Multi-source geographic information data intelligent matching fusion method and system

Abstract

The invention discloses an intelligent matching fusion method and system for multi-source geographic information data, and relates to the technical field of data fusion, and the method comprises the steps: obtaining the multi-source geographic information data of a target mountain, carrying out the data alignment registration, generating a fusion data cube of the target mountain, building a time-space diagram neural network prediction model, and carrying out the prediction of the fusion data cube. Quantifying the hidden danger type risk probability graph and the deformation state graph of the target mountain, and evaluating a hidden danger type risk grading early warning graph of the target mountain; performing clustering division on the hidden danger type risk grading early warning graph of the target mountain by using a clustering algorithm, verifying a multi-source change trend of each high hidden danger type risk area of the target mountain, and obtaining a hidden danger type risk state of the target mountain; and judging whether the hidden danger type risk state of the target mountain exceeds a hidden danger type risk safety threshold, if not, judging that there is no abnormity, and if yes, judging that there is abnormity, and sending out a hidden danger type risk early warning. The method has the advantages that the hidden danger type early warning precision and timeliness are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data fusion, and in particular to a method and system for intelligent matching and fusion of multi-source geographic information data. Background Art

[0002] Intelligent matching and fusion of multi-source geographic information data refers to the matching, integration, and fusion of geographic information data from different sources, formats, and resolutions through intelligent algorithms and technical means, thereby achieving comprehensive, accurate, and unified analysis and utilization of geospatial data. Its main purpose is to enhance the value of data, overcome the differences and inconsistencies between different data sources, and obtain more accurate and effective information.

[0003] The existing intelligent matching and fusion of multi-source geographic information data in mountain hazard type monitoring cannot reflect the dynamic changes of hazard types in a timely manner due to the different update frequencies of different data sources and the lack of real-time performance; at the same time, due to differences in data sources, formats, and resolutions during the data fusion process, it is often impossible to efficiently integrate them, resulting in unsatisfactory fusion effects, affecting the accuracy and reliability of the monitoring results, and causing poor timeliness and high false alarm rates in mountain hazard type monitoring. Summary of the Invention

[0004] In order to solve the above technical problems, a method and system for intelligent matching and fusion of multi-source geographic information data are provided and disclosed. This technical solution solves the problem that the above-mentioned existing intelligent matching and fusion of multi-source geographic information data in mountain hidden danger type monitoring is unable to timely reflect the dynamic changes of hidden danger types due to the different update frequencies of different data sources and insufficient real-time performance; at the same time, due to differences in data sources, formats, and resolutions during the data fusion process, efficient integration is often not possible, resulting in unsatisfactory fusion effects, affecting the accuracy and reliability of the monitoring results, and leading to poor timeliness and high false alarm rates in mountain hidden danger type monitoring.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0006] A method for intelligent matching and fusion of multi-source geographic information data is disclosed, including:

[0007] Based on heterogeneous sensors, multi-source geographic information data of the target mountain is obtained;

[0008] Perform data alignment and registration based on the multi-source geographic information data of the target mountain to generate a fused data cube of the target mountain;

[0009] Based on the fusion data cube of the target mountain, a spatiotemporal graph neural network prediction model is established to quantify the risk probability map and deformation state map of the target mountain's hidden danger types, and to evaluate the risk classification warning map of the target mountain's hidden danger types;

[0010] Using a clustering algorithm, we cluster the target mountain's hidden danger type risk classification warning map, verify the multi-source change trend of each high hidden danger type risk area of ​​the target mountain, and obtain the hidden danger type risk status of the target mountain;

[0011] Determine whether the hidden danger type risk status of the target mountain exceeds the hidden danger type risk safety threshold. If not, it is determined that there is no abnormality. If so, it is determined to be abnormal and a hidden danger type risk warning is issued.

[0012] Furthermore, based on the satellite data of the target mountain, the volume deformation rate of the target mountain in the satellite data is extracted by using phase unwrapping and deformation calculation;

[0013] Based on the UAV aerial photography data, the DEM grid of the target mountain is generated by rasterizing the point cloud according to the ICP registration;

[0014] According to the volume change rate of the target mountain and the DEM grid of the target mountain, a deformation registration error function is established to generate the geometric consistency error between the volume change rate of the target mountain and the DEM grid of the target mountain;

[0015] The singular value decomposition (SVD) is used to solve the geometric consistency error between the target mountain's body shape change rate and the target mountain's DEM grid, and the optimal rotation matrix and translation matrix are obtained.

