Loading...

Table of Content

    13 May 2024, Volume 53 Issue 4
    Real-time Remote Sensing Mapping
    Efficient-communication on-orbit distributed hyperspectral image processing
    Weiying XIE, Zixuan WANG, Yunsong LI
    2024, 53(4):  589-598.  doi:10.11947/j.AGCS.2024.20230594
    Asbtract ( )   HTML ( )   PDF (8336KB) ( )  
    Figures and Tables | References | Related Articles | Metrics

    In recent years, with the increase in the number of on-orbit remote sensing satellites and advancements in hyperspectral imaging technology, there has been a sharp rise in the volume of available hyperspectral data, marking an era of big data applications and data-driven scientific discoveries. However, this substantial and wide-ranging volume of hyperspectral data poses significant challenges for deep learning algorithms to learn and infer on a single node, hindering real-time and efficient intelligent interpretation of information. Therefore, there is a need for comprehensive multi-satellite resource distributed cooperative analysis to address the block effects caused by block processing. However, collaborative processing inherently involves information interaction and transmission, necessitating gradient compression to reduce the transmitted information further, thereby alleviating communication bottlenecks in distributed learning. This paper comprehensively discusses various mainstream efficient communication gradient compression algorithms, specifically focusing on their pros and cons in communication-constrained on-orbit environments, and provides insights into the developmental trends of gradient compression. Through extensive experimental comparisons, we comprehensively evaluate the performance of various gradient compression methods in hyperspectral image processing. These experiments demonstrate the applicability and performance differences of different methods, providing robust references for selecting the most suitable gradient compression methods in practical applications in the future.

    On-orbit processing technology and verification of Luojia-3 01 satellite
    Mi WANG, Beibei GUO, Yingdong PI, Zhiqi ZHANG, Jing XIAO, Rongfan DAI, Shao XIANG
    2024, 53(4):  599-609.  doi:10.11947/j.AGCS.2024.20230357
    Asbtract ( )   HTML ( )   PDF (14704KB) ( )  
    Figures and Tables | References | Related Articles | Metrics

    The on-orbit processing technology of remote sensing satellites stands as a pivotal component driving the achievement of real-time intelligent services in remote sensing technology. In response to the challenges posed by limited on-orbit resources and the growing demands for real-time data processing and information extraction, this paper initially establishes a task-driven framework for on-orbit processing of satellite remote sensing data. The framework places mission requirements at the core and collaborates with space and ground-based resources to establish a comprehensive on-orbit processing system centered around regions of interest (ROIs). In addition, this paper addresses the diverse needs of different task levels and application scenarios by presenting on-orbit processing modes; the basic model driven by location information and the intelligent modes driven by target or scene content and change events. Finally, this paper delves into the critical algorithms necessary for adapting to the resource-constrained on-board environment. Using the on-orbit processing applications (APPs) on the Luojia-3 01 satellite, the performance of these algorithms was verified. Under the framework of task-driven processing and satellite-ground coordination, the Luojia-3 01 satellite successfully achieved high-precision on-orbit geo-positioning, rapid generation of multi-type image products, real-time extraction of information from both static and dynamic targets, and high-rate, near real-time compression of static and dynamic images. The on-orbit processing significantly enhances the efficiency of traditional ground-based processing and effectively supports subsequent real-time intelligent services in satellite remote sensing.

    Fast SAR autofocus based on convolutional neural networks
    Zhi LIU, Shuyuan YANG, Zifan YU, Zhixi FENG, Quanwei GAO, Min WANG
    2024, 53(4):  610-619.  doi:10.11947/j.AGCS.2024.20230281
    Asbtract ( )   HTML ( )   PDF (11943KB) ( )  
    Figures and Tables | References | Related Articles | Metrics

    Autofocus is a key technology for high-resolution synthetic aperture radar imaging. However, traditional SAR autofocus methods require too many iterations, have low computational efficiency, and are unsuitable for on-orbit processing. This paper proposes a fast SAR autofocus method based on convolutional neural networks. This method utilizes CNNs to learn the mapping from defocused images to focused images, mainly designed to correct the azimuth phase errors. It has a real-time performance and is more suitable for on-orbit processing since it does not need to iterate or adjust parameters in the testing phase. Experimental results on real SAR data show that our proposed method has the highest focusing quality and speed.

