Against the backdrop of rapid advances in artificial intelligence and the surging demand for spatio-temporal information applications, spatio-temporal intelligence, a new discipline integrating surveying, navigation, remote sensing, and artificial intelligence (AI) has emerged. Leveraging intelligent sensors for communication, navigation, and remote sensing, cloud computing, and AI technologies, it enables intelligent perception, cognition, and decision support for natural and human activities, with a focus on promoting sustainable development. Surveying and mapping plays a core supporting role, requiring the construction of a four-dimensional spatio-temporal reference frame to ensure the accuracy of spatio-temporal information, the development of “fast, accurate, and agile” intelligent processing technologies to address massive data challenges, and the enhancement of spatio-temporal information comprehension efficiency through multi-dimensional dynamic visualization. The “National One Map” project serves as a typical paradigm for the practical implementation of spatio-temporal intelligence theory. Relying on the Oriental Smart Eye constellation and Tianditu (national geospatial information public service platform), it promotes the unified application of spatio-temporal information in government affairs, public services, and commercial sectors, achieving dynamic updates and intelligent analysis. In the future, spatio-temporal intelligence will deepen its theoretical system, improve the theories of spatio-temporal reference frame construction and data processing, integrate cutting-edge AI technologies, facilitate global spatio-temporal information sharing, provide more precise decision support for urban governance, environmental protection, and other fields, and contribute to the construction of a community with a shared future for mankind.

The GNSS deformation monitoring field faces dual challenges from increasingly complex monitoring scenarios and the shift toward low-cost equipment, where performance degradation occurs under signal obstruction, interference, or long baselines. To overcome these limitations, this study leverages spatiotemporal correlations in regional station networks to develop a high-precision method with GNSS multi-baseline solutions. We derive the analytical solution for this model, theoretically analyze the improvement in float solution precision with increasing numbers of monitoring stations, and validate its efficacy through simulated obstructed-environment tests and real-world application on Hangzhou Bay bridge. Results confirm that the approach significantly improves performance in complex conditions, delivering faster convergence, extended monitoring range, and streamlined parameters—highlighting its scalability and practical potential for engineering applications.

Relativistic effects in satellite navigation stem from the differential motion states between satellites and terrestrial users, manifesting as gravitational frequency shifts and time dilation. These effects are more pronounced for low earth orbit (LEO) satellites due to stronger perturbations from Earth's non-spherical gravity, raising questions about the applicability of existing correction methods. This study reviews the rigorous relativistic correction formula and two approximations used by IGS Analysis Centers and global navigation satellite systems: the traditional formula, which assumes a small orbital eccentricity and only considers Earth's central gravity, and a modified formula that additionally accounts for gravitational perturbations from Earth's higher-order terms. We first evaluate these formulas for GPS, GLONASS, Galileo, and BDS-3 satellites. Subsequently, we analyze the relationship between their correction accuracy and the orbital inclination, semi-major axis, and eccentricity of LEO satellites using simulated and measured data. Results indicate that for MEO satellites (excluding E14/E18), the modified formula reduces the periodic error amplitude from 0.11 ns to 0.05 ns. For BDS-3 IGSO satellites, however, the traditional formula yields a better accuracy of 0.05 ns compared to 0.06 ns from the modified one. For LEO satellites, the accuracy of both formulas decreases significantly and is strongly influenced by orbital parameters. Specifically, correction accuracy decreases with greater orbital inclination, lower orbital altitude, and larger eccentricity, with periodic errors for near-polar LEO satellites reaching up to 1 ns.

Accurate satellite attitude modeling significantly influences both precise satellite clock estimation and precise point positioning (PPP). This work elucidates the mathematical relationships between satellite attitude, antenna phase center correction and phase wind-up effects, following by a comparative analysis of attitude modeling discrepancies between GLONASS-M+ and GLONASS-K satellites. Finally, six-month observations in 2024 are utilized to evaluate the impacts of three attitude strategies on GLONASS satellite clock accuracy and PPP positioning performance. It reveals that the eclipsing GLONASS-K satellites last for 5~6 days on average, substantially shorter than GLONASS-M+ satellites. Currently, GLONASS-K satellite attitude quaternions in IGS analysis centers continue to employ the legacy model or nominal attitude model. When the sun angle above the satellite orbital plane approaches 0°, maximum yaw angle discrepancies reach 90° for the legacy model and 180° for the nominal model. In contrast to the correct attitude model, the nominal and GLONASS-M+ satellite model degrade GLONASS satellite clock accuracy by 29.6% and 3.7%, respectively, with corresponding PPP positioning precision reductions of 15.4% and 3.7%.

