[Purposes] Long-distance oil and gas pipelines cover a wide area,with diverse terrain and landforms along the way and rather complex geological conditions.When passing through areas prone to geological disasters,pipelines are prone to damage,which can cause huge property losses.Therefore,it is particularly important to monitor landslides along oil and gas pipelines. [Methods] This paper takes the area passed by the Changqing oil and gas pipeline as the research area.The DS-InSAR technology is adopted to obtain the surface deformation information in the research area.Based on the analysis of Sentinel-1 data from January 2020 to May 2022,landslide identification is carried out,and combining soil erosion modulus and multiple influencing factors,the random forest interpretability analysis method is used to quantify the relative contributions of landslide inducing factors in the study area. [Findings] A total of 348 landslide hazard points were identified in the study area.Among them,188 landslide hazard points were identified within a 5 km corridor along the Changqing oil and gas pipeline. [Conclusions] The factor contribution analysis results showed that terrain undulation,normalized vegetation index (NDVI),soil erosion modulus and precipitation were the core dominant factors affecting the stability of the slopes in the study area．

[Purposes] To address the issues of insufficient displacement inversion accuracy in landslide monitoring using multi-temporal UAV LiDAR point clouds under conditions of fine registration requirements and lack of stable control points in mountainous areas,a 3D deformation monitoring method for landslides based on bidirectional nearest-neighbor(BNN) matching and local plane fitting is proposed. [Methods] Firstly,multi-temporal ground point clouds are constructed through flight strip adjustment,noise removal,and ground point filtering.Then,strong-matching point pairs are selected via bidirectional nearest-neighbor matching,and horizontal and vertical displacements are decomposed using local plane fitting.Finally,weighted principal component analysis is applied to invert the 3D displacement field and determine the primary sliding direction. [Findings] Experiments conducted on the Tudi Ridge landslide in the Three Gorges Reservoir area demonstrate that,without external control points,the method can stably detect horizontal displacement changes as small as 2~3 cm,with an absolute error of 4~6 mm compared to GNSS measurements,and the displacement direction aligns well with the main sliding direction of the landslide. [Conclusions] Compared with DEM differencing,C2C,and M3C2 methods,this approach effectively reduces the accumulation of registration errors in complex terrain,showing good adaptability and reliability in scenarios lacking control points,thus providing technical support for landslide deformation monitoring and disaster early warning in vegetated areas．

[Purposes] Landslides are widely distributed,frequent,and highly destructive geological disasters,and evaluating their susceptibility is crucial for disaster early warning and risk management.This paper proposes a WOA-BP landslide susceptibility evaluation model that incorporates DInSAR surface deformation features to address the limitations of traditional methods that only rely on static factors and neglect surface deformation characteristics. [Methods] Using Dejiang county,Guizhou province as the study area,the DInSAR technology based on Sentinel-1A satellite ascending data is employed to extract surface deformation features,combined with historical landslide data and 10 static factors,to evaluate landslide susceptibility. [Findings] The results show that compared to BP,LSTM,SVM,and RF models,the WOA-BP model's AUC value improves to 0.800,and its prediction accuracy increases by at least 10.19%. [Conclusions] The proposed model,which includes surface deformation features,effectively corrects nearly 30.9 percentage points of misclassified areas,significantly improving the reliability of landslide susceptibility evaluation and providing a new method for accurate landslide susceptibility assessment．

[Purposes] Research on geological hazards in large-scale water conservancy and hydropower projects located in mountainous and highly vegetated areas remains relatively scarce,and traditional geological hazard investigation methods are difficult to implement in such regions. [Methods] This study takes the Wujiangdu hydropower station as a case study,and combines 20 ascending orbit scenes and 23 descending orbit scenes of L-band Lutan-1 (LT-1)data with strong penetration capability with the SBAS-InSAR technique to monitor surface deformation and identify geological hazards in this inland area.In addition,spatiotemporal deformation characteristic analysis is conducted by integrating optical imagery,rainfall data,and geological fault data. [Findings] The results show that：①A total of 24 geological hazards were identified in the Wujiangdu Hydropower Station area using the ascending and descending orbit data of Lutan-1.Among these hazards,the largest deformation zones were jointly detected by both orbit datasets,with deformation rates reaching 125.96 mm/a for the ascending orbit and 103.92 mm/a for the descending orbit,which verifies the effectiveness of the proposed method for large-scale water conservancy and hydropower projects in mountainous and highly vegetated areas.②The combination of ascending and descending orbit data can effectively overcome the limitations caused by topographic factors and single-orbit datasets in such regions,thereby improving the accuracy of geological hazard identification.③The spatiotemporal evolution analysis of typical potential landslide bodies indicates that rainfall and geological faults are the dominant inducing factors for the formation of geological hazards in the study area. [Conclusions] This method provides scientific data support for geological hazard assessment and precise prevention and control in reservoir areas of large hydropower stations in mountainous regions with high vegetation．