    Geodesy and Navigation
    Development and practice of CHZ-Ⅱ marine gravimeter
    Lintao LIU, Jiangang HE, Jiangning XU, Leijun LIU, Jinyao GAO, Zhongmiao SUN, Taiqi WU, Xinghui LIANG, Ming HU, Junjian LANG, Hongyang HE
    2024, 53(4):  620-628.  doi:10.11947/j.AGCS.2024.20220580
    Asbtract ( )   HTML ( )   PDF (8036KB) ( )  
    Figures and Tables | References | Related Articles | Metrics

    Marine gravimeter is an important instrument to obtain high-precision and high-resolution marine gravity field, which is mainly used in resource exploration, geoid refinement, marine science research and gravity assisted inertial navigation. The existing marine gravimeters in China mainly rely on foreign imports, which not only have high prices, but also have many inconveniences in the introduction and maintenance of the instruments. With the support of the major scientific instrument projects of the Ministry of Science and Technology, the CHZ-Ⅱ spring type marine gravimeter with completely independent intellectual property rights has been successfully developed based on the CHZ marine gravimeter. The new gravimeter uses metal zero length spring as the measuring element, and its unique axisymmetric vertical suspension structure can overcome the cross coupling effect. The application of high-precision servo control technology and strong damping can overcome the influence of carrier disturbance and ensure that the instrument maintains measurement accuracy under different sea conditions. High precision constant temperature technology and temperature compensation technology can eliminate the influence of ambient temperature fluctuation on the measuring accuracy of the instrument. This paper introduces the working principle, system structure, features and test results of the instrument. The static test shows that its monthly drift is better than 3 mGal; the wharf mooring test shows that its dynamic accuracy reaches 0.2 mGal, which can clearly reflect the gravity variation characteristics of sea tide; the ship borne test with a total voyage of more than 50 000 nautical miles shows that the comprehensive measurement accuracy is better than 1 mGal. To sum up, the accuracy and stability of the instrument have reached the level of the same kind of instruments at home and abroad, which can replace the current imported marine gravimeter.

    Analysis of ionospheric disturbance induced by Tonga volcanic eruption on January 15, 2022 based on GPS TEC
    Yiyong LUO, Dawei WU
    2024, 53(4):  629-643.  doi:10.11947/j.AGCS.2024.20220523
    Asbtract ( )   HTML ( )   PDF (12465KB) ( )  
    Figures and Tables | References | Related Articles | Metrics

    On January 15, 2022, the Tonga undersea volcano in the South Pacific Ocean erupted violently, which was the largest volcanic eruption in the past 30 years and produced strong atmospheric fluctuations, providing a rare opportunity for the study of volcanic ionospheric disturbances. Ionospheric disturbances caused by volcanic eruptions are calculated using GPS data near volcanoes, New Zealand, Australia and China, and the characteristics of traveling ionospheric disturbances (TIDs) were analyzed in terms of waveform, frequency, propagation speed and space-time distribution. The ionosonde, sea level monitoring and atmospheric pressure monitoring data are used to further analyze the propagation characteristics of TIDs. The results indicate that the eruption of the Tonga volcano has caused three types of TID in its vicinity: New Zealand, Australia and China. The first type of TIDs were detected in the vicinity of the volcano in the east, west, south and north directions, with a propagation speed of approximately 617~972 m/s. This type of TIDs is highly likely caused by sound waves generated by volcanic eruptions. The Tonga volcanic eruption only causes the second type of TIDs in the east and west directions near the volcano, and its propagation speed is about 472 m/s and 418 m/s, which may be caused by acoustic gravity waves or mixed waves derived from sound waves. The formation mechanism of the second type of TIDs needs further study. The Tonga volcanic eruption triggered the third type of TIDs in New Zealand, Australia and China, with a propagation velocity of about 328~352 m/s. This type of TIDs is closely related to Lamb waves.