To address the issue that medium-long baseline single epoch solutions are easily affected by ionospheric delays, which reduces the reliability of narrow-lane ambiguity resolution, the characteristics of multi-frequency phase combination observations of BDS-3 and Galileo, which feature low total noise levels and ionosphere-reduced (IR) delays, are systematically analyzed. Based on this analysis, a single-epoch real-time kinematic (RTK) positioning optimization method based on the BDS-3/Galileo multi-frequency IR combination is proposed and validated using measured data. The experimental results indicate that the optimal multi-frequency IR combinations for the proposed method are obtained, namely the quad-frequency combination of BDS-3 (2,2, -3,0) and Galileo (4,0, -2, -1) and the penta-frequency combination of BDS-3 (2,2, -3,0,0) and Galileo (4, -2,0, -1,0). The positioning accuracy using BDS-3, Galileo, and BDS-3/Galileo combined systems is superior to that of conventional methods. Specifically, the 3D positioning accuracy for the quad- and penta-frequency combinations improved on average by 18.87%/24.84%/41.47% and 12.97%/25.52%/42.07%, respectively. Additionally, the ambiguity fixing success rate of the BDS-3/Galileo penta-frequency IR combination can maintain over 99.00% in active ionospheric regions. When the cut-off elevation angle is 30°, the positioning availability of BDS-3/Galileo remains at 100%. Therefore, the combined BDS-3/Galileo multi-frequency observations effectively improve positioning availability and reliability.

Accurately extracting the multipath errors on the modeling day is a crucial prerequisite for effective multipath mitigation in PPP-RTK. However, due to the accuracy of correction products and parameter estimation errors, PPP-RTK observation residuals often contain other unmodeled errors that lack spatiotemporal repeatability. These errors are mixed with multipath errors in the frequency domain, making them difficult to remove using conventional frequency-domain-based filtering methods, and if not properly addressed, may be incorporated into the multipath correction model, thereby degrading positioning accuracy. This paper proposes a data-driven multipath error mitigation method. First, the observation residuals are decomposed into a series of reconstructed components (RCs) using the multi-channel singular spectrum analysis (MSSA) technique, where each RC represents an underlying spatiotemporal signal mode. Then, the inherent spatiotemporal characteristics of different error components are utilized to identify and extract the multipath error and establish a multipath correction model. Moreover, to accommodate the special requirements of dynamic monitoring stations, we replace the traditional mean positional parameter with one that reflects the actual deformation trend of the station during residual extraction. Results from PPP-RTK experiments conducted on a real landslide scenario demonstrate that the proposed method achieves positioning accuracies of 0.89, 0.99 and 2.40 cm in the east, north, and up directions, respectively. Compared to traditional wavelet-based sidereal filtering (SF) methods, the proposed approach improves positioning accuracy by approximately 8, 9 and 7 percentage points, respectively.

Subway tunnels, as critical components of urban rail transit systems, are prone to uneven settlement during operation due to geological stress, adjacent construction activities, and long-term loading, which may compromise structural integrity and potentially lead to severe accidents. To address the limitations of conventional methods such as insufficient monitoring frequency and vulnerability to environmental vibrations, this paper proposes a real-time settlement monitoring method for subway tunnels based on inertial-visual measurement. The methodology integrates a hybrid visual-inertial measurement device to simultaneously capture tunnel marker images and collect instrument pose data through built-in inertial measurement units (IMU). By establishing a tunnel settlement calculation model that combines pixel coordinates of feature points with real-time pose parameters, this system achieves high-frequency displacement measurement. Simulation experiments demonstrate the method's capability for high-precision displacement monitoring (accuracy better than 1 mm) under platform vibration conditions, showing an 89.2% error reduction compared with pure vision-based approaches. Practical applications in metro tunnel monitoring reveal that the inertial-visual measurement method significantly outperforms traditional optical techniques in both accuracy and stability, with monitoring results exhibiting strong consistency with total station data. This approach demonstrates promising potential for real-time safety monitoring in long-span tunnel engineering applications.

