A land surveying and mapping data intelligent analysis and management method and system

Abstract

The present invention relates to the field of land surveying and mapping technology, and in particular to a method and system for intelligent analysis and management of land surveying and mapping data. The method comprises: obtaining first surveying and mapping data and second surveying and mapping data of the same surveying and mapping area, and preprocessing the first surveying and mapping data and the second surveying and mapping data; constructing a joint model and a dual-stream network, analyzing the preprocessed first surveying and mapping data and the second surveying and mapping data through the joint model and the dual-stream network, and outputting terrain data; extracting multiple feature values ​​from the first surveying and mapping data according to a preset acquisition frequency to form a surveying and mapping feature sequence; inputting the surveying and mapping feature sequence into a preset vegetation multi-layer state model, and obtaining vegetation vertical spatial distribution information and ground elevation data through the vegetation multi-layer state model; classifying and managing land in the surveying and mapping area based on the vegetation vertical spatial distribution information, and simultaneously correcting the terrain data using the ground elevation data, and outputting the corrected terrain data to improve the accuracy of the terrain data.

Description

Technical Field

[0001] The present invention relates to the field of land surveying and mapping technology, and in particular to a method and system for intelligent analysis and management of land surveying and mapping data. Background Art

[0002] As a key means of acquiring, processing, and analyzing spatial information about land and its appurtenant structures, land surveying and mapping technology provides a reliable basis for understanding various land attributes. It is widely used in land management, urban planning, and engineering construction.

[0003] GNSS, a digital surveying and mapping technology, has become a key technology in modern surveying and mapping due to its significant advantages, such as high precision and strong real-time performance. It can achieve real-time positioning on a global scale, unrestricted by factors such as weather and terrain, greatly facilitating land surveying and mapping. 3D laser scanning technology, with its fast and precise characteristics, can quickly acquire 3D coordinate data on the surfaces of large quantities of objects, forming detailed 3D models and providing rich information for subsequent analysis and processing. Remote sensing measurement technology uses sensors mounted on platforms such as satellites and aircraft to observe the Earth's surface from a distance, acquiring land information over large areas, offering the advantages of high efficiency and a broad scope. Laser radar (LiDAR) technology can penetrate obstacles such as vegetation, making it suitable for surveying and mapping complex terrain.

[0004] Traditional surveying and mapping primarily uses a single surveying method and conducts analysis based on a single data source. This single data source can contain errors and easily accumulates errors during data processing, affecting the accuracy and reliability of surveying and mapping results.

[0005] To this end, we propose an analysis and management method to improve the accuracy of surveying and mapping data through the fusion of multi-source data. Summary of the Invention

[0006] The present invention integrates LiDAR data and SAR data to improve the accuracy of surveying and mapping data. At the same time, it integrates the vertical structural characteristics of vegetation (number of layers and density) to classify the vegetation covering the land, and further realizes the classification of land types based on the classification of vegetation.

[0007] The technical solution proposed by the present invention is: a method for intelligent analysis and management of land surveying and mapping data, the method comprising:

[0008] Acquire first surveying and mapping data and second surveying and mapping data of the same surveying and mapping area, and preprocess the first surveying and mapping data and the second surveying and mapping data; the first surveying and mapping data is LiDAR elevation data; the second surveying and mapping data is Synthetic Aperture Radar elevation data;

[0009] constructing a joint model and a dual-stream network, analyzing the pre-processed first surveying and mapping data and the second surveying and mapping data through the joint model and the dual-stream network, and outputting terrain data;

[0010] Extracting multiple characteristic values ​​from the first surveying and mapping data according to a preset acquisition frequency to form a surveying and mapping characteristic sequence; inputting the surveying and mapping characteristic sequence into a preset vegetation multi-layer state model, and obtaining vegetation vertical spatial distribution information and ground elevation data through the vegetation multi-layer state model;

[0011] Classify and manage the land in the surveying and mapping area based on the vertical spatial distribution information of vegetation, and use the ground elevation data to correct the terrain data and output the corrected terrain data to improve the accuracy of the terrain data;

[0012] The method of correcting terrain data using ground elevation data and outputting the corrected terrain data includes:

[0013] Get the original ground elevation data h g =H ground-truth and the output H of the two-stream network network ;

[0014] H network With H ground-truth Align in space, that is, H is interpolated by Kriging algorithm. ground-truth Interpolate to the grid network output by the dual-stream network, and interpolate to H corr =Kriging(H ground-truth,grid＝d ), where d represents the width of the grid; where H corr represents Kriging interpolation;

[0015] Calculate the residual ΔH=H corr -H network ;

[0016] Build a lightweight UNet and input X in =[ΔH,H network ,▽H network ], output residual correction Among them ▽H network Represents the terrain gradient of the grid network, the UNet includes 4 encoding layers and 1 decoding layer;

[0017] Then, the corrected Where w′ represents the dynamic weight, c represents the slope parameter, which is used to control the rate of change of the dynamic weight; K represents the standard deviation of the elevation of each grid in the grid network output by the dual-stream network; β represents the priority threshold. When K>β, the difference correction is used first.

