Search Results (59)

Building extraction from disaster scenes is critical for emergency response and post-disaster assessment. Unlike conventional static remote sensing imagery, multi-phase disaster imagery contains scenes spanning early, middle, and late disaster stages, where building morphology, class distribution, and boundary characteristics exhibit significant cross-phase heterogeneity. Such phase-dependent variations substantially increase the difficulty of stable semantic segmentation, particularly under complex damage conditions. To address these challenges, we propose MSS-MambaNet for building extraction from multi-phase disaster imagery. A multi-scale architecture is designed to overcome the limitations of single-scale scanning in Mamba, enabling more effective perception of diverse building morphologies. To enhance feature discrimination, a Dual-Domain Cross-Gated Fusion (DDCGF) module is introduced through complementary interactions between spatial and frequency-domain representations. In addition, a Pixel-Aware Dynamic Weighting (PADW) strategy is developed to adaptively emphasize imbalanced foreground pixels and ambiguous boundary regions, thereby improving segmentation consistency under complex disaster conditions. Extensive experiments demonstrate that MSS-MambaNet consistently outperforms state-of-the-art methods, achieving an average mIoU of 92.78% and mF1 of 96.25% with only 12.37 M parameters. These results indicate that the proposed method effectively handles the heterogeneity of multi-phase data, providing a stable and efficient solution for building extraction from multi-phase disaster imagery. Full article

Accurate vessel traffic flow prediction provides an important data basis for intelligent shipping management, including maritime traffic monitoring, navigational risk awareness, waterway organization, and emission-related assessment. Although recent studies have advanced spatiotemporal, graph-based, and hybrid forecasting methods, improving the predictive ability of a conventional ConvLSTM backbone without introducing substantially more complex model structures remains underexplored in grid-based waterway scenarios. This study proposes a lightweight CA-ConvLSTM framework for grid-level vessel inflow and outflow prediction. AIS-derived flow data and MERRA-2 meteorological variables are rasterized onto a common spatial grid and fused at an early stage. A residual dilated convolution module with dilation rates of 1, 2, and 4 is used to extract multi-scale spatial dependencies, and a channel attention mechanism is applied before ConvLSTM-based temporal prediction to adaptively reweight the fused flow-meteorological feature channels. Experiments using AIS and MERRA-2 data from the northern Bohai Strait waterway show that the proposed framework improves baseline ConvLSTM performance. Compared with ConvLSTM, CA-ConvLSTM reduces MSE and MAE by 24.93% and 12.55% for outflow prediction, and by 24.80% and 12.82% for inflow prediction. These results suggest that spatially aligned meteorological fusion, multi-scale spatial feature extraction, and channel-wise feature weighting can effectively enhance ConvLSTM-based grid-level vessel traffic flow prediction without relying on complex model fusion or heavy graph-based architectures. Full article

Background/Objectives: Early and accurate discrimination of neurological conditions, dementia, stroke and healthy aging, remains a critical clinical challenge. Electroencephalography (EEG) is a non-invasive measure of brain dynamics and entropy-based features obtained from multichannel EEG have shown strong discriminative ability. However, existing deep learning approaches do not sufficiently address the combined challenges of small clinical cohorts and high-dimensional entropy feature spaces. In this study, a novel architecture is proposed for multi-class neurological EEG classification under extreme small-sample conditions. Methods: A novel dual-branch Channel-wise Transformer and Attention-Branch Network (EEG-ChTABNet) are pr to classify 19-channel EEG entropy features into three classes (dementia, stroke, healthy control; N = 45; 15 per class). The architecture suggests four new designs. First, the Channel Importance Attention (CIA) block, which adaptively learns to re-weight the importance of electrodes via squeeze-excitation. Second, the dual-branch encoder, which combines the global multi-head self-attention with the local depthwise-separable convolution. Third, the gated sigmoid fusion mechanism. Fourth, the bottleneck residual classification head, to solve overfitting. Eight entropy feature sets: Amplitude-Aware Permutation Entropy (AAPE), Attention Entropy (AttEn), Dispersion Entropy (DisEn), Distribution Entropy (DistrEn), Fluctuation-based Dispersion Entropy (FDispEn), Fuzzy Entropy (FuzEn), Linear Gaussian Estimation of the Conditional Entropy (LinEn), and Symbolic Dynamics (SyDy) were evaluated individually with stratified 5-fold cross-validation on within-fold SMOTE augmentation. Results: EEG-ChTABNet consistently outperformed the baseline Transformer on all 8 feature sets. DisEn and SyDy features yielded peak classification accuracy of 73.3% (AUC: 0.823 and 0.857, respectively) compared to the corresponding baseline of 57.8% and 55.6%. SyDy achieved the best overall AUC of 0.857 and the dementia detection sensitivity was up to 86.7% over multiple feature sets. Conclusions: EEG-ChTABNet shows the effectiveness of channel-adaptive, dual-branch Transformer Designs for EEG-based neurological classification from Small-Sample Entropy Feature Data, and Identifying SyDy and DisEn as the Most Discriminative Feature Representations for Three-Class Neurological EEG Classification. Full article

