Search Results (168)

Aiming at the problems of detail blurring and structural distortion caused by speckle noise, additive white noise and hybrid noise interference in synthetic aperture radar (SAR) images, this paper proposes a Dynamic Adaptive Transformer Network (DAT-Net) integrating a dynamic gated attention module and a frequency-domain multi-expert enhancement module for SAR image denoising. The proposed model leverages a multi-scale encoder–decoder framework, combining local convolutional feature extraction with global self-attention mechanisms to transcend the limitations of conventional approaches restricted to single noise types, thereby achieving adaptive suppression of multi-source noise contamination. Key innovations comprise the following: (1) A Dynamic Gated Attention Module (DGAM) employing dual-path feature embedding and dynamic thresholding mechanisms to precisely characterize noise spatial heterogeneity; (2) A Frequency-domain Multi-Expert Enhancement (FMEE) Module utilizing Fourier decomposition and expert network ensembles for collaborative optimization of high-frequency and low-frequency components; (3) Lightweight Multi-scale Convolution Blocks (MCB) enhancing cross-scale feature fusion capabilities. Experimental results demonstrate that DAT-Net achieves quantifiable performance enhancement in both simulated and real SAR environments. Compared with other denoising algorithms, the proposed methodology exhibits superior noise suppression across diverse noise scenarios while preserving intrinsic textural features. Full article

Automated electroencephalography (EEG) seizure detection is meaningful in clinical medicine. However, current approaches often lack comprehensive feature extraction and are limited by generic classifier architectures, which limit their effectiveness in complex real-world scenarios. To overcome this traditional coupling between feature representation and classifier development, this study proposes DySC-MDE, an end-to-end co-designed framework for seizure detection. A novel multi-domain entropy (MDE) representation is constructed at the feature level based on amplitude-sensitive permutation entropy (ASPE), which adopts entropy-based quantifiers to characterize the nonlinear dynamics of EEG signals across diverse domains. Specifically, ASPE is extended into three distinct variants, refined composite multiscale ASPE (RCMASPE), discrete wavelet transform-based hierarchical ASPE (HASPE-DWT), and time-shift multiscale ASPE (TSMASPE), to represent various temporal and spectral dynamics of EEG signals. At the classifier level, a dynamic synapse classifier (DySC) is proposed to align with the structure of the MDE features. Particularly, DySC includes three parallel and specialized processing pathways, each tailored to a specific entropy variant. These outputs are then adaptively fused through a dynamic synaptic gating mechanism, which can enhance the model’s ability to integrate heterogeneous information sources. To fully evaluate the effectiveness of the proposed method, extensive experiments are conducted on two public datasets using cross-validation. For the binary classification task, DySC-MDE achieves an accuracy of 97.50% and 98.93% and an F1-score of 97.58% and 98.87% in the Bonn and CHB-MIT datasets, respectively. Moreover, in the three-class task, the proposed method maintains a high F1-score of 96.83%, revealing its strong discriminative performance and generalization ability across different categories. Consequently, these impressive results demonstrate that the joint optimization of nonlinear dynamic feature representations and structure-aware classifiers can further improve the analysis of complex epileptic EEG signals, which opens a novel direction for robust seizure detection. Full article

