Precise object detection for unmanned aerial vehicle (UAV) images is a prerequisite for many UAV image applications. Compared with natural scene images, UAV images often have many small objects with few image pixels. These small objects are often obscured, densely distributed, or in complex scenes, which causes great interference to object detection. Aiming to solve this problem, a GhostConv-based lightweight YOLO network (GCL-YOLO) is proposed. In the proposed network, a GhostConv-based backbone network with a few parameters was firstly built. Then, a new prediction head for UAV small objects was designed, and the original prediction head for large natural scene objects was removed. Finally, the focal-efficient intersection over union (Focal-EIOU) loss was used as the localization loss. The experimental results of the VisDrone-DET2021 dataset and the UAVDT dataset showed that, compared with the YOLOv5-S network, the mean average precision at IOU = 0.5 achieved by the proposed GCL-YOLO-S network was improved by 6.9% and 1.8%, respectively, while the parameter amount and the calculation amount were reduced by 76.7% and 32.3%, respectively. Compared with some excellent lightweight networks, the proposed network achieved the highest and second-highest detection accuracy on the two datasets with the smallest parameter amount and a medium calculation amount, respectively.

Deep learning-based landslide detection in optical remote sensing images has been extensively studied. However, several challenges remain. Over time, factors such as vegetation cover and surface weathering can weaken the distinct characteristics of landslides, leading to blurred boundaries and diminished texture features. Furthermore, obtaining landslide samples is challenging in regions with low landslide frequency. Expanding the acquisition range introduces greater variability in the optical characteristics of the samples. As a result, deep learning models often struggle to achieve accurate landslide identification in these regions. To address these challenges, we propose a multi-scale target attention YOLO model (MSTA-YOLO). First, we introduced a receptive field attention (RFA) module, which initially applies channel attention to emphasize the primary features and then simulates the human visual receptive field using convolutions of varying sizes. This design enhances the model’s feature extraction capability, particularly for complex and multi-scale features. Next, we incorporated the normalized Wasserstein distance (NWD) to refine the loss function, thereby enhancing the model’s learning capacity for detecting small-scale landslides. Finally, we streamlined the model by removing redundant structures, achieving a more efficient architecture compared to state-of-the-art YOLO models. Experimental results demonstrated that our proposed MSTA-YOLO outperformed other compared methods in landslide detection and is particularly suitable for wide-area landslide monitoring.

Mining of mineral resources exposes various minerals to oxidizing environments, especially sulfide minerals, which are decomposed by water after oxidation and make the water in the mine area acidic. Acid mine drainage (AMD) from mining can pollute surrounding rivers and lakes, causing serious ecological problems. Compared with traditional field surveys, unmanned aerial vehicle (UAV) technology has advantages in terms of real-time imagery, security, and image accuracy. UAV technology can compensate for the shortcomings of traditional technology in mine environmental surveys and effectively improve the implementat ion efficiency of the work. UAV technology has gradually become one of the important ways of mine environmental monitoring. In this study, a UAV aerial photography system equipped with a Red, Green, Blue (RGB) camera collected very-high-resolution images of the stone coal mining area in Ziyang County, northwest China, and classified the very-high-resolution images by support vector machine (SVM), random forest (RF), and U-Net methods, and detected the distribution of five types of land cover, including AMD, roof, water, vegetation, and bare land. Finally, the accuracy of the recognition results was evaluated based on the land-cover map using the confusion matrix. The recognition accuracy of AMD using the U-Net method is significantly better than that of SVM and RF traditional machine-learning methods. The results showed that a UAV aerial photography system equipped with an RGB camera and the depth neural network algorithm could be combined for the competent detection of mine environmental problems.