[Purposes] Remote sensing image target recognition and knowledge graph technology have become one of the important technical methods for the update and application of real-scene 3D data.However，there are still problems such as low automation level，large amount of data annotation，and inconsistent relationship models and classification systems in the extraction of the association relationships between remote sensing image target entities. [Methods] This paper proposes a technical framework for extracting the relationship between remote sensing image target entities based on the DeepSeek large model.Through the construction of relationship models and category systems and the production of remote sensing image entity relationship data based on prompt engineering，the construction of remote sensing image entity knowledge graphs is completed.To verify the effectiveness of the proposed method，experiments were conducted on a dataset of 3392 high-resolution remote sensing images，focusing on relation extraction for typical complex surface targets. [Findings] The results show that the overall F1 value of the relationship extraction of this method on mainstream large models is above 0.85. [Conclusions] This method can quickly adapt to different remote sensing scenarios，is flexible and convenient to operate，and provides structured knowledge support for tasks such as remote sensing recognition and geographic question answering，promoting the transformation of remote sensing intelligent interpretation from the “feature perception” paradigm to the “knowledge reasoning”paradigm．

[Purposes] Hyperspectral remote sensing imagery is often limited by high acquisition costs and low signal-to-noise ratios.To address these challenges，this paper introduces a Transformer-based spectral implicit function (TSIF) to reconstruct high-quality hyperspectral data from low-spectral-resolution images in a low-cost manner. [Methods] The method models the wavelength-reflectance relationship of each pixel as a continuous function，employs a Transformer-based encoder to fully extract joint spatial-spectral features，and integrates target wavelength information during decoding to achieve fine-grained hyperspectral reconstruction.Additionally，the spectral angle mapper (SAM) term is incorporated into the loss function to enhance the physical consistency of the reconstructed spectra. [Findings] Experiments on two benchmark datasets demonstrate that TSIF achieves superior performance over existing methods，with an improvement of approximately 1 dB in MPSNR and a reduction of more than 0.1 in SAM，confirming its effectiveness in preserving spectral details and recovering spatial structures. [Conclusions] TSIF thus provides an efficient solution for low-cost acquisition of high-quality hyperspectral imagery．

[Purposes] To address the limitations of traditional remote sensing methods in tree species classification accuracy under strong background interference,this study proposes an optimized random forest model integrating the northern goshawk optimization(NGO)algorithm and an attention mechanism (NGO-Attention-RF). [Methods] Taking Fuxin city,Liaoning province as the study area,multi-source remote sensing data were fused to construct a multi-dimensional feature set.The NGO algorithm was employed to optimize the hyperparameters of the random forest,thereby avoiding local optima.An attention mechanism was introduced during node splitting to dynamically adjust feature weights,enabling the model to focus on key features that distinguish tree species. [Findings] Experimental results show that the proposed model achieved an overall accuracy of 85.4%.It outperformed methods such as MLP,CNN,RNN,Transformer,and SVM in terms of accuracy for each tree species category and overall classification metrics,validating the effectiveness of the NGO algorithm and the attention mechanism. [Conclusions] The study reveals the spatial distribution patterns of forest tree species in Fuxin city,elucidates the role of different features in tree species discrimination,and provides a methodological reference for forest monitoring in similar ecological regions．

