Gis-based digital management system and method for land acquisition and demolition of traffic projects

Abstract

The application discloses a GIS-based traffic project land acquisition and demolition digital management system and method, and particularly relates to the technical field of geographic information intelligent monitoring, NDVI vegetation features of satellite images and building contour feature maps of unmanned aerial vehicle data are extracted respectively, unified geographic projection conversion is carried out in combination with sensor coordinate flow, a cross-modal feature alignment mechanism driven by an adversarial network is set, multi-scale spatial correlation features are extracted through wavelet transform, space-time consistency compensation coefficients are dynamically generated in combination with sensor displacement vectors, and a dynamic weight fusion model is used to realize adaptive fusion of multi-source data with centimeter-level precision; the spatial distance between boundary posts and demolition red lines is calculated in real time through a ray tracing algorithm, a dynamic safety threshold is generated in combination with historical dispute rates, the lagging defect of traditional static threshold early warning is broken through, and the precision and timeliness bottleneck problems in land acquisition and demolition management are solved by constructing multi-source geographic data fusion and dynamic early warning.

Description

Technical Field

[0001] The present invention relates to the technical field of geographic information intelligent monitoring, and more particularly to a GIS-based digital management system and method for land acquisition and demolition in transportation projects. Background Art

[0002] With the rapid development of transportation infrastructure construction, land acquisition and demolition have become a core part of project pre-construction work, and the effectiveness of its management directly impacts project progress and social stability. Traditional land acquisition and demolition management relies primarily on multi-source information such as satellite remote sensing imagery, manual mapping data, and paper-based ownership records, using GIS platforms for visualization and basic analysis. However, transportation projects often involve large-scale linear engineering land, and the demolition area has a complex geographical environment with intertwined structures, vegetation cover, and underground pipelines, and ownership boundaries frequently change. While existing technical systems can achieve basic spatial data collection, the inherent heterogeneity of multi-source data in terms of spatial benchmarking, temporal synchronization, and semantic representation leads to fragmented decision-making, seriously affecting the efficiency of demolition progress control and risk prediction.

[0003] Existing technologies have significant deficiencies in multi-source data fusion and dynamic early warning. First, there is a lack of a systematic framework for the unification of spatial benchmarks for multimodal geographic data. Satellite imagery, drone aerial photography, and sensor coordinate streams use independent projection conversion methods, and cross-platform data superposition is prone to centimeter-level cumulative errors, resulting in spatial offsets between the digital model of ownership boundaries and actual boundary markers. Second, there is a lack of a cross-modal feature alignment mechanism, and there is a lack of semantic association modeling between vegetation cover (NDVI), building outlines, and displacement monitoring data, making it difficult to eliminate inherent measurement biases between devices through feature compensation. More prominently, existing early warning models generally use static and dynamic safety thresholds, which are unable to dynamically adjust the judgment boundaries based on demolition progress, environmental disturbances, and historical dispute data, resulting in a surge in false alarm rates and response delays. The above problems seriously restrict the digitalization of land acquisition and demolition management, exacerbating the risk of ownership disputes and engineering compliance risks. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a GIS-based digital management system and method for land acquisition and demolition in transportation projects. By constructing a unified framework of multi-source data geographic references and a cross-modal feature alignment mechanism, it solves the problems of multi-source data error accumulation and risk warning lag in land acquisition and demolition proposed in the above-mentioned background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solutions: a GIS-based digital management system for land acquisition and demolition in transportation projects, comprising:

[0006] The multi-source geographic data acquisition module records the land acquisition and demolition area of ​​the transportation project as the target area and collects multi-source geographic data corresponding to the target area, including satellite image data, drone data, and boundary stake coordinate time series stream data (real-time 3D coordinates of boundary stakes are obtained through GPS or RTK positioning);

[0007] Preferably, the contour clarity score based on ResNet-50 feature extraction is used to obtain the drone data quality index Q UAV , the value range is [0, 1], and the sensor displacement rate V is obtained based on the boundary stake coordinate time series flow data disp ;

[0008] The multi-source geographic data benchmark unification module processes satellite image data through a superpixel segmentation algorithm to generate NDVI vegetation feature maps, uses a transfer learning network to extract building outline feature maps from drone data, synchronously receives boundary stake coordinate time series stream data, performs unified geographic projection coordinate conversion, and outputs multimodal standardized data;

[0009] The feature alignment and compensation module extracts the cross-modal edge feature map E based on the multi-scale spatial correlation between the NDVI vegetation feature map and the building outline feature map. m Generate spatiotemporal consistency compensation coefficients through adversarial networks, and correct the spatial offset of multi-source data by combining the displacement vectors in the sensor coordinate time series flow;

[0010] The dynamic weight fusion and credibility correction module uses a nonlinear function to dynamically assign weights to satellite imagery data and drone data based on the spatiotemporal consistency compensation coefficient and drone data quality indicators. It fuses data through an extended Kalman filter, triggers a particle filter fault-tolerant iteration, and integrates multi-source geographic data to output a centimeter-level accurate dynamic demolition model, which is a collection of 3D spatial data of the target area.

[0011] The 3D conflict warning module constructs a 3D voxel space of the land acquisition redline BIM model and the dynamic demolition model. It uses a ray tracing algorithm to calculate the minimum real-time distance between each boundary stake and the land acquisition redline. It also generates dynamic safety thresholds for target areas based on historical dispute rates and real-time distances, triggering multi-level alarm signals and freezing permissions.