[0016] Using the optimal rotation matrix and translation matrix, the original point cloud data of the UAV aerial photography data is transformed into the coordinate system of the target point cloud, the transformed point cloud of the UAV aerial photography data is calculated, and the transformed DEM grid of the target mountain is determined;

[0017] The coordinate difference between the transformed point cloud of the UAV aerial photography data and the original point cloud data of the human-machine aerial photography data is compared to calculate the transformation deformation rate of the target mountain.

[0018] Furthermore, a grid joint field is formed based on the transformed DEM grid of the target mountain and the transformed deformation rate of the target mountain to generate the deformation-DEM joint field of the target mountain;

[0019] Step 202: Gaussian kernel mapping is performed based on the position of the sensor in the multi-source geographic information data of the target mountain to obtain a spatial code based on the multi-source geographic information data of the target mountain;

[0020] Use sine and cosine functions to map the timestamps in the target mountain multi-source geographic information data to the same unit time range, extract the trend characteristics of the target mountain multi-source geographic information data, and obtain the time code of the target mountain multi-source geographic information data; the trend characteristics include: climate change, temperature change

[0021] Based on the spatial coding of the multi-source geographic information data of the target mountain and the temporal coding of the multi-source geographic information data of the target mountain, a multi-dimensional information spatiotemporal state matrix of the target mountain is constructed.

[0022] Furthermore, based on the geological data in the multi-source geographic information data of the target mountain, a high-resolution lithologic distribution map of the target mountain is established;

[0023] Based on the rainfall observation data of the target mountain, the rainfall vector of the target mountain is marked according to the time series regression analysis;

[0024] The rainfall vector of the target mountain is weighted according to the spatial distance between the sensor position of the multi-source geographic information data of the target mountain and the corresponding grid point of the two-dimensional geographic area of ​​the target mountain, and the rainfall field of the target mountain is calculated according to the weighted formula;

[0025] Based on the high-resolution lithologic distribution map of the target mountain, the hydraulic conductivity and rainfall infiltration of the target mountain are determined, a dynamic geological and meteorological field function is established, and the stability index field and pore water pressure field of the target mountain are generated.

[0026] Furthermore, bicubic interpolation is used to synchronize the data sampling rates of the high-resolution lithologic distribution map of the target mountain and the rainfall vector of the target mountain;

[0027] Based on the multi-head attention of the Transformer model, the target mountain's deformation-DEM joint field is used as a query, and the target mountain's rainfall field, stability index field, and pore water pressure field are used as key values ​​to generate multi-dimensional features of the target mountain.

[0028] Based on the fusion of the multi-dimensional characteristics of the target mountain and the multi-dimensional information spatiotemporal state matrix of the target mountain, the data is divided according to the corresponding grid points of the two-dimensional geographical area of ​​the target mountain to form a fusion data cube of the target mountain.

[0029] Furthermore, based on the corresponding grid cells of the two-dimensional geographical area of ​​the target mountain as nodes, the fused data cube of the corresponding target mountain as attributes, the deformation Euclidean distance, elevation gradient and lithology consistency of the deformation-DEM joint field between each node are used as physical constraint edges;

[0030] Based on the ST-GNN spatiotemporal graph neural network model, the deformation Euclidean distance, elevation gradient and lithology consistency of the deformation-DEM joint field between each node are used as physical constraint edges as input. The topological mask is used to mark the corresponding grid cells of the two-dimensional geographical area of ​​the target mountain connected by the physical constraint edges as the original attention scores corresponding to the nodes. The softmax function is input to generate the weights of the original attention scores as dynamic spatiotemporal constraint edges to generate a dynamic graph of the target mountain.

[0031] Furthermore, based on the dynamic map of the target mountain, the risk probability prediction layer of hidden danger type and the deformation prediction layer are input;

[0032] Based on the hidden danger type risk probability prediction head layer, the convolution layer is used to compress the number of channels of the dynamic image of the target mountain, and then input into the batch normalization layer. The LeakyReLU activation function is applied to generate a compressed feature map of the target mountain. The depthwise separable convolution is input to extract local spatial features and obtain the spatial feature map of the target mountain.

