Search Results (2,455)

Obtaining multiple MRI contrasts for each patient prolongs scan acquisition time, increases healthcare costs, and may not always be feasible due to patient specific constraints. Deep learning-based MRI contrast synthesis offers a potential solution, yet most existing approaches are evaluated on preprocessed public benchmarks that do not reflect real-world clinical variability. In this study, we propose a fusion U-Net transformer framework for bidirectional T1-weighted ↔ T2-weighted brain MRI synthesis trained and evaluated exclusively on retrospectively acquired clinical data. The proposed architecture integrates multiscale convolutional feature extraction with axial attention mechanisms and a transformer bottleneck for efficient global context modeling. A fusion refinement block is incorporated to mitigate skip connection artifacts. An adversarial training strategy with the least squares GAN objective and a hybrid loss combining L1 reconstruction and structural similarity (SSIM) is employed to promote both pixel-level accuracy and perceptual fidelity. The model is evaluated using SSIM and PSNR metrics alongside qualitative expert assessment conducted by two board-certified radiologists. For both synthesis directions, the framework achieves competitive quantitative performance against baseline models under the challenging conditions of clinical data. Expert evaluation confirms high anatomical fidelity and clinically acceptable image quality across both synthesis directions. These results indicate that the proposed framework represents a promising approach for multi-contrast MRI synthesis in clinically heterogeneous data environments. 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

Laser powder bed fusion (LPBF) can fabricate intricate metal components but is prone to defects, such as porosity and cracks, that degrade performance. We present an in situ monitoring framework that fuses structure-borne acoustic emission (AE) and coaxial two-color pyrometry acquired synchronously at 1 MHz. Modality-specific encoders are pretrained separately, their latent representations are exported, and a lightweight feature-level fusion classifier with two binary heads predicts crack-like and porosity-like indications. Evaluation uses a held-out grouped experiment/build-machine-part split with independent Archimedes density and micro-CT ground truth. On the held-out test set, the fused model achieved F1 = 0.974 for crack-like detection and F1 = 0.987 for porosity-like detection, with AUROC = 0.998 and 0.993, respectively. Recall was 1.00 for both heads, corresponding to false-positive rates of 11.18% for crack-like and 0.945% for porosity-like indications. These results support synchronized AE-pyrometry fusion as a promising high-sensitivity in situ screening approach for LPBF. A later matched within-framework ablation campaign was also performed under stricter checkpoint-screening rules to compare AE + PY + Aux, AE + PY, AE-only, and PY-only variants under a common grouped-split protocol. Together, these results support multimodal monitoring while highlighting the need for explicit coupon/geometry-stratified reporting and for separately architecture-optimized unimodal baselines. Full article

With the proliferation of edge computing in IoT and smart security, there is a growing demand for large-scale encrypted traffic anomaly detection. However, the opaque nature of encrypted traffic makes it difficult for traditional detection methods to balance efficiency and accuracy. To address this challenge, this paper proposes FMTF, a Multi-Scale Feature Fusion method based on Federated Learning for encrypted traffic anomaly detection. FMTF constructs graph structures at three scales—spatial, statistical, and content—to comprehensively characterize traffic features. At the spatial scale, communication graphs are constructed based on host-to-host IP interactions, where each node represents the IP address of a host and edges capture the communication relationships between them. The statistical scale builds traffic statistic graphs based on interactions between port numbers, with nodes representing individual ports and edge weights corresponding to the lengths of transmitted packets. At the content scale, byte-level traffic graphs are generated, where nodes represent pairs of bytes extracted from the traffic data, and edges are weighted using pointwise mutual information (PMI) to reflect the statistical association between byte occurrences. To extract and fuse these multi-scale features, FMTF employs the Graph Attention Network (GAT), enhancing the model’s traffic representation capability. Furthermore, to reduce raw-data exposure in distributed edge environments, FMTF integrates a federated learning framework. In this framework, edge devices train models locally based on their multi-scale traffic features and periodically share model parameters with a central server for aggregation, thereby optimizing the global model without exposing raw data. Experimental results demonstrate that FMTF maintains efficient and accurate anomaly detection performance even under limited computing resources, offering a practical and effective solution for encrypted traffic identification and network security protection in edge computing environments. Full article