In dynamic, degenerate, and large-scale cluttered environments, loop closure detection methods based solely on point cloud processing exhibit poor robustness. Moreover, existing methods generally suffer from weak translation sensitivity and low computational efficiency. To address these challenges, this paper proposes a bag-of-words with stable static point cloud clusters-based loop closure detection method. Firstly, the degradation of the preprocessed point cloud is evaluated from the environmental structure perspective, and a robust point cloud cluster classification scheme is designed to obtain the stable static point cloud clusters to weaken the interference of dynamic targets. Subsequently, to reduce the redundancy in loop closure information, the fuzzy comprehensive evaluation algorithm is used to adaptively filter the key frames. Finally, based on the stable static point cloud cluster and keyframe selection results, a bag-of-words with point cloud cluster local descriptors-based loop closure detection algorithm is proposed. The relative spatial relationship and attribute relationship between stable point cloud clusters are used to improve the translation and rotation sensitivity of bag information, so as to ensure the actual performance of loop closure detection in degenerate and cluttered scenes. Experimental results demonstrate that the proposed method can robustly detect the correct loop closure relationship in the measured scene, and the non-loop closure frame error detection rate is only 5.56%, with a single-keyframe processing time of 0.052 8 s. Compared with three similar methods BoW3D, ISC, and SGLC, the average improvement in the loop closure frame correct detection rate reaches 75.73%, the average reduction in the non-loop closure frame error detection rate is 81.93%, the processing has strong real-time performance, and it exhibits stronger robustness and applicability.

Global navigation satellite system (GNSS) can provide high-precision positioning services. However, in complex urban environments, the presence of multipath effects and non-line-of-sight (NLOS) signals leads to a mismatch between GNSS observation quality and prior stochastic models, significantly degrading positioning performance. Methods based on fisheye cameras can utilize sky-view information to mitigate the impact of NLOS observations, but existing solutions are mostly limited to semantic segmentation applications and fail to fully exploit the high-dimensional environmental features in images. To address this issue, this paper proposes a neural network-based GNSS stochastic model generation method using fisheye images. The proposed method employs neural networks to extract high-dimensional environmental features from images that reflect GNSS observation conditions and tightly integrates GNSS and image features in a cross-attention layer to predict the stochastic model of satellite observations. Experimental results demonstrate that the proposed method can effectively capture the correlation between fisheye images and GNSS observation environments, accurately inflating the variance of abnormal observations. Moreover, in scenarios where fisheye images are affected by errors, the method can leverage GNSS feature information to reduce the impact of image errors on prediction results. When further applied to a RTK/IMU integrated navigation system, the proposed method improves positioning accuracy by 32.9%, verifying that the proposed method can significantly reduce the influence of abnormal observations and enhance system performance in complex urban environments.

This study addresses the issues of material confusion and the susceptibility of thin and elongated structures to fragmentation in road extraction from remote sensing images under complex plateau environments. We propose an improved road extraction model, SRENet, which incorporates a spatial relationship enhancer (SRE) and a connectivity loss (Cnt_Loss). The core contributions of this work are as follows: ① The spatial relationship enhancer is designed to explicitly model the topological structure of roads through key point graph convolution, significantly improving the connectivity detection capability in curved and occluded areas. ② A dual-branch heterogeneous architecture was constructed with a specially designed heterogeneous feature fusion module to achieve complementary enhancement between semantic features and spatial details, thereby improving extraction capability for low-contrast roads with material and environmental similarities. ③ A connectivity constraint loss function is proposed to suppress mis-segmentation in narrow and fragmented regions through geometry-driven optimization. Based on a dual-branch deep neural network, this method achieves multi-scale feature complementarity through the heterogeneous feature fusion module and optimizes road geometric features using the Cnt_Loss. The research results demonstrate that SRENet achieves IoU scores of 0.700 2 and 0.660 4 on the JL1 and DGRD datasets, respectively, representing improvements of 0.011 6 and 0.025 2 over existing models. The model also demonstrates outstanding performance in optimizing road connectivity, such as significantly reducing the number of fractures in curved sections and areas occluded by roadside trees. The proposed Cnt_Loss function effectively addresses the problem of missing detections in roads with weak boundaries through geometric constraint mechanisms. This method provides a new solution for road extraction from high-resolution remote sensing images.

Hyperspectral image classification is a key technology for achieving fine-grained recognition of ground objects. With the advancement of imaging technology, the spatial resolution of hyperspectral images acquired by UAV platforms has significantly improved, bringing new opportunities and challenges to fine-grained land cover classification. However, existing deep neural networks still exhibit insufficiently comprehensive feature learning for high spatial-resolution hyperspectral images under small-sample conditions. To address this issue, this paper proposes a mixed feature optimization method of convolutional neural networks (CNN) and vision Transformer (ViT), including three aspects: adaptive spatial-spectral feature learning, bidirectional feature integration and multi-segment feature interaction enhancement. First, multi-scale 3D spatial-spectral features and 2D local selfattention features are incorporated into a cascaded residual structure to achieve global-local multi-scale spatial-spectral feature extraction, enhancing the feature richness. Then, spatial and channel features are integrated from two directions to extract correlations across both dimensions, thereby complementing and enhancing the features extracted by CNN and ViT. After fusing these multi-stage features, they are fed into a factorized second order pooling layer to address the issues of large discrepancies and insufficient interaction among multi-stage features. Finally, the fine-grained fused features are input into a fully connected layer for classification. Small-sample classification experiments were conducted on three hyperspectral image datasets with high spatial resolution, namely LongKou, HanChuan, and HongHu. Only five samples per land-cover class are used for model training. The proposed method achieved classification accuracies of 94.00%, 83.24%, and 87.63%, respectively, demonstrating its effectiveness under small-sample conditions.