[Purposes] High-density point cloud data provided by global ecosystem dynamics investigation(GEDI)LiDAR have been applied to the retrieval of forest canopy height over large areas.However,in complex terrain,GEDI data suffer from insufficient systematic bias calibration,which affects the accuracy of forest canopy height retrieval. [Methods] This study innovatively proposes a GEDI data bias calibration method by utilizing UAV LiDAR sample plot observation data to construct a random forest-based bias calibration model,thereby addressing the challenge of GEDI data bias calibration. [Findings] Validation results demonstrate that the proposed bias calibration model is highly effective in reducing systematic bias caused by terrain effects.After bias calibration,the R² increased from 0.29 to 0.80,and the RMSE decreased substantially from 11.80 m to 1.70 m. [Conclusions] The forest canopy height product retrieved from bias-calibrated GEDI data exhibits good temporal transferability and stability．

[Purposes] Interferometric synthetic aperture radar (InSAR)atmospheric water vapor monitoring is based on the principle of differential interferometry,with the inversion result representing the differential precipitable water vapor (ΔPWV),i.e.,a relative measure of PWV.To derive the absolute InSAR PWV values,this study proposes a refined atmospheric water vapor monitoring method by integrating InSAR,BeiDou Navigation Satellite System (BDS),and ERA5 reanalysis data. [Methods] By applying the StaMPS-InSAR technique to 25 Sentinel-1A images over the Jiaodong peninsula region,ΔPWV was retrieved.The InSAR-derived ΔPWV was then spatiotemporally calibrated using BDS PWV and ERA5 PWV to obtain absolute PWV values. [Findings] Results indicate that the correlation coefficients between InSAR PWV and PWV measurements from Qingdao sounding stations,FULS,and RUSH stations all exceed 0.96,with respective root mean square errors (RMSE)of 3.44,2.58,and 2.39 mm.Comprehensive validation demonstrated that InSAR PWV exhibits no significant systematic bias,with overall accuracy exceeding 3.5 mm.Compared with the single BDS calibration approach,the proposed joint BDS-ERA5 calibration method improves accuracy by 16.5%to 28.4%,demonstrating the reliability of the calibration strategy in complex coastal environments. [Conclusions] The proposed method yields absolute PWV with meter-level spatial resolution,effectively capturing fine-scale spatiotemporal variations.This approach provides significant reference value for investigating the distribution and evolution of PWV during extreme precipitation events in complex terrain regions．

[Purposes] This study aims to evaluate the accuracy and applicability of soil moisture products derived from the microwave radiation imager (MWRI)onboard the Chinese Fengyun-3D (FY-3D)satellite over Henan province,providing a scientific basis for the operational application of domestic satellite remote sensing data in regional agricultural drought monitoring and water resource management. [Methods] Using in-situ observations from the Henan automatic soil moisture monitoring network at 0～10 cm depth in 2024 as reference data,the FY-3D ascending and descending soil moisture products are systematically validated after pixel representativeness screening and overpass time matching,and spatial consistency cross-comparison with contemporaneous SMAP products was also conducted. [Findings] The FY-3D soil moisture products exhibited consistent temporal trends with ground observations.The descending product demonstrated significantly better accuracy than the ascending product,with a bias of 0.002 0 m³·m^-3,a root mean square error (RMSE)of 0.027 9 m³·m^-3,and a correlation coefficient of 0.833 6.Pronounced seasonal characteristics were observed： errors were lower in winter and spring (October to April)with RMSE generally below 0.04 m³·m^-3,while errors increased in summer and autumn,peaking in July.Spatially,the correlation coefficient between FY-3D and SMAP products reached 0.720 1,with the highest accuracy observed in the homogeneous agricultural areas of the central Henan Plain (Bias< 0.03 m³·m^-3),while larger errors occurred in the hilly and mountainous regions of western and southern Henan,as well as around water bodies. [Conclusions] The FY-3D soil moisture products demonstrate promising potential for operational applications in the major agricultural areas of Henan Province,particularly during winter and spring and over homogeneous plain farmlands.For summer-autumn periods and regions with complex underlying surfaces,integrated application with multi-source data is recommended．

