Aiming at the problems of imperfect feature selection and poor stability in traditional GNSS elevation time series prediction models, a combined forecasting model based on variational mode decomposition (VMD) and extreme gradient boosting (XGBoost) algorithm is proposed. The model obtains the reconstructed signal through multiple VMD sub-models, and inputs it into the XGBoost model as a feature for forecasting of the original time series. To verify the performance of the forecasting model, the experiment selects the vertical time series data of 4 observatories for the forecasting experiment, the experimental results show that the VMD model can accurately extract the features. Compared with the VMD-CNN-LSTM model, the experimental results of VMD-XGBoost show that the MAE values are reduced by 19.74%~35.90% and the RMSE values are reduced by 22.22%~31.14%. The forecasting results have higher stability and are highly correlated to the original time series, which can better predict the Targeted time series. Therefore, the forecasting method can be applied to GNSS vertical time series forecasting.

With the increase of spatial density and time span of GNSS observations, more and more abundant information including the time-dependent tectonic deformation has been presented in GNSS coordinate time series, which makes the traditional function model of GNSS coordinate timeseries fitting no longer have universal applicability. To overcome this problem, we propose an adaptive algorithm to extract the characteristics of tectonic deformation in GNSS coordinate timeseries. Firstly, a modified functional model of coordinate time series is constructed to characterize the variation characteristics of tectonic motion of GNSS stations and the adaptive algorithm of Bayesian framework is introduced to solve the time node of linear rate change as well as the posterior probability density function of parameters to obtain the optimal solution of parameters. Secondly, the coordinate time series of different noise levels and different piecewise linear rate differences are simulated. The accuracy of the linear rate and time nodes are analyzed. The results show that with the current GNSS daily positioning accuracy for the horizontal component, the calculated linear rate accuracy is 0.1 mm/a, the tectonic motion of more than 0.4 mm/a can be distinguished. The resolution of the tectonic motion change time node is related to the difference value of the tectonic motion, and its accuracy is 1.0 year when the difference value is 0.4 mm/a. is 1.0 years. Finally, the effectiveness of the method is verified by using the GNSS continuous observations on the northeastern corner of the Tibetan Plateau. At the forefront of the northeastern corner of the Tibetan Plateau, the regional tectonic kinematic variation characteristics that occurred in 2017 and the obstruction of northeastward movement are extracted.

Gravity disturbance compensation technology is an important way to improve the positioning accuracy of high precision inertial navigation system. This paper analyzes the error characteristics and frequency characteristics of the influence of gravity disturbance on the inertial navigation system. The gravity compensation method of inertial navigation system using multispectral gravity disturbance is studied. The results show that the horizontal gravity disturbance can cause the navigation error in the form of Schuler oscillation, and the amplitude of the navigation error is directly proportional to the amplitude of the gravity disturbance. The north or east gravity disturbance component can not only cause the position and velocity error in its own direction, but also cause the position and velocity error in the east or north direction due to the coupling effect between horizontal channels. For the carrier running at low speed, the high frequency signal of horizontal gravity disturbance has a more and more significant impact on the position of inertial navigation. In order to compensate the navigation error caused by horizontal gravity disturbance, a gravity compensation method for determining multispectral gravity disturbance based on EIGEN-6C4 model and residual terrain model technology is proposed,which can effectively recover the high-frequency signal of gravity disturbance. The effectiveness of this gravity compensation method is verified by dynamic test, the compensation method can reduce the error oscillation trend of inertial navigation, and the position positioning accuracy of the vehicle and shipborne two-axis rotary modulation inertial navigation system after compensation is improved by 13.2% and 17.9%, respectively.

Due to the particularity of water medium and weak underwater communication conditions, the configuration of underwater unmanned vehicles (UUVs) directly affects the accuracy of UUVs cooperative localization. In the former work, the bending of underwater acoustic ray and sound velocity variation are not considered, and the change of incident angle caused by the distance variation between UUVs is ignored. The model of UUVs cooperative localization with underwater acoustic errors is proposed, and the optimal configuration of UUVs is analyzed by introducing the concept of horizontal dilution precision (HDOP). Selecting the sound velocity profile in Nankai trough, Japan as the underwater velocity field. The cooperative localization accuracy of five UUVs configurations at different depths are analyzed,and the relationship between UUVs cooperative localization results and the ratio of horizontal distance of UUVs and water depths researched. The experimental results show that the optimal UUVs separation angle is 90° in traditional method. When the underwater acoustic ranging error is considered, the separation angle of UUVs is optimal at 25°~30°. The higher the ratio of horizontal distance of UUVs and water depth, the higher the accuracy of UUVs cooperative localization. Considering the limitation of incident angle, it is recommended that the ratio is 1.5~2.0, i.e, the incident angle is 55°~63°, as the optimal configuration of UUVs cooperative localization. Compared with the traditional configuration design method, the configuration with underwater acoustic error improves the accuracy of UUVs cooperative localization by 24.3% on average.

