Search Results (218)

Extremely low overlap leads to severely scarce local geometric correspondences across frame pairs. Pure geometric descriptors—encoding merely low-level shape signatures—inherently fail to impose sufficient constraints for reliable transformation estimation when matches become critically sparse, rendering registration fundamentally fragile. While recent red-green-blue-depth (RGB-D) attempts have explored visual augmentation, they predominantly rely on low-level chromatic statistics or shallow convolutional neural network (CNN) features, underutilizing the rich hierarchical semantics inherent in RGB imagery. We present V3Reg, a robust registration framework that pioneers the integration of large-scale vision foundation models (DINOv3) with adaptive cross-modal fusion. Specifically, we extract mid-to-deep semantic features (Layer 11) from DINOv3 to transcend low-level texture limitations, and propose a Task-Aware Channel-Wise Gated Adaptive Fusion (TACGAF) module that dynamically calibrates geometric-visual contributions via registration-error-guided channel-wise gating. To rigorously evaluate ultra-low-overlap robustness, we reconstruct RGBD-ZeroMatch, a benchmark with controllable overlap ratios ranging from 1% to 20%. Extensive experiments demonstrate that V3Reg achieves 99.6% Feature Matching Recall and 96.3% Registration Recall on standard benchmarks. Notably, it maintains 50.2% Registration Recall at merely 5% overlap, outperforming prior methods by over 18 percentage points. Full article

Modern power systems are becoming increasingly complex due to the rapid integration of renewable energy sources, the widespread use of nonlinear power-electronic devices, and the deployment of microgrids operating in parallel with conventional power grids. These evolving conditions intensify the occurrence of diverse and highly complex power quality disturbances (PQDs), demanding accurate and computationally efficient monitoring strategies. This paper presents a novel multi-stage hierarchical framework for PQD detection and classification, comprising an initial training stage with a dedicated 1D Convolutional Neural Network (1D-CNN), a transfer learning stage, and a subsequent fine-tuning stage. The proposed approach operates directly on raw voltage waveforms, eliminating the need for any signal preprocessing, as the CNN performs internal feature extraction. The framework is evaluated using a comprehensive dataset that includes synthetic signals, Matlab/Simulink (version R2022a) time-domain simulations, and real voltage sag events. Additionally, up to 29 types of disturbances, including complex multi-event combinations defined by the IEEE-1159 Standard, are generated using the PQ-SyDa toolbox. The proposed model achieves an F1-score of 97.8% using a three-cycle analysis window and further improves to 98.86% when five cycles are used. These results highlight the robustness and generalization capability of the proposed approach for the real-time PQD monitoring task in modern electrical networks. Full article

Microbial classification and taxonomic information are fundamental to microbiological studies. Raman microspectroscopy, a rapid and non-destructive single-cell analytical technique, captures intrinsic molecular fingerprints reflecting cellular biochemical composition, thereby enabling microbial classification at the single-cell level. However, current Raman-based classification frameworks allow accurate identification only for micro-organisms already represented in reference databases. These approaches often fail or yield errors for uncharacterized microorganisms. To address this limitation, we collected 6600 single-cell Raman spectra from 11 microbial species, including bacteria and fungi, and developed deep learning models for rapid classification. A hierarchical clustering (HC) framework based on Raman features extracted by a one-dimensional convolutional neural network (1D-CNN) was constructed and compared with phylogenetic trees derived from rRNA gene sequences. 1D-CNN achieved high classification performance with an overall accuracy of 99.7%. Notably, the Raman HC tree exhibited clear concordance with phylogenetic structures, particularly at the higher taxonomic levels. Validation using five independent unknown strains demonstrated that the Raman HC tree consistently positioned these strains near their closest phylogenetic relatives, in strong agreement with sequence-based analyses. Collectively, these findings highlight the potential of single-cell Raman spectroscopy with deep learning as an alternative and complementary framework for microbial taxonomic analysis, particularly for previously uncharacterized microorganisms. Full article