    The improved max-flow/min-cut weight algorithm for InSAR phase unwrapping
    Yandong GAO, Yikun JIA, Shijin LI, Yu CHEN, Huaizhan LI, Nanshan ZHENG, Shubi ZHANG
    2024, 53(4):  644-652.  doi:10.11947/j.AGCS.2024.20220633
    Asbtract ( )   HTML ( )   PDF (8819KB) ( )  
    Figures and Tables | References | Related Articles | Metrics

    InSAR has been widely used in high-precision DEM inversion. The accuracy of phase unwrapping is one of the key steps affecting the accuracy of DEM acquisition, however, the areas with large-gradient changes has been the core issue affecting the accuracy of the unwrapping results. To address this issue, a phase unwrapping max-flow/min-cut algorithm (PUMA) based on the improved weights of the potential function is proposed in this paper. Firstly, the problem of unreasonable weight setting of the PUMA model is studied, and the priori information of the phase gradient change is obtained by using the external existing DEM, and the window maximum absolute phase gradient values are substituted into the corresponding potential function equation to obtain the weight value. Then, through the threshold adjustment of the potential function weight setting, the problem of unwrapping errors caused by the inability of PUMA potential function to function due to unreasonable potential function weight setting is solved, and thus the phase unwrapping accuracy in the region of large gradient change is improved. Finally, the proposed method is validated by simulated data and real TanDEM-X InSAR data, and compared with existing phase unwrapping methods. The results show that the proposed algorithm can improve the deconvolution accuracy by at least 44.93% in the simulated data, and the proposed algorithm can obtain a larger range of effective unwrapping results than the existing algorithm in the region of large gradient variations in the real data.

    Preliminary analysis to positioning precision and crustal movement of BDS-3 data recorded by the China seismic experiment site
    Tian HE, Guojie MENG, Weiwei WU, Xiaoning SU, Guoqiang ZHAO, Congmin WEI, Zhihua DONG
    2024, 53(4):  653-665.  doi:10.11947/j.AGCS.2024.20230044
    Asbtract ( )   HTML ( )   PDF (15603KB) ( )  
    Figures and Tables | References | Related Articles | Metrics

    The BDS-3 navigation satellite system of China was completely accomplished in July 2020, then starting to provide services for global users. GNSS stations in the China seismic experiment site (CSES) have been receiving BDS-3 satellite data ever since, and accumulated observational data for more than 2 years, providing an important platform in acquiring observational data for the application of BDS-3 in exploring crustal movement in Sichuan-Yunan area. To assess the current precision of the BDS-3 positioning and its performance in crustal movement monitoring, we first evaluate the quality of BDS-3 observational data according to the relationship of multi-path effect and signal-to-noise ration with elevation angles. Using GAMIT/GLOBK (version 10.7), we have processed the simultaneously recorded BDS-3 and GPS data separately to obtain coordinate time series for each station. Fitting the time series of three coordinate components separately with a functional model which encompasses linear, annual, semi-annual and other terms by the maximum probability estimation method, we obtain velocity, amplitude and phase for E, N and U components of the coordinate time series. Furthermore, we evaluate the precision of BDS-3 and GPS by comparative analysis of the fitting results. Finally we discuss the possible factors which could influence the positioning precision of BDS-3, and regional features of the horizontal velocity field derived from BDS-3 observations. The results show that the quality of raw data for BDS-3 observations is comparable with GPS data. The data fitting for BDS-3 and GPS time series shows that the average value of root mean square (RMS) of the BDS-3 residual time series are 4.42, 4.25 and 8.34 mm for the E, N and U components, respectively, larger than those of GPS data. Systematic difference is identified of about 2 mm/a in E direction between velocity fields of BDS-3 and GPS. The velocity fields, and the annual and semi-annual signals derived from BDS-3 and GPS data do not show obvious differences in regional distribution. We think that currently the factors affecting the positioning precision of BDS-3 are the relatively lower precision of satellite orbit and clock difference products, the shortage of empirical solar pressure and satellite antenna phase center correction for BDS-3 and so on. The difference between the velocity fields of BDS-3 and GPS is due to the inconsistence of their reference frames. We expect that, with the continuous accumulation of BDS-3 observational data and the improvement of the data processing models, the positioning precision of BDS-3 will enhance with time, and the BDS-3 will be used independently from GPS to provide geodetic products of high-precision for monitoring crustal movement at CSES.