[0033] Based on the jump connection between the compression feature map of the target mountain and the spatial feature map of the target mountain, the Sigmoid activation function is input to output the risk probability map of the hidden danger type of the target mountain;

[0034] Based on the deformation prediction head layer, transposed convolution is used to perform spatial upsampling on the dynamic image of the target mountain to generate an upsampled feature map of the target mountain. This map is then substituted into the temporal convolution layer to mark the temporal change trend of the upsampled feature map of the target mountain and obtain the temporal feature map of the target mountain.

[0035] The multi-channel convolutional layer is used to project the temporal feature map of the target mountain into the Tanh activation function, and the deformation prediction value of the target mountain is output to obtain the deformation state map of the target mountain.

[0036] Furthermore, grid resampling is performed based on the risk probability map of the hidden danger type of the target mountain and the deformation state map of the target mountain, and the grid cells corresponding to the two-dimensional geographical area of ​​the target mountain are unified into the UTM coordinate system;

[0037] Based on the hidden danger type risk probability map and the target mountain deformation state map, randomly mark the pixels in the area with hidden danger type risk probability and deformation rate as the initial seeds;

[0038] According to the known risk type and deformation rate of the target mountain, the propagation radius of the initial seed in the risk probability map of the target mountain's hidden danger type and the deformation state map of the target mountain is determined;

[0039] According to the hidden danger type risk probability map of the target mountain and the propagation radius of the initial seed in the deformation state map of the target mountain, density reachable expansion iteration is performed, with 2 times the standard deviation of the elevation gradient as the first stopping iteration condition and the graphic boundary as the second stopping iteration condition to obtain the hidden danger type risk-deformation area set of the target mountain.

[0040] Furthermore, based on the hidden danger type risk-deformation area set of the target mountain, the correlation factors affecting the hidden danger type-deformation are determined;

[0041] Based on random forest, the hidden danger type of the target mountain is used as the leaf node, the correlation factors affecting the hidden danger type-deformation in the fused data cube of the target mountain is used as the branch threshold, the fused data cube of the hidden danger type risk-deformation area set of the target mountain is used as input, and the hidden danger type risk index of the target mountain is used as output.

[0042] Furthermore, a multi-source geographic information data intelligent matching and fusion system is disclosed, which is characterized by including:

[0043] Data acquisition module, data fusion module, hidden danger type early warning map module, hidden danger type status analysis module, early warning module;

[0044] The data acquisition module is used to obtain multi-source geographic information data of the target mountain based on heterogeneous sensors;

[0045] The data fusion module is wirelessly connected to the data acquisition module, and is used to perform data alignment and registration based on multi-source geographic information data of the target mountain to generate a fused data cube of the target mountain;

[0046] The hidden danger type warning map module is electrically connected to the data fusion module. The hidden danger type warning map module is used to establish a spatiotemporal graph neural network prediction model based on the fusion data cube of the target mountain, quantify the hidden danger type risk probability map and deformation state map of the target mountain, and evaluate the hidden danger type risk classification warning map of the target mountain;

[0047] The hidden danger type status analysis module is electrically connected to the hidden danger type early warning map module. The hidden danger type status analysis module is used to use a clustering algorithm to perform clustering division on the hidden danger type risk classification early warning map of the target mountain, verify the multi-source change trend of each high hidden danger type risk area of ​​the target mountain, and obtain the hidden danger type risk status of the target mountain;

[0048] The early warning module is electrically connected to the hidden danger type status analysis module. The early warning module is used to determine whether the hidden danger type risk status of the target mountain exceeds the hidden danger type risk safety threshold. If not, it is determined that there is no abnormality. If so, it is determined to be abnormal and a hidden danger type risk warning is issued.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] The application provides a multi-source geographic information data intelligent matching fusion mountain hidden danger type early warning mountain hidden danger type scheme, multi-source geographic information data is acquired through a heterogeneous sensor network, a high-precision data cube is constructed by combining an innovative space-time alignment registration technology and a tensor fusion method, dynamic quantitative evaluation of hidden danger type risks is realized by using a space-time graph neural network and a multi-modal clustering algorithm, and finally intelligent early warning is carried out based on a non-equilibrium thermodynamic criterion. Its beneficial effects are that the early warning accuracy and timeliness are significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 The application discloses a multi-source geographic information data intelligent matching fusion method.