[0018] The corrected raster elevation data obtained after decoding will replace the raster elevation data output by the dual-stream network.

[0019] Preferably, the preprocessing includes spatiotemporal reference alignment and error model calibration;

[0020] The spatiotemporal alignment includes:

[0021] The first surveying and mapping data and the second surveying and mapping data are converted to a unified coordinate system, and the first surveying and mapping data is subjected to elevation conversion, specifically: H e =H L1 +N g , where H e Indicates the converted elevation data, H L1 Indicates the elevation data before conversion, N g Indicates the height of the geoid level;

[0022] rasterizing the first surveying and mapping data into a grid network, and resampling the second surveying and mapping data using a bilinear interpolation algorithm and converting the data into the same grid network;

[0023] The error model calibration includes:

[0024] Calibrate the first surveying data noise R L1 =σ L 2 , where σ L Indicates the first surveying and mapping data error;

[0025] Calibrate the second mapping data noise R S =σ S 2 , where σ S Indicates the second surveying and mapping data error.

[0026] Preferably, the constructing of the joint model and the dual-stream network, analyzing the pre-processed first surveying and mapping data and the second surveying and mapping data by the joint model and the dual-stream network, and outputting the terrain data, includes:

[0027] Step 2.1: Build a joint model based on the Kalman filter algorithm, fuse the first surveying and mapping data with the second surveying and mapping data, and obtain the elevation data vector of the surveying and mapping area, specifically including:

[0028] Step 2.1.1: Assume that the elevation of the terrain in the survey area does not change within the time period T. Then the elevation vector of the kth grid in the time period T is Among them, H k Represents the elevation value of the kth grid network;

[0029] Assume the state equation is:

[0030] in, represents the LiDAR elevation vector of the kth grid, A represents the SAR elevation vector of the kth grid; k represents the identity matrix, w k represents the process noise vector, Q represents the process noise covariance, σ process represents the terrain change error;

[0031] Step 2.1.2: After fusing the LiDAR elevation vector and the SAR elevation vector, the elevation vector of the kth grid is The observation matrix

[0032] The first surveying and mapping data noise and the second surveying and mapping data noise constitute the measurement noise covariance matrix

[0033] Let Kalman gain K k =P k|k-1 H T (HP k|k-1 H T +R) -1 ; Among them, the updated error covariance matrix P k =(IK k H)P k|k-1 ;P k|k-1 Represents the prediction error covariance matrix of the kth grid;

[0034] The Kalman gain after substituting the observation matrix and noise covariance matrix is:

[0035] The updated elevation vector is

[0036] Preferably, the constructing of the joint model and the dual-stream network, analyzing the pre-processed first surveying and mapping data and the second surveying and mapping data by the joint model and the dual-stream network, and outputting the terrain data, further includes:

[0037] Constructing a two-stream network based on the updated elevation vector and outputting terrain data includes the following sub-steps:

[0038] Step 2.2.1. Obtain LiDAR elevation vector and SAR elevation vector and P k ;

[0039] Generate 6-channel input tensor for each grid Among them, the first and second channels are used to input the updated LiDAR elevation vector and the first mapping data error; the third and fourth channels are used to input the updated SAR elevation vector and the second mapping data error; the fifth and sixth channels are used to input the diagonal elements of the updated error covariance matrix;

[0040] The input block is constructed in a 5×5 area to capture the terrain continuity, specifically X p ={X(i′+k″,j′+l)} 2 k″,l＝-2 ∈R 5×5×M , where X(i′+k″,j′+l) represents the adjacent grids of grid X(i′,j′), and M represents the number of layers of the grid network in the vertical direction;

[0041] Step 2.2.2: Construct a convolutional network 1, which includes 3 convolutional layers, each of which includes 3×3 convolutions. The convolutional network 1 extracts LiDAR elevation vector features from the input tensor and forms a LiDAR elevation vector feature set F after normalization. L ;

[0042] Step 2.2.3: Construct a convolutional network 2, which includes 3 convolutional layers, each of which includes 3×3 convolutions. The SAR elevation vector features are extracted from the input tensor through the convolutional network 2, and after normalization, the SAR elevation vector feature set F is formed. S ;

[0043] Step 2.2.4: Construct convolutional network 3, and resample the SAR elevation vector features to the grid network to obtain a new SAR elevation vector feature set F S ′, then calculate the attention weight: Attn = sigmoid(Conv([F L ; F S ′]));

[0044] Fusion F L and F S Get fusion feature F LS =Attn⊙F L +(1-Attn)⊙F′ S ;

[0045] Step 2.2.5: Construct a terrain decoder to decode the fusion features and obtain the corresponding raster fusion elevation data. Output the raster fusion elevation data and the raster network of the surveying and mapping area through the output layer. The raster fusion elevation data and the raster network of the surveying and mapping area constitute the terrain data H. network , the terrain decoder consists of three 3×3 convolutional layers and one output layer.