[Purposes] Semantic change detection aims to identify and distinguish different types of land surface changes from multi-temporal remote sensing images，holding significant application value in fields such as urban planning，environmental monitoring，and disaster assessment.However，traditional supervised methods heavily rely on pixel-level annotated data，which severely limits their practical deployment.In recent years，self-supervised learning has provided a promising approach to mitigate annotation dependency by leveraging large-scale unlabeled data for pre-training. [Methods] To this end，this paper proposes a self-supervised framework termed GLSNet for semantic change detection tasks.Built upon an encoder-decoder architecture，GLSNet incorporates joint global multi-scale contrastive learning and local feature matching contrastive learning during the pre-training stage to learn semantically consistent representations across temporal phases. [Findings] Experiments on the SECOND and Hi-UCD mini dataset demonstrate that GLSNet achieves superior performance in semantic change detection，particularly exhibiting significant advantages over existing methods in few-shot scenarios. [Conclusions] These results indicate that the proposed approach offers an efficient and practical solution for remote sensing semantic change detection．

[Purposes] 3D modeling of aerial and ground image collaboration is a significant challenge in real-scene modeling.This paper focuses on Fahai temple and White Pagoda temple in Beijing as research objects and proposes a 3D Gaussian splatting (3DGS)method based on aerial and ground image collaboration,which aiming to solve the high-quality reconstruction of scene with drastic changes in aerial and ground perspectives. [Methods] Firstly,the ground multi-modal heterogeneous equipment array acquisition strategy is implemented to achieve comprehensive and high-precision near-ground images of complex scenes.Secondly,a stepwise rigid constraint resolution method is employed to accomplish integrated aerial-ground joint aerial triangulation.Finally,a combined modeling strategy integrating block-based parallel training,multi-source camera viewpoint balancing,generative viewpoint augmentation,and spatio-temporal consistency filtering is utilized to reconstruct a high-fidelity and seamless 3D Gaussian splatting model. [Findings] The results demonstrate that the proposed method outperforms mainstream modeling approaches in terms of model completeness,geometric accuracy and visual details.Specifically,the peak signal-to-noise ratio (PSNR)is increased by 1.143 7 dB,the structural similarity index measure (SSIM)is improved by 0.022 3,and the learned perceptual image patch similarity (LPIPS)is reduced by 0.018 7. [Conclusions] The proposed method effectively addresses the challenge of high-quality 3D modeling under large viewpoint variations,which offers a new technological pathway for digital preservation of cultural heritage and fine-grained urban management．

[Purposes] In complex environments like urban canyons,GNSS signals are susceptible to multipath effects and blockages.Traditional extended Kalman filter algorithms often suffer from IMU bias estimation errors and filter divergence due to a lack of robust outlier handling and state constraints.This paper proposes an RTK/INS integrated positioning method based on adaptive regularization constraints. [Methods] A multi-level quality control mechanism is constructed： first,a motion consistency check model is employed to detect GNSS outliers online； second,the IGG-Ⅲ robust estimation theory is introduced to calculate equivalent weight factors.These factors are mapped to adaptive regularization parameters to dynamically adjust the constraint intensity on IMU bias states,thereby actively mitigating the impact of abnormal observations. [Findings] Vehicle experiments demonstrate that,compared with the traditional EKF loose coupling,the proposed method significantly enhances system stability.The positioning accuracy in the east and north directions is improved by 57.4% and 44.6%,respectively,with the overall root mean square error reduced by 45.0%. [Conclusions] The proposed method effectively ensures positioning accuracy and robustness in harsh observation environments．

[Purposes] To investigate the impact of ionospheric disturbances during geomagnetic storms on GNSS static positioning accuracy，this paper analyzes the ionospheric characteristics and GNSS positioning performance during the geomagnetic storm in November 2023. [Methods] Based on observation data from Kunming CORS reference stations，the ionospheric delay is computed using the uncombined precise point positioning (PPP) algorithm.The vertical total electron content (VTEC) interpolated from global ionospheric maps (GIM)is adopted to determine the upper and lower bounds of the disturbance background via the sliding interquartile range method，so as to identify and extract ionospheric disturbances during the storm.The effects of ionospheric disturbances on GNSS static positioning accuracy are evaluated using indicators including synchronous loop and asynchronous loop closure errors. [Findings] The experimental results show that signal scintillation caused by ionospheric irregularities during the geomagnetic storm is the main factor degrading GNSS static positioning accuracy.Under the 4 h observation duration，the relative accuracies of the three positioning schemes fail to meet the 1×10^-6 requirement for Class C networks. [Conclusions] The results clarify the ionospheric response characteristics and their significant influence on GNSS static positioning during the November 2023 geomagnetic storm，which provides guidance for GNSS static positioning surveys．

