Search Results (820)

High-resolution remote sensing image (HRRSI) scene classification faces challenges such as significant target scale variations, complex background interference, and the difficult spatial parsing of dense objects (such as tightly packed buildings in dense residential areas or scattered aircraft on aprons), while existing models struggle to balance computational efficiency and classification accuracy. To address these issues, this paper proposes a lightweight Multi-Scale Selective Dilated Attention Residual Network (MS-DARNet). The model utilizes a Multi-branch Dilated Feature Extraction (MDFE) module, employing parallel convolutional branches with varying dilation rates to dynamically expand the receptive field and collaboratively extract multi-scale features without increasing parameter counts. Furthermore, a Context-Position Aware Attention (CPAA) module is introduced, combining a large kernel decomposition strategy to suppress irrelevant background noise with direction-aware feature aggregation to retain precise spatial coordinates for dense objects. Extensive experiments on the AID, NWPU-RESISC45, and RSD-WHU46 datasets show that MS-DARNet achieves superior classification accuracies of 97.78%, 94.53%, and 94.55%, respectively. Concurrently, it maintains a significantly low complexity of just 2.50 M parameters and 0.5940 GMACs. These findings demonstrate that MS-DARNet effectively achieves an optimal balance between lightweight architecture and exceptional classification performance for complex remote sensing scenes. Full article

Breast cancer histopathological classification is critical for diagnosis and treatment planning, yet manual assessment remains time-consuming and subject to inter-observer variability. Although deep learning approaches have advanced automated analysis, image-level data splitting may introduce data leakage, and spatial-domain architectures lack explicit multi-scale frequency modeling. This study proposes MSWA-ResNet, a Multi-Scale Wavelet Attention Residual Network that embeds recursive discrete wavelet decomposition within residual blocks to enable frequency-aware and scale-aware feature learning. The model is evaluated on the BreakHis dataset using a strict patient-level protocol with 70/30 patient-wise splitting, five-fold stratified cross-validation, ensemble prediction, and hierarchical aggregation from patch to patient level. MSWA-ResNet achieves 96% patient-level accuracy at 100×, 200×, and 400× magnifications, and 92% at 40×, with F1-scores of 0.97 and 0.94, respectively. At 200× and 400×, accuracy improves from 0.92 to 0.96 and F1-score from 0.94 to 0.97 over baseline CNNs while maintaining 11.8–12.1 M parameters and 2.5–4.8 ms inference time. Grad-CAM demonstrates improved localization of diagnostically relevant regions, indicating that explicit multi-scale frequency modeling enhances accurate and interpretable patient-level classification. Full article

Flexural stiffness identification of prestressed concrete beams plays an important role in evaluating the mechanical performance and damage condition of bridge structures and has become a critical research direction in bridge health monitoring. Accordingly, this paper presented a Physics-Informed Neural Network (PINN)-based method for flexural stiffness identification. In the physical modeling framework, point source forces in the beam-column equation (BCE) were represented by approximating the Dirac delta function with Gaussian functions. This strategy alleviated the convergence issue of the loss function caused by singular behavior and enabled the formulation of a unified governing equation for multi-point loading scenarios. To eliminate the long-term deflection caused by non-load-related factors and self-weight, the BCE was expressed in incremental form. The resulting nondimensional equation was adopted as the target constraint for PINN training to alleviate multi-scale challenges. Furthermore, the residual-based adaptive refinement (RAR) strategy was incorporated during network training to improve computational efficiency and identification accuracy. The proposed method was validated through nine numerical cases without linear relationships and three experimental cases. The results indicate that, even with limited measurement data and under the tested noise levels, the proposed framework can achieve satisfactory flexural stiffness identification under the tested loading conditions. This suggests that the proposed method has promising potential for flexural stiffness identification and may be useful in bridge structural health monitoring under sparse-data conditions. Full article