Compared with the extended Kalman filtering (EKF) algorithm, factor graph optimization (FGO) shows better stability, flexibility and expansibility for the asynchronous and dynamic change of sensor information carried by AUV. This paper compares the performance of FGO and EKF algorithms applied to AUV multi-sensor information fusion and location firstly, and then proposes an AUV multi-source information fusion location method based on robust factor graph to solve the problem that abnormal observations of sensors affect the positioning accuracy of FGO algorithm in complex underwater environment. Dynamic covariance scaling (DCS) strategy was used to reduce the weight of the gross error factor, and the algorithm was verified by simulating the observed gross error based on the sea data. The ordinary FGO algorithm and DCS anti-difference FGO algorithm were used to calculate the data disturbed by gross error. The statistical results show that the proposed algorithm reduces the plane position error by 14.6% compared with that without anti-error processing, and has good robustness for abnormal observation.

Global navigation satellite system interferometric reflectometry (GNSS-IR) is an emerging remote sensing technique for earth observation, which can be applied to monitor soil moisture and has a prosperous application prospect. Aiming at the modeling problem of soil moisture inversion, we developed a GNSS-IR soil moisture inversion model integrating particle swarm optimization (PSO) and convolutional neural network (CNN). The metrics extracted from two frequency signal-to-noise ratio (SNR) observation data of GPS satellites were used as the input, and particle swarm optimization algorithm was used to optimize the hyperparameters of the convolutional neural network. Detailed modeling was carried out with the site of P041. Its root mean square error is 0.015 0, which is 60%, 27%, 31% and 21% lower than that based on single satellite linear, multi-satellite linear, conventional CNN and back propagation (BP) models; The applicability of the model was verified by COPR, P183 and P341 sites. The results indicate that the integrated GNSS-IR soil moisture inversion model based on PSO-CNN can effectively restrain the influence of the land surface environmental factors within consideration of multi-source observation data.

In high-precision satellite positioning, multi-frequency observation can be used to construct combinations such as long wavelength, ionospheric-free delay and geometric-free distance, which can effectively solve the problems of carrier phase ambiguity, ionospheric delay and cycle slip. In this paper, a multi-frequency minimum noise coefficient combination method is proposed in terms of real coefficient combination, which takes the minimum noise coefficient after combination as the basic principle, and analyzes the minimum noise coefficient combination between ionospheric-free and ionospheric-combined case.In the process of minimum noise coefficient combination, the more frequency points used, the higher the observation accuracy after the combination.Ionospheric-free minimum noise coefficient combination mainly considered the influence of observed noise, and compared with single frequency observation, the optimal combination of dual-frequency, triple-frequency, four-frequency, five-frequency and six-frequency improve the observation accuracy by 40%, 53%, 58%, 62% and 64%, respectively.Ionospheric-combined minimum noise coefficient combination takes into account the effect of ionospheric delay and reduces the ionospheric coefficient by appropriately weakening the effect of noise coefficient.According to actual signal results, the improvement effect is basically consistent with the change trend of theoretical analysis.The actual data is used to analyze the positioning accuracy of BDS PPP in decimeter-level satellite-based augmentation service, and the results show that the positioning accuracy of minimum noise coefficient combination is slightly better than that of undifferenced and uncombined observations.

In close-range photogrammetry, manually turning objects and taking highly overlapping images from multiple views is significant to complete 3D model reconstruction. However, the background of the images is inconsistent with the object in the process of turning, leading to the failure of bundle adjustment, and the deformation and incompleteness of object models. This paper proposes a complete 3D reconstruction method based on multi-view geometric consistency. We analyze the distribution pattern of rotation unit quaternion to judge whether the object is turned in the image pair and group the images. Based on the grouping results and the geometric consistency of feature point matching, we reconstruct object regions without the interference in the background of the images, improving the precision of aerial triangulation and obtaining complete mesh models. Experiments show that our method could improve the accuracy and completeness of 3D models effectively when the object has been overturned.

