Haiyang Li

Landslide deformation monitoring is crucial for mitigating geological hazards and ensuring public safety in disaster-prone regions. Remote sensing imagery offers unparalleled advantages in capturing large-scale, high-resolution terrain data for landslide monitoring, but its effectiveness is hindered by challenges in accurately segmenting complex terrain features and delineating boundaries. This study proposes EP-UNet++, a hybrid architecture that integrates UNet++, Pyramid Vision Transformer (PVT), and EfficientNet through multi-level cross-model fusion, cross-resolution alignment, and 11-channel input adaptation. The framework effectively combines local and global features, substantially enhancing the accuracy and generalization of landslide remote sensing image segmentation. Additionally, Local Interpretable Model-agnostic Explanations (LIME) analysis is employed to interpret the contribution of each input factor to the model’s predictions. Experimental results show that the EP-UNet++ achieves a mean Intersection over Union (mIoU) of 74.15, a mean Dice coefficient (mDice) of 85.16, and specificity of 0.979. It achieves an improvement of roughly 4% in mIoU and 2% in mDice over the second-best model, Swin-UNet. The model also exhibits faster convergence during training and robust generalization for landslide pixel detection. These results highlight the EP-UNet++ as a highly effective approach for precise and reliable landslide remote sensing image segmentation, with potential benefits for cost-effective environmental monitoring and disaster mitigation.

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.

Bridge deformation monitoring is essential for safeguarding infrastructure against dynamic loads. However, conventional Kalman filter (KF) fusion methods suffer from suboptimal noise covariance tuning that lacks global optimization under non-stationary conditions. To address these issues, this study proposes a novel differential evolution-enhanced extended KF (DE-EKF) method, which combines global optimization techniques with traditional filtering models for improved sensor fusion. This method leverages the DE algorithm to optimize the noise covariance matrices dynamically, thereby enhancing the adaptability of the EKF to varying measurement conditions and ensuring more accurate fusion of global navigation satellite systems and accelerometer data. Experimental results demonstrate that DE-EKF significantly outperforms existing models, with root mean square error and mean absolute error reductions of 68.9% and 54.1% compared to unscented KF respectively, underscoring its superior accuracy, stability, and robustness in real-world applications. This method provides a promising direction for advancing multi-sensor data fusion in deformation monitoring, with the potential for broader applications in geodetic systems, structural health monitoring, and large-scale infrastructure management.

Deep learning models for bi-temporal remote sensing change detection increasingly rely on attention-based context modeling, yet most designs apply a fixed computation pattern to all regions. This uniform processing is inefficient because large homogeneous areas typically dominate high-resolution imagery, while only a small fraction contains structurally complex changes. This letter presents an Entropy-Guided Adaptive Window Attention network (EWACD), a lightweight framework that allocates computation according to local change complexity. The core component is a Dynamic Entropy-Guided Hybrid (DEGH) block, which estimates a change-entropy prior from bi-temporal feature differences and uses a differentiable router to select among three branches with distinct receptive fields: a low-cost depthwise convolution branch for low-entropy regions and two window-attention branches for medium- and high-entropy regions. In this way, EWACD concentrates global context modeling on ambiguous areas while preserving high throughput on easy backgrounds. Extensive experiments on LEVIR-CD and WHU-CD demonstrate a strong accuracy–efficiency trade-off. EWACD achieves 87.20% F1 on LEVIR-CD and 90.02% F1 on WHU-CD, with only 0.28 GFLOPs and 156 FPS. The results indicate its practicality for real-time edge-oriented deployment.