[0052] Figure 2 The application discloses a multi-source geographic information data intelligent matching fusion system. DETAILED DESCRIPTION

[0053] The following description is used to disclose the application so that those skilled in the art can implement the application. The preferred embodiments in the following description are only used as examples, and other obvious modifications can be thought of by those skilled in the art.

[0054] Referring to Figure 1 It is shown that the application discloses a multi-source geographic information data intelligent matching fusion method, comprising:

[0055] Step one, based on a heterogeneous sensor, multi-source geographic information data of a target mountain is acquired; the multi-source geographic information data of the target mountain comprises but is not limited to satellite data, unmanned aerial vehicle aerial photography data, real-time rock mass state data, geological and meteorological data.

[0056] Step two, data alignment registration is performed according to the multi-source geographic information data of the target mountain, and a fusion data cube of the target mountain is generated;

[0057] The step two comprises the following contents:

[0058] Step 201, based on satellite data of the target mountain, the body deformation rate of the target mountain in the satellite data is extracted by using phase unwrapping and deformation calculation;

[0059] Based on unmanned aerial vehicle aerial photography data, ICP registration point cloud rasterization is performed to generate a DEM grid of the target mountain;

[0060] According to the body deformation rate of the target mountain and the DEM grid of the target mountain, a deformation registration error function is established, and geometric consistency error between the body deformation rate of the target mountain and the DEM grid of the target mountain is generated;

[0061] The geometric consistency error between the body deformation rate of the target mountain and the DEM grid of the target mountain is solved by using singular value decomposition SVD, and an optimal rotation matrix and a translation matrix are obtained.

[0062] Using the optimal rotation matrix and translation matrix, the original point cloud data of the UAV aerial photography data is transformed into the coordinate system of the target point cloud, the transformed point cloud of the UAV aerial photography data is calculated, and the transformed DEM grid of the target mountain is determined;

[0063] Compare the coordinate difference between the transformed point cloud data of the UAV aerial photography data and the original point cloud data of the human-machine aerial photography data to calculate the transformation deformation rate of the target mountain;

[0064] A grid joint field is established based on the transformed DEM grid of the target mountain and the transformed deformation rate of the target mountain to generate a deformation-DEM joint field of the target mountain;

[0065] Step 202: Gaussian kernel mapping is performed based on the position of the sensor in the multi-source geographic information data of the target mountain to obtain a spatial code based on the multi-source geographic information data of the target mountain;

[0066] Use sine and cosine functions to map the timestamps in the target mountain multi-source geographic information data to the same unit time range, extract the trend characteristics of the target mountain multi-source geographic information data, and obtain the time code of the target mountain multi-source geographic information data; the trend characteristics include: climate change, temperature change

[0067] Based on the spatial coding of the multi-source geographic information data of the target mountain and the temporal coding of the multi-source geographic information data of the target mountain, a multi-dimensional information spatiotemporal state matrix of the target mountain is constructed;

[0068] Step 203: Create a high-resolution lithologic distribution map of the target mountain based on the geological data in the multi-source geographic information data of the target mountain.

[0069] Based on the rainfall observation data of the target mountain, the rainfall vector of the target mountain is marked according to the time series regression analysis;

[0070] The rainfall vector of the target mountain is weighted according to the spatial distance between the sensor position of the multi-source geographic information data of the target mountain and the corresponding grid point of the two-dimensional geographic area of ​​the target mountain, and the rainfall field of the target mountain is calculated according to the weighted formula;

[0071] Based on the high-resolution lithologic distribution map of the target mountain, the hydraulic conductivity and rainfall infiltration of the target mountain are determined, a dynamic geological and meteorological field function is established, and the stability index field and pore water pressure field of the target mountain are generated;

[0072] Step 204: Synchronize the data sampling rate of the high-resolution lithologic distribution map of the target mountain and the rainfall vector of the target mountain using bicubic interpolation;

[0073] Based on the multi-head attention of the Transformer model, the target mountain's deformation-DEM joint field is used as a query, and the target mountain's rainfall field, stability index field, and pore water pressure field are used as key values ​​to generate multi-dimensional features of the target mountain.

