High-accuracy astro-geodetic vertical deflection plays a key role in gravity model improvement as well as regional geoid refinement, et al. Now, the digital zenith camera is most widely used for astro-geodetic vertical deflection measurement, which suffers from some shortcomings, such as a complicated structure, large size and heavy weight, which makes it difficult to manufacture. A high cost and inconvenient transport are additional shortcomings. In this paper, the image total station is utilized for high-accuracy and automatic vertical deflections. The measurement principal and data processing method are briefly introduced, and 23 night field tests were conducted in Henan and Shaanxi province, using the image total station TS60 made by Leica company. The results shows that 12 minutes observations produced precisions of 0.18″~0.23″, which improved to 0.13″~0.19″ by increasing the observation time to 96 minutes, and the accuracy reaches 0.2″. Unlike digital zenith cameras, the image total station has merits of both high accuracy and efficiency, and even own the capability of single-person measurement, so it has a broad prospect of application.

To conform to the trend of multi-frequency and multi-GNSS, PPP-RTK is gradually transforming its data processing mode from the ionosphere-free method to the undifferenced and uncombined one. Existing studies on undifferenced and uncombined PPP-RTK mainly focus on the code division multiple access (CDMA) systems, whereas conducting frequency division multiple access (FDMA) PPP-RTK is challenging due to the effects of inter-frequency bias. This work proposes a regional PPP-RTK model which is capable of processing multi-frequency CDMA and FDMA data at undifferenced and uncombined level. The model adopts the recently proposed integer estimability theory to realize FDMA PPP-RTK in a network of homogeneous receivers, thereby ensuring the rigor of ambiguity resolution. We carry out experiments by collecting GPS, BDS, Galileo, GLONASS data with a sampling rate of 30 seconds from the Hong Kong continuously operating reference stations. Results on the network side indicate that, due to the strong correlation between different products, it is not wise to assess the precision of individual product, but necessary to analyze the combined product. After forming the combined product of satellite clock, satellite phase bias, and ionospheric delay, the precision reaches to millimeter level, thus precise enough for user correction. Results on the user side indicate that the time to first fix of GPS, BDS, and Galileo single-system PPP-RTK is 5, 1, and 3 minutes, respectively. Once the ambiguities are successfully fixed, the positioning errors converge to centimeter level. Instantaneous ambiguity resolution is achievable when combining GLONASS and GPS, and the positioning accuracy is improved by, as compared to GPS-only case, 9%, 12%, 14% (on the east, north, up component). Integrating BDS with GPS also achieves instantaneous ambiguity resolution and obtains an accuracy improvement of 29%, 22%, 18% compared to GPS-only case. Combining additionally Galileo observables further improves the accuracy by 12%, 8%, 16%. A slight improvement of 4%, 3%, 8% is obtained by adding GLONASS observables to carry out quad-system PPP-RTK.

Aiming at the error caused by the uncertainty of the height of the antenna phase center of the station and the tidal level during the tide level retrieval by GNSS-R, an optimization algorithm based on robust estimation is proposed. By determining the optimal weight of the observation value, the gross inversion value of the water surface height is weakened, and the high-precision station height is solved, thereby improving the accuracy of GNSS-R tide level inversion. The tide level inversion experiments based on the data of the four IGS GNSS continuously operating tracking stations HNLC, SC02, TDAM, and TPW2 show that the station height calculated by this method is optimized by 1.01 cm, 1.31 cm, 16.2 cm, and 1.22 cm respectively compared with the traditional inversion method. The root mean square error of the inverted tide level is reduced by 26.7%, 34.4%, 84%, and 31.6%, respectively. At the same time, this paper also proves that the station height calculated from the tidal level base level obtained by this method can be applied to the subsequent tide level inversion without the tide gauge station or the tide gauge station data interruption.