[0012] Explanation: The three-dimensional conflict warning module takes the dynamic demolition model and sensor coordinate time series stream data as input, constructs the three-dimensional voxel space mapping relationship between the land acquisition red line BIM model and the dynamic demolition model, and combines the ray tracing algorithm to calculate the spatial distance between the sensor boundary stakes and the geographic red line in real time, and dynamically adjusts the dynamic safety threshold parameters based on the displacement rate, providing three-dimensional spatial data support for the accurate identification of land encroachment behavior and hierarchical triggering of authority management, ensuring that land acquisition and demolition management is upgraded from static rule execution to dynamic risk adaptive decision-making.

[0013] Preferably, the dynamic safety threshold is calculated as follows:

[0014]

[0015] Among them, α, β, k, and V0 are preset parameters that trigger multi-level alarm signals.

[0016] Preferably, if the risk diffusion coefficient is higher than the level threshold for three consecutive cycles, the influence weight of the historical dispute rate in the threshold calculation is increased synchronously to ensure that the dynamic security threshold is dynamically adapted to the risk situation.

[0017] Preferably, the risk diffusion coefficient RDC is calculated as follows:

[0018]

[0019] Among them, k is the sensor displacement rate sensitivity coefficient, V0 is the displacement rate reference value; HDR is the historical dispute rate, which reflects the proportion of the chaotic duration of the target area ownership in the total monitoring time in the historical period;

[0020] Preferably, the operation process of the feature alignment and compensation module includes:

[0021] Based on the NDVI vegetation feature map and building outline feature map in the multimodal standardized data, the multi-scale spatial correlation features of the NDVI vegetation feature map and the building outline feature map are extracted by wavelet transform as the cross-modal edge feature map E m ;

[0022] Construct an adversarial network model, input the NDVI vegetation feature map and the building outline feature map into the adversarial network model for semantic mapping relationship learning, and output the spatiotemporal consistency compensation coefficient K comp The spatiotemporal consistency compensation coefficient is used to quantify the strength of spatial alignment error correction of multi-source data. The value range is 0-10, and a larger value indicates a higher required compensation strength.

[0023] The covariance matrix of the displacement vector in the sensor coordinate time series stream is calculated, and the spatiotemporal consistency compensation coefficient is dynamically corrected in combination with the covariance matrix of the displacement vector to generate compensated multimodal standardized data.

[0024] Preferably, in the dynamic weight fusion and credibility correction module, the weight of the satellite image data is adjusted by an exponential decay function, which decreases as the spatiotemporal consistency compensation coefficient increases, and the decay rate is controlled by a preset decay factor; the decay factor is used to adjust the sensitivity of the compensation coefficient to weight changes, ensuring that the weight of the satellite image data is quickly reduced in high compensation coefficient scenarios;

[0025] The drone data weight is generated by normalizing the product of the drone data quality index and the spatiotemporal consistency compensation coefficient to ensure that high-quality drone data has a dominant weight when the compensation coefficient is high. The quantification method of satellite image data weight and drone data weight meets the following requirements:

[0026] The weight of satellite image data decreases monotonically as the spatial and temporal consistency compensation coefficient increases;

[0027] The drone data weight is positively driven by both the spatiotemporal consistency compensation coefficient and the data quality index;

[0028] The weight distribution result satisfies W Satellite +W UAV ≤1.

[0029] Preferably, the extended Kalman filter fusion refers to: constructing a state vector and an observation vector based on dynamically allocated satellite image data weights and drone weights, the state vector containing the spatial coordinates of satellite image data and drone data, and the observation vector being the projection value of multi-source data in a unified coordinate system; describing the data evolution law through a state transfer matrix, and adjusting the process noise covariance matrix according to dynamic weights, the higher the satellite image data weight, the lower the corresponding noise covariance value; using the observation matrix to map the state vector to the observation space, and combining the observation noise covariance matrix to iteratively update the fusion result, giving priority to retaining the characteristics of high-weight data.

[0030] Preferably, at the sensor displacement rate V disp When the preset threshold is exceeded, the particle filter fault-tolerant iteration is triggered, including:

[0031] Particle generation and propagation: Generates several particles, each containing satellite image data, drone data, and a hypothetical state of the sensor displacement rate. The particle distribution range is dynamically adjusted according to the sensor displacement rate.

[0032] Weight update and resampling: Particle weights are calculated based on the deviation between the observed value and the particle. The smaller the deviation, the higher the weight. Low-weight particles are eliminated and new particles are added to ensure that the particle set focuses on the high-probability state area.

[0033] Fusion output: Generate the final fusion result based on the weighted average of particle weights and output the dynamic demolition model.

[0034] Preferably, the three-dimensional spatial data corresponding to the dynamic demolition model is divided into several grids, and the confidence parameter of each grid is marked with a value range of 0-1; the reliability of the corresponding grid position data is characterized by the particle filter fault-tolerant iterative calculation of the dynamic weight fusion and credibility correction module; a particle set is generated based on the dynamic demolition model and the sensor displacement rate, the particle weight is updated by observing the deviation and resampling is performed, and the confidence parameter is generated by counting the particle weights in the grid unit.