Remote tower operations require air traffic controllers to maintain continuous visual monitoring and integrate information from panoramic displays, radar data, flight strips, and voice communication. Such screen-mediated and sustained surveillance tasks may lead to covert fatigue, which is difficult to capture using a single physiological or behavioral signal. To address this issue, this study proposes a Gated EEG–Eye Fusion Network (GEEF-Net) for window-level fatigue detection in remote tower controllers. EEG and eye-tracking signals were synchronously collected during simulated remote tower tasks and segmented into 5 s windows with a 2 s step. For each window, 53 EEG features and 47 eye-tracking features were extracted to construct a 100-dimensional multimodal representation. GEEF-Net adopts a lightweight modality-gating mechanism to adaptively weight EEG and eye-tracking representations before fatigue classification. Under the main subject-dependent validation setting, GEEF-Net achieved an Accuracy of 0.883, an F1-score of 0.788, and a ROC-AUC of 0.944, outperforming EEG-only, eye-only, and early-fusion baselines in most overall metrics. The gating analysis indicated that eye-tracking features received a higher average weight than EEG features, suggesting the importance of visual behavior in remote tower fatigue detection. Cross-subject validation showed that individual differences remain a major challenge, while few-shot subject-specific calibration improved model adaptation when limited target-subject samples were available. These findings suggest that EEG–eye-tracking fusion with lightweight modality gating is a feasible approach for fatigue detection in simulated remote tower tasks. However, larger datasets and operationally realistic validation considering shift work, circadian effects, and operational pressure are still required before the approach can be considered operationally reliable. Full article

Mechanical Wear particle recognition is an important approach for equipment health monitoring and fault early warning. However, flow-field disturbances and high-speed particle motion in high-speed fluid environments can lead to image degradation, non-stationary electrostatic signals, and insufficient reliability of single-source recognition methods. Therefore, this study proposes a wear particle recognition method based on multi-source information fusion for high-speed fluid environments. The method establishes a multi-scale electrostatic sensing model to characterize the coupling relationship among particle material properties, motion states, and electrostatic response characteristics. Empirical mode decomposition and independent component analysis are combined for adaptive electrostatic signal denoising, and a Transformer network is used to extract multi-domain features. Meanwhile, an ECA-CNN model with an efficient channel attention mechanism is introduced to enhance the feature representation of degraded particle images. On this basis, a meta-learning-based sample-adaptive decision fusion framework is developed to achieve dynamic and complementary fusion of electrostatic and visual information. The experimental results demonstrate that the proposed method exhibits excellent recognition accuracy and robustness in the tested high-speed fluid environment of 10 m/s, achieving a fusion recognition accuracy of 96.0%, which is significantly superior to single-source recognition methods. Ablation experiments further show that removing the global scaling factor, guidance loss, interpolation loss, and category-specific weight generator decreases the average recognition accuracy by 0.7%, 1.2%, 0.4%, and 1.8%, respectively, confirming the contribution of each key module to fusion recognition performance. These findings provide a new technical approach for the online intelligent recognition of wear particles under high-speed fluid conditions and offer theoretical support and methodological guidance for condition monitoring, health assessment, and intelligent operation and maintenance of large-scale equipment. Full article