Atmospheric weighted mean temperature (T_m) is an important parameter for retrieving precipitable water vapor (PWV) from GNSS signals. However, current empirical T_m models are difficult to capture the diurnal variation of T_m, which limited its application in high temporal resolution GNSS-PWV monitoring. The T_m information with high temporal resolution can be obtained by atmospheric reanalysis data, which is need to use high-precision T_m lapse rate model for vertical elevation correction. Aiming at the shortages of current T_m lapse rate models, which only single gridded point data is used for modeling, we used MERRA-2 reanalysis data over 6 a period from 2012 to 2016 to develop T_m lapse rate grid model considering the time-varying lapse rate with horizontal resolutions of 1°×1.25°, 2°×2.5° and 4°×5° based on sliding window algorithm, named as CTm-H1, CTm-H2 and CTm-H3 model, respectively. Both MERRA-2, GGOS atmospheric gridded data and radiosonde data from 2017 are treated as reference values to assess the performance of CTm-H models. The results are compared with the united T_m lapse rate model of China, named as united model. The results show that CTm-H models show the similar performance when compared with MERRA-2 gridded data, before MERRA-2 surface gridded data were corrected to each layer height of MERRA-2 pressure level gridded data by CTm-H models, CTm-H models show significant advantages when the height difference between two kinds of T_m data is large. In terms of RMS, CTm-H models have improved by approximately 30% against united model. CTm-H models show the similar performance when compared with radiosonde data, before MERRA-2 surface gridded data and GGOS atmospheric gridded products were corrected to the height of radiosonde data by CTm-H models, respectively. In terms of RMS, CTm-H models have improved by approximately 3% and 5% against united model, respectively. CTm-H and united models show significant advantages compared with the condition without vertical correction, especially in western China. In summary, CTm-H models have a good performance in China, which is provided real-time high-precision T_m elevation correction for any location of the near-earth space range (the height range from 0 to 10 km) in China without any in situ meteorological parameters, thus, which have potential application for real-time high-precision GNSS-PWV retrieval in China.

At present, the position and attitude parameters of panoramic images are mostly solved by point features, while the line features commonly existing in the scene have not been fully utilized. In this paper, a solution method based on point-line feature combination is proposed, which can not only be used to solve the position and attitude of panoramic image in the scene with missing point features, but also improve the accuracy and robustness in the scene with sufficient point features. The line feature in this method is represented by any two points on the line, which does not require the corresponding relationship between the panoramic image and the 3D scene, so it is easy to select and has great practicability. Firstly, the direct linear transform (DLT) is used to construct the point-line feature combination model of panoramic image, and the simplified model is obtained for horizontal and vertical lines; Then, using the simulated road scene, the applicability of the model is analyzed from the two aspects:different combinations of feature points and lines, large attitude angle, and the tolerance is analyzed by manually introducing different types and magnitude of point-line errors; Finally, this method is applied to the fusion of panoramic image and LiDAR points. It is proved that this method of point-line feature combination is better than the simple point method in accuracy, robustness and tolerance.

Aiming at the problems of large affine deformation, serious occlusion, and obvious viewpoint difference, a method of robust matching is proposed to solve the problems of multiple solutions and a number of mismatches in UAV image sequences matching with significantly different viewpoints. First, the improved dual-head communication D2-Net convolutional neural network is used to extract the learning features of the image sequences. In the subsequent matching search stage of the corresponding image points, a coarse-to-fine matching purification strategy is designed to solve the problem that the unique matching point is interfered by many potential feasible points, which achieves the robust matching and greatly reduces matching time cost. The proposed algorithm was tested using multiple sets of sequence images of different scenes in the HPatches dataset and field-collected images with large different viewpoints, and compared with the representative ASIFT method based on the hand-crafted and some methods based on deep learning. The results show that the proposed method can extract robust affine invariant deep learning features, and has advantages in terms of the number of correct matching points, the correct rate of matching points, the RMSE of matching points and the cost of matching time.