[Purposes] To assess the risk of regional land subsidence and clarify the future development trends of regional land subsidence，it is necessary to predict land subsidence and discuss the effectiveness of various prediction methods. [Methods] In the paper，based on long-term level monitoring data，land subsidence characteristics and its relationships with groundwater level in central urban area of Ningbo were analyzed，then some deep learning models including back propagation(BP) neural network，long short-term memory (LSTM)and attention-mechanism-based long short-term memory (AM-LSTM)were used for land subsidence prediction in the study area，the corresponding prediction errors and prediction results were compared and analyzed. [Findings] The results show that： ①The prediction effect of cumulative settlement data is significantly better than that of settlement rate，and monitoring data volume and prediction error show a roughly linear relationship.②After the groundwater was banned from extraction，more than 80% of groundwater monitoring points in the study research area show weak correlation with ground subsidence，and the prediction error of land subsidence increases when groundwater level factors are taken into account.③After removing individual monitoring points，The average relative errors for BP and LSTM predictions are 8.4%～15.8% and 7.1%～7.7%，respectively.The predicted trends of land subsidence development by BP and LSTM are basically consistent，but the LSTM predictions are more stable.④When the amount of monitoring data and the sliding window size is limited，the prediction error of the AM-LSTM model incorporating attention mechanisms has essentially not decreased. [Conclusions] This study demonstrates that using the deep learning methods for subsidence time series prediction is feasible based on long-term level monitoring data，however，abnormal monitoring point quantities and prediction results vary with the different prediction models.When selecting prediction models，sample verification and error analysis should be considered．

[Purposes] To address the limitations of single monitoring techniques in terms of spatial coverage and stability for surface deformation monitoring of recently dredged and filled sites，a subsidence monitoring and prediction approach integrating GNSS and InSAR data with deep learning is proposed. [Methods] Taking an artificial reclaimed area in Doumen district，Zhuhai city，as the study area，the SBAS-InSAR technique was first applied to 22 Sentinel-1A images acquired from July 2022 to February 2024 to obtain subsidence time series and spatial distribution characteristics.Subsequently，a Kalman filter was employed to fuse GNSS and InSAR observations，generating deformation time series with improved stability.On this basis，a long short-term memory(LSTM)network optimized using the Adam algorithm was constructed for subsidence prediction，and the model performance was evaluated using MAE，RMSE，and R². [Findings] The results indicate that subsidence in the study area exhibits significant spatial heterogeneity.After data fusion，the standard deviations of deformation time series at three GNSS monitoring points decreased from 4.939 77， 2.276 89，and 2.660 43 mm to 3.863 44，2.000 49，and 2.298 29 mm，respectively.Compared with predictions based solely on InSAR data，the prediction errors using the fused data were reduced by 11.27%，13.61%，and 12.38%. [Conclusions] The proposed method effectively improves the stability of subsidence time series and the accuracy of subsidence prediction，providing technical support for subsidence monitoring and risk warning in coastal engineering areas．

[Purposes] To address the challenges of low accuracy in traditional bottom detection methods using side scan sonar data,which leads to difficulties such as image stitching problems and shape distortion,this study aims to achieve high-accuracy and stable real-time bottom detection for side scan sonars. [Methods] Accordingly,a dual-channel one-dimensional real-time bottom detection model based on the U-Net architecture (D1D-UNet)is proposed.The model takes dual-channel sequential data from the port and starboard sides of the side-scan sonar as input,enhancing its ability to capture both global and local features of the sonar data,thereby significantly improving bottom detection accuracy. [Findings] Experiments conducted on measured datasets from Meizhou Bay and Bayuquan demonstrate that the model achieved maximum accuracies of 97.26% and 96.16%,respectively.Compared with existing mainstream methods,the detection accuracy on the three test sets of Meizhou Bay improved by 5.35,6.58,and 12.05 percentage point. [Conclusions] This study provides an effective solution for real-time bottom detection using side scan sonars and further expands the application scope of side-scan sonars．