Cervical cancer, primarily caused by human papillomavirus (HPV) infection, remains a major cause of cancer-related mortality among women worldwide, making early detection through cytological screening essential. However, manual analysis of cytology images is time-consuming and subject to variability, while recent deep learning approaches, particularly transformer-based architectures, often require high computational resources, limiting their use in resource-constrained settings. In this study, we propose ECAViTNet, a lightweight hybrid CNN–Transformer architecture for cervical cytology image classification that balances accuracy and efficiency. The model integrates Efficient Channel Attention modules for adaptive feature recalibration, residual connections for stable optimization, MobileViT blocks to capture local and global dependencies, and gated multi-scale fusion mechanisms to enhance feature representation, along with progressive downsampling and skip connections to preserve fine-grained details. The proposed approach was evaluated on the SIPaKMeD dataset, achieving a test accuracy of 96.42% with only 982,491 parameters and a macro-average F1-score of 0.96 and a weighted-average F1-score of 0.96, while maintaining balanced class-wise performance and reduced computational cost compared to recent methods. These results demonstrate that ECAViTNet is an effective and efficient solution for automated cervical cytology classification, with strong potential for deployment in mobile health systems and low-resource clinical environments. Full article

Accurate recognition and segmentation of seafloor topographic units is of great significance for marine surveying and engineering applications. Efficient segmentation of multibeam bathymetric point clouds typically requires projecting them into two-dimensional images. However, segmentation methods based on single-modality images suffer from incomplete information representation and insufficient model adaptability, which often lead to blurred boundaries, false positives, and missed detections, thereby limiting segmentation accuracy. To address these challenges, this study proposes a seafloor topography recognition and segmentation method based on YOLO11n-seg with bimodal image feature fusion, from the perspectives of image generation and model optimization, aiming to improve segmentation accuracy and robustness. First, an early fusion strategy for bimodal images is adopted. Two types of images generated from point clouds via continuous curvature tension spline interpolation are concatenated at the input level, fusing local texture details with absolute water depth information, thereby enhancing the model’s ability to perceive topographic features. Second, a lightweight Efficient Channel Attention (ECA) module is embedded after the Spatial Pyramid Pooling-Fast (SPPF) module of the backbone network. This module adaptively calibrates channel weights, reinforcing the contribution of the grayscale channel to the final segmentation decision. Finally, a weighted BCE-Dice joint loss function is constructed to mitigate class imbalance between flat seabed and topographic regions, while also optimizing boundary segmentation accuracy. Experimental results on a self-constructed multibeam image dataset demonstrate that the proposed method achieves an mAP@50 of 92.8%, representing an absolute improvement of 7.6 percentage points over the baseline model. Notably, the model has only 2.84 M parameters, maintaining a lightweight profile. Full article

Coal-mine-gas-monitoring data exhibits missing phenomena due to the harsh underground operating environment. Accurate imputation of missing values in gas-monitoring sequences serves as a key data foundation for guaranteeing the continuity of gas data, enhancing the reliability of disaster early warning, and improving the accuracy of mine safety situation analysis and judgment. Aiming at the prevalent random and segmented missing issues in coal-mine-gas-monitoring time-series data, and the limitation that existing imputation methods struggle to accurately capture the nonlinear spatiotemporal correlations and long-range temporal dependencies of such data, this study proposes a missing data imputation method for coal mine gas time-series data based on the Spatio-Temporal Graph Attention Network—Echo State Network (ST-GAT-ESN). Firstly, this method extracts temporal features of the gas concentration sequence using a Gated Recurrent Unit (GRU). Subsequently, it models multiple monitoring points as graph nodes through a Graph Attention Network (GAT), constructs an adjacency matrix based on airflow propagation relationships, and adaptively learns the spatial dependency weights between monitoring points to realize the deep fusion of spatiotemporal features. Finally, it designs a dual-channel Echo State Network (ESN), synchronously inputs the spatiotemporal fusion features of the missing regions before and after, efficiently fits the nonlinear evolutionary trend of the data by virtue of the echo state property of the reservoir, and solves the output layer weights through ridge regression to achieve accurate imputation of missing values. Experimental results demonstrate that, compared with the single-ST-GAT-ESN, ESN, and ARIMA models, the proposed method achieves the optimal imputation performance in both random and segmented missing scenarios within the missing rate range of 5–50%. The three evaluation metrics—Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE)—are reduced by 30–80% compared with the benchmark models. Moreover, the imputation curve achieves the best fitting performance with the ground-truth curve at a 50% segmented missing rate. This study confirms that the ST-GAT-ESN model effectively enhances the adaptability and robustness to complex missing patterns via spatiotemporal collaborative modeling and a dual-channel fusion mechanism, providing a high-precision and highly stable technical solution for ensuring the integrity of coal-mine-gas-monitoring data, and also provides theoretical references and engineering insights for the missing-value processing of industrial time-series monitoring data. Full article