[0035] Preferably, the confidence parameter is obtained as follows:

[0036] Step S1, particle set initialization: generate initial particles after the initial fusion of the extended Kalman filter, the particles obey the Gaussian distribution, control the particle propagation range based on the sensor displacement rate, and output the initialized particle set;

[0037] Step S2, particle weight update: input the particle set, unfused satellite image data and drone data to calculate the predicted value, map the particle state to the observation space through the nonlinear observation function, generate the predicted coordinates; calculate the Euclidean distance between the predicted coordinates and the observed coordinates;

[0038] Step S3, resampling and confidence parameter generation: input the updated particle set and particle weights; eliminate particles with weights below the threshold through resampling operations; copy high weight factors according to the weight ratio to maintain the total number of particles unchanged;

[0039] Step S4, after confidence calculation: the dynamic demolition model is divided into grids, the weights of all particles in each grid are counted, and the confidence parameters are generated after normalization.

[0040] To achieve the above objectives, the present invention provides the following technical solutions: a GIS-based digital management method for land acquisition and demolition in transportation projects, comprising:

[0041] Step 1: Multi-source geographic data collection: Collect multi-source geographic data corresponding to the target area, including satellite image data, drone data, and boundary stake coordinate time series stream data; Based on the contour clarity score extracted by ResNet-50 features, obtain the drone data quality index Q UAV , based on the boundary stake coordinate time series data, the sensor displacement rate V is obtained disp ;

[0042] Step 2: Unify the benchmarks of multi-source geographic data, generate NDVI vegetation feature maps through superpixel segmentation algorithms, use transfer learning networks to extract building outline feature maps, synchronously receive boundary stake coordinate time series stream data, perform unified geographic projection coordinate conversion, and output multimodal standardized data;

[0043] Step 3: Feature alignment and compensation: The multi-scale spatial correlation features of the NDVI vegetation feature map and the building outline feature map are extracted through wavelet transform as the cross-modal edge feature map E m ; Construct an adversarial network model, input the NDVI vegetation feature map and the building outline feature map for semantic mapping relationship learning, and output the spatiotemporal consistency compensation coefficient K comp ; Combine the covariance matrix of the displacement vector to dynamically correct the spatial offset of the spatiotemporal consistency compensation coefficient;

[0044] Step 4: Dynamic weight fusion and credibility correction: Dynamically allocate satellite image data weights and drone data weights based on the spatiotemporal consistency compensation coefficient and drone data quality index; fuse data through the extended Kalman filter, trigger the particle filter fault-tolerant iteration, and output the dynamic demolition model.

[0045] The technical effects and advantages of the present invention are as follows:

[0046] The present invention solves the problems of insufficient precision, accumulation of spatiotemporal registration errors and static threshold misjudgment in cross-modal data fusion of traditional methods by constructing a unified framework of multi-source data geo-references and a cross-modal feature alignment mechanism, thereby realizing centimeter-level dynamic perception and intelligent decision-making in land acquisition and demolition management. The specific methods are as follows: feature decoupling technology based on superpixel segmentation and transfer learning significantly enhances the semantic consistency of satellite, UAV and sensor data, and eliminates spatial dislocation interference of heterogeneous data; generates spatiotemporal consistency compensation coefficients through adversarial networks, combines dynamic weight fusion and particle filter fault-tolerant iteration to improve the accuracy of multi-source data fusion; introduces a dynamic security threshold model that links ray tracing algorithms with historical dispute rates to achieve real-time early warning and adaptive response to three-dimensional spatial conflicts. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a structural diagram of the digital management system for land acquisition and demolition of transportation projects of the present invention.

[0048] Figure 2 The present invention provides a flow chart for obtaining confidence parameters. DETAILED DESCRIPTION

[0049] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0050] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0051] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present disclosure, its application, or uses.

[0052] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0053] Example 1, see Figure 1 The digital management system structure diagram of land acquisition and demolition for transportation projects is provided. Figure 1 The GIS-based digital management system for land acquisition and demolition in transportation projects shown includes:

[0054] The multi-source geographic data acquisition module records the land acquisition and demolition area of ​​the transportation project as the target area and collects multi-source geographic data corresponding to the target area, including satellite image data, drone data, and boundary stake coordinate time series stream data (real-time 3D coordinates of boundary stakes are obtained through GPS or RTK positioning);

[0055] Explanation: Satellite image data processing: Use superpixel segmentation algorithms to cut satellite images into high-precision puzzle pieces, divide land blocks, calculate the NDVI index (which determines green coverage by the difference in light reflected by vegetation), and generate vegetation maps; this is used to identify which areas are farmland and forest land, avoiding the accidental removal of vegetation protection areas;

[0056] Drone data processing: Pre-trained AI models are used to analyze high-definition drone photos and extract building outlines (such as house boundaries and floor structures). This is used to accurately determine the distribution of buildings within the demolition area to avoid omissions or misjudgments.

