value analyses. Results: AGIRA achieves 92.0% AUC, 84.2% accuracy, and 79.0% F1-score on RSNA Pneumonia, together with 57.5% zero-shot accuracy and 53.7% image-to-text P@5 on CheXpert 5 × 200, while updating only 7.54% of the complete model. On an independently annotated 600-unit report-routing set, the deterministic parser obtains macro-F1 0.953 and exact multi-label routing accuracy 0.918. Under the identical 34,080-record cohort, AGIRA remains above GLoRIA-ViT, MLIP, and BCC on all reported downstream metrics. Adding finite-scale descriptors to conventional intensity/texture/uncertainty features increases cross-validated AUROC for identifying anatomy-alignment-sensitive cases from 0.671 to 0.731 for zero-shot classification and from 0.651 to 0.704 for retrieval. Mild anatomical crop perturbations (±5%) change CheXpert P@5 and zero-shot accuracy by less than one percentage point, whereas severe 10% perturbations and explicit anatomy swaps cause progressively larger degradation. Conclusions: AGIRA improves cross-dataset benchmark transfer and retrieval while preserving parameter efficiency. The post hoc fractal/multifractal analysis contributes measurable incremental explanatory information about anatomy-alignment-sensitive model behavior beyond conventional descriptors without entering model training, and the added controlled experiments support parser reliability, cohort fairness, parameter robustness, and tolerance to moderate anatomical-localization noise. Full article

Accurate detection of sub-millimeter defects in reactor core-plate cotter-pin holes is essential for nuclear safety. However, underwater inspection images often suffer from low signal-to-noise ratios, weak boundary responses, and pseudo-edge interference, resulting in unstable localization of defects. Existing deformable and attention-based detectors remain vulnerable to sampling drift and semantic–boundary inconsistency under such conditions. To address these challenges, an Edge-Geometry-Guided Deformable Detection Network (EGD-Net) is proposed for underwater defect detection. EGD-Net introduces an edge-geometry-constrained deformable sampling mechanism that embeds edge-confidence priors into deformable convolution to improve boundary-aware feature sampling. A cross-level semantic–geometric alignment strategy is designed to enhance the interaction between defect semantics and geometric boundary cues, while a top-down feedback recalibration mechanism improves multi-scale response consistency for weak defects. Experiments on the Core-Plate Pin-Hole Defect (CPHD) dataset demonstrate that EGD-Net achieves the highest AP@[0.5:0.95] on both datasets while maintaining competitive or superior Precision, Recall, and F1-score while reducing engineering center error under a fixed operating point. Performance across the two complementary domains suggests its robustness to variations between coupon images and practical underwater inspection scenes. These results indicate that EGD-Net provides a reliable solution for boundary-sensitive localization of underwater sub-millimeter defects in nuclear inspection. Full article

Pixel-level annotation of bi-temporal high-resolution imagery is costly because annotators must distinguish genuine changes from pseudo-changes caused by illumination, seasonality, shadows, and residual misregistration. From a temporal-symmetry perspective, unchanged regions approximately preserve cross-temporal semantic correspondence, whereas genuine building changes introduce localized symmetry breaking between the two acquisition times. This paper presents BSCNet, a semi-supervised framework for binary building change detection that jointly models boundary-sensitive differences and scene-dependent scale preferences. A shared-weight MixTransformer extracts multi-level bi-temporal features. The Edge-Aware Optimization Module suppresses spatially invariant channel responses, enhances residual spatial cues, and predicts a Sobel-supervised edge map. The Parallel Selective Context Module aggregates depthwise-separable branches with different receptive fields and produces an image-level scale distribution. The Multi-scale Edge-Consistent Mean Teacher framework aligns the final prediction, intermediate edge representation, and scale-selection distribution between an exponential-moving-average teacher and the student. Experiments on WHU-CD and LEVIR-CD under 5%, 10%, and 20% labeled-data settings show consistent improvements over RCL, C2F-SemiCD, and CutMix-CD. With 5% labeled data, BSCNet achieves F1/IoU scores of 88.57%/79.49% on WHU-CD and 88.88%/79.98% on LEVIR-CD. An additional UAV-CD evaluation examines transfer to 0.06 m low-altitude UAV imagery containing both building and land changes; under 5% supervision, BSCNet obtains an F1/IoU of 68.07%/51.60%. Progressive ablations confirm complementary gains from the boundary, scale, and consistency components. Full article