Accurate forecasting of energy consumption in sensor-rich environments remains challenging due to strong inter-sensor dependencies, temporal variability, and heterogeneous sensor behavior. This paper proposes a lightweight and interpretable multi-sensor fusion framework for short-term energy consumption forecasting. The heterogeneous sensor dataset is first preprocessed to handle missing values, outliers, and temporal misalignment, followed by synchronization of the multivariate signals on a common timeline to enable consistent learning. The proposed framework systematically investigates multiple strategies for exploiting information from synchronized multi-sensor data without performing explicit feature elimination or time-lag engineering. In particular, three fusion paradigms are considered: (i) Early Fusion, where all sensor measurements are jointly used as input features for a multivariate regression model; (ii) Late Fusion, where individual sensor predictors are trained independently and their outputs are combined using reliability-based weighting; and (iii) an attention-inspired fusion strategy, in which adaptive weights are assigned to sensor-level predictions based on their predictive reliability estimated from training errors and normalized via a softmax function. In addition, classical machine learning models including Random Forest (RF), Support Vector Regression (SVR), and Gradient Boosting (GB) are evaluated under the same experimental conditions to provide a consistent benchmark. Experimental results on a real-world building energy monitoring dataset consisting of nine heterogeneous sensors demonstrate that multi-sensor fusion approaches consistently improve forecasting performance compared to single-model baselines. Among the evaluated strategies, Late Fusion provides stable performance across strongly correlated loads, while the attention-inspired fusion strategy exhibits improved robustness when handling sensors with varying predictive reliability. To ensure robustness and reproducibility, results are reported using multiple chronological validation splits, with performance evaluated in terms of RMSE, MAE, and $R^2$ along with statistical measures including standard deviation and confidence intervals. The proposed framework provides a practical balance between predictive accuracy, interpretability, and computational efficiency, making it suitable for smart building energy management and real-world deployment scenarios. Full article

The global poultry hatching industry faces severe challenges of resource waste and animal ethics issues due to the routine culling of day-old male chicks. Meanwhile, early sex identification of 4-day-incubated chicken embryos is limited by low accuracy, as embryos at this stage have weak, low-contrast blood vessels that are highly susceptible to interference from the eggshell’s texture. To address these issues, this paper proposes a non-destructive early sex identification method for chicken embryos based on an improved MobileViT-V3 model. Taking the lightweight hybrid architecture MobileViT-V3 as the backbone, we embedded a Micro Feature Enhancement module (MFE-Module) in Stage 3 to strengthen the extraction of fine vascular details, and a Multi-Scale Adaptive Attention Fusion module (MSAAF-Module) in Stage 4 to realize adaptive weighted screening of multi-source features. Experiments on the self-constructed dataset of 4-day-incubated embryos show that the improved model achieves a test set classification accuracy of 92.26%, with an F1-score of 92.15%, a recall rate of 92.12%, and a Kappa coefficient of 0.845. It outperforms mainstream models such as YOLOv12, ShuffleNetV2, ConvNeXt-T, ResNet, and Swin-ViT, with only 2.98 M parameters and an inference speed of 97.6 FPS, well exceeding the 30 FPS real-time requirement of industrial sorting lines and showing high potential for practical industrial deployment. This method provides a new scheme for non-destructive, high-precision, and high-efficiency early sex identification in poultry hatching. Full article