[0057] Sensor data processing: Receives sensor signals installed on land boundary stakes (similar to GPS locators) and obtains three-dimensional coordinates (longitude, latitude, altitude) and timestamps in real time to monitor whether land boundaries have moved;

[0058] Furthermore, after obtaining multi-source geographic data, the drone data quality index Q is obtained based on the contour clarity score extracted by ResNet-50 features. UAV , the value range is [0, 1], and the sensor displacement rate V is obtained based on the boundary stake coordinate time series flow data disp ;

[0059] The multi-source geographic data benchmark unification module processes satellite image data through a superpixel segmentation algorithm to generate NDVI vegetation feature maps, uses a transfer learning network to extract building outline feature maps from drone data, synchronously receives boundary stake coordinate time series stream data, performs unified geographic projection coordinate conversion, and outputs multimodal standardized data;

[0060] The feature alignment and compensation module extracts the cross-modal edge feature map E based on the multi-scale spatial correlation between the NDVI vegetation feature map and the building outline feature map. m Generate spatiotemporal consistency compensation coefficients through adversarial networks, and correct the spatial offset of multi-source data by combining the displacement vectors in the sensor coordinate time series flow;

[0061] The dynamic weight fusion and credibility correction module uses a nonlinear function to dynamically assign weights to satellite imagery data and drone data based on the spatiotemporal consistency compensation coefficient and drone data quality indicators. It fuses data through an extended Kalman filter, triggers a particle filter fault-tolerant iteration, and integrates multi-source geographic data to output a centimeter-level accurate dynamic demolition model, which is a collection of 3D spatial data of the target area.

[0062] The three-dimensional conflict warning module constructs a three-dimensional voxel space of the land acquisition red line BIM model and the dynamic demolition model, and calculates the minimum real-time distance between each boundary stake and the land acquisition red line through a ray tracing algorithm; it generates a dynamic safety threshold for the target area based on the historical dispute rate and real-time distance, triggers a multi-level alarm signal and freezes permissions; the land acquisition red line BIM model refers to a digital three-dimensional space model constructed based on building information modeling (BIM) technology, which accurately marks the legal boundary line (ie "red line") within the land acquisition scope of the transportation project, and integrates multi-dimensional data such as geographic coordinates, land ownership information, engineering design parameters and compliance requirements.

[0063] What needs to be explained in the present invention is that the three-dimensional conflict warning module takes the dynamic demolition model and sensor coordinate time series stream data as input, constructs the three-dimensional voxel space mapping relationship between the land acquisition red line BIM model and the dynamic demolition model, and combines the ray tracing algorithm to calculate the spatial distance between the sensor boundary stakes and the geographic red line in real time, and dynamically adjusts the dynamic safety threshold parameters based on the displacement rate, providing three-dimensional spatial data support for the accurate identification of land encroachment behavior and hierarchical triggering of authority management, ensuring that land acquisition and demolition management is upgraded from static rule execution to dynamic risk adaptive decision-making.

[0064] In a possible embodiment, if the risk diffusion coefficient is higher than the level threshold for three consecutive cycles, the influence weight of the historical dispute rate in the threshold calculation is simultaneously increased to ensure that the dynamic security threshold is dynamically adapted to the risk situation.

[0065] Furthermore, the risk diffusion coefficient RDC is calculated as follows:

[0066]

[0067] Among them, k is the sensor displacement rate sensitivity coefficient, V0 is the displacement rate reference value; HDR is the historical dispute rate, which reflects the proportion of the chaotic duration of the target area ownership in the total monitoring time in the historical period;

[0068] Explanation: The dynamic safety threshold refers to a safety distance parameter dynamically generated by the three-dimensional conflict warning module based on real-time data in the GIS-based digital management system for land acquisition and demolition of transportation projects. It is used to determine whether the spatial relationship between the boundary stakes and the land acquisition red line BIM model in the target area constitutes a potential conflict risk. The dynamic safety threshold is based on the minimum real-time distance between the boundary stakes and the land acquisition red line, combined with the historical dispute rate and sensor displacement rate, and is calculated and dynamically adjusted through a ray tracing algorithm; it is adaptively updated as the risk situation changes. For example, when the risk diffusion coefficient (based on the product of the real-time distance and the historical dispute rate) exceeds the preset level threshold, the sensitivity coefficient of the sensor displacement rate is increased proportionally to form the final dynamic safety threshold, ensuring accurate identification and graded warning of land encroachment.

[0069] It should be explained in the present invention that the operation process of the feature alignment and compensation module includes:

[0070] Based on the NDVI vegetation feature map and building outline feature map in the multimodal standardized data, the multi-scale spatial correlation features of the NDVI vegetation feature map and the building outline feature map are extracted by wavelet transform as the cross-modal edge feature map E m ;

[0071] Explanation: A three-level two-dimensional discrete wavelet transform is performed on the NDVI vegetation feature map and the building outline feature map, respectively. Each image is decomposed into a low-frequency component containing the overall outline and average brightness, and a high-frequency component containing edge and detail information. The high-frequency component is further subdivided into components in the horizontal, vertical, and diagonal directions. Each high-frequency component is normalized using the Frobenius norm to ensure that the scales of the components are consistent. The normalized NDVI high-frequency component and the building outline high-frequency component are element-wise multiplied (Hadamard product) to extract the edge features of the NDVI vegetation feature map and the building outline feature map in the same direction. The cross-modal edge features in the three directions are summed to obtain the final cross-modal edge feature map.

[0072] Construct an adversarial network model, input the NDVI vegetation feature map and the building outline feature map into the adversarial network model for semantic mapping relationship learning, and output the spatiotemporal consistency compensation coefficient K comp The spatiotemporal consistency compensation coefficient is used to quantify the strength of spatial alignment error correction of multi-source data. The value range is 0-10, and a larger value indicates a higher required compensation strength.