To address the issues of inaccurate positioning, weak feature extraction capability, and poor cross-domain adaptability in the detection of surface defects of steel materials, this paper proposes an improved YOLOv11-EMD algorithm and integrates a multi-stage transfer learning framework to achieve high-precision, robust, and low-cost industrial defect detection. Specifically, the InnerEIoU loss function is introduced to improve the accuracy of bounding box regression, the multi-scale dilated attention (MSDA) module is integrated to enhance the multi-scale feature fusion capability, and the Cross-Stage Partial Network with 3 Convolutions and Kernel size 2 Dynamic Convolution (C3k2_DynamicConv) module is embedded to improve the expression of and adaptability to complex defects. To address the problem of performance degradation when the model migrates between different data domains, a multi-stage transfer learning framework is constructed, combining source domain pre-training and target domain fine-tuning strategies to improve the model’s generalization ability in scenarios with changing data distributions. On the comprehensive dataset constructed of NEU-DET and Severstal steel defect images, YOLOv11-EMD achieved a precision of 0.942, a recall of 0.868, and an mAP@50 of 0.949, which are 3.5%, 0.8%, and 1.6% higher than the original model, respectively. On the cross-scenario mixed dataset composed of NEU-DET and GC10-DET data, the mAP@50 was 0.799, outperforming mainstream detection algorithms. The multi-stage transfer strategy can shorten the training time by 3.2% and increase the mAP by 8.8% while maintaining accuracy. The proposed method improves the defect detection accuracy, has good generalization and engineering application potential, and is suitable for automated quality inspection tasks in diverse industrial scenarios. Full article

Wind turbine power curve modeling is essential for wind power forecasting, turbine performance monitoring, and predictive maintenance. However, SCADA data often contain anomalies (e.g., curtailment, sensor faults), degrading the accuracy of power curve predictions. This paper presents a parameter-transfer learning strategy within a preprocessing and modeling framework that jointly optimizes anomaly detection (iForest, LOF, DBSCAN) and WTPC regressors (MLP, RF, GP) via a multi-metric objective adaptable to specific modeling requirements. In the source domain, hyperparameters are explored with randomized search, and in the target domain, transferred settings are refined with Bayesian optimization. Applied to real SCADA from different locations and turbine models, the approach achieves a 90% reduction in optimization iterations and consistently improves target domain performance according to the objective, with no observed loss when comparable source and target turbines differ in site or rated power. Gains are larger for more similar source–target pairs. Overall, the approach yields a practical model-agnostic pipeline that accelerates preprocessing and modeling while preserving or improving fit, particularly for newly installed turbines with limited data. Full article

Individual recognition of Hu sheep is a core requirement for precision livestock management, significantly improving breeding efficiency and fine management. However, traditional machine vision methods face challenges such as high annotation time costs, the inability to quickly annotate new sheep, and the need for manual intervention and retraining. To address these issues, this study proposes a solution that integrates automatic annotation and transfer learning, developing a sheep face recognition algorithm that adapts to complex farming environments and can quickly learn the characteristics of new Hu sheep individuals. First, through multi-view video collection and data augmentation, a dataset consisting of 82 Hu sheep and a total of 6055 images was created. Additionally, a sheep face detection and automatic annotation algorithm was designed, reducing the annotation time per image to 0.014 min compared to traditional manual annotation. Next, the YOLOv10n-CF-Lite model is proposed, which improved the recognition precision of Hu sheep faces to 92.3%, and the mAP@0.5 to 96.2%. To enhance the model’s adaptability and generalization ability for new sheep, transfer learning was applied to transfer the YOLOv10n-CF-Lite model trained on the source domain (82 Hu sheep) to the target domain (10 new Hu sheep). The recognition precision in the target domain increased from 91.2% to 94.9%, and the mAP@0.5 improved from 96.3% to 97%. Additionally, the model’s convergence speed was improved, reducing the number of training epochs required for fitting from 43 to 14. In summary, the Hu sheep face recognition algorithm proposed in this study improves annotation efficiency, recognition precision, and convergence speed through automatic annotation and transfer learning. It can quickly adapt to the characteristics of new sheep individuals, providing an efficient and reliable technical solution for the intelligent management of livestock. Full article

Hyperspectral imaging and diffraction imaging technologies, owing to their non-destructive nature, high efficiency, and superior resolution, have found widespread application in agricultural diagnostics. This review synthesizes recent advancements in the deployment of these two technologies across various agricultural domains, including the detection of plant diseases and pests, crop growth monitoring, and animal health diagnostics. Hyperspectral imaging utilizes multi-band spectral and image data to accurately identify diseases and nutritional status, while combining deep learning and other technologies to improve detection accuracy. Diffraction imaging, by exploiting the diffraction properties of light waves, facilitates the detection of pathogenic spores and the assessment of cellular vitality, making it particularly well-suited for microscopic structural analysis. The paper also critically examines prevailing challenges such as the complexity of data processing, environmental adaptability, and the cost of instrumentation. Finally, it envisions future directions wherein the integration of hyperspectral and diffraction imaging, through multisource data fusion and the optimization of intelligent algorithms, holds promise for constructing highly precise and efficient agricultural diagnostic systems, thereby advancing the development of smart agriculture. Full article