[0074] Based on the fusion of the multi-dimensional characteristics of the target mountain and the multi-dimensional information spatiotemporal state matrix of the target mountain, the data is divided according to the grid points corresponding to the two-dimensional geographical area of ​​the target mountain, and the fusion data cube of the target mountain is constructed as follows:

[0075]

[0076] Among them, xy,t is the multi-dimensional state vector of the x,y grid point of the target mountain at the tth time node, d xy is the elevation value of the x,y grid point of the transformed DEM grid of the target mountain. is the transformation deformation rate of the x,y grid point of the target mountain, l xy is the lithologic code of the x,y grid point of the target mountain, r xy,t is the rainfall at the x,y grid point of the target mountain at the tth time node, is the multi-dimensional information spatiotemporal state matrix of the xth and yth grid points of the target mountain at the tth time node, and MLP() is the fully connected network function of the Transformer model.

[0077] When using, combine the contents in steps 201 to 204:

[0078] As a further content: Since the target mountain is located in a remote mountainous area, when the data of the current target mountain is incomplete, multimodal learning can be performed based on the geological characteristics or similar characteristics of the current target mountain using previous data cases to improve the accuracy of the target mountain state prediction when the data is incomplete. The source of the data can be generated by an adversarial network or based on an existing database, and is well known to technical personnel in this field, so no further description will be given here.

[0079] This solution uses InSAR phase unwrapping and UAV point cloud ICP registration to extract mountain deformation rate and DEM grids, respectively, and uses SVD optimization to establish a joint deformation-DEM field. A spatiotemporal state matrix is ​​constructed using Gaussian kernel spatial encoding and sinusoidal time encoding. A lithologic distribution map is generated by combining geological data, and a rainfall field is established through time series regression analysis to determine the stability index field and pore water pressure field. The Transformer multi-head attention mechanism is substituted, and key-value features such as the rainfall field and the stability index field are integrated to construct a multi-dimensional fusion data cube using the joint deformation-DEM field as a query. The beneficial effects include improving data registration accuracy, enabling multi-physics field collaborative analysis through dynamic coupling analysis, and providing dynamic decision-making indicators for hazard type warnings.

[0080] Step 3: Based on the fused data cube of the target mountain, a spatiotemporal graph neural network prediction model is established to quantify the risk probability map and deformation state map of the target mountain’s hidden danger types, and to evaluate the target mountain’s hidden danger type risk classification warning map;

[0081] The step three includes the following:

[0082] Step 301: Based on the grid cells corresponding to the two-dimensional geographic area of ​​the target mountain as nodes, the fused data cube of the corresponding target mountain as attributes, and the deformation Euclidean distance, elevation gradient, and lithology consistency of the deformation-DEM joint field between each node as physical constraint edges;

[0083] Based on the ST-GNN spatiotemporal graph neural network model, the deformation Euclidean distance, elevation gradient, and lithology consistency of the deformation-DEM joint field between each node are used as physical constraint edges as input. The topological mask is used to mark the corresponding grid cells of the two-dimensional geographical area of ​​the target mountain connected by the physical constraint edges as the original attention score corresponding to the node. The softmax function is input to generate the weight of each original attention score as the dynamic spatiotemporal constraint edge to generate a dynamic graph of the target mountain.

[0084] Step 302: Based on the dynamic image of the target mountain, input the first layer of risk probability prediction of hidden danger type and the first layer of deformation prediction;

[0085] Based on the hidden danger type risk probability prediction head layer, the convolution layer is used to compress the number of channels of the dynamic image of the target mountain, and then input into the batch normalization layer. The LeakyReLU activation function is applied to generate a compressed feature map of the target mountain. The depthwise separable convolution is input to extract local spatial features and obtain the spatial feature map of the target mountain.

[0086] Based on the jump connection between the compression feature map of the target mountain and the spatial feature map of the target mountain, the Sigmoid activation function is input to output the risk probability map of the hidden danger type of the target mountain;

[0087] Based on the deformation prediction head layer, transposed convolution is used to perform spatial upsampling on the dynamic image of the target mountain to generate an upsampled feature map of the target mountain. This map is then substituted into the temporal convolution layer to mark the temporal change trend of the upsampled feature map of the target mountain and obtain the temporal feature map of the target mountain.