Background: Accurate segmentation of white blood cells (WBCs) in peripheral blood smear images is a fundamental step in computational hematology, enabling downstream tasks such as classification, morphological assessment, and quantitative analysis. However, reliable segmentation remains challenging due to staining variability, complex cellular morphology, overlapping structures, and limited availability of high-quality annotations. Aim and Methods: The aim of this study is to develop a robust and fully automated segmentation framework for white blood cells (WBCs) in microscopic blood smear images, providing a reliable foundation for subsequent computational analysis. We propose TransNet–SAM2, a hybrid deep learning architecture that integrates hierarchical transformer-based feature extraction with a foundation-model-based decoder for prompt-free segmentation. Specifically, a Swin Transformer backbone is employed to capture multi-scale contextual representations, which are subsequently aligned and fused through a feature adaptation module. The fused features are directly injected into the SAM2 mask decoder, replacing conventional prompt-based conditioning and enabling fully automatic segmentation. In addition, a weakly supervised self-training strategy is incorporated to utilize partially annotated data, improving model generalization while reducing annotation requirements. The proposed framework is evaluated using a clinically curated dataset from Dicle University, the publicly available Raabin-WBC dataset, and an additional external leukemic blast validation dataset (ALL-IDB) to assess robustness under both routine and atypical hematological conditions. Results: TransNet-SAM2 achieved a Dice coefficient of 0.95 ± 0.01 and IoU of 0.90 on internal testing, significantly outperforming U-Net (0.89), Mask R-CNN (0.90), and SAM2 (0.92) (p < 0.05). In cross-dataset evaluation (Dicle training, Raabin-WBC testing), the framework maintained strong performance (Dice: 0.91, IoU: 0.84), demonstrating robustness to domain shifts. Ablation studies confirmed each component’s contribution, with the full model improving Dice by 6% over a CNN baseline. Qualitative analysis showed accurate boundary delineation even with cell overlap and background clutter. Conclusions: These findings indicate that the proposed framework provides a promising and scalable framework for WBC segmentation. While the current study focuses on segmentation, future work will investigate integration with classification and radiomics pipelines, as well as validation on more diverse clinical datasets, including bone marrow and leukemia samples. Full article

Background/Objectives: To evaluate the effect of detector architecture and dataset characteristics on intracranial hemorrhage (ICH) subtype localization on noncontrast head CT, with emphasis on bidirectional cross-dataset generalization. Methods: This retrospective study analyzed two publicly available datasets: the Brain Hemorrhage Extended (BHX) dataset and the RSNA 2019+ dataset. Models were trained and internally validated on one dataset and externally tested on the other dataset in both directions: BHX-to-RSNA+ and RSNA+-to-BHX. Six representative deep learning detectors, including CNN-based one-stage and two-stage detectors and a Swin Transformer-based RT-DETR (Swin-RT-DETR) variant, were evaluated. Localization performance was assessed using mean average precision at a bounding-box intersection-over-union threshold of 0.5 (mAP@50), bounding-box Dice similarity coefficient (BB-DSC), and bounding-box intersection-over-union (BB-IoU). Image-level and patient-level analyses were performed, with Bonferroni correction applied for statistical comparisons. Dataset characterization analyses were performed to compare subtype prevalence, bounding-box geometry, lesion burden, annotation density, and spatial distribution. Results: Under internal validation, Swin-RT-DETR achieved competitive or superior performance across several ICH subtypes, but its advantage was subtype-dependent rather than uniform. Faster R-CNN with a ResNeXt101 backbone achieved comparable IVH performance and higher IPH BB-DSC and BB-IoU, whereas Swin-RT-DETR performed better for SAH, SDH, and EDH. External validation showed substantial performance degradation across architectures, subtypes, and validation directions. Absolute BB-DSC reductions for Swin-RT-DETR ranged from approximately 0.54–0.79 in the BHX-to-RSNA+ direction and 0.17–0.74 in the RSNA+-to-BHX direction. Similar degradation patterns were observed at the patient level. Statistical comparisons showed fewer significant model-level differences under external validation, suggesting attenuation of architecture-specific advantages under domain shift. Dataset characterization analysis demonstrated differences in subtype distribution, bounding-box geometry, lesion burden, annotation density, and spatial localization patterns between BHX and RSNA+. Conclusions: ICH subtype localization performance is strongly influenced by dataset characteristics, annotation heterogeneity, and domain shift. Although Transformer-based hierarchical feature extraction showed subtype-dependent advantages under internal validation, these advantages diminished under bidirectional external validation. These findings highlight the need for dataset characterization, external validation, patient-level evaluation, and task-specific clinical benchmarks before automated ICH localization models can be considered for real-world clinical integration. Full article