Quickly extracting road networks from high-resolution remote sensing images is crucial in mapping, urban planning, and GIS databases updating. Semi-automatic road extraction, as the main method of road surveying and mapping, is a labor-intensive task. In order to reduce the cost of manual intervention and improve extraction efficiency, this paper proposes a fast road centerline extraction algorithm based on geodesic distance field. First, the optimal circular template is proposed to automatically estimated the road width and adjust the manual seeds to road center based on the morphological gradient map, and the road saliency map is calculated according to the local color features inside the templates. Second, we propose the soft road center kernel density based on road saliency map which overcomes the difficulty of threshold presetting of road segmentation in traditional road center kernel density estimation. Most importantly, a geodesic distance field is proposed to quickly extract the geodesic curve between two consecutive seeds, which dramatically increase the efficiency of our algorithm. Finally, we introduce the mean filter into our scheme to smooth the road centerlines. Extensive experiments and quantitative comparisons show that the proposed algorithm can greatly reduce manual intervention without losing much accuracy, and significantly improve the efficiency of road extraction. Furthermore, the proposed algorithm takes almost the same time to extract any length of road centerline given fixed image size, and no hyperparameters need to be set. The algorithm behaves good experience in human-computer interaction.

With the rapid development of intelligent transportation and refined navigation technology, the requirements of coverage, accuracy, richness and freshness for road maps are becoming higher and higher. As a significant element in connectivity of urban road network, U-Turn has become an important part of road data renewal. It is high-cost and long-term to update by existing professional surveying and mapping models, resulting in poor reality of U-Turn data. In this paper, using vehicle GNSS trajectory big data, an automatic U-Turn information collecting approach is proposed. The turning round point pairs and turning round behaviors are first extracted through trajectory tracking. Then, DBSCAN algorithm is used to extract turning round clusters. Next, a support vector machine model is built based on traffic flow proportion to eliminate illegal turning round behaviors. Finally, U-Turn positions and spatial structures are identified according to distribution characteristics of U-Turn clusters. Taking vehicle GNSS trajectory data of DiDi in Wuhan as an example, the experiment detects 183 road sections in Jianghan District. The recognition recall rate of U-Turn structures was 88.3%, and the precision rate was 87.6%. At the same time, the horizontal and vertical position accuracy of U-turn positions are 3.40 m and 5.90 m, respectively. Results show that the proposed method can effectively collect the position and structural category of U-Turn from vehicle GNSS trajectory big data, and can provide a promising solution for short-term and low-cost collection of U-Turn data.

The Chinese Loess Plateau has been widely acknowledged as one of the world's mostly eroded areas and thus characterized by a fragmented and complex terrain. During the past decades, various monitoring methods, such as field investigations, erosion pins, manipulation experiments and tracer studies, have been employed to monitor soil erosion on the Loess Plateau. However, due to the limitation in the monitoring range of the above methods, previous soil erosion studies have been primarily undertaken at an erosion plot scale, while the catchment scale erosion monitoring was seriously lacking. In recent years, emerging remote sensing technologies, such as airborne light detection and ranging (LiDAR), have provided a promising means for an effective monitoring of soil erosion process over a large area (i.e. the catchment scale). However, little was known about the uncertainty of topographic changes detected by the airborne LiDAR for topographically complex areas, and thus the ability of airborne LiDAR to monitor soil erosion remained unclear. In the study, four airborne LiDAR flights were undertaken during a period without topographic change using an unmanned aerial vehicle (UAV) platform to acquire point clouds for a typical slope-gully system (consisting of hillslopes and gully slopes) in a small catchment (i.e. Dongzhuanggou) of the gullied Loess Plateau. Digital elevation models (DEMs) were produced based on the acquired point clouds using the triangle irregular network (TIN) algorithm. The uncertainty of topographic change detections were then derived as the DEM of difference (DoD) through subtracting the DEMs derived using the point clouds acquired by different flights from one another. The spatial pattern of the DoD uncertainty for the slope-gully system was investigated, while the ability of airborne LiDAR to detect soil erosion was assessed through comparing the magnitude of DoD (DoD_ua) with soil erosion rates collected from literature. Results showed that ① The DoD_ua for different flight combination was generally insignificant, while the spatial pattern of DoD uncertainty derived based on different flight combinations was similar, with apparent difference only emerging in certain places. ② The DoD_ua on the hillslope ranged between 0.023 m and 0.034 m, which was much lower than that of gully areas (0.057 m~0.077 m). The peak values of DoD_ua were normally found on steep-sloping gully walls. The area percentage decreased with the increase of DoD_ua, with a<0.05 m DoD_ua occupying over 40% while a>0.3 m DoD_ua accounting for less than 7% of the study area. ③ In terms of a comparison of DoD_ua and measured soil erosion rates, the UAV LiDAR was found to be able to detect soil erosion of permanent gullies at an event scale and deep-seated/shallow landslides. The results also showed that the UAV LiDAR may be able to monitor erosion of shallow (ephemeral) gullies, and was not able to monitor rill erosion. Our results provided a useful reference for the catchment-scale soil erosion monitoring and erosion process studies over topographically complex areas.