Automated landslide change detection using remote sensing imagery is critical for rapid disaster response. However, landslide change detection using bi-temporal optical imagery is frequently degraded by cross-region domain shifts and by the elongated, anisotropic morphology of landslide boundaries, leading to substantial pseudo-change alarms. To suppress pseudo-changes and improve cross-region robustness, we propose a DEM-assisted topography-conditioned and orientation-adaptive Siamese network (DEMO-Net) that injects topographic inductive bias through terrain-conditioned feature modulation and orientation-adaptive convolutions. Specifically, DEM-derived multi-channel priors are encoded to predict spatially varying FiLM parameters that recalibrate shallow optical features, suppressing spurious changes while preserving discriminative cues. In addition, we introduce an adaptive-oriented attention convolution that leverages a DEM-derived aspect to guide sparse multi-orientation aggregation via shared-kernel transformation, enabling direction-aware receptive-field alignment for elongated and direction-varying landslide structures without costly global attention. Experiments on the GVLM benchmark under a 5-fold site-wise cross-region protocol show that DEMO-Net achieves 85.17% F1 and 74.26% mIoU, outperforming the strongest CNN baseline FC-EF by 5.05% and 7.20%, respectively. These results demonstrate the effectiveness of jointly leveraging terrain-conditioned calibration and physically consistent orientation-aligned feature extraction for robust cross-region landslide change detection.

Bridge deformation monitoring plays a crucial role in ensuring structural safety and service longevity, particularly under complex environmental and loading conditions. However, existing prediction methods often struggle to accurately model the nonlinear and spatially correlated relationships among heterogeneous sensor data. To address these challenges, this study proposes a modified graph attention network (MGAT) that integrates dynamic edge weighting and physics‐constrained attention to fully exploit the spatial relationships among accelerometer, environmental, and GNSS data, enabling physically consistent and highly accurate bridge deformation prediction. Experimental results demonstrate that MGAT improves prediction accuracy by 27% compared with traditional approaches, effectively maintaining deformation trend continuity. These findings demonstrate the potential of MGAT for real‐time bridge deformation monitoring and structural health management.

Accurate monitoring of ground deformation is crucial for ensuring the safety and stability of hydraulic structures. Current deformation monitoring techniques often face challenges such as limited accuracy and robustness, particularly in complex environments. In this study, we propose a comprehensive method for Global Navigation Satellite System (GNSS) deformation monitoring in hydraulic structures by integrating the strengths of Gated Recurrent Units (GRUs) and Autoregressive Attention mechanisms. This approach enables efficient modeling of long-term dependencies while focusing on critical time steps, thereby enhancing prediction accuracy and robustness, especially in multi-step forecasting tasks. Experimental results show that the proposed GRU–Attention model achieves millimeter-level multi-step prediction accuracy, with predictions closely matching actual deformation data. Compared to the traditional method, the GRU–Attention model improves prediction accuracy by approximately 37%. The model’s attention mechanism effectively captures both short-term variations and long-term trends, ensuring accurate predictions even in complex scenarios. This research advances the field of GNSS deformation monitoring for hydraulic structures, providing valuable insights for engineering decision-making and risk management, ultimately contributing to enhanced infrastructure safety.

Bridge deformation monitoring is critical for ensuring the structural integrity and safety of bridges, as it enables early detection of potential issues that may lead to failure. However, current methods often struggle to effectively handle complex, multi-source data, leading to challenges in accurately predicting deformation behaviors. To addres this issue, this study proposes a Two-Stage Encoder Transformer (TSE-Transformer) that integrates the Global Navigation Satellite System (GNSS), accelerometers, and environment sensor data for accurate displacement prediction. The TSE-Transformer integrates multi-source sensor data through a custom-designed encoder and a Pytorch Encoder module, which together enhance its ability to model complex, non-linear relationships and capture long-range dependencies in the data. The approach is validated through experiments and analyzed using Local Interpretable Model-agnostic Explanations (LIME) to determine the contribution of each sensor to the prediction model. Compared to the eXtreme Gradient Boosting (XGBoost), the proposed method demonstrates an approximate 50% improvement in accuracy. This research highlights the potential of the TSE-Transformer model to enhance the accuracy and reliability of real-time structural health monitoring, offering a robust solution for future advancements in structural health monitoring systems.