Deepfake technology poses significant threats to information authenticity and social trust, necessitating effective detection methods. However, existing detection approaches predominantly rely on high-complexity network architectures that, while accurate in controlled environments, suffer from prohibitive computational costs that hinder deployment in resource-constrained scenarios such as social media platforms. To address this efficiency-accuracy dilemma, we propose a lightweight face forgery detection method that systematically learns multi-scale forgery traces. Our approach features a four-stage lightweight architecture that hierarchically extracts features from local textures to global semantics, mimicking the human visual system. Within each stage, a multi-scale dynamic perception mechanism divides feature channels into parallel groups equipped with lightweight attention modules to capture forgery cues spanning pixel-level anomalies, local structures, regional patterns, and semantic inconsistencies. Furthermore, rather than relying on conventional feature fusion that risks suppressing subtle artifacts, we introduce a novel Context-Guided Dynamic Convolution. This mechanism uses mid-level spatial anomalies as active anchors to dynamically modulate high-level semantic filters, with the goal of mitigating the disconnect between semantic content and forgery evidence. Our model achieves strong performance, yielding an AUC of 91.98% on FaceForensics++ and 93.50% on DeepFake Detection Challenge, outperforming current state-of-the-art lightweight methods. Furthermore, compared to heavy Vision Transformers, our model achieves a superior performance-efficiency trade-off, requiring only 3.06 M parameters and 1.36 G FLOPs, making it highly suitable for real-time, resource-constrained deployment. Full article

Accurate landslide segmentation from remote sensing imagery is important for geohazard assessment and emergency response, yet it remains challenging because landslide regions are often spectrally confused with bare soil, riverbeds, shadows, and disturbed surfaces while also suffering from severe foreground–background imbalance. To address these issues, we propose an Efficient Spectrally Guided Hierarchical Fusion Network (SGH-Net) for multi-modal landslide segmentation. Instead of directly concatenating heterogeneous inputs at the image level, SGH-Net adopts an asymmetric encoder–decoder design in which a pretrained EfficientNet-B4 extracts RGB features, while two lightweight guidance encoders capture complementary multispectral band and DEM-derived terrain cues. These guidance features are progressively injected into the RGB backbone through multi-stage Guided Attention Blocks, enabling selective feature recalibration and reducing cross-modal interference. In addition, a hybrid Dice–Focal loss is used to alleviate class imbalance. Experiments on the Landslide4Sense dataset show that SGH-Net achieves the best overall performance among the compared methods under the adopted evaluation protocol, reaching 81.15% IoU and a 77.86% F1-score. Compared with representative multi-modal baselines, the proposed method delivers more accurate boundary delineation and fewer false alarms while maintaining favorable model complexity. These results indicate that modality-guided hierarchical fusion is an effective and efficient strategy for multi-modal landslide segmentation. Full article

Detecting surface defects on steel products is crucial for maintaining quality standards in industrial manufacturing. However, existing detection algorithms face several challenges, including the difficulty of capturing multi-scale defect characteristics with fixed receptive fields, insufficient utilization of defect edge and frequency domain features, and simplistic feature fusion strategies. In response to the above challenges, this paper proposed the Multi-Scale Frequency-Enhanced YOLO (MSFE-YOLO) algorithm that integrates multi-scale frequency domain enhancement with defect-aware attention mechanisms. First, a Multi-Scale Frequency-Enhanced Convolution (MSFC) module was constructed, which extracted multi-scale spatial features in parallel through depth-adaptive dilated convolutions, explicitly modeled high-frequency edge information using the Laplacian operator, and achieved adaptive fusion of multi-branch features via learnable weights. Second, a Cross-Stage Partial with Multi-Scale Defect-Aware Attention (C2MSDA) module was designed, integrating Sobel operator-based edge perception, multi-scale spatial attention, and adaptive channel attention to collaboratively enhance features across spatial, channel, and edge domains through a gated fusion strategy. Finally, an Adaptive Feature Fusion Enhancement (AFFE) module was proposed to achieve adaptive aggregation of multi-level features through a data-driven weight generation network and cross-scale feature interaction mechanism. Experimental results on the NEU-DET and GC10-DET datasets demonstrated that MSFE-YOLO achieved the mAP@0.5 of 79.8% and 66.7%, respectively, which were 1.7% and 2.1% higher than the benchmark model YOLOv11s respectively, while maintaining an inference speed of 89.3 FPS, which satisfied the real-time detection requirements in industrial scenarios. Full article