There are many complex degradation phenomena in nighttime light remote sensing image of Luojia01-1, such as cloud, glow, reduced resolution, and so on. The existing deep learning-based networks for image quality improvement often only aim at a certain type of degradation problems,and do not make full use of the priori information of the image, the interpretability of the training and learning process is poor, and the type of degradation removed is single. Therefore, aiming at the problem of image quality improvement with multiple complex degradation features, an interpretable a priori guided nighttime light remote sensing image quality improvement method with multiple degradation features is proposed. Firstly, the degradation process and performance are analyzed, and a comprehensive degradation model of glow, cloud noise, and spatial resolution degradation is derived. Taking the model as a priori guide, three kinds of data sets of cloud glow, and resolution are constructed. In terms of network structure, for three kinds of degradation, a residual dense convolution neural network including channel attention module and pixel attention module is designed, and the ratio sparse constraint loss function is used to further improve the clarity of the image. The experiment is carried out with the nighttime light remote sensing image of Luojia01-1. The results show that the method proposed in this paper can effectively remove the influence of cloud and fog and glow. After processed, the image light edge information is clearer, the spatial resolution is improved, and the image quality is improved obviously.

Due to the semantic annotation of point cloud scene by manual annotation is time-consuming and label cost process, and the annotation accuracy is not high, the point cloud processing is not suitable for large-scale scenes, this paper proposed an active learning annotation method of point cloud based on ranked batch-mode. This method firstly downsampling the original point cloud, then an improved recursive feature addition method is used to filter out the optimal feature subset from a huge feature set, and a ranked batch-mode sampling algorithm is adopted to iteratively select and manually label fraction of unlabeled points. The semantic annotation of the down-sampled point cloud is completed by creating a minimum manual annotation training set, and finally the original point cloud data is annotated using the neighborhood equal-weight label propagation algorithm. Experiments on three outdoor large scene point clouds show that the method in this paper only needs to manually label 7.50%, 7.35%, and 5.83% of the point clouds to complete the labeling of the down-sampled point clouds. In addition, comparative experiments show that this method is superior to other methods in labeling accuracy and reducing labor costs, and save a lot of labor costs for point cloud semantic annotation work.

Deep learning is an important method for extracting impervious surfaces (IS), which has the advantages of high accuracy and generalization. However, the training of the models relies on a huge of train samples. Especially in large-scale and high-resolution IS mapping, it is time-consuming and laborious to obtain sufficient and high-quality training samples. Therefore, this study proposes a method to automatically extract IS based on multi-source remote sensing images and open-source data. Firstly, training samples are automatically obtained from crowdsourced OpenStreetMap data, and then the noise samples are weighted with open-source IS maps to reduce the negative influence of label noise on model training. Moreover, an ultra-lightweight CNN model with three branches was constructed to generate 10 m IS products by integrating optical, SAR and terrain data. In this paper, the method was validated in Vietnam. The results show that the overall accuracy and Kappa coefficient of the method proposed are 91.01% and 0.82, respectively, which are better than the currently released IS products. The research results of this paper can provide basic technology and data support for the sustainable development and ecological environment protection of tropical and subtropical cities in the Lancing-Mekong River basin.

With the rapid development of sensor technology, change detection based on multi-sensor optical remote sensing images has become a research hotspot in the field of remote sensing. Due to the differences of sensor imaging, different patterns of manifestation are shown in multi-sensor optical remote sensing images for one scene, leading to a more obvious problem of "pseudo change". Therefore, an object-level change detection method for multi-sensor optical remote sensing images combining UNet++ and multi-stage difference module is proposed in this paper. Firstly, multi-scale feature extraction difference (MFED) module is proposed by this method to enhance the ability of the model to identify "pseudo change". On this basis, multi-scale feature outputs by UNet++ network are used for multi-angle meticulous depiction. Adaptive evidence credibility indicator (AECI) is proposed as well. At last, image segmentation and Dempster-Shafer (DS) theory are combined to design weighted Dempster-Shafer evidence fusion (WDSEF), so as to achieve mapping from pixel-level output of deep network to object-level results. Experiment was conducted to four sets of high-resolution multi-sensor optical image datasets from different regions, and contrastive analysis was conducted to multiple methods of advanced deep learning. The results revealed that, the overall accuracy (OA) and F₁ score of the proposed method reached more than 91.92% and 63.31%, respectively, under different conditions of spatial resolution and temporal phase difference, which were significantly better than the comparison methods in both visual analysis and quantitative evaluation.