Currently, mainstream zero-watermark approaches fail to adequately capture the unique spectral-spatial multidimensional characteristics of remote sensing images. This limitation renders them vulnerable to targeted attacks. Additionally, their reliance on third-party intellectual property management organizations introduces potential risks of data tampering and undermines mutual trust in transactions. To this end, a zero-watermark copyright protection method for remote sensing images based on coupled neural P system and blockchain is proposed. Firstly, the non-subsampled shearlet transform is used to obtain the corresponding low-frequency components by multi-scale decomposition of the R, G, and B bands of the remote sensing image, respectively. Secondly, a coupled neural P system model of multilayer perception is constructed to simulate the coupled interaction of neurons to extract the spatiotemporal dynamic features of each low-frequency component, and the external inputs of the model are optimized according to the multi-scale morphological gradient to enhance the spatial correlation of the features. Thirdly, an encrypted feature image is generated using an asymmetric Tent map, which is then scrambled and subjected to XOR operations to form the zero-watermark. Finally, a decentralized copyright registration framework is established by integrating Hyperledger Fabric and the Inter Planetary File System. This framework leverages smart contracts to facilitate the on-chain storage and automated verification of copyright information. The experimental results show that the normalized correlation coefficient of the proposed method stably stays above 0.99 in the face of different degrees of geometric, non-geometric, and combinatorial attacks, demonstrating high robustness and attack resistance.

Accurate matching and integration of multi-source building vector data are crucial for urban spatial analysis and applications. However, the prevalent spatial inconsistency significantly hinders matching accuracy and data fusion quality. Existing spatial alignment methods either adopt a “matching-then-alignment” pipeline that suffers from a circular dependency between alignment quality and matching accuracy and thus fails to fundamentally improve matching conditions, or utilize global/local transformation models that struggle to accurately correct entity-level nonlinear distortions and often introduce geometric deformations that interfere with subsequent morphology-based matching criteria. To address these limitations, this paper proposes a novel entity-level conformal spatial alignment model for multi-source building matching optimization, employing an innovative “vector-raster-vector” collaborative workflow. This approach independently executes prior to the matching process: first, it extracts building centroids to construct a Delaunay triangulation representing spatial structures and rasterizes it; second, it applies a global-local progressive feature matching strategy in the raster domain to efficiently identify high-confidence corresponding points; subsequently, it constructs a continuous displacement field based on reliable correspondences. Crucially, this displacement field drives each source building polygon to undergo an overall rigid translation, accurately correcting positional deviations while strictly preserving their inherent geometric shapes, ultimately ensuring spatial validity through topological conflict detection and resolution mechanisms. Experimental results demonstrate that this method significantly enhances spatial consistency among multi-source data, with an average Hausdorff distance reduction of 18.23% and substantial improvements in F₁ scores ranging from 1.09 to 6.65 percentage points across various downstream matching algorithms. The experiments confirm the effectiveness and application potential of this method as a preprocessing strategy for enhancing the accuracy and quality of multi-source building vector data matching and integration.

Live-streaming geographic information services have great application potential in fields such as emergency command, construction management, and environmental monitoring. However, traditional methods struggle to meet users' comprehensive demands for on-site geographic information that is all-encompassing, high-fidelity, analyzable, and highly timely. Therefore, this study establishes a VR-panoramic-based remote immersive perception method and equipment system for on-site live situations. First, a panoramic camera mounting head was built for use in complex geographic environments. Second, a remote transmission mechanism for geographic information based on panoramic video was designed. Third, a display-analyze-explore multi-level enhanced remote immersive sensing method was proposed for live field scenarios. Fourth, VR panoramic remote immersive sensing equipment and a prototype system were developed. Finally, case experiment analysis is conducted. The results show that the method presented in this paper enables remote users to perceive and analyze on-site geographic information in an immersive manner, thereby realizing multi-level on-site live geographic information services in complex environments, with the delay time of the live broadcast stabilized at approximately 6 seconds. The audio and video are clear and smooth, which has a significant advantage compared to other methods in terms of immersion, real-time capabilities, and geographic specialization. It provides an example reference for UAV VR panorama technology to empower mapping and geographic information services.