This paper addresses stability issues in modern power grids arising from extensive integration of power electronic converters, which introduce complex multi-time-scale interactions. A symbolic simplification method is proposed to accurately model grid-connected converter dynamics, significantly reducing computational complexity through transfer function approximations and yielding efficient reduced-order models. An impedance-based approach utilizing impedance ratio (IR) is developed for stability assessment under active-reactive (PQ) and active power-AC voltage (PV) control strategies. The impacts of Phase-Locked Loop (PLL) and proportional-integral (PI) controllers on system stability are analysed, with a particular focus on quantifying remote converter interactions and delineating stability boundaries across varying network strengths and configurations. Furthermore, time-scale separation effectively simplifies Multi-Voltage Source Converter (MVSC) systems by minimizing inner-loop dynamics. Validation is conducted through frequency response evaluations, IR characterizations, and eigenvalue analyses, demonstrating enhanced accuracy, particularly with the application of lead–lag compensators within the critical 50–250 Hz frequency band. Time-domain simulations further illustrate the adaptability of the proposed models and reduction methodology, providing an effective and computationally efficient tool for stability assessment in converter-dominated power networks. Full article

Microseismic monitoring has become a critical technology for hydraulic fracturing in unconventional oil and gas reservoirs, owing to its high temporal and spatial resolution. It plays a pivotal role in tracking fracture propagation and evaluating stimulation effectiveness. However, the automatic picking of first-arrival times and accurate source localization remain challenging under complex noise conditions, which constrain the reliability of fracture parameter inversion and reservoir assessment. To address these limitations, we propose a hybrid approach that combines optimized variational mode decomposition (OVMD), wavelet thresholding denoising (WTD), and an adaptive fractal box-counting dimension algorithm for enhanced first-arrival picking and source localization. Specifically, OVMD is first employed to adaptively decompose seismic signals and isolate noise-dominated components. Subsequently, WTD is applied in the multi-scale frequency domain to suppress residual noise. An adaptive fractal dimension strategy is then utilized to detect change points and accurately determine the first-arrival time. These results are used as inputs to a particle swarm optimization (PSO) algorithm for source localization. Both numerical simulations and laboratory experiments demonstrate that the proposed method exhibits high robustness and localization accuracy under severe noise conditions. It significantly outperforms conventional approaches such as short-time Fourier transform (STFT) and continuous wavelet transform (CWT). The proposed framework offers reliable technical support for dynamic fracture monitoring, detailed reservoir characterization, and risk mitigation in the development of unconventional reservoirs. Full article

Synthetic Aperture Radar (SAR) plays a pivotal role in remote-sensing target detection. However, domain shift caused by distribution discrepancies across sensors, coupled with the scarcity of target-domain samples, severely restricts the generalization and practical performance of SAR detectors. To address these challenges, this paper proposes a few-shot SAR target-detection framework tailored for cross-sensor scenarios (CS-FSDet), enabling efficient transfer of source-domain knowledge to the target domain. First, to mitigate inter-domain feature-distribution mismatch, we introduce a Multi-scale Uncertainty-aware Bayesian Distribution Alignment (MUBDA) strategy. By modeling features as Gaussian distributions with uncertainty and performing dynamic weighting based on uncertainty, MUBDA achieves fine-grained distribution-level alignment of SAR features under different resolutions. Furthermore, we design an Adaptive Cross-domain Interactive Coordinate Attention (ACICA) module that computes cross-domain spatial-attention similarity and learns interaction weights adaptively, thereby suppressing domain-specific interference and enhancing the expressiveness of domain-shared target features. Extensive experiments on two cross-sensor few-shot detection tasks, HRSID→SSDD and SSDD→HRSID, demonstrate that the proposed method consistently surpasses state-of-the-art approaches in mean Average Precision (mAP) under 1-shot to 10-shot settings. Full article

