Robust segmentation in adverse weather conditions is crucial for autonomous driving. However, these scenes struggle with recognition and make annotations expensive, resulting in poor performance. As a result, the Segment Anything Model (SAM) was recently proposed to finely segment the spatial structure of scenes and to provide powerful prior spatial information, thus showing great promise in resolving these problems. However, SAM cannot be applied directly for different geographic scales and non-semantic outputs. To address these issues, we propose SAM-EDA, which integrates SAM into an unsupervised domain adaptation mean-teacher segmentation framework. In this method, we use a “teacher-assistant” model to provide semantic pseudo-labels, which will fill in the holes in the fine spatial structure given by SAM and generate pseudo-labels close to the ground truth, which then guide the student model for learning. Here, the “teacher-assistant” model helps to distill knowledge. During testing, only the student model is used, thus greatly improving efficiency. We tested SAM-EDA on mainstream segmentation benchmarks in adverse weather conditions and obtained a more-robust segmentation model.

Unmanned aerial vehicles (UAVs) are widely used in many industries. The use of UAV images for surveying requires that the images contain high-precision localization information. However, the accuracy of UAV localization can be compromised in complex GNSS environments. To address this challenge, this study proposed a scheme to improve the localization accuracy of UAV sequences. The combination of traditional and deep learning methods can achieve rapid improvement of UAV image localization accuracy. Initially, individual UAV images with high similarity were selected using an image retrieval and localization method based on cosine similarity. Further, based on the relationships among UAV sequence images, short strip sequence images were selected to facilitate approximate location retrieval. Subsequently, a deep learning image registration network, combining SuperPoint and SuperGlue, was employed for high-precision feature point extraction and matching. The RANSAC algorithm was applied to eliminate mismatched points. In this way, the localization accuracy of UAV images was improved. Experimental results demonstrate that the mean errors of this approach were all within 2 pixels. Specifically, when using a satellite reference image with a resolution of 0.30 m/pixel, the mean error of the UAV ground localization method reduced to 0.356 m.

With the continuous development and popularization of sensor-fusion technology for mobile robots, the application of camera and light detection and ranging (LiDAR) fusion perception has become particularly important. Moreover, the calibration of extrinsic parameters between the camera and LiDAR is a crucial prerequisite for fusion. Although traditional target-based calibration methods have been widely adopted, their cumbersome operation and high costs necessitate the development of more efficient and flexible calibration methods. To address this problem, this study proposed a two-stage calibration method based on motion and edge matching. In the first stage, the preliminary estimation of the extrinsic parameters between the camera and LiDAR was performed by matching visual odometry and LiDAR odometry using a hand–eye target method. In the second stage, the calibration results from the first stage were further refined by matching the image edges and discontinuous depth point clouds. The calibration system was then tested in both simulated and actual environments. The experimental results showed that this method, which did not require specially structured targets, could achieve highly flexible and robust automated calibration. Compared to other advanced methods, the accuracy of the proposed method was higher.

In traditional 3D reconstruction using UAV images, only radiance information, which is treated as a geometric constraint, is used in feature matching, allowing for the restoration of the scene’s structure. After introducing radiance supervision, NeRF can adjust the geometry in the fixed-ray direction, resulting in a smaller search space and higher robustness. Considering the lack of NeRF construction methods for aerial scenarios, we propose a new NeRF point sampling method, which is generated using a UAV imaging model, compatible with a global geographic coordinate system, and suitable for a UAV view. We found that NeRF is optimized entirely based on the radiance while ignoring the direct geometry constraint. Therefore, we designed a radiance correction strategy that considers the incidence angle. Our method can complete point sampling in a UAV imaging scene, as well as simultaneously perform digital surface model construction and ground radiance information recovery. When tested on self-acquired datasets, the NeRF variant proposed in this paper achieved better reconstruction accuracy than the original NeRF-based methods. It also reached a level of precision comparable to that of traditional photogrammetry methods, and it is capable of outputting a surface albedo that includes shadow information.