[Purposes] Multi-beam bathymetry system has the advantages of high efficiency and high resolution in ocean depth measurement and terrain acquisition,but due to changes in sea conditions,sound velocity profile errors and equipment noise and other factors,bathymetry data often contains a large number of random noise and outlier values,which seriously restricts the quality of water depth data and the accuracy of subsequent terrain modeling.In view of the insufficient adaptability of traditional filtering and regular discrimination methods to complex noise,this paper proposes an intelligent denoising method applied to multi-beam bathymetry data. [Methods] This method is based on convolutional neural network-bidirectional long short-term memory (CNN-BiLSTM)to model multi-beam water depth time series data,by learning the time series characteristics and spatial continuity of water depth changes,to achieve accurate prediction of normal water depth trends,and to achieve accurate prediction of normal water depth trends.The predicted residual is used as the implicit basis for noise discrimination and suppression,so as to complete the automatic removal of depth noise. [Findings] Experimental results show that compared with traditional denoising methods,the proposed method has better robustness and generalization capabilities in complex seafloor terrain and high-noise environments,can effectively reduce water depth measurement errors,improve the overall quality of multi-beam bathymetry data,and provides a new intelligent data processing idea for high-precision seabed topography mapping. [Conclusions] The CNN-BiLSTM model proposed in this article makes up for the limitations of a single filter.Compared with traditional threshold filtering and statistical filtering methods,this method does not require manual setting of complex parameters,can effectively suppress random noise while retaining real terrain details,and is suitable for large-scale multi-beam bathymetry data processing in complex underwater environments．

[Purposes] To address missed detections caused by the small size，weak feature representation，and complex-background interference of safety helmets，reflective vests，and other personal protective equipment in railway UAV imagery，this paper proposes a detection method integrating human spatial-structure priors and shallow feature enhancement. [Methods] Based on YOLO11n，prior regions are constructed according to the spatial relationships between personnel，safety helmets，and reflective vests.A region-weighted classification loss is designed to strengthen the learning of protective equipment within anatomically reasonable regions.Meanwhile，a high-resolution shallow feature branch is introduced to improve small-object representation. [Findings] On the self-built dataset，the proposed method achieved a P of 92.3%，R of 89.2%，mAP _0.5 of 94.3%，and mAP _0.5：0.95 of 50.7%，outperforming the baseline by 8.6，9.1，8.7，and 9.0 percentage points，respectively. [Conclusions] The proposed method effectively improves the detection accuracy of personnel and protective equipment in railway UAV imagery，providing technical support for construction safety localization and railway safety management．

[Purposes] This study aims to analyze the spatial matching mechanism between the greenery coverage ratio and the green view index of urban roads,addressing the issue of volume-perception mismatch in green space layouts. [Methods] High-precision data from over 2000 roads in Beijing were acquired using a bionic visual perception survey system and high-resolution remote sensing.Pearson correlation coefficients were employed to identify the greenery response scale,and a coupling coordination degree (CCD)model was used to quantitatively evaluate the synergy between the two indicators. [Findings] ①Road GVI and GCR are significantly positively correlated,with the correlation peaking at the 10 m buffer scale,defining the core sensitivity range for visual greenery.② The city's road network overall exhibits a level of primary coordination,with distinct type-based differentiation： expressways are characterized as GVI-lagging due to shielding by elevated structures and noise barriers,while county and township roads are GCR-lagging due to limited land-use conditions,and arterial roads tend toward synergistic development.③The CCD shows significant spatial heterogeneity at the sub-district level,where multi-node areas and ecological conservation zones outperform the central city； high-density built-up environments are the core areas restricting the conversion of greenery volume into perception. [Conclusions] This research reveals the scale effects and bottlenecks in the conversion of greenery volume to visual perception,providing a scientific reference for the precision planning of garden cities and the improvement of human settlement quality．

[Purposes] Traditional outlier detection methods process each station and direction independently，ignoring the spatio-temporal cross-correlation of GNSS monitoring data，and thus fail to identify multivariate collaborative anomalies.This paper aims to exploit dynamic couplings among monitoring series to improve such identification. [Methods] This paper develops a vector autoregressive (VAR)model integrating three-dimensional displacements (E，N，U) to analyze dynamic couplings among monitoring series.An adaptive criterion based on model residuals distinguishes true deformation from observation outliers. [Findings] In simulations，the F1-score of gross error detection with the proposed method improved by approximately 20% on average compared with the traditional 3-times mean error method.In field data from 13 stations on the Jiashao Bridge，the average RMSE in the E，N，and U directions decreased by approximately 60%，85%，and 75%，respectively，compared with the original series，and the overall RMSE was significantly lower than that of other traditional methods. [Conclusions] The method significantly improves data quality，providing a better foundation for long-term health monitoring．