[0046] Preferably, the extracting of multiple feature values ​​from the first surveying and mapping data according to a preset acquisition frequency to form a surveying and mapping feature sequence includes:

[0047] Taking the vegetation falling event as the time node, multiple elevation feature values ​​are extracted from the first surveying and mapping data to form a surveying and mapping feature sequence A i =(x L1k ti 1,x L1k ti 2,…,x L1k ti m );

[0048] Multiple surveying and mapping feature sequences constitute a surveying and mapping feature sequence A={A i |i=1,2,3,4}; where ti represents the i-th time node; x L1k_ti_1 It represents the LiDAR elevation feature value of the kth grid sampled for the first time before time node ti; m represents the number of acquisitions.

[0049] Preferably, the step of inputting the surveying and mapping feature sequence into a preset vegetation multi-layer state model and obtaining vegetation vertical spatial distribution information and ground elevation data through the vegetation multi-layer state model includes:

[0050] Constructing a vegetation layer relationship model: ρ j+1 =ρ j ·e -λΔh +ε, where ρ j represents the vegetation density of the jth layer, Δh represents the vertical distance between layers, λ represents the attenuation ratio, and ε represents the error value;

[0051] Set inter-layer timing constraints W j+1 (tn')=W j+1 (ti)+ΔW(tn'), where W j+1 (tn') represents the laser echo amplitude of the j+1 layer of vegetation when the jth layer of vegetation falls at time node tn', and ΔW(tn') represents the increment of the laser echo amplitude due to the reduction of the shielding of the upper layer of vegetation; where n'=i+1;

[0052] Construct a hierarchical graph model, specifically including:

[0053] The height and density of each layer of vegetation are used as nodes;

[0054] The vegetation layer relationship model and temporal constraints between each layer of vegetation are used as edges;

[0055] Construct the maximum joint probability P(θ|D)∝P(D|θ)·P(θ), where P(D|θ) represents the LiDAR measurement data and the state equation The residual of the output data; P(θ) represents the constraints provided by the inter-layer relationship model; where, P(θ|D) represents the maximum joint probability;

[0056] Using Markov Chain Monte Carlo MCMC sampling, iteratively update the parameter θ:

[0057]

[0058] Among them, σ step represents the MCMC step size; θ (n+1) Represents the parameters after n iterations, represents the iterated covariance;

[0059] Assume the acceptance probability Accept the updated parameters with probability α, otherwise keep the original parameters;

[0060] Perform ground information inversion, including:

[0061] If the bottom elevation of the jth vegetation layer is known to be Assume that the elevation of the k′th layer satisfies Δh k′ represents the vertical distance between the k′th layer and the k′-1th layer;

[0062] Based on the surveying and mapping feature sequence, the LiDAR elevation feature value x is obtained when all the ground vegetation is removed, that is, when the ground is exposed. L1k_ti_1 , correct x L1k_ti_1 Get ground elevation in

[0063] Preferably, the land in the surveying and mapping area is classified and managed based on the vertical spatial distribution information of vegetation, including:

[0064] Perform GMM fitting on the LiDAR laser echo of each grid to extract the number of layers and parameters of each layer;

[0065] Count the number of vegetation layers penetrated by LiDAR in each season to form a penetration layer dataset;

[0066] If at time T, the vegetation density of the jth layer of vegetation is ρ j (T) drops more than the threshold value compared to T-1, and this is marked as shedding; the elevation of the jth layer of vegetation If the change compared to T-1 exceeds the elevation change threshold, it is judged that the vegetation structure has changed; the time when the j-th layer of vegetation first falls off is recorded. The first shedding time of multiple layers of vegetation constitutes the shedding time series D;

[0067] The number of layers, layer density and bottom elevation are used to form a vertical space vector. The first drop-off time is added to the vertical space feature vector to form a space-time vector. The space-time vector is normalized to form a space-time feature vector X. ST ;

[0068] The classification model is constructed by XGBoost algorithm. ST Input into the pre-trained classification model to classify vegetation

[0069] The present invention also provides a land surveying and mapping data intelligent analysis and management system, comprising: a processor and a memory and a communication module connected to the processor, and the system is used to execute the land surveying and mapping data intelligent analysis and management method.

[0070] The present invention also provides a computer-readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the method for intelligent analysis and management of land surveying and mapping data.

[0071] Beneficial effects of the present invention:

[0072] 1. The present invention complements the high precision of LiDAR with the strong penetration ability of SAR, and combines it with the dynamic fusion of dual-stream networks to solve the data blind spot problem of traditional single sensors in multi-layer vegetation areas (such as rainforests) and improve terrain data.

[0073] 2. The present invention adjusts the weights of LiDAR and SAR data in real time based on the covariance matrix of Kalman filtering (for example, high occlusion areas rely on SAR), avoiding local error amplification caused by fixed weights and adapting to the surveying and mapping of complex terrain. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 The present invention provides a flow chart of a method for intelligent analysis and management of land surveying and mapping data. DETAILED DESCRIPTION

[0075] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are for illustrative purposes only, and those skilled in the art will readily appreciate other obvious variations. The basic principles of the present invention defined in the following description may be applied to other embodiments, variations, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the present invention.

[0076] It is to be understood that the term "one" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element may be one, while in another embodiment, the number of the elements may be multiple, and the term "one" should not be understood as a limitation on the quantity.