Accurate forecasting of respiratory virus-associated hospitalization rates per 100,000 population is essential for healthcare capacity planning, yet remains challenging during the COVID-19 era due to abrupt distribution shifts and symptom overlap among influenza-like illnesses caused by multiple pathogens. We propose a two-stage deep learning framework that disentangles stable pre-pandemic seasonal dynamics from COVID-19-induced excess hospitalizations. A lightweight GRU is first trained on pre-pandemic surveillance data to model baseline influenza/RSV-driven seasonality, after which an excess model learns from the residual series and integrates multiple online search trends (flu, COVID-19, and fever) using a standard multi-head self-attention mechanism. While we use COVID-19-era data as a case study, the proposed baseline–excess decomposition is not disease-specific and is intended to generalize to future large-scale respiratory outbreaks or pandemics that induce abrupt regime shifts. Experiments on U.S. weekly respiratory hospitalization rate data curated from CDC surveillance networks (AME) show that the proposed approach achieves strong accuracy on a chronological COVID-era split (2020–2025), reaching $R^2=0.907$ with MAPE = 19.22%. Beyond point forecasts, we further evaluate an expanding-window rolling-origin protocol and report calibrated prediction intervals via split conformal prediction, supporting deployment-oriented uncertainty quantification. By decoupling baseline and excess components and fusing behavioral trend signals in a disciplined manner, this framework improves predictive performance under regime shift while providing interpretable excess estimates for timely situational awareness and healthcare resource planning. Full article

Accurate forecasting of fatal construction accidents is critical for proactive safety management; however, accident time series exhibit strong non-stationarity, nonlinear dynamics, and multi-scale temporal patterns that challenge conventional models. This study proposes a hybrid STL–BiGRU framework that integrates Seasonal–Trend decomposition using Loess (STL) with a Bidirectional Gated Recurrent Unit (BiGRU) network to deliver robust and interpretable forecasts tailored to construction safety needs. STL first decomposes the original monthly accident series (January 2012–December 2024, OSHA) into trend, seasonal, and residual components, reducing structural complexity and mitigating non-stationarity. Independent BiGRU models are then trained on each component to capture bidirectional temporal dependencies, and final forecasts are reconstructed through component aggregation. Comparative experiments against Gated Recurrent Units (GRUs), Long Short-Term Memory (LSTM), Recurrent Neural Networks (RNNs), Support Vector Regression (SVR), Autoregressive Integrated Moving Average (ARIMA), and their STL-enhanced variants demonstrate that the proposed STL–BiGRU model achieves superior performance across both short-term and medium-term horizons. The model achieves the lowest error levels, with a short-term Root Mean Squared Error (RMSE) of 6.8522 and a medium-term RMSE of 7.0568, and shows consistent improvements in Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE). Results indicate that multi-scale decomposition combined with bidirectional deep learning provides a practical, forward-looking tool. It helps regulators and contractors anticipate high-risk periods, optimize resource allocation, and reduce fatal accidents through targeted preventive measures. Full article

High-resolution remote sensing imagery encompasses complex land cover types and rich textural details, whilst temporal variations often manifest as subtle feature differences and unstable structural patterns. This renders traditional change detection methods ineffective at accurately characterizing genuine alterations, frequently leading to underdetection, false positives, and ambiguous boundaries. To address these challenges, this paper proposes a Transformer-Based Multi-level Adaptive Fusion Network. It is built upon the DeepLabV3+ encoder–decoder framework, in which a shared-weight ResNet-101 is adopted as the backbone for dual-temporal feature extraction, with the final residual block of layer 4 cropped to extract deeper semantic features at a higher spatial resolution. The Adaptive Window–Attention Feature Fusion Module (AWAFM) adaptively models local and global differences across temporal phases, enhancing sensitivity to genuine changes. The Dual Strip Pool Fusion Module (DSPFM) enhances sensitivity to directional structural variations through horizontal and vertical strip pooling. The Progressive Multi-Scale Feature Fusion Module (PMFFM) progressively aggregates deep and shallow features via semantic residual transmission. To further suppress misleading suppression caused by complex textures, the Transformer-Enhanced Reverse Attention Fusion Module (TRAFM) explicitly models long-range dependencies, effectively mitigating false change responses. On the LEVIR-CD dataset, it achieves state-of-the-art performance, with a PA and an IoU of 92.36% and 90.13%, respectively. On the SYSU-CD dataset, PA and IoU reach 88.96% and 86.15%, demonstrating TMAFNet’s stability and superiority in scenarios involving complex ground surface disturbances, weak textural variations, and large-scale structural changes. Full article