Accurate bus passenger flow prediction is a critical prerequisite for optimizing resource allocation in intelligent bus systems and improving public transport service quality. However, due to dynamic fluctuations in passenger travel demand and the complex topological structure of bus networks, effectively capturing the inherent spatiotemporal dependencies of passenger flow remains a significant challenge. To address this issue, this paper proposes GAT-Informer, a hybrid deep learning model for short-term bus passenger flow prediction. Unlike conventional methods that mainly rely on fixed physical adjacency or single-view spatial correlations, the proposed model incorporates domain knowledge into graph construction through a four-dimensional associated-node mechanism. Specifically, four types of features, namely line connectivity, spatial proximity, OD correlation, and station functional similarity, are used to identify relevant associated nodes and construct a passenger flow interaction spatiotemporal graph that better reflects inter-station dependencies. Based on this graph, a Graph Attention Network (GAT) is introduced to adaptively learn spatial interaction features and differentiated influence weights among associated stations. The spatially enhanced features extracted by GAT are then fed into the Informer network, where probabilistic sparse self-attention and hierarchical timestamp encoding are employed to efficiently capture long-term temporal dependencies and periodic fluctuation patterns of passenger flow. Experimental results based on bus passenger flow data from Foshan City show that the proposed GAT-Informer model significantly outperforms benchmark models, including LSTM-Transformer, CNN-GRU, and STGCN, across different prediction horizons, validating its effectiveness and improved predictive performance in bus passenger flow prediction. Full article

Parkinson’s disease (PD) is the second most prevalent neurodegenerative disorder worldwide, and its early diagnosis remains a major challenge due to reliance on subjective clinical assessments. This study proposes a bio-inspired computational framework for automatic PD detection that draws explicit architectural inspiration from two biological systems: the hierarchical tonotopic organization of the human auditory cortex, which motivates the design of a 1D Convolutional Neural Network (CNN) for vocal biomarker analysis, and the basal ganglia–cerebellar motor control circuit, which motivates the selection and design of features extracted from Archimedean spiral drawing tasks. Unlike previous studies that apply standard machine learning techniques without grounding architectural choices in biological mechanisms, the proposed framework establishes a direct mapping between neural processing pathways and model design decisions. A Support Vector Machine (SVM) classifier evaluated on the Kaggle vocal dataset achieved 87% test accuracy with no overfitting, outperforming AdaBoost, Random Forest, KNN, XGBoost, and Decision Trees in terms of generalization. The 1D CNN applied to UCI spiral drawing data achieved 85% test accuracy, with overfitting behavior addressed through architectural regularization strategies including early stopping. A conceptual multimodal fusion architecture integrating both modalities is proposed as a direction for future experimental validation; it was not implemented or experimentally validated within the present study. The primary novelty of the framework resides in this explicit biomimetic grounding, which distinguishes it from existing performance-driven approaches. Results confirm that biologically grounded computational models constitute promising objective decision-support tools for early PD diagnosis. Full article

In tennis, where performance is governed by complex kinetic chain interactions, objective skill classification is vital for coaching and talent identification. This study presents a hierarchical deep learning framework leveraging synchronized bilateral Inertial Measurement Unit (IMU) data from 39 participants (11 elite, 28 amateur). The proposed system successfully distinguishes expertise levels across a total of 4594 strokes, including augmented samples. A hybrid Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-BiLSTM) architecture was developed to autonomously extract spatiotemporal features from the raw kinematic signals of forehand, backhand, service, and volley strokes. The proposed model achieved an accuracy of 95.54%, significantly outperforming both traditional machine learning and state-of-the-art deep learning benchmarks. Qualitative t-distributed Stochastic Neighbor Embedding (t-SNE) analyses revealed that elite athletes form highly homogeneous clusters in the feature space. Furthermore, quantitative Asymmetry Index assessments confirmed that professionals exhibit superior bilateral coordination stability. These findings demonstrate that the proposed end-to-end system offers a robust, field-applicable solution for identifying technical excellence. It provides coaches with reliable digital biomarkers, thereby overcoming the limitations of subjective visual observation. Full article