[0077] refer to Figure 1 The technical solution provided by the present invention is: a method and system for intelligent analysis and management of land surveying and mapping data, comprising the following steps:

[0078] Step 1: Acquire first surveying and mapping data and second surveying and mapping data of the same surveying and mapping area, and preprocess the first surveying and mapping data and the second surveying and mapping data.

[0079] The first surveying and mapping data is LiDAR elevation data, and the second surveying and mapping data is Synthetic Aperture Radar (SAR) elevation data.

[0080] Preprocessing includes spatiotemporal reference alignment and error model calibration; the spatiotemporal alignment includes:

[0081] The first surveying and mapping data and the second surveying and mapping data are converted to a unified coordinate system, and the first surveying and mapping data is subjected to elevation conversion, specifically: H e =H L1 +N g , where H e Indicates the converted elevation data, H L1 Indicates the elevation data before conversion, N g Indicates the height of the geoid level;

[0082] rasterizing the first surveying and mapping data into a grid network, and resampling the second surveying and mapping data using a bilinear interpolation algorithm and converting the data into the same grid network;

[0083] The error model calibration includes:

[0084] Calibrate the first surveying data noise R L1 =σ L 2 , where σ L Indicates the first surveying and mapping data error;

[0085] Calibrate the second mapping data noise R S =σ S 2 , where σ S Indicates the second surveying and mapping data error.

[0086] Step 2: Constructing a joint model and a dual-stream network, analyzing the pre-processed first and second surveying and mapping data through the joint model and the dual-stream network, and outputting terrain data; including the following steps:

[0087] Step 2.1: Build a joint model based on the Kalman filter algorithm, fuse the first surveying and mapping data with the second surveying and mapping data, and obtain the elevation data vector of the surveying and mapping area, specifically including:

[0088] Step 2.1.1: Assume that the elevation of the terrain in the survey area does not change within the time period T. Then the elevation vector of the kth grid in the time period T is Among them, H k Represents the elevation value of the kth grid network;

[0089] Assume the state equation is:

[0090] in, represents the LiDAR elevation vector of the kth grid, A represents the SAR elevation vector of the kth grid; k represents the identity matrix, w k represents the process noise vector, Q represents the process noise covariance, σ process represents the terrain change error;

[0091] Step 2.1.2: After fusing the LiDAR elevation vector and the SAR elevation vector, the elevation vector of the kth grid is The observation matrix

[0092] The first surveying and mapping data noise and the second surveying and mapping data noise constitute the measurement noise covariance matrix

[0093] Let Kalman gain K k =P k|k-1 H T (HP k|k-1 H T +R) -1 ; Among them, the updated error covariance matrix P k =(IK k H)P k|k-1 ;P k|k-1 Represents the prediction error covariance matrix of the kth grid;

[0094] The Kalman gain after substituting the observation matrix and noise covariance matrix is:

[0095] The updated elevation vector is

[0096] Constructing a two-stream network based on the updated elevation vector and outputting terrain data includes the following sub-steps:

[0097] Step 2.2.1. Obtain LiDAR elevation vector and SAR elevation vector as well as

[0098] Generate 6-channel input tensor for each grid Among them, the first and second channels are used to input the updated LiDAR elevation vector and the first mapping data error; the third and fourth channels are used to input the updated SAR elevation vector and the second mapping data error; the fifth and sixth channels are used to input the diagonal elements of the updated error covariance matrix;

[0099] The input blocks are constructed in 5×5 areas to capture the terrain continuity, specifically:

[0100] X p ={X(i′+k″,j′+l)} 2 k″,l＝-2 ∈R 5×5×M , where X(i′+k″,j′+l) represents the adjacent grids of grid X(i′,j′), and M represents the number of layers of the grid network in the vertical direction;

[0101] Step 2.2.2: Construct a convolutional network 1, which includes 3 convolutional layers, each of which includes 3×3 convolutions. The convolutional network 1 extracts LiDAR elevation vector features from the input tensor and forms a LiDAR elevation vector feature set F after normalization. L ;

[0102] Step 2.2.3: Construct a convolutional network 2, which includes 3 convolutional layers, each of which includes 3×3 convolutions. The SAR elevation vector features are extracted from the input tensor through the convolutional network 2, and after normalization, the SAR elevation vector feature set F is formed. S ;

[0103] Step 2.2.4: Construct convolutional network 3, and resample the SAR elevation vector features to the grid network to obtain a new SAR elevation vector feature set F S ′, then calculate the attention weight: Attn = sigmoid(Conv([F L ; F S ′]));

[0104] Fusion F L and F S Get fusion feature F LS =Attn⊙F L +(1-Attn)⊙F′ S ;

[0105] Step 2.2.5: Construct a terrain decoder to decode the fusion features and obtain the corresponding raster fusion elevation data. Output the raster fusion elevation data and the raster network of the surveying and mapping area through the output layer. The raster fusion elevation data and the raster network of the surveying and mapping area constitute the terrain data H.network , the terrain decoder consists of three 3×3 convolutional layers and one output layer.