In tool wear monitoring, the efficient fusion of multi-source sensor signals poses significant challenges due to their inherent heterogeneous characteristics. In this paper, we propose a Multi-Sensor Multi-Domain feature fusion Transformer (MSMDT) model that achieves precise tool wear prediction through innovative feature engineering and cross-modal self-attention mechanisms. Specifically, we first develop a physics-aware feature extraction framework, where time-domain statistical features, frequency-domain energy features, and wavelet packet time–frequency features are systematically extracted for each sensor type. This approach constructs a unified feature matrix that effectively integrates the complementary characteristics of heterogeneous signals while preserving discriminative tool wear signatures. Then, a position-embedding-free Transformer architecture is constructed, which enables adaptive cross-domain feature fusion through joint global context modeling and local feature interaction analysis to predict tool wear values. Experimental results on the PHM2010 demonstrate the superior performance of MSMDT, outperforming state-of-the-art methods in prediction accuracy. Full article

The existing deep learning-based Airborne Laser Scanning (ALS) point cloud semantic segmentation methods require a large amount of labeled data for training, which is not always feasible in practice. Insufficient training data may lead to over-fitting. To address this issue, we propose a novel Multi-branch Feature Extractor (MFE) and a three-stage transfer learning strategy that conducts pre-training on multi-source ALS data and transfers the model to another dataset with few samples, thereby improving the model’s generalization ability and reducing the need for manual annotation. The proposed MFE is based on a novel multi-branch architecture integrating Neighborhood Embedding Block (NEB) and Point Transformer Block (PTB); it aims to extract heterogeneous features (e.g., geometric features, reflectance features, and internal structural features) by leveraging the parameters contained in ALS point clouds. To address model transfer, a three-stage strategy was developed: (1) A pre-training subtask was employed to pre-train the proposed MFE if the source domain consisted of multi-source ALS data, overcoming parameter differences. (2) A domain adaptation subtask was employed to align cross-domain feature distributions between source and target domains. (3) An incremental learning subtask was proposed for continuous learning of novel categories in the target domain, avoiding catastrophic forgetting. Experiments conducted on the source domain consisted of DALES and Dublin datasets and the target domain consists of ISPRS benchmark dataset. The experimental results show that the proposed method achieved the highest OA of 85.5% and an average F1 score of 74.0% using only 10% training samples, which means the proposed framework can reduce manual annotation by 90% while keeping competitive classification accuracy. Full article

To address the challenges posed by weak early fault signal features, strong noise interference, low diagnostic accuracy, poor reliability when using single information sources, and the limited availability of high-quality samples in practical applications for permanent magnet synchronous motor (PMSM) bearings, this paper proposes an early bearing fault diagnosis method based on Hippopotamus Optimization Variational Mode Decomposition (HO-VMD) and weighted evidence fusion of current–vibration signals. The HO algorithm is employed to optimize the parameters of VMD for adaptive modal decomposition of current and vibration signals, resulting in the generation of intrinsic mode functions (IMFs). These IMFs are then selected and reconstructed based on their kurtosis to suppress noise and harmonic interference. Subsequently, the reconstructed signals are demodulated using the Teager–Kaiser Energy Operator (TKEO), and both time-domain and energy spectrum features are extracted. The reliability of these features is utilized to adaptively weight the basic probability assignment (BPA) functions. Finally, a weighted modified Dempster–Shafer evidence theory (WMDST) is applied to fuse multi-source feature information, enabling an accurate assessment of the PMSM bearing health status. The experimental results demonstrate that the proposed method significantly enhances the signal-to-noise ratio (SNR) and enables precise diagnosis of early bearing faults even in scenarios with limited sample sizes. Full article