[Purposes] To address the challenges of crop identification in hilly regions caused by complex terrain and fragmented fields,this study proposes an efficient intelligent crop identification method. [Methods] By integrating high-resolution Planet satellite imagery and UAV imagery,a “satellite-UAV”collaborative observation system was constructed.Precise samples are generated using UAV imagery.Based on multi-temporal data,a fusion of a two-dimensional convolutional neural network (CNN/U-Net)and a three-dimensional temporal network (3D-CNN/ConvLSTM)is employed to jointly explore the spatial-spectral-temporal features of crops,enabling intelligent crop identification. [Findings] The model achieved an overall accuracy of 87.51%(Kappa coefficient： 0.831).Wheat identification performed the best with a mapping accuracy of 96.32%,followed by rice and maize (>88%),while rapeseed reached 86.44%.Crop distribution exhibited distinct altitudinal characteristics： rice is concentrated in valleys below 360 m,whereas maize is widely distributed across hilly areas between 250～550 m. [Conclusions] The proposed method provides a reliable technical pathway for precise crop monitoring in complex terrain,offering practical value for advancing precision agriculture and regional agricultural management．

[Purposes] This paper aims to address the issues of unclear error accumulation mechanisms caused by multi-station serial connections and the difficulty in predicting accuracy under long-distance conditions in tunnel chain-type machine vision deformation monitoring systems. [Methods] This paper constructs a refined chained visual error propagation model that considers the station-by-station propagation and coupling of multi-source errors.It analyzes the comprehensive impact mechanism of calibration residual errors，sub-pixel positioning noise，environmental disturbances，and registration errors of overlapping points between adjacent measurement stations on the accuracy of remote measurement points，and simultaneously adopts the Topcon MS05AX total station for automated monitoring and accuracy evaluation. [Findings] Experimental results show that the internal consistency accuracy of chain-based machine vision is superior to 0.30 mm in both horizontal and vertical directions.The STD value generally shows an upward trend with increasing distance，and the overall repeatability in the vertical direction is slightly better than that in the horizontal direction.The RMS value in the horizontal direction is 1.04 mm，and the RMS value in the vertical direction is 1.13 mm.The predictive ability of the proposed error model for the attenuation trend of far-end accuracy is verified，revealing the inherent mechanism that vertical repeatability is superior to horizontal repeatability，and environmental disturbances are the main contributors to errors at long distances. [Conclusions] The error propagation model established in this paper can effectively explain the accuracy degradation pattern of the chain-based vision system in tunnel environments.The overall accuracy of the system can meet the requirements for monitoring structural deformations in subways，but the cumulative effect of errors over long distances is significant.It is recommended to adopt an error suppression strategy that combines regular joint calibration with redundant benchmark adjustments．

[Purposes] The pre-assembly test of tunnel segments is a necessary task for tunnel construction.Traditional manual methods have problems of low efficiency and large errors.The existing virtual assembly method of segments don't fully consider the constraint relationship between the splicing surface and the bolt hole.Based on this,this paper proposes a plane extraction and point cloud registration method that consider the consistency of the splicing surface and the alignment of the bolt hole positions. [Methods] It designs an automated segment virtual assembly process,through multi-stage noise reduction filtering,coarse registration,contact surface extraction,geometric constraint fine registration,single-ring virtual assembly,multi-ring dynamic virtual assembly,achieves high-precision virtual assembly of segments. [Findings] The experimental analysis and engineering application show that the method is superior to the traditional method in terms of accuracy,efficiency,and stability.It solves the problem of easy misalignment within and between rings,and the average joint seam can reach 1.2 mm. [Conclusions] It effectively improves the quality and efficiency of segment pre-assembly,and has high engineering application value．