[0106] Step 3: extract multiple feature values ​​from the first surveying and mapping data according to the preset acquisition frequency to form a surveying and mapping feature sequence; specifically, the following steps: taking the vegetation falling event as the time node, extract multiple elevation feature values ​​from the first surveying and mapping data to form a surveying and mapping feature sequence A i =(x L1k_ti_1 ,x L1k_ti_2 ,…,x L1k_ti_m );

[0107] Multiple surveying and mapping feature sequences constitute a surveying and mapping feature sequence A={A i |i=1,2,3,4}; where ti represents the i-th time node; x L1k_ti_1 It represents the LiDAR elevation feature value of the kth grid sampled for the first time before time node ti; m represents the number of acquisitions.

[0108] Input the surveying and mapping feature sequence into the preset vegetation multi-layer state model, and obtain the vegetation vertical spatial distribution information and ground elevation data through the vegetation multi-layer state model; specifically, the following steps are included:

[0109] Constructing a vegetation layer relationship model: ρ j+1 =ρ j ·e -λΔh +ε, where ρ j represents the vegetation density of the jth layer, Δh represents the vertical distance between layers, λ represents the attenuation ratio, and ε represents the error value;

[0110] Set inter-layer timing constraints W j+1 (tn')=W j+1 (ti)+ΔW(tn'), where W j+1 (tn') represents the laser echo amplitude of the j+1 layer of vegetation when the jth layer of vegetation falls at time node tn', and ΔW(tn') represents the increment of the laser echo amplitude due to the reduction of the shielding of the upper layer of vegetation; where n'=i+1;

[0111] Construct a hierarchical graph model, specifically including:

[0112] The height and density of each layer of vegetation are used as nodes;

[0113] The vegetation layer relationship model and temporal constraints between each layer of vegetation are used as edges;

[0114] Construct the maximum joint probability P(θ|D)∝P(D|θ)·P(θ), where P(D|θ) represents the LiDAR measurement data and the state equation The residual of the output data; the prior P(θ) represents the constraints provided by the inter-layer relationship model; where,

[0115] Using Markov Chain Monte Carlo MCMC sampling, iteratively update the parameter θ:

[0116]

[0117] Among them, σ step represents the MCMC step size; θ (n+1) Represents the parameters after n iterations, represents the iterated covariance;

[0118] Assume the acceptance probability Accept the updated parameters with probability α, otherwise keep the original parameters;

[0119] Perform ground information inversion, including:

[0120] If the bottom elevation of the jth vegetation layer is known to be Assume that the elevation of the k′th layer satisfies Δh k′ represents the vertical distance between the k′th layer and the k′-1th layer;

[0121] Based on the surveying and mapping feature sequence, the LiDAR elevation feature value x is obtained when all the ground vegetation is removed, that is, when the ground is exposed. L1k_ti_1 , correct x L1k_ti_1 Get ground elevation in

[0122] For example, vegetation is distributed vertically, with known height intervals for each layer, such as 0-5m for the first layer, 5-10m for the second layer, and so on. Vegetation fall reduces the upper cover and exposes the underlying vegetation structure.

[0123] In this example, it is assumed that the LiDAR laser can penetrate three layers of vegetation, and the bottom elevation of the third layer of vegetation is After the third layer falls off in autumn (t2), the LiDAR laser penetrates to the fourth layer, and the bottom elevation of the fourth layer is By time series change Δh4=9-6.5=2.5m. In this embodiment, λ=0.1m -1 , ΔW(t i+1 ) is a fixed value, ΔW(t i+1 )=15%,MCMC step size is 0.05m,vertical distance Δh between layers k The prior value of is 2.0±0.3m. Iteratively calculate the parameters of the 5th and 6th layers (density and bottom elevation) and the ground elevation h g .

[0124] The obtained parameters such as the density of each layer of vegetation and the bottom elevation can be used to analyze the vegetation types and coverage of the land in the surveying area.

[0125] In this step, the purpose of setting up a hierarchical graph model is to abstract the physical data of each vegetation layer (vertical distance between layers, vegetation density, bottom elevation) and terrain parameters (ground elevation) into nodes in the hierarchical graph model, which can clarify the dependencies between various variables; for example, the bottom height of the third layer affects the top height of the fourth layer; LiDAR data and SAR data are uniformly encoded and integrated into the graph model, solving the problem of data heterogeneity while reducing the risk of single data failure.

[0126] The hierarchical graph model can capture the dynamic evolution of parameters in each layer. For example, after the leaves fall in autumn, the density of the upper vegetation decreases, exposing the lower layer. The hierarchical graph model automatically updates the parameters of each layer. When new data is input, only the relevant nodes need to be locally updated (correcting the vertical distance between vegetation layers in the third layer, etc.).

[0127] In the judgment of land type, we no longer rely on the black box model, but instead rely on the combination of node parameters in the hierarchical graph model (for example, if the number of layers is greater than 3 and the bottom elevation is greater than 5m, it is forest land) to enhance the credibility of the classification results.

[0128] Step 4: Classify and manage the land in the surveying area based on the vertical spatial distribution information of vegetation, and use the ground elevation data to correct the terrain data and output the corrected terrain data to improve the accuracy of the terrain data.