The digital wave, represented by new technologies such as big data, IoT, and artificial intelligence, is sweeping the globe, driving all industries toward digitalization and intelligent transformation. Digital twins are becoming an indispensable opportunity for new infrastructure initiatives. As geotechnical engineering constitutes a critical component of new infrastructure, its corresponding digital transformation is essential to align with these initiatives. However, due to the difficulty of modeling, the demand for computing resources, interdisciplinary integration, and other issues, current digital twin applications in geotechnical engineering remain in their nascent stage. This paper delineates the developmental status of geotechnical digital twin technology in China, and it focuses on the advantages and disadvantages of digital twins in five application fields, identifying key challenges, including intelligent sensing and interconnectivity of multi-source heterogeneous physical entities, integrated sharing of 3D geological models and structural models, unified platforms for lifecycle information management, standardization of digital twin data protocols, and theoretical frameworks for digital twin modeling. Furthermore, this study systematically expounds future research priorities across four dimensions: intelligent sensing and interoperability technologies for geotechnical engineering; knowledge graph development and model-based systems engineering; integrated digital twin entity technologies combining 3D geological bodies with engineering structures; and precision enhancement, temporal extension, and spatial expansion of geotechnical digital twins. This paper systematically reviews the application status of digital twin technology in geotechnical engineering for the first time, reveals the common technical challenges in cross-domain implementation, and proposes a theoretical framework for digital twin accuracy improvement and spatiotemporal expansion for geotechnical engineering characteristics, which fills the knowledge gap in the adaptability of existing research in professional fields. These insights aim to provide references for advancing digitalization, intelligent transformation, and sustainable development of geotechnical engineering. Full article

The accurate and large-scale mapping of seagrass meadows is essential, as these meadows form primary habitats for marine organisms and large sinks for blue carbon. Image data available for mapping these habitats are often scarce or are acquired through multiple surveys and instruments, resulting in images of varying spatial and spectral characteristics. This study presents an unsupervised domain adaptation (UDA) strategy that combines histogram-matching with the transformer-based SegFormer model to address these challenges. Unoccupied aerial vehicle (UAV)-derived imagery (3-cm resolution) was used for training, while orthophotos from airplane surveys (12.5-cm resolution) served as the target domain. The method was evaluated across three Danish estuaries (Horsens Fjord, Skive Fjord, and Lovns Broad) using one-to-one, leave-one-out, and all-to-one histogram matching strategies. The highest performance was observed at Skive Fjord, achieving an F1-score/IoU = 0.52/0.48 for the leave-one-out test, corresponding to 68% of the benchmark model that was trained on both domains. These results demonstrate the potential of this lightweight UDA approach to generalization across spatial, temporal, and resolution domains, enabling the cost-effective and scalable mapping of submerged vegetation in data-scarce environments. This study also sheds light on contrast as a significant property of target domains that impacts image segmentation. Full article

To address the problems of feature extraction, cost of obtaining labeled samples, and large differences in domain distribution in bearing fault diagnosis on variable operating conditions, an unsupervised domain-adaptive bearing fault diagnosis method based on migration learning using MSACNN-IJMMD-DANN (multi-scale and attention-based convolutional neural network, MSACNN, improved joint maximum mean discrepancy, IJMMD, domain adversarial neural network, DANN) is proposed. Firstly, in order to extract fault-type features from the source domain and target domain, this paper establishes a MSACNN based on multi-scale and attention mechanisms. Secondly, to reduce the feature distribution difference between the source and target domains and address the issue of domain distribution differences, the joint maximum mean discrepancy and correlation alignment approaches are used to create the metric criterion. Then, the adversarial loss mechanism in DANN is introduced to reduce the interference of weakly correlated domain features for better fault diagnosis and identification. Finally, the method is validated using bearing datasets from Case Western Reserve University, Jiangnan University, and our laboratory. The experimental results demonstrated that the method achieved higher accuracy across different migration tasks, providing an effective solution for bearing fault diagnosis in industrial environments with varying operating conditions. Full article