Monitoring surface subsidence in mining areas is essential for geological disaster early warning and safe production. Existing geometric difference methods heavily rely on the local consistency of multi-temporal point clouds. When horizontal displacement and vertical subsidence are coupled, horizontal movements often cause local misalignments, leading to spatial deviations and discrete anomalies in vertical estimations. To address this issue, this paper proposes DL-C2C, a deep learning model for subsidence extraction from bi-temporal ground point clouds. Within a unified framework, the model introduces horizontal displacement as an auxiliary constraint into the vertical solving process, effectively improving the stability of vertical subsidence estimation through continuous cross-temporal alignment and correlation updating. For feature extraction, DL-C2C employs a PointConv multi-scale pyramid combined with a proposed scale-adaptive Transformer to enhance cross-scale information interaction under sparse and non-uniform sampling conditions. Furthermore, the network constructs dynamic local associations through iterative alignment within a recursive framework, and introduces diffusion-based residual correction at the fine-scale stage to compensate for detail errors at subsidence basin boundaries and in data-missing regions. Experiments on simulated and real-world datasets—covering aeolian sand and mountainous gully landforms—demonstrate that the method achieves mining 3D error (M3DE) of 0.16 cm and 0.22 cm in simulated scenarios. In real-world mining area validations, compared to existing methods, DL-C2C significantly reduces discrete anomalous points, yields an error distribution closer to zero, and exhibits superior performance in boundary transition continuity and non-subsidence area stability. In conclusion, this model provides reliable technical support for large-scale, high-precision intelligent monitoring of geological disasters in mining areas. Full article

Vigna umbellate, a typical edible and medicinal crop, is rich in polyphenolic compounds with antioxidant, antibacterial, anti-inflammatory, and lipid-regulating activities. However, traditional methods for polyphenol content detection rely on chemical analysis, which is cumbersome and time-consuming, making it difficult to meet the demands of high-throughput rapid detection. Although hyperspectral imaging technology offers the potential for non-destructive and rapid detection, existing analytical methods are often limited by issues such as high spectral band redundancy, insufficient feature extraction, and inadequate model stability, which constrain prediction accuracy and practical application potential. To address this, this study proposes a multi-scale residual convolutional neural network (MS-RCNN) based on competitive adaptive reweighted sampling (CARS) for feature band selection, combined with near-infrared hyperspectral imaging technology, to construct a rapid and non-destructive prediction model for the polyphenol content of Vigna umbellata. The model employs a parallel multi-scale convolutional module to extract spectral features with different receptive fields, and incorporates residual connections and adaptive pooling mechanisms to enhance feature reuse and robustness. Experiments compared the performance of partial least squares regression (PLSR), least squares support vector machine (LS-SVM), multi-scale convolutional neural network (MS-CNN), and MS-RCNN models. The results indicate that the MS-RCNN model based on CARS screening achieved the best prediction performance, with a coefficient of determination (R²) of 0.9467, a root mean square error of prediction (RMSEP) of 0.0448, and a residual predictive deviation (RPD) of 4.33. Compared with the optimal PLSR and LSSVM models, its R² values were improved by 0.2078 and 0.1119, respectively. In summary, the MS-RCNN model proposed in this study enables rapid, non-destructive, and accurate prediction of polyphenol content in Vigna umbellata, providing an efficient technical approach for quality detection of edible and medicinal crops. Full article

Stock price prediction is vital for quantitative investment but challenging due to multi-source data complexity, including endogenous, exogenous, and noise components. Standard deep learning models rely on end-to-end modeling of raw market data, failing to disentangle these distinct drivers and hindering prediction accuracy. To address this, we propose MMCAD-Net, a novel model based on time series decomposition. It first decomposes the original stock series into an exogenous cyclical component, endogenous temporal component and residual component, thereby disentangling the mixed temporal patterns. Subsequently, deep feature extraction and information refinement are applied to each component: multi-scale convolutions capture diverse patterns in the cyclical component; multi-level convolutional networks refine local and global features in the temporal component; and an attention mechanism sifts for potentially informative signals within the residuals. Finally, a multi-source feature aggregation mechanism fuses all enhanced information. Experiments on real-world stock market datasets demonstrate that MMCAD-Net surpasses mainstream models in both prediction accuracy and efficiency. Ablation studies further confirm the necessity and effectiveness of each core module. Full article