[0129] The classification and management of land in the surveying and mapping area based on the vertical spatial distribution information of vegetation includes the following steps:

[0130] Perform GMM fitting on the LiDAR laser echo of each grid to extract the number of layers and parameters of each layer;

[0131] Count the number of vegetation layers penetrated by LiDAR in each season to form a penetration layer dataset;

[0132] If at time T, the vegetation density of the jth layer of vegetation is ρ j (T) drops more than the threshold value compared to T-1, and this is marked as shedding; the elevation of the jth layer of vegetation If the change compared to T-1 exceeds the elevation change threshold, it is judged that the vegetation structure has changed; the time when the j-th layer of vegetation first falls off is recorded. The first shedding time of multiple layers of vegetation constitutes the shedding time series D;

[0133] The number of layers, layer density and bottom elevation are used to form a vertical space vector. The first drop-off time is added to the vertical space feature vector to form a space-time vector. The space-time vector is normalized to form a space-time feature vector X.ST ;

[0134] The classification model is constructed by XGBoost algorithm. ST Input the pre-trained classification model to classify vegetation. For example, if the number of vegetation layers is greater than 4, the vegetation density is greater than 0.7, and the bottom elevation is greater than 5m, it can be judged as a forest; if the number of vegetation layers is 1 and the bottom elevation is less than 1m, it is judged as grassland.

[0135] The method of correcting the terrain data using the ground elevation data and outputting the corrected terrain data includes the following steps:

[0136] Get the original and the output H of the two-stream network network ;

[0137] H network With H ground-truth Align in space, that is, H is interpolated by Kriging algorithm. ground-truth Interpolate to the grid network output by the dual-stream network, and interpolate to H corr =Kriging(H ground-truth,grid＝d ), where d represents the width of the grid;

[0138] Calculate the residual ΔH=H corr -H network ;

[0139] Build a lightweight UNet and input X in =[ΔH,H network ,▽H network ], output residual correction Among them ▽H network Represents the terrain gradient of the grid network, the UNet includes 4 encoding layers and 1 decoding layer;

[0140] Then, the corrected Where w′ represents the dynamic weight,

[0141] c represents the slope parameter, which is used to control the rate of change of the dynamic weight; K represents the standard deviation of the elevation of each grid in the grid network output by the dual-stream network; β represents the priority threshold. When K>β, the difference correction is used first.

[0142] The corrected raster elevation data obtained after decoding will replace the raster elevation data output by the dual-stream network.

[0143] The present invention also provides a land surveying and mapping data intelligent analysis and management system, comprising: a processor and a memory and a communication module connected to the processor, and the system is used to execute the land surveying and mapping data intelligent analysis and management method.

[0144] The present invention also provides a computer-readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the method for intelligent analysis and management of land surveying and mapping data.

[0145] The present invention also provides a computer-readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the method for intelligent analysis and management of land surveying and mapping data.

[0146] In some preferred embodiments, the method further includes the step of managing the dual-stream network output and the corrected raster elevation data, specifically:

[0147] Add identification label 1 and timestamp 1 to the raster network data output by the dual-stream network, and add identification label 2 and timestamp 2 to the corrected raster elevation data;

[0148] The versions of the raster network data and the corrected raster elevation data are distinguished according to the timestamp 1 and timestamp 2; the data types are distinguished according to the identification tag 1 and identification tag 2;

[0149] Recording the correction process of the grid network data through the correction log, including the replacement time of the elevation data and the corresponding grid information, wherein the grid information includes the grid width and the coordinates of the grid center point;

[0150] Import the corrected raster elevation data into PostGIS to support SQL queries, and publish WMS or WFS services through GeoServer to support team calls;

[0151] Use the data lake LakeFS to store different versions of raster network data and corrected raster elevation data.

[0152] The embodiments disclosed in the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. The embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part, and / or installed from a removable medium. When the computer program is executed by the central processing unit (CPU), the above-mentioned functions defined in the method of the present application are executed. It should be noted that the computer-readable medium mentioned above in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more wire segments, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take a variety of forms, including, but not limited to, an electromagnetic signal, an optical signal, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, electrical wire, optical fiber cable, RF, etc., or any suitable combination thereof.

[0153] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of the systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or portion of code that contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two boxes shown in succession can actually be executed substantially in parallel, or they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, as well as combinations of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified functions or operations, or can be implemented using a combination of dedicated hardware and computer instructions.

[0154] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are only examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and explained in the embodiments. Without departing from the principles, the implementation methods of the present invention may be subject to any changes or modifications.