    Fitting improvement of PPP receiver tracking error stochastic model for high latitudes
    Yingzong LIN, Xiaomin LUO, Junfeng DU
    2024, 53(4):  666-676.  doi:10.11947/j.AGCS.2024.20230323
    Asbtract ( )   HTML ( )   PDF (8154KB) ( )  
    Figures and Tables | References | Related Articles | Metrics

    Ionospheric scintillation refers to the phenomenon of rapid random fluctuations in the amplitude and phase of radio signals which can result in increasing of measurement noise and signal loss of lock. At present, the receiver tracking error stochastic (RTES) model can effectively reduce the influence of ionospheric scintillation on GNSS precision point positioning (PPP). However, the RTES model relies on the data products from specialized ionospheric scintillation monitoring receivers (ISMR). Compared with geodetic receivers widely distributed around the world, the number of ISMR monitoring stations is very limited, and the acquisition of ISMR data products is difficult. Using the GPS observation and scintillation data recorded by the Canadian high arctic ionospheric network (CHAIN) during 2014—2022, this study proposes a PPP stochastic model suitable for geodetic GNSS receivers in high latitudes, referred to as the high latitude receiver tracking error (HL_RTES) model. The HL_RTES model uses S4c index and rate of total electron content index (ROTI) to estimate the receiver tracking error variance, and the GPS positioning accuracy is improved by reweighting the observed values. The GPS observations of CHAIN station from February 1, 2023 to February 28, 2023 are used to conduct single-frequency mimics kinematic PPP experiment. Experimental results show that the performance of HL_RTES model and RTES model are comparable, and both can improve the PPP positioning accuracy under ionospheric scintillation; compared with EAS model, the monthly average RMS improvement rates of HL_RTES model in the horizontal, vertical and 3D directions are 33.3%, 42.8% and 38.3% respectively. In addition, using the data from the IGS stations INVK, KIRU, SCOR and URAL on February 15, 2023 to conduct experiment, it is found that the HL_RTES model can significantly mitigate the effects of high latitudes scintillation on PPP; compared with the EAS model, the improvement rates of HL_RTES model based PPP on the 3D RMS of the four stations are 57.2%, 32.9%, 43.8% and 31.4%, respectively.

    Photogrammetry and Remote Sensing
    Knowledge graph-guided deep network for high-resolution remote sensing image scene classification
    Yansheng LI, Minlang WU, Yongjun ZHANG
    2024, 53(4):  677-688.  doi:10.11947/j.AGCS.2024.20230125
    Asbtract ( )   HTML ( )   PDF (11844KB) ( )  
    Figures and Tables | References | Related Articles | Metrics

    Thanks to the rapid development of deep network theory and methods, deep networks have gradually become the mainstream technology for remote sensing image scene classification tasks. However, existing deep network-based remote sensing image scene classification methods are highly dependent on a large number of manually labeled training samples and cannot effectively integrate and utilize the rich prior knowledge in the remote sensing field. In order to improve the utilization of domain knowledge while reducing the dependence on labeled samples, this paper proposes a knowledge graph-guided deep network learning method for high-resolution remote sensing image scene classification. First, this paper constructs a land cover concept knowledge graph that includes various sources of knowledge in the field to more flexibly and conveniently apply domain prior knowledge. Furthermore, through the knowledge graph representation learning method, the semantic categories of remote sensing scenes in the land cover concept knowledge graph are expressed as semantic vectors to form a semantic benchmark for remote sensing scene categories. In the knowledge-guided learning stage, the cross-modal alignment constraint between the scene category semantic vector and the shallow visual feature vector of the deep network is applied to guide the shallow part of the deep network to more effectively learn shared features of different categories of remote sensing image scenes, while in the deep part of the deep network, it is still guided by scene category labels to learn discriminative features of different remote sensing scenes. In the testing stage, the optimized deep network model can complete high-precision remote sensing image scene classification without relying on any prior knowledge. The experimental results on the currently largest publicly available remote sensing image scene classification dataset show that the proposed knowledge-guided learning method can obtain optimal classification performance at different training sample ratios such as 10%, 30%, and 50% compared with existing methods. Under the condition of 10% sample ratio, our proposed method can achieve an improvement of 5.11% in overall accuracy (OA) compared with baseline deep networks.