The growing number of on-orbit spacecraft and the increasing volume of telemetry data have made intelligent anomaly detection in multi-channel telemetry essential for mission operations. Current spacecraft anomaly detection methods primarily rely on statistical models or time-series deep learning approaches, which often fail to explicitly model spatiotemporal dependencies across multiple telemetry channels. This shortcoming limits their ability to capture the dynamically evolving and intricately coupled relationships between variables. To overcome this limitation, a Progressive Spatiotemporal Graph (PSTG) model is proposed for anomaly detection in multi-channel spacecraft telemetry. PSTG employs a multi-scale patch embedding module to extract hierarchical semantic features from multi-channel time series, effectively reducing the dimensionality of the spatiotemporal graph. It constructs a sparse adjacency matrix using a multi-head attention mechanism that integrates intra-channel temporal dynamics, inter-channel spatial correlations, and cross-channel spatiotemporal interactions. An improved multi-head graph attention network then captures pairwise dependencies among nodes within the adjacency matrix. As a result, PSTG encodes rich spatiotemporal representations derived from intricate variable interactions, enabling accurate, real-time prediction of multi-channel telemetry. Furthermore, a dynamic thresholding mechanism is incorporated into PSTG to perform online anomaly detection based on prediction residuals. Extensive experiments on real-world spacecraft telemetry data collected over 84 months show that PSTG outperforms eleven state-of-the-art benchmark methods in almost all cases across multiple evaluation metrics. Finally, visualizations of the learned adjacency and attention matrices are presented to interpret the spatiotemporal modeling process, providing operators with actionable insights into the detected anomalies and facilitating root cause analysis. Full article

CNNs can effectively extract features with low computational costs, achieving significant progress in hyperspectral image classification. However, due to the limited receptive field of CNNs, they have difficulty in capturing the multi-scale structural and global contextual information. Moreover, the class imbalance in hyperspectral images often causes the model to focus disproportionately on certain spectral bands, thereby reducing the average accuracy. To address these challenges, a method called the Cascaded Spatial-Frequency Convolutional Network (CSFCNet) was proposed for hyperspectral image classification. It integrates rich spatial-domain information and frequency-domain information by jointly modeling both domains. Specifically, a Dual Spatial Fourier Convolution (DSF-Conv) module was proposed to project feature maps into parallel spatial and frequency representations. In the Spatial pathway, input features are grouped and processed with multi-scale convolutions to extract hierarchical structures; in the Fourier pathway, frequency-domain convolutions can aggregate the global context. Subsequently, a group-cascaded structure connects the DSF-Conv modules with residual connections, alleviating the class imbalance problem by promoting more balanced contributions from different spectral components. Additionally, we introduce a Lightweight Local Attention module to enhance the feature discrimination. Furthermore, experiments on three datasets achieved competitive accuracies, demonstrating the effectiveness of CSFCNet. Ablation studies further verify the effectiveness of the core components within the network. Full article

Parallel robots are widely applied in many fields because of their unique advantages. To ensure their operational safety and reduce maintenance costs, designing an accurate and reliable fault diagnosis method is essential. Focusing on the 2RRU-RRS parallel robot, this paper proposes an intelligent fault diagnosis method based on a multi-scale convolutional residual network integrated with an Efficient Channel Attention mechanism (MS-ECA-ResNet). Firstly, to fully retain the time-frequency features of the signals, the one-dimensional vibration signals are converted into two-dimensional images using the Continuous Wavelet Transform (CWT). Secondly, a multi-scale convolutional feature extraction structure is designed to enhance the model’s feature extraction ability at different time scales. Furthermore, the ECA mechanism is introduced into the residual network to reinforce important feature channels and suppress noise interference. Comparative experiments, noise environment experiments, and ablation experiments were conducted on a 2RRU-RRS parallel robot experimental platform with a vibration signal dataset. The results demonstrate that the proposed method achieves superior diagnostic accuracy and robustness compared to typical deep learning models, particularly in maintaining high performance under simulated noise conditions. This provides a preliminary validation of the method’s effectiveness in capturing fault-related impacts, offering a potential technical reference for the health monitoring of parallel robots in real-world scenarios. Full article