The core challenge of high-precision medical image segmentation lies in modeling the multi-scale fractal self-similarity of human abdominal organs and cardiac structures, especially the blurred boundaries and weak features of low-contrast tissues. Existing CNNs and Transformers fail to simultaneously capture both global fractal topology and high-frequency fractal details, thereby limiting segmentation performance. To address this, we propose MIS-DFH, a dual-branch CNN–Transformer hybrid model that integrates Hybrid Feature Branches, Multi-Fusion Dense Frequency Skip Connections, and hierarchical Deep Supervision, achieving superior multi-scale feature extraction and segmentation performance. Experiments on the Synapse abdominal CT and ACDC cardiac MRI datasets show that MIS-DFH outperforms all compared state-of-the-art methods. Notably, it achieves 79.90% mean DSC and 20.06 mm HD95 on Synapse, representing a 5.2% DSC improvement and 34.3% HD95 reduction over MSLAU-Net, with consistent gains on ACDC. These results validate the model’s superior segmentation accuracy and clinical application value. Full article

Change detection in open-pit mining areas from remote sensing imagery is of great importance for mining supervision, ecological monitoring, and restoration planning. Nevertheless, mining-related changes usually exhibit multi-scale patterns, irregular boundaries, and fragmented spatial distributions, which make accurate detection difficult. Existing CNN- and Transformer-based methods often cannot effectively balance global context perception and local detail preservation, resulting in incomplete boundary extraction and insufficient sensitivity to subtle changes. To overcome these limitations, we propose GLMECD-Net, a Global–Local Multi-scale Cross-fusion Enhanced Change Detection Network for remote sensing image change detection in open-pit mining areas. Specifically, a Siamese encoder is used to extract hierarchical bi-temporal features, while a Global–Local Feature Mixing Embedding (GLME) module is introduced to jointly capture long-range contextual information and local spatial details. Furthermore, multi-scale feature aggregation and cross-temporal feature fusion are employed to improve change representation and boundary recovery. Experimental results on mining area datasets show that the proposed method achieves 71.66% Precision, 83.78% OA, 77.53% F1-score, and 53.82% IoU. The results demonstrate that GLMECD-Net provides effective and robust performance for detecting complex and subtle changes in open-pit mining areas. Full article

In impedance imaging, the incompatibility and nonlinearity of the inverse problem lead to problems such as blurred boundaries and severe artifacts in the reconstructed images, making it difficult to meet the requirements for precise identification of multi-layer tissue structures in the legs. To this end, this paper proposes a post-processing algorithm for leg EIT that integrates the boundary attention mechanism, with a Wasserstein generative adversarial network as the training framework, cyclic residual U-Net as the generator, and the boundary attention module embedded in the RecurrentBlock. This leads to adaptive enhancement of the ability to extract organizational boundary features through a three-path fusion of spatial attention, channel attention, and learnable Laplacian edge enhancement. A leg anatomy prior constraint loss function was designed, integrating six constraints—pixel loss, edge loss, hierarchical tissue constraint, total variation regularization, structural similarity loss, and histogram matching—to guide the reconstruction results to conform to the multi-layered tissue structure features of the leg. A simulation dataset of leg sections containing multiple tissues such as skin, fat, muscle, bone, blood vessels, and nerves was constructed, and the pre-reconstructed images were obtained using the hybrid total variation regularization algorithm as the network input. The simulation results show that, under noise-free and different signal-to-noise ratio conditions, the proposed BAM-R2UNet algorithm achieves the best performance in RMSE, SSIM and PSNR metrics compared with HTV, DnCNN and standard U-Net algorithms, can remove artifacts, accurately restore the boundary and conductivity distribution of leg tissues, and has stronger anti-noise robustness. Full article

Accurate forecasting of agricultural product prices is crucial for informed decision-making in agricultural markets; however, such time series are inherently characterized by non-stationarity, multi-scale dynamics, and substantial noise, posing significant challenges to conventional methods. To overcome these limitations, this study proposes a novel hybrid framework, termed TOC-CNN-BiLSTM-SA, built upon a “quadratic decomposition–clustering–optimization” paradigm. Specifically, a composite CEEMDAN–K-means++–VMD approach is first employed to hierarchically decompose the raw price series via coarse decomposition, feature clustering, and refined decomposition, enabling effective noise suppression and multi-scale feature extraction. Subsequently, a deep learning architecture integrating Convolutional Neural Networks (CNNs), Bidirectional Long Short-Term Memory networks (BiLSTM), and a self-attention mechanism is developed, where CNN captures local patterns, BiLSTM models bidirectional temporal dependencies, and the attention mechanism enhances global feature representation. Furthermore, the Tornado Optimizer with Coriolis force (TOC) is introduced to adaptively tune key hyperparameters, thereby improving model robustness and generalization capability. Empirical results based on wheat price data from Henan Province, China, demonstrate that the proposed model achieves outstanding predictive performance, with RMSE, MAE, MAPE, and R² values of 4.425, 3.9372, 0.16%, and 99.97%, respectively, significantly outperforming existing benchmark models. These research indicate that the proposed framework effectively captures complex price dynamics and offers a reliable and practical solution for agricultural price forecasting. Full article