Reliable property prediction and process selection in laser powder bed fusion are hindered by small, set-level datasets in which key morphology descriptors are intermittently missing, limiting both generalization and actionable co-design. A hybrid multimodal surrogate strategy is introduced that couples engineered process physics features with morphology proxies through a deployable two-stage embedding module and gradient-boosted tree regressors. Set-resolved inputs are assembled from L-PBF parameters, linear energy density and related energy-density variants, pore and prior-β grain summary statistics, and stress–strain-derived descriptors, followed by missingness-aware feature filtering, median imputation, and 5-fold GroupKFold evaluation grouped by set_id, with morphology embeddings learned on training folds and predicted when absent. Across six targets, the final deployable models achieve an RMSE/R² of 11.07 MPa/0.895 (yield), 13.88 MPa/0.873 (UTS), 0.677%/0.861 (elongation), and 2.38 GPa/0.663 (modulus), while roughness and hardness remain challenging (RMSE 2.31 μm and 16.54 HV; R² about 0.12 and 0.11). These surrogates enable constraint-aware candidate generation that identifies a concise set of manufacturing recipes balancing strength and surface objectives under uncertainty-aware screening. The resulting framework provides a practical blueprint for multimodal, small-data additive manufacturing studies and can be extended to richer microstructure measurements and prospective validation to accelerate functional and biomedical alloy development. Full article

Radiation pneumonitis (RP), one of the most common and severe complications in locally advanced non-small cell lung cancer (LA-NSCLC) patients following thoracic radiotherapy, presents significant challenges in prediction due to the complexity of clinical risk factors, incomplete multimodal data, and unavailable slice-level annotations in pre-radiotherapy CT images. To address these challenges, we propose a multimodal fusion framework based on Dual-Layer Attention-Based Adaptive Bag Embedding Multiple-Instance Learning (DAAE-MIL) for accurate RP prediction. This study retrospectively collected data from 995 LA-NSCLC patients who received thoracic radiotherapy between November 2018 and April 2025. After screening, Subject datasets (n = 670) were allocated for training (n = 535), and the remaining samples (n = 135) were reserved for an independent test set. The proposed framework first extracts pre-radiotherapy CT image features using a fine-tuned C3D network, followed by the DAAE-MIL module to screen critical instances and generate bag-level representations, thereby enhancing the accuracy of deep feature extraction. Subsequently, clinical data, radiomics features, and CT-derived deep features are integrated to construct a multimodal prediction model. The proposed model demonstrates promising RP prediction performance across multiple evaluation metrics, outperforming both state-of-the-art and unimodal RP prediction approaches. On the test set, it achieves an accuracy (ACC) of 0.93 and an area under the curve (AUC) of 0.97. This study validates that the proposed method effectively addresses the limitations of single-modal prediction and the unknown key features in pre-radiotherapy CT images while providing significant clinical value for RP risk assessment. Full article