[0073] Explanation: The input data of the adversarial network model is the cross-modal edge feature map Em and the sensor displacement vector v vs =[Δx, Δy, Δz] T, where Δx, Δy, Δz represent the displacement in three-dimensional space; the encoding layer of the generator network: 5 layers of convolution (kernel size 3×3, step size 2), E m Compressed into a 128-dimensional hidden vector; the fusion layer of the generator network: the hidden vector is combined with the displacement vector v vs Splicing, mapped to scalar output through the fully connected layer; the generator network output is normalized by the Sigmoid function and expanded to the range of 0-10, and the spatiotemporal consistency compensation coefficient K is calculated by the following formula comp :

[0074] K comp =10·σ(W5RELU(W4·Flatten(E m )+W3·v disp ))

[0075] Where σ(·) is the Sigmoid function, W5, W4, and W3 are weight matrices, and W3∈R128×3 is used to convert the 3D sensor displacement vector v disp Mapped to 128-dimensional latent space features, capturing the influence of displacement direction and intensity on the compensation coefficient; W4∈R128×(H×W×3), Flatten(E m ) is to convert E m Flatten and map to a 128-dimensional latent space feature function to extract the contribution of vegetation-building spatial correlation to the compensation coefficient; W5∈R1×128 is used to compress the 128-dimensional latent space feature into a scalar value to complete the final mapping of the compensation coefficient;

[0076] Explanation: The size of the cross-modal edge feature map Em is H×W×3, where H represents the number of pixels in the vertical direction (height) of the cross-modal edge feature map, corresponding to the downsampling size of the input image (satellite or drone image) after wavelet transform; W represents the number of pixels in the horizontal direction (width) of the cross-modal edge feature map; and 3 represents the high-frequency components in three directions, corresponding to the horizontal, vertical, and diagonal edge information of the wavelet decomposition.

[0077] Calculate the covariance matrix of the displacement vector in the sensor coordinate time series stream, dynamically correct the spatiotemporal consistency compensation coefficient based on the covariance matrix of the displacement vector, and generate compensated multimodal standardized data;

[0078] Explanation: Based on the sensor coordinate time series stream output by the multi-source geographic data acquisition module, the three-dimensional displacement vector sequence of adjacent moments is first calculated to quantify the displacement changes of the sensor in the east, north and elevation directions in the UTM coordinate system; the covariance matrix is ​​generated by statistically analyzing the direction and intensity distribution of the displacement vector to characterize the spatial correlation characteristics of the sensor displacement; then the main displacement direction corresponding to the maximum eigenvalue in the covariance matrix is ​​extracted to determine the dominant trend direction of the sensor displacement, and the correction amount is dynamically adjusted along this direction in combination with the spatiotemporal consistency compensation coefficient (a parameter in the range of 0 to 10 output from the adversarial network) to generate the final spatial offset parameter; the spatial offset parameter corrects the spatial offset error of multi-source data (satellites, drones, sensors) at the centimeter level along the main displacement direction to ensure the spatial alignment accuracy of the land acquisition red line BIM model and the dynamic demolition model, and solve the risk of misjudgment caused by sensor displacement or data acquisition errors.

[0079] Furthermore, in the dynamic weight fusion and credibility correction module, the satellite image data weight is adjusted by an exponential decay function, which decreases as the spatiotemporal consistency compensation coefficient increases, and the decay rate is controlled by a preset decay factor. The decay factor is used to adjust the sensitivity of the compensation coefficient to weight changes, ensuring that the satellite image data weight decreases rapidly in high compensation coefficient scenarios.

[0080] The drone data weight is generated by normalizing the product of the drone data quality index and the spatiotemporal consistency compensation coefficient to ensure that high-quality drone data has a dominant weight when the compensation coefficient is high. The quantification method of satellite image data weight and drone data weight meets the following requirements:

[0081] The weight of satellite image data decreases monotonically as the spatial and temporal consistency compensation coefficient increases;

[0082] The drone data weight is positively driven by both the spatiotemporal consistency compensation coefficient and the data quality index;

[0083] The weight distribution result satisfies W Satellite +W UAV ≤1.

[0084] Furthermore, the extended Kalman filter fusion refers to: constructing a state vector and an observation vector based on dynamically allocated satellite image data weights and drone weights, wherein the state vector contains the spatial coordinates of the satellite image data and the drone data, and the observation vector is the projection value of the multi-source data in a unified coordinate system; describing the data evolution law through the state transfer matrix, and adjusting the process noise covariance matrix according to the dynamic weight, the higher the satellite image data weight, the lower the corresponding noise covariance value; using the observation matrix to map the state vector to the observation space, and combining the observation noise covariance matrix to iteratively update the fusion result, giving priority to retaining the characteristics of high-weight data.

[0085] Furthermore, at the sensor displacement rate V disp When the preset threshold is exceeded, the particle filter fault-tolerant iteration is triggered, including:

[0086] Particle generation and propagation: Generates several particles, each containing satellite image data, drone data, and a hypothetical state of the sensor displacement rate. The particle distribution range is dynamically adjusted according to the sensor displacement rate.

[0087] Weight update and resampling: Particle weights are calculated based on the deviation between the observed value and the particle. The smaller the deviation, the higher the weight. Low-weight particles are eliminated and new particles are added to ensure that the particle set focuses on the high-probability state area.

[0088] Fusion output: Generate the final fusion result based on the weighted average of particle weights and output the dynamic demolition model.