Accurate and rapid transient voltage stability assessment is crucial for the safe and stable operation of new energy bases in desert and grassland regions. Existing deep learning methods fail to adequately capture the high-dimensional dynamic coupling features of transient voltage signals in large-scale renewable energy bases with UHVDC transmission, and suffer from poor performance under class-imbalanced sample conditions. This paper proposes a transient voltage stability assessment method utilizing continuous wavelet transform (CWT) time–frequency images and a deep residual network (ResNet-50). CWT with the Morlet wavelet basis converts voltage time-series signals into multi-scale time–frequency images to simultaneously capture temporal and frequency-domain transient features. An improved focal loss (FL) function is introduced to dynamically adjust category weights based on actual sample distribution, enhancing model robustness under extreme class imbalance. The proposed method is validated on a modified IEEE 39-bus system incorporating the Qishao UHVDC line and wind/photovoltaic integration in Northwest China, using 1490 simulation samples under diverse fault scenarios. Results demonstrate that the proposed CWT-ResNet achieves 98.88% accuracy, 94.74% precision, 100% recall, and 97.29% F1-score, outperforming SVM, 1D-CNN, and 1D-ResNet baselines. Under 5 dB noise conditions, the method maintains over 90% accuracy, demonstrating strong noise robustness. Full article

High-precision 3D building data are pivotal for smart city development, urban planning, and disaster management. However, large-scale building extraction from airborne LiDAR point clouds remains challenging due to semantic ambiguity, uneven point density, and complex architectural structures. To address these limitations, we propose a novel framework integrating geometric topology perception with cross-dimensional attention mechanisms within a Sparse Voxel Convolutional Neural Network (SPVCNN). The key contributions include: (1) an enhanced LaserMix++ multi-scale hybrid augmentation strategy featuring cross-scene block replacement, ground normal–constrained rotation, and non-uniform scaling; (2) a dual-branch SPVCNN architecture embedding a collaborative module of Geometric Self-Attention (GSA) and Cross-Space Residual Attention (CSRA) to preserve topological consistency and enable cross-dimensional feature interaction; and (3) a Boundary Enhancement Module (BEM) specifically designed to resolve boundary ambiguity and overlapping predictions. Evaluated on a 177 km² dataset covering Washington, D.C., our method significantly outperforms the baseline SPVCNN, improving accuracy by 12.04 percentage points (0.8212 to 0.9416) and Intersection over Union (IoU) by 9.96 percentage points (0.866 to 0.9656). Furthermore, it surpasses mainstream networks such as Cylinder3D and MinkResNet by over 50% in absolute accuracy gain. These results demonstrate the effectiveness of synergistically combining geometric perception with adaptive attention for robust building extraction from large-scale LiDAR data. Full article

Aileron fault diagnosis in fixed-wing unmanned aerial vehicles (UAVs) faces significant challenges due to strong noise, multi-modal coupling, and limited fault samples. This paper presents a hybrid fault diagnosis framework that integrates variational mode decomposition (VMD) with a cross-attention-based feature fusion mechanism. First, residual signals are generated from UAV kinematic models and decomposed into multi-scale intrinsic mode functions (IMFs) using VMD to extract multiscale frequency-localized features. An integrated framework is then constructed, where Transformer encoders capture the global features and bidirectional long short-term memory (BiLSTM) networks extract local temporal dynamics. To effectively combine these complementary features, a cross-attention fusion module is designed to focus on the discriminative time-frequency features. Furthermore, a hybrid pooling strategy integrating max pooling and attention pooling is introduced to enhance classification robustness. Experiments on the AirLab failure and anomaly (ALFA) dataset demonstrate that the proposed method achieves 95.12% accuracy with improved fault separability, outperforming VMD + BiLSTM (87.66%), VMD + Transformer (86.89%), Transformer + BiLSTM (84.83%), Transformer (72.24%), CNN + LSTM (94.05%), and HDMTL (94.86%). Full article