    Design of optical remote sensing satellite onboard processing system based on model definition
    Wei LIU, Songlin LIU, Zibo GUO, Kai LIU, Lizhe ZHANG
    2024, 53(4):  689-699.  doi:10.11947/j.AGCS.2024.20220658
    Asbtract ( )   HTML ( )   PDF (8337KB) ( )  
    Figures and Tables | References | Related Articles | Metrics

    In this paper, a design method of optical remote sensing satellite on-board processing system based on model definition and model-based systems engineering (MBSE) is proposed, and a model paradigm of on-board processing task behavior of “hardware resource-operator module-general processing core-typical application” is constructed. At the hardware level, the on-board heterogeneous embedded computing platform is adopted to integrate the design of multiple processors, and complete high-speed signal transmission and data processing through standardized high-speed data interconnection. In order to support changes in device scale and data processing complexity, good scalability through network topology should be achieved. At the software level, in view of the requirements of frequent data read and write operations, a large number of repeated computing operations, and general computing and acceleration of convolutional neural networks in on-board intelligent processing tasks, a configurable operator module composed of the instruction set of the on-board platform, the general operator of the on-board algorithm and the components of the on-board intelligent network is designed, and the hardware IP core of the specific algorithm can be quickly implemented based on the module. Simulation experiments show that this method can adapt the best software and hardware solutions according to the needs of satellite platforms and on-board processing tasks, and effectively improve the utilization rate of computing resources. The proposed real-time streaming compression coding scheme combined with cloud detection significantly improves the compression performance. The designed lightweight target detection and recognition method achieves a computing resource efficiency of 91.5%. Taking the raw data rates of GF-1 and GF-7 as examples, the overall computing resource utilization rate increased by 16.51% and 17.77% respectively compared with the conventional solution. The method has realized the modeling of the development process, the definable working mode, and the sharing of logical resources of on-board processing system, which is a reasonable optimization choice to adapt to the miniaturization and rapid deployment of satellites.

    Urban impervious surface extraction based on the deep features of high-resolution remote sensing image and ensemble learning
    Xuetao LI, Pancheng WANG, Yongnian ZENG
    2024, 53(4):  700-711.  doi:10.11947/j.AGCS.2024.20220389
    Asbtract ( )   HTML ( )   PDF (15019KB) ( )  
    Figures and Tables | References | Related Articles | Metrics

    The effective extraction of urban impervious surface is important on the application of high-resolution remote sensing. Focusing on existing issues, an urban impervious surface extraction method is proposed based on U-Net combining with ensemble machine learning. The impervious surface areas with different density are selected as the experimental areas. Firstly, the deep features of high-resolution images are extracted by U-Net with the GF-2 multispectral data. Then, the urban impervious surface is extracted by using the ensemble learning with stacking mechanism. The experimental results show that the ensemble learning based on deep features of high-resolution remote sensing image can obtain high accuracy of urban impervious surface extraction. In the experimental areas with different density of urban impervious surface, the overall accuracy is not less than 91.66%, and Kappa is not less than 0.83. The commission error is 7.83%~9.39%; the omission error is 7.22%~14.88%. Compared with the ensemble learning, random forest and support vector machine based on image spectral features, the overall accuracy and Kappa are increased in experimental areas with relatively sparse, medium dense, dense and complex distribution of impervious surface. The commission and omission errors are significantly reduced. This indicates that the deep features can effectively improve the mapping accuracy and user accuracy of integrated learning to extract impervious surface. Compared with U-Net and SegNet, the overall accuracy and Kappa are increased by in the four experimental areas with relatively sparse, medium dense, dense and complex distribution of impervious surface. The commission errors are significantly reduced. The integration ensemble learning with deep learning can effectively improve the mapping accuracy and user accuracy of impervious surface extraction. In general, the ensemble learning based on the deep features of high-resolution remote sensing image can obtain higher accuracy of urban impervious surface extraction, which has application prospects in urban land use/cover classification.

    Lightweight SAR target detection based on channel pruning and know-ledge distillation
    Qihao HUANG, Guowang JIN, Xin XIONG, Limei WANG, Jiahao LI
    2024, 53(4):  712-723.  doi:10.11947/j.AGCS.2024.20220605
    Asbtract ( )   HTML ( )   PDF (15100KB) ( )  
    Figures and Tables | References | Related Articles | Metrics

    Lightweight SAR target detection algorithm is of great significance for the rapid detection of ground objects in SAR images. Aiming at the low precision of lightweight detection algorithm, a lightweight SAR target detection method combining channel pruning and knowledge distillation was designed. In this method, the importance of the corresponding feature channels is identified by sparse training of the scaling factor γ of the batch normalization layer in the complex network, and then the secondary channels are cut. After fine-tuning training, it is used as a teacher network to construct a knowledge distillation framework to guide the training of lightweight model and improve the detection accuracy of light weight model. The YOLOv5-6.1 algorithm was used to build a detection framework, and the training and detection experiments were carried out on the reconstituted MSAR and SSDD multi-class target datasets. The results show that the proposed method can improve the accuracy of SAR target detection while maintaining the lightweight model size of only 3.73 MB, which verifies the effectiveness of the proposed method.