[0089] Furthermore, the three-dimensional spatial data corresponding to the dynamic demolition model is divided into several grids, and the confidence parameters of each grid are marked with a value range of 0-1; the reliability of the corresponding grid position data is characterized by the particle filter fault-tolerant iterative calculation of the dynamic weight fusion and credibility correction module; a particle set is generated based on the dynamic demolition model and the sensor displacement rate, the particle weight is updated by observing the deviation and resampling is performed, and the confidence parameters are generated by counting the particle weights in the grid unit.

[0090] For further information, see Figure 2 The confidence parameter acquisition flow chart is as follows:

[0091] Step S1, particle set initialization: After the initial fusion of the extended Kalman filter, N = 1000 particles are generated, and the state of each particle is x i =[x sat ,x uav ,V disp ], where x sat with x uav They represent the predicted values ​​of the three-dimensional coordinates of satellite image data and drone data in the UTM coordinate system respectively; the particles obey the Gaussian distribution, and the covariance matrix is Ensure that the particle distribution is focused on the corrected high probability area, Δ offset Indicates the error offset of each coordinate axis;

[0092] Example: If the horizontal error of satellite image data is ±2cm (Δ_x=Δ_y=0.02m), and the elevation error is ±5cm (Δ_z=0.05m), then the covariance matrix is: Σ=diag(0.02 2 , 0.02 2 , 0.05 2 );

[0093] Based on the sensor displacement rate V disp Control the particle propagation range: the higher the rate, the wider the particle distribution range; output the initialized particle set

[0094] Step S2, particle weight update: input particle set The unfused satellite image data and drone data are used to calculate the predicted value, and the particle state is mapped to the observation space through the nonlinear observation function h(·) to generate the predicted coordinate h(x i )=[h sat (x i ),h uav (x i )];h sat (x i ),h uav (x i ) represent the satellite image data observation model and the drone data observation model, respectively; they are used to extract geospatial features from satellite image data and drone data and map them to a unified coordinate system (such as UTM) for comparison with particles; the observation coordinate z is the weighted average projection value of the satellite image and drone data in the UTM coordinate system, and the weight is dynamically adjusted according to the time difference between the data acquisition timestamp and the current time;

[0095] Calculate the Euclidean distance d between the predicted coordinate and the observed coordinate z i =||zh(x i )||2; Use exponential decay function to calculate the weight of each particle:

[0096]

[0097] Among them, σ = 0.05m is the observation error tolerance parameter, which controls the decay rate of the weight as the deviation increases; the updated particle set is output;

[0098] Step S3, resampling and confidence parameter generation: input the updated particle set and particle weights After resampling, particles with weights below the threshold are eliminated; high-weight factors are replicated according to the weight ratio to maintain the total number of particles N = 1000; new particles are randomly generated within the range of the dynamic demolition model to fill the gaps;

[0099] Step S4, after confidence calculation: the dynamic demolition model is divided into grids, i represents the particle index, N1 represents the total number of particles in the grid, such as 1cm×1cm; the weight sum of all particles in each grid (H, W) is counted and normalized to generate the confidence parameter:

[0100]

[0101] Where Ω(H,W) represents the set of particles covering the grid (H,W); the output is the gridded confidence parameter set.

[0102] Rationality analysis of dynamic weight allocation: Traditional fixed-weight methods cannot adapt to dynamic changes in data quality (such as fluctuations in the clarity of drone images affected by weather). Data reliability must be quantified using a combination of a spatiotemporal consistency compensation coefficient and a quality indicator. The compensation coefficient reflects the degree of data alignment, while the quality indicator reflects the credibility of the data itself. Combining the two can avoid bias in single-parameter decision-making and improve the rationality of weight allocation.

[0103] Rationality analysis of extended Kalman filter fusion: The standard Kalman filter assumes that the noise obeys Gaussian distribution and the system is linear, which cannot handle the nonlinear noise and dynamic change characteristics of weights of multi-source data in land acquisition and demolition scenarios; by adaptively adjusting the process noise covariance through weights, the noise interference of low-weight data is preferentially suppressed to ensure the spatial consistency of data fusion.

[0104] Analysis of the rationality of particle filter fault-tolerant iteration: Sudden sensor anomalies (such as strong electromagnetic interference causing displacement rate jumps) will destroy the linear assumption of Kalman filtering, and nonlinear filtering methods need to be introduced for supplementary correction; particle filtering approximates complex noise distribution through non-parametric sampling, which can effectively handle abnormal jitter of sensor data and ensure the stability of data fusion in extreme scenarios.

[0105] The dynamic demolition model is a collection of three-dimensional spatial data of the target area. It divides the three-dimensional spatial data into several grids and marks the confidence parameters of each grid. The output data provides a collision detection priority basis for the ray tracing algorithm of the three-dimensional conflict warning module. The dynamic demolition model converges the multi-source data error to the centimeter level (measured RMSE ≤ 0.8cm) through the collaborative calculation of the extended Kalman filter and the particle filter, providing high-precision spatial base data with credibility assessment for the three-dimensional conflict warning module, solving the problem of excessively high false alarm rate caused by data noise in traditional methods.