Building patterns are important urban structures that reflect the effect of the urban material and social-economic on a region. Previous researches are mostly based on the graph isomorphism method and use rules to recognize building patterns, which are not efficient. The knowledge graph uses the graph to model the relationship between entities, and specific subgraph patterns can be efficiently obtained by using relevant reasoning tools. Thus, we try to apply the knowledge graph to recognize linear building patterns. First, we use the property graph to express the spatial relations in proximity, similar and linear arrangement between buildings; secondly, the rules of linear pattern recognition are expressed as the rules of knowledge graph reasoning; finally, the linear building patterns are recognized by using the rule-based reasoning in the built knowledge graph. The experimental results on a dataset containing 1286 buildings show that the method in this paper can achieve the same precision and recall as the existing methods; meanwhile, the recognition efficiency is improved by 5.98 times.

Map generalization is essentially a similar transformation process, and the quantitative analysis of this particular process is of great practical significance. In this paper, the universal functional relationship between scale change and median Hausdorff distance of river elements is obtained via large sample statistics. Firstly, randomly selecting river samples from a 1∶10 000 vector river database, and a multi-scale river element dataset is constructed via manual simplification. The median Hausdorff distance is used to measure the multi-scale similarity of river elements, and a quantitative functional relationship between scale change and median Hausdorff distance is presented via curve fitting. Subsequently, a preliminary applicative exploration of applying the functional relationship to the control of the map generalization process is presented, and the automatic determination of thresholds for BS (bend-simplify) algorithm, DP (Douglas-Peucker) algorithm, and WA (weighted-area) algorithm is achieved. Experimental results show the strong useability of the proposed functional relationship, which contribute to the full automation of the existing semi-automatic map generalization algorithms.

With the rapid development of geospatial science, there are more and more researches towards geospatial intelligent perception. Based on geospatial 3D model as priori data, an instance object perception and localization method based on semantic information and geo-registration is proposed. First, GNSS and IMU (inertial measurement unit) are used to obtain its initial position and orientation, which are then used to render the images from 3D models. Second, the panoptic segmentation and the match network are together used to segment and match the 3D model rendered image and the truth image. Third, the matching results between two images are used to restore the camera motion, which then is used to refine the position of the camera with the geographic coordinate information of the 3D models. Finally, the objects can be detected and localized using the results of instance segmentation, depth information and the refined camera position. The method was tested on different types of 3D models and images. The results demonstrate that the proposed method can improve the matching and localization accuracy, and can detect and localize objects effectively.

Geospatial text contains rich location information, which provides important support for the location of geographic entities. The extraction of geographic entities and spatial relationships is the key to obtaining location information. Aiming at the construction of the geospatial relation corpus, we take the sentences containing the spatial relation as the unit from the Encyclopedia of China Geography, and complete the construction of the geospatial relation corpus by marking the spatial relation in the sentence. For the pipeline relation extraction model which ignores the correlation between geographic entities and spatial relations, we use enhanced representation through knowledge integration (ERNIE) and BiLSTM+self-attention mechanism+BiLSTM (BAB) layers to improve the CasRel model to achieve joint extraction of geographic entities and spatial relationships, and solve the extraction of overlapping spatial relationships in geospatial texts by cascading annotation. Experiments show that on the DuIE dataset and our constructed geospatial corpus, compared with the CasRel joint extraction model, the F₁ value of our model is increased by 4.81% and 1.97%, respectively, and the extraction effect of overlapping spatial relationships is effectively improved.

A large number text including spatial information exist widely on the Internet. In order to solve the problems of inconsistent spatial semantic modeling methods and inappropriate fuzzy processing methods for describing the location of events in those text, this paper uses the square discrete grid to establish a structured semantic expression model, and uses a unified form to express three basic semantics (direction, distance and topology). The convolution method is used to quantify the fuzzy concepts in the spatial semantics, and the uncertain semantic description is projected to the geographical space, and finally the geographical location of the event is determined through the multi-sentence spatial semantics. Experiments show that: ① The structured semantic representation model can be applied to semantics with various types of spatial information, and can determine the geographic range of unknown events when multi-semantic joint modeling and merging; ② The credibility of semantic location is related to the semantic type, the type of reference entity, the number of reference entities, the proportion of correct semantics and other factors. When the number of reference entities is large, the geographical location range of events can be determined under the condition that the number of correct semantics is less than that of wrong.