Citrus trees exhibit severe alternate bearing, resulting in significant annual yield fluctuations and posing substantial challenges to orchard management planning. Accurate citrus bud counting provides an effective solution by supplying essential data for tree-level and orchard-level yield prediction. However, citrus buds are extremely small (5–10 mm in diameter) and are frequently occluded by leaves during the flowering stage, which makes precise detection highly challenging in complex orchard environments. To address these challenges, this paper proposes a Contextual and Frequency-Aware Detector (CFADet) for robust citrus bud detection. Specifically, an Enhanced Feature Fusion (EFF) module is introduced in the neck to refine multi-scale feature aggregation and strengthen information flow for small targets. A Contextual Boundary Enhancement Module (CBEM) is designed to capture surrounding contextual cues and enhance boundary representation through dimensional interaction and max-pooling operations. To suppress background interference, a Frequency-Aware Module (FAM) is developed to adaptively recalibrate frequency components in the amplitude spectrum, thereby enhancing target features while reducing background noise. In addition, Spatial-to-Depth Convolution (SPDConv) is employed to reconstruct the backbone to preserve fine-grained bud features while reducing model parameters. Experimental results show that CFADet achieves 81.1% precision, 80.9% recall, 81.0% F1-score, and 87.8% mAP, with stable real-time performance on mobile devices in practical orchard scenarios. This study presents a preliminary investigation into robust citrus bud detection in real-world orchard environments and provides a promising technical foundation for intelligent orchard monitoring and early yield estimation, while further validation on larger and more diverse datasets is still required. Full article

Accurate segmentation of structural cracks is a core prerequisite for quantifying crack parameters, assessing damage severity, and providing early warning of structural safety. However, different types of structures exhibit significant individual variations in features such as color, texture, and brightness. Consequently, commonly used image segmentation algorithms struggle to establish a universal mathematical model, making it challenging to robustly identify and precisely segment crack targets amidst multi-feature disparities. To address the issue, this paper proposes a crack-segmentation algorithm based on bimodal image fusion and brightness piecewise linear enhancement (CSA-BB), and further enables parameter extraction and crack monitoring. The algorithm utilizes the complementary properties of visible-light and pseudo-color images for bimodal image fusion, thereby enhancing the detailed features of cracks. Furthermore, a brightness piecewise linear function has been devised that automatically selects appropriate parameters for image enhancement of structural cracks across varying background brightness. Subsequently, the crack region is effectively segmented using the bottom-hat transform and the OTSU algorithm. Ultimately, the crack’s safety level is determined from the acquired crack parameters, thereby enabling effective monitoring and assessment of the crack development process. In this paper, the proposed method achieves the best segmentation performance with a Dice coefficient of 0.4511 and a Jaccard index of 0.2981. Compared to the second-best algorithm, it yields significant improvements of 26.9% and 34.5%, respectively, demonstrating higher consistency with the ground truth. Moreover, superior computational efficiency and robustness are achieved, fulfilling the operational demands of real-world engineering environments. Full article

Background/Objectives: An accurate preoperative assessment of prostate cancer malignancy is crucial for risk stratification and selection of the optimal treatment strategy. This study assessed the concordance of Gleason scores between MRI–TRUS fusion biopsy and radical prostatectomy specimens, and identified clinical and histopathological factors associated with post-procedural Gleason score upgrading. Methods: This retrospective analysis involved patients who underwent transperineal MRI–TRUS fusion biopsy followed by radical prostatectomy from 2020 to 2025. Concordance, upgrading, and downgrading of the Gleason score were assessed by comparing biopsy results with the final histopathological examination. Clinical parameters (age, PSA level, prostate volume, and PSA density) and histopathological features of biopsies (Gleason score and percentage of prostate lobes affected by cancer) were analyzed. Multivariate logistic regression models were stratified by PSA level (<10 ng/mL and >10 ng/mL). Results: Gleason score concordance was found in 53.1% of the 603 patients analyzed, upgrading in 29.9%, and downgrading in 17.1%. Higher Gleason scores on biopsy were independently associated with a lower risk of upgrading in the entire cohort and in both PSA subgroups. Larger tumor extent on biopsy was associated with a lower risk of upgrading, with heterogeneous dependencies between prostate lobes. The other clinical parameters showed no independent association with upgrading. Conclusions: Gleason score upgrading remains common after radical prostatectomy. The risk of this progression is primarily related to the histopathological features of the biopsy rather than to baseline clinical parameters, reflecting the limitations of biopsy as a sampling method and the biological heterogeneity of prostate cancer. Full article