with the development of military-civilian-commercial satellites in recent years, many practical problems such as the lack of initiative and the lack of personalization in remote sensing data services have becoming increasingly serious, which restrict the intelligent application of remote sensing in multiple fields. Aiming at the problems, this paper designs an active recommendation method of remote sensing information based on user portrait model. Firstly, an extensible user portrait model was constructed that included five theme elements, including time, space, sensor, resolution and product level; Secondly, the weight, interval length and distribution characteristics of each element in the model are expressed and calculated in detail; Finally, combined with the weight and the interest characteristic of each element, the recommendation degree and the correlation degree between the data to be distributed and the interest characteristic value are calculated, so as to realize the ordered active recommendation of remote sensing data. Taking two users of the National Disaster Reduction Center of China and Beijing Anti-drug Brigade as examples, the experimental results based on real demand orders in past three years show that the distribution characteristics of the theme elements can objectively reflect the actual needs of users, and the average recommendation accuracy is better than 94%. The research results provide a model method for engineering realization of remote sensing data personalized service and intelligent recommendation.

As a classical local weighting least-square method, the geographic weighted regression (GWR) model always suffers from the space sparsity of samples and the local multicollinearity of predictors, which results in the uncertainties of the model results show spatial heterogeneity. By constructing accuracy evaluation metric of posterior standard error based on the covariance propagation law, this study proposed an uncertainty measuring and constraining method for geographic weighted regression model and validated this method using the instance of ground PM_2.5 concentration remote sensing mapping. After uncertainty constraint, the results show the fitted accuracy and sample-based/site-based/regional-based cross validation accuracy for GWR model with different parameters are all improved; the sign error of regression coefficients caused by local multicollinearity are also corrected; the outlier and negative values in the GWR predicted values can also be effectively detected which improve the reliability of the ground PM_2.5 concentration mapping results. The proposed method can effectively guarantee the stability and effectiveness of GWR results.

Point of interest (POI) is one of the main elements of electronic maps, navigation and other applications. Its data quality directly affects the level of intelligence of geographic information services. In view of the non-professional collection characteristics of data on public geographic information platforms such as OpenStreetMap (OSM), the POI data labels often have quality problems such as missing labels or incorrect labels. Thus, there is an urgent need for intelligent inference of POI labels. The conventional neural network model predicts multi-category data labels directly from a single level, which does not consider the problem of the uneven distribution of POI categories. The labels predicted by neural network tend to data categories which contain larger data volume, where the learning algorithm is difficult to generalize small-scale sample rules. This paper takes into account the massive gaps in the data scale between different POI categories, proposing a neural network prediction method based on multi-level POI category organization. Through the hierarchical aggregation of small sample categories, the structured POI category tree is established, achieving a relatively balanced category division of the data scales at different levels of the tree, which supports the high-precision prediction of labels. Experiments show that based only on the POI location information, the accuracy of this method is higher than those of the traditional methods.

Aiming at matching the high sampled GNSS trajectory data with complex urban road networks, a matching method based on path increment is proposed. The method consists of two parts:combined filtering and incremental matching. Firstly, the road network is simplified through combined filtering, and then the matching process is carried out with the road paths as the increments. During the matching process at the intersection point, the similarity evaluation scheme integrating distance factor and curvature is adopted. In order to verify the effectiveness of the method, several high sampled trajectory data with different complexity are selected for experiments. The method is compared with two existing matching methods, including the curvedness feature constrained map matching method and hidden Markov model (HMM). The results show that the proposed method not only performs better in accuracy and efficiency, but also can suppress the occurrence of matching errors in various complex sections.