[0106] Example 2: This embodiment of the present invention differs from Example 1 in that the multi-source geographic data acquisition module dynamically adjusts the data acquisition strategy based on the terrain characteristics of the target area, increasing the drone image acquisition frequency and the boundary stake coordinate sampling density when the terrain complexity is high; at the same time, the convolution kernel parameters of the feature extraction algorithm are optimized according to the terrain undulation to improve the ability to capture edge details; if the quality of the collected data does not meet expectations, a data re-acquisition mechanism is triggered to ensure accuracy;

[0107] A specific implementation method may be: calculating the slope change rate of the target area in real time; when the slope change rate is ≥15% / second, determining it as high-relief terrain and triggering the drone aerial photography frequency to increase to 5 frames / second and the boundary stake coordinate sampling points to be encrypted to 10 per meter; dynamically adjusting the ResNet-50 convolution kernel size according to the terrain undulation; switching to a 5×5 kernel to extract edge features when the undulation is ≥30°; and generating a return-and-retake command for the drone if the optimized drone data quality indicator QUAV is <0.8.

[0108] Furthermore, the feature alignment and compensation module performs three-level wavelet decomposition on the NDVI vegetation feature map and the building outline feature map, extracts the low-frequency components after the second-level decomposition, maps them to a 128-dimensional latent space through a residual connection layer, and splices them with the sensor displacement vector; the generator network outputs the spatiotemporal consistency compensation coefficient Kcomp based on the latent space features, and the discriminator constrains the Kcomp error within ±0.5 by optimizing the cross-modal edge feature consistency; the offset is calculated along the main direction of the sensor displacement according to Kcomp, and the spatial offset of the multi-source data is corrected.

[0109] Furthermore, the three-dimensional conflict warning module adjusts the authority control strategy based on the real-time distance relationship between boundary stakes and land acquisition red lines, restricting operation permissions when the potential conflict risk is high; combines historical data with real-time status to assess the probability of conflict, triggers the dynamic update of the land acquisition red line model, and pushes alarm signals to the management terminal in a hierarchical manner;

[0110] The specific implementation method may be: calculating the minimum real-time distance D between the boundary stake and the land acquisition red line. min , when D min When the value is <0.8×dynamic security threshold, low-privilege user operations are frozen; based on the historical dispute rate HDR and the real-time distance, the conflict probability is generated, Pconflict=0.6HDR+0.4·(1-Dmin / dynamic security threshold); when Pconflict≥60%, the land acquisition red line BIM model is locally updated every hour; according to Pconflict>80%, 60-80%, and <60%, level I, II, and III alarms are pushed to the corresponding terminals.

[0111] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. The GIS-based digital management system for land acquisition and demolition of transportation projects is characterized by: include: Multi-source geographic data acquisition module, which collects multi-source geographic data corresponding to the target area, including satellite image data, drone data, and boundary stake coordinate time series stream data; The drone data quality index Q is obtained based on the contour clarity score extracted by ResNet-50 features. UAV , based on the boundary stake coordinate time series data, the sensor displacement rate V is obtained disp ; The multi-source geographic data benchmark unification module generates NDVI vegetation feature maps through superpixel segmentation algorithms, uses a transfer learning network to extract building outline feature maps, synchronously receives boundary stake coordinate time series stream data, performs unified geographic projection coordinate conversion, and outputs multimodal standardized data; The feature alignment and compensation module extracts the multi-scale spatial correlation features of the NDVI vegetation feature map and the building outline feature map through wavelet transform as the cross-modal edge feature map E m ; Construct an adversarial network model, input the NDVI vegetation feature map and the building outline feature map for semantic mapping relationship learning, and output the spatiotemporal consistency compensation coefficient K comp ; Combine the covariance matrix of the displacement vector to dynamically correct the spatial offset of the spatiotemporal consistency compensation coefficient; Dynamic weight fusion and credibility correction module dynamically allocates satellite image data weights and drone data weights based on spatiotemporal consistency compensation coefficients and drone data quality indicators; By using an extended Kalman filter to fuse data, a particle filter fault-tolerant iteration is triggered to output a dynamic demolition model; the weight of the satellite image data is adjusted by an exponential decay function, which decreases as the spatiotemporal consistency compensation coefficient increases, and the decay rate is controlled by a preset decay factor; The attenuation factor is used to adjust the sensitivity of the compensation coefficient to the weight change, ensuring that the weight of satellite image data decreases rapidly in high compensation coefficient scenarios; the drone data weight is generated by normalizing the product of the drone data quality index and the spatiotemporal consistency compensation coefficient, ensuring that high-quality drone data occupies the dominant weight when the compensation coefficient is high; the quantification method of satellite image data weight and drone data weight meets the following requirements: the satellite image data weight decreases monotonically as the spatiotemporal consistency compensation coefficient increases; the drone data weight is positively driven by both the spatiotemporal consistency compensation coefficient and the data quality index; the weight distribution result meets W Satellite +W UAV ≤1; The three-dimensional conflict warning module constructs a three-dimensional voxel space of the land acquisition red line BIM model and the dynamic demolition model, and calculates the minimum real-time distance between each boundary stake and the land acquisition red line through a ray tracing algorithm; it generates a dynamic safety threshold for the target area based on the historical dispute rate and real-time distance, triggering a multi-level alarm signal.

2. The GIS-based digital management system for land acquisition and demolition of transportation projects according to claim 1 is characterized in that: The extended Kalman filter fusion method involves constructing a state vector and an observation vector based on dynamically assigned satellite imagery data weights and drone weights. The state vector contains the spatial coordinates of the satellite imagery data and drone data, and the observation vector is the projection value of the multi-source data in a unified coordinate system. The data evolution law is described by the state transition matrix, and the process noise covariance matrix is ​​adjusted according to the dynamic weights. The higher the satellite imagery data weight, the lower the corresponding noise covariance value. The state vector is mapped to the observation space using the observation matrix, and the fusion result is iteratively updated in combination with the observation noise covariance matrix, giving priority to retaining the features of high-weight data.