[Purposes] This paper constructs a railway geographic network model integrating geographical attributes and topological characteristics，evaluates the changes of network structure and function under different attack strategies，and analyzes the vulnerability of railway network structure，so as to provide a scientific basis for improving the anti-interference ability of railway transportation system. [Methods] Based on the GIS technology and complex network theory，the Chinese railway geographic network is constructed using the Space L modeling method.Topological characteristics such as clustering coefficient，degree distribution，and betweenness centrality are analyzed.Taking the relative size of the giant connected component and network global efficiency as evaluation indicators，simulation strategies including random attack，initial degree/betweenness attack，and recalculated degree/betweenness attack are designed using Python and NetworkX to systematically evaluate the response of the network under node and edge failures. [Findings] The Chinese railway geographic network exhibits both small-world and scale-free characteristics，with significant spatial clustering effects.Intentional attacks cause more severe damage to network connectivity and global efficiency than random attacks.Among them，the recalculated betweenness attack reduces the giant connected component by nearly 100%in the early stage，and the network global efficiency approaches zero in the middle and late stages of the attack.Under random attacks，the network maintains relatively strong resilience. [Conclusions] The vulnerability of the railway geographic network is highly dependent on the distribution of key nodes and edges，and the recalculated betweenness attack can more accurately identify network weak links．

[Purposes] The real-scene 3D China has been initially established.This paper aims to deeply analyze its application development status，further explore its application value，and fully release the potential of data. [Methods] Based on the data of 304 real-scene 3D application cases released by the Ministry of Natural Resources，provincial and municipal information release platforms from 2024 to 2025，it systematically analyzes the scene distribution，regional differences，participating entities and evolution patterns of real-scene 3D applications in China by using spatial statistical analysis，comprehensive comparative analysis and other methods. [Findings] The results show that assisting government decision-making and supporting natural resource management are the core application scenarios，accounting for over 60%，which have contributed to the rapid growth of the digital economy.The application leading trend in the eastern region is remarkable，while the characteristic applications in the central and western regions are accelerating their rise.Public institutions have become the core implementing entities，and a multi-party collaborative model has initially taken shape. [Conclusions] In the future，it is necessary to promote the continuous deepening and expansion of real-scene 3D applications through regional collaboration，mechanism innovation，and technological integration，so as to better serve the high-quality development of China's economy and society．

[Purposes] Aiming at difficulties in data collection at land-water boundaries and insufficient full-region precision of single sensors for the Yangtze River bank collapse monitoring，this paper proposes a ground-air integrated modeling method coupling multi-source laser point clouds and close-range photogrammetry. [Methods] An “air-ground-water” integrated data acquisition system combining handheld SLAM，terrestrial laser scanning and UAV close-range photogrammetry is established.Heterogeneous point clouds are aligned via control point coarse registration and NDT-ICP two-step fine registration，and an extended point cloud constrained bundle block adjustment model with hierarchical weighting is constructed. [Findings] Tests on the Jingjiang River reach show the integrated model has a horizontal mean square error of 0.046 m and vertical mean square error of 0.072 m，with vertical accuracy improved by 49.3%versus pure image reconstruction. [Conclusions] The method overcomes single-sensor monitoring defects，supports automatic slope profile extraction and quantitative stability analysis，and provides high-precision technical support for early warning and precise management of bank collapse hazards．

[Purposes] Addressing the lack of research on the detection of linear targets using synthetic aperture sonar (SAS) and the difficulty of obtaining precise contours with traditional bounding box detection，this paper aims to develop a precise detection method suitable for shallow-buried linear targets. [Methods] By analyzing the slender structural characteristics of linear targets，a pixel-level semantic segmentation strategy was employed to construct a detection model for shallow-buried linear targets based on the DeepLabv3+ semantic segmentation network.A dataset was constructed using real SAS images，and the model was trained using Kaiming initialization，the SGD optimizer，and a dynamic learning rate strategy. [Findings] Experimental results show that the model achieved a precision of 0.812，a recall of 0.756，a mean pixel accuracy of 0.824，and a mean intersection over union of 0.791，outperforming comparison models such as U-Net，which demonstrates the ability to accurately segment the edge contours of linear targets and effectively distinguish targets from the seabed background. [Conclusions] This method effectively enables the automated and precise detection of shallow-buried linear targets，providing a methodological foundation for the automated detection of such targets in large-area seabed topography images．