    Cartography and Geoinformation
    Spatial co-location pattern mining based on graph structure
    Jinghan WANG, Tinghua AI, Hao WU, Haijiang XU, Guangyue LI
    2024, 53(4):  724-735.  doi:10.11947/j.AGCS.2024.20230012
    Asbtract ( )   HTML ( )   PDF (12868KB) ( )  
    Figures and Tables | References | Related Articles | Metrics

    Under the first law of geography, spatial co-location patterns reflect the dependency of different geographic elements’ distribution, satisfying the association discovery of big spatial data analysis. Spatial co-location pattern mining needs to consider the spatial conjunction mechanisms, detect spatial neighborhood relationships and search high-frequency patterns with metrics such as support. The common co-location mining methods usually combine geometric computation and logical reasoning, which resulting in the need to correct geometric neighborhoods while mining higher-order co-location patterns. Considering that the topological information contained in graph data is suited to spatial co-location pattern, this study proposes a graph structure-based co-location pattern mining method that completes the geometric proximity detection in one step, and then completes the logical co-location pattern discrimination by subgraph search in the graph database. Firstly, we construct the adjacency graph based on the Delaunay triangle network and use an adaptive adjacency filter to eliminate invalid connections. Second, the N+1 elements of candidate co-location patterns are obtained recursively from the N elements through continuous joining, pruning, and growing of subgraphs. Finally, the spatial co-location patterns are determined by calculating the support metrics and compared with predefined thresholds. Based on the concept of continuous graph traversal, this study improves the generality of spatial co-location pattern mining in complex scenarios. Experiments show that this method is more efficient than traditional algorithms, with better results in multivariate spatial co-location pattern mining.

    A building aggregation method based on deep clustering of graph vertices
    Zhanlong CHEN, Xiechun LU, Yongyang XU
    2024, 53(4):  736-749.  doi:10.11947/j.AGCS.2024.20230316
    Asbtract ( )   HTML ( )   PDF (14173KB) ( )  
    Figures and Tables | References | Related Articles | Metrics

    Building element aggregation is pivotal for simplifying spatial structures in cartographic generalization. Conventional rule-based aggregation methods often cannot simultaneously consider the morphological and distributional characteristics of the features, because they are heavily influenced by preset algorithm parameters and lack flexibility in the cartographic generalizing process. To fill these limitations, this paper proposes a building aggregation model based on deep clustering of graph vertices. The model utilizes the Delaunay triangulation network to construct a representation graph model of building groups and combines an autoencoder and graph convolutional network to learn the subdivided triangles’ geometric shapes and spatial distribution features. A self-supervised learning approach is employed to cluster and classify the triangles into the categories of “retain” and “delete”. Consequently, it aggregates buildings intelligently in an end-to-end manner without relying on predefined samples. The experimental results demonstrate that the proposed method reduces reliance on preset aggregation parameters while simultaneously considering building elements’ morphology and distribution features. The aggregation process exhibits a certain degree of flexibility, resulting in aggregated buildings better aligning with the requirements of map visualization.

    Deep learning-based spatio-temporal prediction and uncertainty assessment of urban PM2.5 distribution
    Huimin LIU, Chenwei ZHANG, Kaiqi CHEN, Min DENG, Chong PENG
    2024, 53(4):  750-760.  doi:10.11947/j.AGCS.2024.20230071
    Asbtract ( )   HTML ( )   PDF (13594KB) ( )  
    Figures and Tables | References | Related Articles | Metrics