Background: The steady rise of prostate cancer in Saudi Arabia signals a critical public health shift that requires immediate investment in early detection and prevention to mitigate a future clinical crisis. Accurate diagnosis using multiparametric MRI and PI-RADS scoring remains challenging, as interpretations are highly experience-dependent and subspecialized radiologists are limited. Methods: To address this gap, this study introduces a novel Dual-Stream Attention Network designed to automate the classification of low-risk (PIRADS 2-3) versus high-risk (PIRADS 4-5) lesions from T2-weighted MRI. Leveraging a ResNet50 backbone, the architecture employs parallel streams for Local and Global Feature Processing, each enhanced by a Channel-Spatial Attention module to highlight diagnostically relevant regions. These features are integrated through a Cross-Stream Fusion mechanism and a gate-controlled Adaptive Feature Fusion module to optimize multi-scale information. The model was developed and validated on a regional dataset of 3850 images from Jazan Specialist Hospital and Prince Mohammed bin Naser Hospital. This research provides a standardized, high-precision diagnostic path tailored to the Saudi Arabian population, conducted under institutional review board approval (No. 25138). Results: The proposed dual-stream attention network achieved an accuracy of 97.8% on the validation set and 96.4% on the test set, demonstrating high performance and generalization capabilities in classifying prostate lesions from Saudi patient populations. Conclusions: The proposed dual-stream architecture with novel attention and fusion mechanisms demonstrates high effectiveness for prostate cancer classification from T2-weighted MRI in Saudi clinical settings. This represents the first deep learning model specifically trained and validated on Saudi Arabian prostate MRI data, with the potential to address the shortage of specialized expertise and improve diagnostic efficiency in the Kingdom. Full article

Osteoarthritis (OA) is a complex degenerative joint disease for which early diagnosis and clear molecular characterization remain limited. DNA methylation has been increasingly recognized as an important regulatory factor in OA pathogenesis. In this study, we proposed an integrative computational framework combining statistical analysis, machine learning, deep learning, and functional genomics to identify and validate OA-associated genes and methylation biomarkers for diagnostic and biological interpretation. Candidate CpG sites were obtained using two complementary strategies: differential methylation analysis and selection of loci located near transcription start sites of previously reported OA-related genes. Key features were further refined using support vector machine recursive feature elimination and random forest algorithms. Based on the selected loci, we developed a feature-fusion diagnostic model that combines Transformer and convolutional neural networks with adaptive weighting to capture both global dependency structures and local methylation patterns. A panel of 220 methylation sites demonstrated stable and reproducible diagnostic performance in an independent cohort. Functional annotation and pathway analysis highlighted several established OA-associated genes, including TGFBR2, SMAD3, PPARG, and MAPK3, and suggested INHBB as a potential novel effector gene, with additional support for AMH and INHBE involvement. Overall, this study presents a robust methylation-based framework for identifying key OA-associated genes and provides new insights into the epigenetic mechanisms underlying OA. Full article