Landslides are major geological hazards that pose serious threats to life and property, particularly in the high-mountain canyon regions of Sichuan, Yunnan, and southeastern Tibet. Displacement prediction plays a critical role in disaster prevention and mitigation. In recent years, machine learning methods based on InSAR data have achieved significant breakthroughs in landslide forecasting. However, models relying solely on a single data-driven approach may fail to fully capture the complex physical mechanisms of landslides, affecting both the reliability and interpretability of predictions. Therefore, developing effective landslide displacement prediction models is essential. The paper introduces a model designed to forecast the landslide displacement using Variational Mode Decomposition (VMD), Bayesian Optimization (BO), and Gated Recurrent Units (GRU). First, wavelet analysis is employed to identify the trend component in the landslide displacement data. Then, the total displacement is separated into its trend and periodic components through the application of the Variational Mode Decomposition (VMD) technique. A wide range of influencing factors is introduced, and Utilizing Grey Relational Analysis, we evaluate the interplay between contributing factors and all components of landslide displacement, both trend and periodic. Prediction models incorporate the trend and periodic terms, alongside the contributing factors, as input variables. The overall displacement is computed by summing the trend and periodic terms series using the Mianshawan landslide as a case study, experimental studies were conducted with landslide data from January 2019 to December 2022 with a Root Mean Squared Error (RMSE) of 0.402, Mean Absolute Error (MAE) of 0.187, Mean Absolute Percentage Error (MAPE) of 2.05%, and a coefficient of determination (R²) of 0.998. These findings indicate that, compared to traditional methods, our model delivers remarkable improvements in performance, offering higher prediction accuracy and greater reliability in the landslide forecasting task for the Mianshawan area.

The application prospects of photorealistic 3D reconstruction are broad in smart cities, cultural heritage preservation, and related domains. However, existing methods face persistent challenges in balancing reconstruction accuracy, computational efficiency, and robustness, particularly in complex scenes characterized by reflective surfaces, vegetation, sparse viewpoints, or large-scale structures. In this study, an enhanced 3D Gaussian Splatting (3DGS) framework that integrates three key innovations is proposed: (i) a depth-aware regularization module that leverages metric depth priors from the pre-trained Depth-Anything V2 model, enabling geometrically informed optimization through a dynamically weighted hybrid loss; (ii) a gradient-driven adaptive densification mechanism that triggers Gaussian adjustments based on local gradient saliency, reducing redundant computation; and (iii) a neighborhood density-based floating artifact detection method that filters outliers using spatial distribution and opacity thresholds. Extensive evaluations are conducted across four diverse datasets—ranging from architectures, urban scenes, natural landscapes with water bodies, and long-range linear infrastructures. Our method achieves state-of-the-art performance in both reconstruction quality and efficiency, attaining a PSNR of 34.15 dB and SSIM of 0.9382 on medium-sized scenes, with real-time rendering speeds exceeding 170 FPS at a resolution of 1600 × 900. It demonstrates superior generalization on challenging materials such as water and foliage, while exhibiting reduced overfitting compared to baseline approaches. Ablation studies confirm the critical contributions of depth regularization and gradient-sensitive adaptation, with the latter improving training efficiency by 38% over depth supervision alone. Furthermore, we analyze the impact of input resolution and depth model selection, revealing non-trivial trade-offs between quantitative metrics and visual fidelity. While aggressive downsampling inflates PSNR and SSIM, it leads to loss of high-frequency detail; we identify 1/4–1/2 resolution scaling as an optimal balance for practical deployment. Among depth models, Vitb achieves the best reconstruction stability. Despite these advances, memory consumption remains a challenge in large-scale scenarios. Future work will focus on lightweight model design, efficient point cloud preprocessing, and dynamic memory management to enhance scalability for industrial applications.

When the in-orbit geometric calibration of optical satellite cameras is not performed in a precise or timely manner, optical remote sensing satellite images (ORSSIs) are produced with inaccurate camera parameters. The internal orientation (IO) biases of ORSSIs caused by inaccurate camera parameters show a discontinuous distorted characteristic and cannot be compensated by a simple orientation model. The internal geometric quality of ORSSIs will, therefore, be worse than expected. In this study, from the ORSSI users’ perspective, a feasible internal geometric quality improvement method is presented for ORSSIs with image reorientation. In the presented method, a sensor orientation model, an external orientation (EO) model, and an IO model are successively established. Then, the EO and IO model parameters are estimated with ground control points. Finally, the original image is reoriented with the estimated IO model parameters. Ten HaiYang-1C coastal zone imager (CZI) images, a ZiYuan-3 02 nadir image, a GaoFen-1B panchromatic image, and a GaoFen-1D panchromatic image, were tested. The experimental results showed that the IO biases of ORSSIs caused by inaccurate camera parameters could be effectively eliminated with the presented method. The IO accuracies of all the tested images were improved to better than 1.0 pixel.