The GaoFen-14 satellite is primarily utilized for global high-precision positioning and generating 1:10,000 scale geographic information products, making it one of the most accurate stereo mapping satellites in China. With a long-stitched CCD (charge couple device) consisting of nine chips for the forward view camera and six chips for the backward view camera, it is crucial to validate the satellite’s geometric performance and achieve high uncontrolled global positioning precision. In this study, two calibration models were proposed for the on-orbit calibration of the GaoFen-14 satellite: one accounting for the inter-chip geometry constraint and another without the constraint. Correspondingly, five calibration schemes (A, B, C, D, and E) were designed, with varying settings of the external calibration parameters, interior parameters, and utilization of the optical axis measurement data. Schemes A and B utilized the same set of alignment angles for the forward and backward images and differed in the set of interior calibration parameters. Scheme A took a single set of interior calibration parameters for all chips within the same view, while Scheme B set independent interior calibration parameters for each chip in both the forward and backward views considering the inter-chip geometry constraint. Schemes C, D, and E set two groups of alignment angles for the forward and backward views, respectively. In addition, Schemes C, D, and E utilized GaoFen-14′s specific real-time recording data of the optical axis, incorporating a slight revision of the alignment angle calibration value using dichotomous search. The interior models of Schemes A and C were identical, and those of Schemes B and D were the same. In comparison with scheme D, scheme E only lacks the inter-chip geometry constraint. These five schemes were taken to validate the most suitable exterior model and interior model for the GaoFen-14 satellite. Calibration experiments were conducted using original multichip images and the stitched images from the Zhongwei, Ningxia, and Songshan (Henan) test fields. Additionally, four sets of data from around the world are used to verify the calibration effect. The results demonstrated significant improvements in uncontrolled positioning accuracy for all five calibration schemes, transitioning from the 100 m level to the meter or even the submeter level. However, Schemes A and B exhibited unstable controlled positioning accuracy, with fluctuations ranging from 1 m to 5 m. In contrast, Schemes C, D, and E achieved substantial enhancements in uncontrolled positioning accuracy, maintaining stability with variations ranging from 0.94 to 1.14 m in the X direction, 0.41 to 0.48 m in the Y direction, and 0.77 to 0.81 m in the Z direction. The positions of the CCD probes obtained from Schemes C and D demonstrated consistency, with over 80% of the probes falling within 0.1 pixels, and all of the sampled probes within 0.3 pixels. However, Scheme D behaves better in describing the geometric deformation of each CCD chip and keeping the integrity of the long-stitched CCD simultaneously. These experimental results validate the GaoFen-14 satellite’s ability to achieve stable uncontrolled positioning accuracy worldwide and a stable geometric structure between multichips. For the GaoFen-14 satellite, it is more appropriate to adopt two sets of alignment angles for the forward and backward views, respectively. As the satellite’s on-orbit operation time increases, it is recommended to employ Scheme D for monitoring inconsistent local geometric deformation in future calibration work.

Novel view synthesis using neural radiance fields (NeRFs) for remote sensing images is important for various applications. Traditional methods often use implicit representations for modeling, which have slow rendering speeds and cannot directly obtain the structure of the 3D scene. Some studies have introduced explicit representations, such as point clouds and voxels, but this kind of method often produces holes when processing large-scale scenes from remote sensing images. In addition, NeRFs with explicit 3D expression are more susceptible to transient phenomena (shadows and dynamic objects) and even plane holes. In order to address these issues, we propose an improved method for synthesizing new views of remote sensing images based on Point-NeRF. Our main idea focuses on two aspects: filling in the spatial structure and reconstructing ray-marching rendering using shadow information. First, we introduce hole detection, conducting inverse projection to acquire candidate points that are adjusted during training to fill the holes. We also design incremental weights to reduce the probability of pruning the plane points. We introduce a geometrically consistent shadow model based on a point cloud to divide the radiance into albedo and irradiance, allowing the model to predict the albedo of each point, rather than directly predicting the radiance. Intuitively, our proposed method uses a sparse point cloud generated with traditional methods for initialization and then builds the dense radiance field. We evaluate our method on the LEVIR_NVS data set, demonstrating its superior performance compared to state-of-the-art methods. Overall, our work provides a promising approach for synthesizing new viewpoints of remote sensing images.