Haul roads in surface mining are critical infrastructure directly influencing operational productivity, safety, and costs. However, these networks change frequently due to ongoing mining activities, making traditional mapping methods impractical for large-scale or rapidly evolving sites. Remote sensing imagery offers a scalable alternative, yet complex backgrounds, variable road widths, and spectral similarities between roads and surrounding surfaces make accurate extraction challenging. This study proposes HRM-Net, a hybrid transformer–CNN autoencoder framework for automated extraction of mine haul roads from remote sensing imagery. HRM-Net introduces inception-like patch embedding to capture local contextual information and employs a manifold-constrained hyper-connection strategy in the attention and fusion blocks to enhance information flow across the architecture. This hierarchical design enables progressive learning of discriminative semantic representations across multiple spatial resolutions, critical for road extraction in cluttered mining environments. Trained and evaluated on diverse mine sites, HRM-Net achieved 92.53% overall accuracy, 85.12% F1-score, 75.57% mIoU, 83.57% precision, and 86.94% recall, outperforming state-of-the-art transformer-based and CNN-based segmentation models. Furthermore, model interpretability was analyzed through linear probing and boundary alignment evaluations. Results demonstrate that discriminative features emerge at early network stages and are effectively preserved throughout the architecture, while boundary predictions exhibit superior consistency compared to existing approaches. Full article

Skin cancer represents a significant public health challenge, necessitating early detection and timely treatment for optimal management. Timely and accurate evaluation of skin lesions is crucial, as delays can lead to more severe outcomes. However, identifying skin lesions accurately can be challenging due to differences in color, shape, and the various types of imaging equipment used for diagnosis. While recent studies have demonstrated the potential of ensemble convolutional neural networks (CNNs) for early diagnosis of skin disorders, these models are often too large and inefficient for processing contextual information. Although lightweight networks like MobileNetV3 and EfficientNet have been developed to reduce parameters and enable deep neural networks on mobile devices, their performance is limited by inadequate feature representation depth. To mitigate these limitations, we propose a new hybrid attention dual-stream deep learning model for skin lesion detection. Our model uses one training process to preprocess the images and splits the task into two branches. Each branch extracts different features using multi-stage and multi-branch attention techniques, improving the model’s ability to detect skin lesions accurately. The first branch processes the original image using a convolutional layer integrated with three novel attention modules: Enhanced Separable Depthwise Convolution (SCAttn), stage attention, and branch attention. The second branch utilizes Contrast Limited Adaptive Histogram Equalization (CLAHE) to enhance the input image, improving local contrast and revealing finer details. The integration of CLAHE with SCAttn modules leverages enhanced local contrast to capture more nuanced features while maintaining computational efficiency. A classification module receives the concatenated hierarchical characteristics that were taken from both branches. Utilizing the PAD2020 and ISIC 2019 datasets, we assessed the proposed model and obtained an accuracy rate of 98.59% for PAD2020, surpassing the state-of-the-art performance by 2%, and stable performance accuracy for the ISIC 2019 dataset. This illustrates how well the model can integrate several attention mechanisms and feature enhancement methods, providing a reliable and effective means of detecting skin cancer. Full article

Deep learning models represent one of the most advanced and effective approaches in predictive modeling. Their hierarchical architectures enable the extraction of complex, non-linear feature relationships and the identification of latent patterns within data, making them highly suitable for tasks involving high-dimensional or unstructured inputs. However, these models are computationally demanding, requiring significant processing resources and time. Furthermore, their predictive performance is largely contingent upon the availability of large-scale datasets. In this study, a Deep Green Framework is employed for the prediction of two computer vision tasks. CIFAR-10 and CIFAR-00 have been taken for image classification. Fifteen convolutional neural network (CNN) variants categorized into light-weight and heavy-weight are trained for the prediction of these two datasets. Based on energy footprint, time, memory usage, Top-1 accuracy, Top-3 accuracy, model size, and model parameters. The study highlights that MobileNetV3-Small produces the best outcomes when compared to other trained models having low task latency and higher efficiency, making it highly suitable for edge environments where resources are scarce. Full article