Claims

1. A method for intelligent analysis and management of land surveying and mapping data, characterized in that: The method comprises: Acquire first and second surveying data of the same surveying area, and preprocess the first and second surveying data; the first surveying data is LiDAR elevation data; the second surveying data is Synthetic Aperture Radar elevation data; constructing a joint model and a dual-stream network, analyzing the pre-processed first surveying and mapping data and the second surveying and mapping data through the joint model and the dual-stream network, and outputting terrain data; Extracting multiple characteristic values ​​from the first surveying and mapping data according to a preset acquisition frequency to form a surveying and mapping characteristic sequence; inputting the surveying and mapping characteristic sequence into a preset vegetation multi-layer state model, and obtaining vegetation vertical spatial distribution information and ground elevation data through the vegetation multi-layer state model; Classify and manage the land in the surveying and mapping area based on the vertical spatial distribution information of vegetation, and use the ground elevation data to correct the terrain data and output the corrected terrain data to improve the accuracy of the terrain data; The method of correcting terrain data using ground elevation data and outputting the corrected terrain data includes: Get the original ground elevation data h g =H ground-truth and the output H of the two-stream network network ; H network With H ground-truth Align in space, that is, H is interpolated by Kriging algorithm. ground-truth Interpolate to the grid network output by the dual-stream network, and interpolate to H corr =Kriging(H ground-truth,grid＝d ), where d represents the width of the grid; where H corr represents Kriging interpolation; Calculate the residual ΔH=H corr -H network ; Build a lightweight UNet and input X in =[ΔH,H network ,▽H network ], output residual correction in Represents the terrain gradient of the grid network, the UNet includes 4 encoding layers and 1 decoding layer; Then, after correction Where w′ represents the dynamic weight, c represents the slope parameter, which is used to control the rate of change of the dynamic weight; K represents the standard deviation of the elevation of each grid in the grid network output by the dual-stream network; β represents the priority threshold. When K>β, the difference correction is used first. The corrected raster elevation data obtained after decoding will replace the raster elevation data output by the dual-stream network.

2. The method for intelligent analysis and management of land surveying and mapping data according to claim 1, characterized in that: The preprocessing includes spatiotemporal reference alignment and error model calibration; The spatiotemporal alignment includes: The first surveying and mapping data and the second surveying and mapping data are converted to a unified coordinate system, and the first surveying and mapping data is subjected to elevation conversion, specifically: H e =H L1 +N g , where H e Indicates the converted elevation data, H L1 Indicates the elevation data before conversion, N g Indicates the height of the geoid level; rasterizing the first surveying and mapping data into a grid network, and resampling the second surveying and mapping data using a bilinear interpolation algorithm and converting the data into the same grid network; The error model calibration includes: Calibrate the first surveying data noise R L1 =σ L 2 , where σ L Indicates the first surveying and mapping data error; Calibrate the second surveying data noise R S =σ S 2 , where σ S Indicates the second surveying and mapping data error.

3. The method for intelligent analysis and management of land surveying and mapping data according to claim 2, characterized in that: The constructing of the joint model and the dual-stream network, analyzing the pre-processed first surveying and mapping data and the second surveying and mapping data by the joint model and the dual-stream network, and outputting the terrain data, includes: Step 2.1: Build a joint model based on the Kalman filter algorithm, fuse the first surveying and mapping data with the second surveying and mapping data, and obtain the elevation data vector of the surveying and mapping area, specifically including: Step 2.1.1: Assume that the elevation of the terrain in the survey area does not change within the time period T. Then the elevation vector of the kth grid in the time period T is Among them H k Represents the elevation value of the kth grid network; Assume the state equation is: in, represents the LiDAR elevation vector of the kth grid, A represents the SAR elevation vector of the kth grid; k represents the identity matrix, w k represents the process noise vector, Q represents the process noise covariance, σ process represents the terrain change error; Step 2.1.2: After fusing the LiDAR elevation vector and the SAR elevation vector, the elevation vector of the kth grid is The observation matrix The first surveying and mapping data noise and the second surveying and mapping data noise constitute the measurement noise covariance matrix Let Kalman gain K k =P k|k-1 H T (HP k|k-1 H T +R) -1 ; Among them, the updated error covariance matrix P k =(IK k H)P k|k-1 ;P k|k-1 Represents the prediction error covariance matrix of the kth grid; The Kalman gain after substituting the observation matrix and noise covariance matrix is: The updated elevation vector is 4. The method for intelligent analysis and management of land surveying and mapping data according to claim 3, characterized in that: The constructing of the joint model and the dual-stream network, analyzing the pre-processed first surveying and mapping data and the second surveying and mapping data by the joint model and the dual-stream network, and outputting the terrain data further includes: Constructing a two-stream network based on the updated elevation vector and outputting terrain data includes the following sub-steps: Step 2.2.

1. Obtain LiDAR elevation vector and SAR elevation vector and P k ; Generate 6-channel input tensor for each grid Among them, the first and second channels are used to input the updated LiDAR elevation vector and the first mapping data error; the third and fourth channels are used to input the updated SAR elevation vector and the second mapping data error; the fifth and sixth channels are used to input the diagonal elements of the updated error covariance matrix; The input block is constructed in a 5×5 area to capture the terrain continuity, specifically X p ={X(i′+k″,j′+l)} 2 k″,l＝-2 ∈R 5×5×M , where X(i′+k″,j′+l) represents the adjacent grids of grid X(i′,j′), and M represents the number of layers of the grid network in the vertical direction; Step 2.2.2: Construct a convolutional network 1, which includes 3 convolutional layers, each of which includes 3×3 convolutions. The convolutional network 1 extracts LiDAR elevation vector features from the input tensor and forms a LiDAR elevation vector feature set F after normalization. L ; Step 2.2.3: Construct a convolutional network 2, which includes 3 convolutional layers, each of which includes 3×3 convolutions. The SAR elevation vector features are extracted from the input tensor through the convolutional network 2, and after normalization, the SAR elevation vector feature set F is formed. S ; Step 2.2.4: Construct convolutional network 3, and resample the SAR elevation vector features to the grid network to obtain a new SAR elevation vector feature set F′. S , and then calculate the attention weight: Attn = sigmoid(Conv([F L ; F′ S ])); Fusion F L and F S Get fusion feature F LS =Attn⊙F L +(1-Attn)⊙F′ s ; Step 2.2.5: Construct a terrain decoder to decode the fusion features and obtain the corresponding raster fusion elevation data. Output the raster fusion elevation data and the raster network of the surveying and mapping area through the output layer. The raster fusion elevation data and the raster network of the surveying and mapping area constitute the terrain data H. network , the terrain decoder consists of three 3×3 convolutional layers and one output layer.