Against the background of highly interconnected global capital markets and rapidly propagating cross-lingual information streams, traditional anomaly detection paradigms based solely on single-modality numerical time-series sensors are insufficient for forward-looking risk sensing. From the perspective of artificial intelligence-driven sensing, this study proposes a multilingual semantic–numerical collaborative Transformer framework to construct a unified multimodal financial sensing architecture for intelligent anomaly sensing and risk perception. Within the proposed sensing paradigm, multilingual texts are conceptualized as semantic sensors that continuously emit event-driven sensing signals, while market prices, trading volumes, and order book dynamics are modeled as heterogeneous numerical sensor streams reflecting behavioral market sensing responses. These heterogeneous sensors are jointly integrated through a cross-modal sensor fusion architecture. A cross-modal temporal alignment attention mechanism is designed to explicitly model dynamic lag structures between semantic sensing signals and numerical sensor responses, enabling temporally adaptive sensor-level alignment and fusion. To enhance sensing robustness, a multilingual semantic noise-robust encoding module is introduced to suppress unreliable textual sensor noise and stabilize cross-lingual semantic sensing representations. Furthermore, a semantic–numerical collaborative risk fusion module is constructed within a shared latent sensing space to achieve adaptive sensor contribution weighting and cross-sensor feature coupling, thereby improving anomaly sensing accuracy and robustness under complex multimodal sensing environments. Extensive experiments conducted on real-world multi-market financial sensing datasets demonstrate that the proposed artificial intelligence-driven sensing framework significantly outperforms representative statistical and deep learning baselines. The framework achieves a Precision of $0.852$, Recall of $0.781$, F1-score of $0.815$, and an AUC of $0.892$, while substantially improving early warning time in practical risk sensing scenarios. In cross-market transfer settings, the proposed sensing architecture maintains stable anomaly sensing performance under bidirectional domain shifts, with AUC consistently exceeding $0.86$, indicating strong structural generalization across heterogeneous sensing environments. Ablation analysis further verifies that temporal sensor alignment, semantic sensor denoising, and collaborative cross-sensor risk coupling contribute independently and synergistically to the overall sensing performance. Overall, this study establishes a scalable multimodal intelligent sensing framework for dynamic financial anomaly sensing, providing an effective artificial intelligence-driven sensing solution for cross-market risk surveillance and adaptive financial signal sensing. Full article

Anti-unmanned aerial vehicle (Anti-UAV) detection is critical for airspace security, yet existing single-modality approaches suffer from severe performance degradation under adverse illumination, thermal crossover, and extreme scale variation. In this paper, we propose CSFADet, a dual-modal detection framework that jointly exploits visible and infrared imagery through four tightly integrated modules. First, a Cross-Spectral Feature Alignment (CSFA) module performs early-stage spectral calibration by computing cross-modal query–value attention maps, generating modality-aware channel descriptors that re-weight and concatenate the two spectral streams. Second, a Dual-path Texture Enhancement Module (DTEM) enriches fine-grained spatial details via cascaded convolutions with residual connections. Third, a Dual-path Cross-Attention Module (DCAM) introduces a feature-shrinking token generation strategy followed by symmetric cross-attention branches with learnable scaling factors, Squeeze-and-Excitation recalibration, and a $1\times 1$ convolution fusion head, enabling deep bidirectional interaction between modalities. Fourth, a Dual-path Information Refinement Module (DIRM) embeds Adaptive Residual Groups (ARGs) that cascade Multi-modal Spatial Attention Blocks (MSABs) with channel and dynamic spatial attention, culminating in a Multi-scale Scale-aware Fusion Refinement (MSFR) unit that employs three parallel multi-head attention branches with a Scale Reasoning Gate and Channel Fusion Layer to produce scale-discriminative enhanced features. Experiments on the public Anti-UAV300 benchmark show that CSFADet achieves 91.4% mAP@0.5 and 58.7% mAP@0.5:0.95, surpassing fifteen representative detectors spanning single-stage, two-stage, YOLO-family, and Transformer-based categories. Ablation studies confirm the complementary contributions of each module, and heatmap visualizations verify the model’s capacity to focus on small, distant UAV targets under challenging conditions. Full article

Early detection is critical for improving outcomes in gastric cancer, yet lesion recognition in capsule endoscopy is challenged by interference from different gastric anatomical sites, patient posture changes, and gastric peristalsis. This study aims to prompt a robust deep learning model to address these challenges. A dual-network model, named DuDeM (DualNet Detection Model), was developed by integrating a ResNet50-based convolutional branch with a CapsuleNet branch incorporating dynamic routing. The convolutional branch extracts local lesion features that are transmitted to primary capsules, while dynamic routing enables adaptive matching between capsule layers to establish local–global feature associations. An attention-weighted strategy is applied for feature fusion. The model was trained using capsule endoscopy images from nine hospitals in China and public datasets, and its performance was compared with eight representative models, with ablation analyses validating key components. Results showed that DuDeM achieved an area under the curve (AUC) of 0.981 and an F1-score of 0.979, with sensitivity, specificity, and precision all exceeding 97%, and performance degradation limited to within 3% under mild image perturbations. These findings suggest that DuDeM enables reliable early gastric cancer (EGC) recognition and may support large-scale capsule endoscopy screening in clinical practice. Full article