Accurate brain tumor segmentation from magnetic resonance imaging (MRI) is crucial for brain tumor diagnosis, clinical treatment decisions, and advancing research. CNNs and Transformers have dominated this area, but CNNs struggle with long-range modeling, whereas Transformers are limited by the high computational costs of self-attention. Recently, Mamba has garnered significant attention due to its remarkable performance in long sequence modeling. However, the original Mamba architecture, designed primarily for 1D sequence modeling, fails to effectively capture the spatial and structural relationships essential for brain tumor segmentation. In this paper, we propose FocusMamba, a Mamba-based model inspired by human visual observation patterns, which jointly enhances local detail modeling and global contextual understanding. FocusMamba consists of three components: (i) a novel hierarchical and tri-directional Mamba unit that elevates attention from the global to the window level, reinforcing local semantic feature extraction, while simultaneously achieving window-level interactions to maintain broader global awareness, (ii) a large kernel convolution unit that captures long-range dependencies within whole-volume features, overcoming the limitations of Mamba’s single-scale context modeling, and (iii) a fusion unit that enhances the overall feature representation by fusing information from different levels. Extensive experiments on the BraTS 2023 and BraTS 2020 datasets demonstrate that FocusMamba achieves superior segmentation performance compared with several advanced methods. Full article

To address the challenge of accurately distinguishing UAVs and birds in a low-altitude detection field, this paper proposes a classification algorithm of UAVs and birds based on L/K dual-band micro-Doppler spectrograms and Mamba. We establish a dual-band radar detection model for unmanned aerial vehicles (UAVs) and birds, provide a method for characterizing the Doppler parameters of the echo signals, and research a UAV and bird target classification network model that integrates micro-Doppler and Mamba. Based on a dual-branch encoding framework, we use Patch block decomposition to design a classification model to serialize the two-dimensional spectrogram of the echo signal, and introduce the Mamba state-space backbone network to extract the long-term sequence features of the target’s micro-motion. The main breakthrough of the proposed classification algorithm lies in the feature fusion stage, where a late fusion strategy is adopted to integrate the dual-path high-level representation measures, fully leveraging the sensitivity of the K-band to high-frequency textures and the scale complementarity of the L-band. Then, according to the joint loss function of mutual learning and contrastive learning, we improve the model’s prediction consistency and representation discriminability. Through experimental testing, the results show that the proposed method can effectively classify UAVs and birds, and compared with other algorithms, the accuracy rate reaches 97.5%. Full article

With the widespread adoption of various types of Unmanned Aerial Vehicles (UAVs), their non-compliant operations pose a severe challenge to public safety, necessitating the urgent identification and detection of UAV targets. However, in complex backgrounds, UAV targets exhibit small-scale dimensions and low contrast, coupled with extremely low signal-to-noise ratios. This forces conventional target detection methods to confront issues such as feature convergence, missed detections, and false alarms. To address these challenges, we propose a Multi-View Feature Fusion Network (MVFF) that achieves precise identification of small, low-contrast UAV targets by leveraging complementary multi-view information. First, we design a collaborative view alignment fusion module. This module employs a cross-map feature fusion attention mechanism to establish pixel-level mapping relationships and perform deep fusion, effectively resolving geometric distortion and semantic overlap caused by imaging angle differences. Furthermore, we introduce a view feature smoothing module that employs displacement operators to construct a lightweight long-range modeling mechanism. This overcomes the limitations of traditional convolutional local receptive fields, effectively eliminating ghosting artifacts and response discontinuities arising from multi-view fusion. Additionally, we developed a small object binary cross-entropy loss function. By incorporating scale-adaptive gain factors and confidence-aware weights, this function enhances the learning capability of edge features in small objects, significantly reducing prediction uncertainty caused by background noise. Comparative experiments conducted on a multi-perspective UAV dataset demonstrate that our approach consistently outperforms existing state-of-the-art methods across multiple performance metrics. Specifically, it achieves a Structure-measure of 91.50% and an F-measure of 85.14%, validating the effectiveness and superiority of the proposed method. Full article