3. The GIS-based digital management system for land acquisition and demolition of transportation projects according to claim 2 is characterized in that: At the sensor displacement rate V disp When the preset threshold is exceeded, the particle filter fault-tolerant iteration is triggered, including: Particle generation and propagation: Generates several particles, each containing satellite image data, drone data, and a hypothetical state of the sensor displacement rate. The particle distribution range is dynamically adjusted according to the sensor displacement rate. Weight update and resampling: Particle weights are calculated based on the deviation between the observed value and the particle. The smaller the deviation, the higher the weight. Low-weight particles are eliminated and new particles are added to ensure that the particle set focuses on the high-probability state area. Fusion output: Generate the final fusion result based on the weighted average of particle weights and output the dynamic demolition model.

4. The GIS-based digital management system for land acquisition and demolition of transportation projects according to claim 1 is characterized in that: The three-dimensional spatial data corresponding to the dynamic demolition model is divided into several grids, and the confidence parameter of each grid is marked, with a value range of 0-1, which represents the reliability of the corresponding grid position data; A particle set is generated based on the dynamic demolition model and the sensor displacement rate. The particle weights are updated through observation deviations and resampling is performed. The confidence parameters are generated by statistically analyzing the particle weights within the grid cells.

5. The GIS-based digital management system for land acquisition and demolition of transportation projects according to claim 4 is characterized in that: The confidence parameter is obtained as follows: Step S1, particle set initialization: generate initial particles after the initial fusion of the extended Kalman filter, the particles obey the Gaussian distribution, control the particle propagation range based on the sensor displacement rate, and output the initialized particle set; Step S2, particle weight update: input the particle set, unfused satellite image data and drone data to calculate the predicted value, map the particle state to the observation space through the nonlinear observation function, generate the predicted coordinates; calculate the Euclidean distance between the predicted coordinates and the observed coordinates; Step S3, resampling and confidence parameter generation: input the updated particle set and particle weights; eliminate particles with weights below the threshold through resampling operations; copy high weight factors according to the weight ratio to maintain the total number of particles unchanged; Step S4, after confidence calculation: the dynamic demolition model is divided into grids, the weights of all particles in each grid are counted, and the confidence parameters are generated after normalization.

6. The GIS-based digital management system for land acquisition and demolition of transportation projects according to claim 1 is characterized in that: The multi-source geographic data acquisition module dynamically adjusts the data acquisition strategy according to the terrain characteristics of the target area, increasing the drone image acquisition frequency and the boundary stake coordinate sampling density when the terrain complexity is high; at the same time, it optimizes the convolution kernel parameters of the feature extraction algorithm according to the degree of terrain undulation to improve the ability to capture edge details; if the quality of the collected data does not meet expectations, the data re-acquisition mechanism is triggered.

7. The GIS-based digital management system for land acquisition and demolition of transportation projects according to claim 1 is characterized in that: After the dynamic safety threshold is generated, the risk diffusion coefficient of the target area is calculated based on the product of the real-time distance and the historical dispute rate. When the risk diffusion coefficient exceeds the preset level threshold, the sensor displacement rate weight parameter of the dynamic safety threshold is proportionally increased to form the final dynamic safety threshold after adaptive adjustment. The dynamic safety threshold is calculated as follows: Among them, α, β, k, V0 are preset parameters, D min It indicates the minimum real-time distance between the boundary stake and the land acquisition red line, HDR indicates the historical dispute rate, and triggers a multi-level alarm signal.

8. A GIS-based digital management method for land acquisition and demolition of transportation projects, used to implement the digital management system for land acquisition and demolition of transportation projects as described in claim 1, characterized in that: include: Step 1: Multi-source geographic data collection: Collect multi-source geographic data corresponding to the target area, including satellite image data, drone data, and boundary stake coordinate time series stream data; The drone data quality index Q is obtained based on the contour clarity score extracted by ResNet-50 features. UAV , based on the boundary stake coordinate time series data, the sensor displacement rate V is obtained disp ; Step 2: Unify the benchmarks of multi-source geographic data, generate NDVI vegetation feature maps through superpixel segmentation algorithms, use transfer learning networks to extract building outline feature maps, synchronously receive boundary stake coordinate time series stream data, perform unified geographic projection coordinate conversion, and output multimodal standardized data; Step 3: Feature alignment and compensation: The multi-scale spatial correlation features of the NDVI vegetation feature map and the building outline feature map are extracted through wavelet transform as the cross-modal edge feature map E m ; Construct an adversarial network model, input the NDVI vegetation feature map and the building outline feature map for semantic mapping relationship learning, and output the spatiotemporal consistency compensation coefficient K comp ; Combine the covariance matrix of the displacement vector to dynamically correct the spatial offset of the spatiotemporal consistency compensation coefficient; Step 4: Dynamic weight fusion and credibility correction: Dynamically allocate satellite image data weights and drone data weights based on the spatiotemporal consistency compensation coefficient and drone data quality indicators; By extending the Kalman filter to fuse data, the particle filter fault-tolerant iteration is triggered and a dynamic demolition model is output.

Images