    The goal of predicting PM2.5 concentration is to achieve a comprehensive perception of the PM2.5 distribution in the study area based on limited observations. Ideal prediction models are required to ensure both high accuracy and high reliability of the results. However, most of the existing studies prioritize the efforts to improve accuracy, which ignores the uncertainty of results caused by data and model. This greatly limits the reliability and potential availability of high-precision prediction results, making it difficult to assist practical applications such as air pollution control effectively. To overcome this problem, this paper proposes a PM2.5 concentration spatiotemporal distribution prediction model with coupled uncertainty assessment. The prediction module, mainly based on graph convolutional and recurrent networks, achieves high-precision prediction of PM2.5 concentration. Meanwhile, the uncertainty quantification module based on adversarial learning strategies and variational autoencoder is constructed to synchronously reveal the uncertainty level of the prediction results. Extensive evaluations of real-world dataset show that the proposed model can effectively balance the accuracy and reliability of PM2.5 concentration prediction results, providing scientific decision-making support for environmental management.

    A method for hydrological information extraction from historical maps combining SAM large model and mathematical morphology
    Fei ZHAO, Zhaozheng LI, Quan GAN, Zuyu GAO, Zhanchu WANG, Qingyun DU, Zhensheng WANG, Yang SHEN, Wei PAN
    2024, 53(4):  761-772.  doi:10.11947/j.AGCS.2024.20230308
    Asbtract ( )   HTML ( )   PDF (15872KB) ( )  
    Figures and Tables | References | Related Articles | Metrics

    Historical maps record rich historical geographic information, which can help understand the laws of historical movement and provide reference for contemporary development. Different from modern maps, remote sensing images and other data, the historical map has been preserved for a long time, and there are some problems such as small number of reservations and low image accuracy. Map symbols are also different from modern maps, so the information is difficult to be extracted efficiently. Aiming at this problem, this study proposes an intelligent method for extracting hydrological information from historical maps based on the experimental data of topographic map of ditches and channels along the Yellow River in Ningxia province. Firstly, the datasets are constructed by clustering and mathematical morphology methods combined with symbolic syntax. Then, the general large model SAM structure is improved and the weight is optimized by transfer learning. Finally, the historical map hydrological information is automatically extracted by improved SAM. Comparing the experimental results with other models, it shows that the extraction results of this method have clear boundaries, complete contours, and the highest accuracy and accuracy. At the same time, the extraction results are compared with the current situation of the hydrological in the region. It is found that most of the rivers and ditches in history are now diverted, offset or disappeared, and the lake area is greatly reduced. The method in this paper is improved based on the SAM general large model, which verifies the availability of the large model in the map field and provides a new idea for map information extraction.

    Summary of PhD Thesis
    Time-dependent road network model and its application
    Xiuquan LI
    2024, 53(4):  773-773.  doi:10.11947/j.AGCS.2024.20230086
    Asbtract ( )   HTML ( )   PDF (1099KB) ( )  
    References | Related Articles | Metrics
    Research on smartphone IMU/geomagnetism/Wi-Fi FTM fusion localization problems
    Meng SUN
    2024, 53(4):  774-774.  doi:10.11947/j.AGCS.2024.20230090
    Asbtract ( )   HTML ( )   PDF (1484KB) ( )  
    References | Related Articles | Metrics
    Single-epoch ambiguity resolution methods for multi-frequency GNSS
    Yuzhao LI
    2024, 53(4):  775-775.  doi:10.11947/j.AGCS.2024.20230091
    Asbtract ( )   HTML ( )   PDF (1185KB) ( )  
    References | Related Articles | Metrics
    Research on multi-view index method of urban scene data
    Xiangqiang MIN
    2024, 53(4):  776-776.  doi:10.11947/j.AGCS.2024.20230094
    Asbtract ( )   HTML ( )   PDF (1201KB) ( )  
    References | Related Articles | Metrics
    Research on evaluation of coordinated development of Jing-Jin-Ji region based on multisource spatiotemporal data
    Jianwan JI
    2024, 53(4):  777-777.  doi:10.11947/j.AGCS.2024.20230095
    Asbtract ( )   HTML ( )   PDF (1244KB) ( )  
    References | Related Articles | Metrics
    Dispatching policy optimizing of cruise taxi in a multiagent-based deep reinforcement learning framework
    Xiangyuan MA
    2024, 53(4):  778-778.  doi:10.11947/j.AGCS.2024.20230098
    Asbtract ( )   HTML ( )   PDF (1203KB) ( )  
    References | Related Articles | Metrics