5. The method for intelligent analysis and management of land surveying and mapping data according to claim 4, characterized in that: The step of extracting multiple feature values ​​from the first surveying and mapping data according to a preset acquisition frequency to form a surveying and mapping feature sequence includes: Taking the vegetation falling event as the time node, multiple elevation feature values ​​are extracted from the first surveying and mapping data to form a surveying and mapping feature sequence A i =(x L1k_ti_1 ,x L1k_ti_2 ,...,x L1k_ti_m ); Multiple surveying and mapping feature sequences constitute a surveying and mapping feature sequence A={A i |i=1,2,3,4}; where ti represents the i-th time node; x L1k_ti_1 It represents the LiDAR elevation feature value of the kth grid sampled for the first time before time node ti; m represents the number of acquisitions.

6. The method for intelligent analysis and management of land surveying and mapping data according to claim 5, characterized in that: The step of inputting the surveying and mapping feature sequence into a preset vegetation multi-layer state model and obtaining vegetation vertical spatial distribution information and ground elevation data through the vegetation multi-layer state model includes: Constructing a vegetation layer relationship model: ρ j+1 =ρ j ·e -λΔh +ε, where ρ j represents the vegetation density of the jth layer, Δh represents the vertical distance between layers, λ represents the attenuation ratio, and ε represents the error value; Set inter-layer timing constraints W j+1 (tn')=W j+1 (ti)+ΔW(tn'), where W j+1 (tn') represents the laser echo amplitude of the j+1 layer of vegetation when the jth layer of vegetation falls at time node tn', and ΔW(tn') represents the increment of the laser echo amplitude due to the reduction of the shielding of the upper layer of vegetation; where n'=i+1; Construct a hierarchical graph model, specifically including: The height and density of each layer of vegetation are used as nodes; The vegetation layer relationship model and temporal constraints between each vegetation layer are used as edges; Construct the maximum joint probability P(θ|D)∝P(D|θ)·P(θ), where P(D|θ) represents the LiDAR measurement data and the state equation The residual of the output data; P(θ) represents the constraints provided by the inter-layer relationship model; where, P(θ|D) represents the maximum joint probability; Using Markov Chain Monte Carlo MCMC sampling, iteratively update the parameter θ: Among them, σ step represents the MCMC step size; θ (n+1) Represents the parameters after n iterations, represents the iterated covariance; Assume the acceptance probability Accept the updated parameters with probability α, otherwise keep the original parameters; Perform ground information inversion, including: If the bottom elevation of the jth vegetation layer is known to be Assume that the elevation of the k′th layer satisfies Δh k′ represents the vertical distance between the k′th layer and the k′-1th layer; Based on the surveying and mapping feature sequence, the LiDAR elevation feature value x is obtained when all the ground vegetation is removed, that is, when the ground is exposed. L1k_ti_1 , correct x L1k_ti_1 Get ground elevation in 7. The method for intelligent analysis and management of land surveying and mapping data according to claim 6, characterized in that: Classify and manage land in the surveying and mapping area based on the vertical spatial distribution information of vegetation, including: Perform GMM fitting on the LiDAR laser echo of each grid to extract the number of layers and parameters of each layer; Count the number of vegetation layers penetrated by LiDAR in each season to form a penetration layer dataset; If at time T, the vegetation density of the jth layer of vegetation is ρ j (T) drops more than the threshold value compared to T-1, and this is marked as shedding; the elevation of the jth layer of vegetation If the change compared to T-1 exceeds the elevation change threshold, it is determined that the vegetation structure has changed; the time when the j-th layer of vegetation first falls off is recorded. The first shedding time of multiple layers of vegetation constitutes the shedding time series D; The number of layers, layer density and bottom elevation are used to form a vertical space vector. The first drop-off time is added to the vertical space feature vector to form a space-time vector. The space-time vector is normalized to form a space-time feature vector X. ST ; The classification model is constructed by XGBoost algorithm. ST Input into the pre-trained classification model to classify vegetation.

8. A land surveying and mapping data intelligent analysis and management system, comprising: A processor and a memory and a communication module connected to the processor, characterized in that the system is used to execute a land surveying and mapping data intelligent analysis and management method as described in any one of claims 1-7 above.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the land surveying and mapping data intelligent analysis and management method according to any one of claims 1 to 7.

Images