[0088] The multi-channel convolutional layer is used to project the temporal feature map of the target mountain into the Tanh activation function, output the deformation prediction value of the target mountain, and obtain the deformation state map of the target mountain;

[0089] When used, combine the content from 301 to step 302:

[0090] As a further content, the hazard type refers to, but is not limited to: target mountain rockfall, target landslide, target mountain debris flow or mud and rock flow;

[0091] This solution constructs a spatiotemporal graph neural network (ST-GNN) prediction model based on a fused data cube. Using two-dimensional mountain grid cells as nodes and fused data as attributes, it constructs physical constraint edges through deformation Euclidean distance, elevation gradient, and lithology consistency. It also generates dynamic spatiotemporal constraint edges using topological masks and an attention mechanism. The model outputs a hazard type risk probability map and a deformation state map through a hazard type risk probability prediction head layer (depthwise separable convolution + skip connection) and a deformation prediction head layer (transposed convolution + temporal convolution), respectively, ultimately generating a hazard type risk graded warning map. This approach has the following benefits: improving the accuracy of mountain hazard type risk prediction, and simultaneously outputting hazard type risk probability maps and deformation state maps, providing data support for disaster prevention decision-making.

[0092] Step 4: Use a clustering algorithm to perform clustering on the target mountain's hidden danger type risk classification warning map, verify the multi-source change trend of each high hidden danger type risk area of ​​the target mountain, and obtain the hidden danger type risk status of the target mountain;

[0093] The step 4 includes the following contents:

[0094] Step 401: perform grid resampling based on the hidden danger type risk probability map of the target mountain and the deformation state map of the target mountain, and unify the grid cells corresponding to the two-dimensional geographical area of ​​the target mountain into the UTM coordinate system;

[0095] Based on the hidden danger type risk probability map and the target mountain deformation state map, randomly mark the pixels in the area with hidden danger type risk probability and deformation rate as the initial seeds;

[0096] According to the known risk type and deformation rate of the target mountain, the propagation radius of the initial seed in the risk probability map of the target mountain's hidden danger type and the deformation state map of the target mountain is determined;

[0097] According to the hidden danger type risk probability map of the target mountain and the propagation radius of the initial seed in the deformation state map of the target mountain, density reachable expansion iteration is performed, with 2 times the standard deviation of the elevation gradient as the first stopping iteration condition and the graphic boundary as the second stopping iteration condition to obtain the hidden danger type risk-deformation area set of the target mountain.

[0098] Step 402: Determine the correlation factors affecting the hidden danger type-deformation based on the hidden danger type risk-deformation area set of the target mountain;

[0099] Based on random forest, the target mountain hidden danger type is used as the leaf node, the correlation factors affecting the hidden danger type and deformation in the fusion data cube of the target mountain are used as the branch threshold, the fusion data cube of the hidden danger type risk and deformation area set of the target mountain is used as the input, and the hidden danger type risk index of the target mountain is used as the output;

[0100] When using, combine the contents in steps 401 to 402:

[0101] As a further step, the initial seeds are dynamically generated based on risk probability and deformation rate. The risk-deformation area set is iteratively divided through density propagation constrained by elevation gradient. The contribution weights of geological, deformation, hydrological and other factors are quantified using random forest. Finally, a risk classification and disaster factor association model is output. Compared with the traditional fixed threshold warning method, the false alarm rate of mountain hidden danger types is reduced. At the same time, the contribution of each factor to the hidden danger type can be analyzed, providing a quantitative numerical basis for targeted disaster prevention.

[0102] Step 5: Determine whether the hidden danger type risk status of the target mountain exceeds the hidden danger type risk safety threshold. If not, determine that there is no abnormality. If so, determine that there is an abnormality and issue a hidden danger type risk warning.

[0103] Reference Figure 2 As shown, a multi-source geographic information data intelligent matching and fusion system is disclosed, including:

[0104] Data acquisition module, data fusion module, hidden danger type early warning map module, hidden danger type status analysis module, early warning module;

[0105] The data acquisition module is used to obtain multi-source geographic information data of the target mountain based on heterogeneous sensors;

[0106] The data fusion module is wirelessly connected to the data acquisition module, and is used to perform data alignment and registration based on multi-source geographic information data of the target mountain to generate a fused data cube of the target mountain;

[0107] The hidden danger type warning map module is electrically connected to the data fusion module. The hidden danger type warning map module is used to establish a spatiotemporal graph neural network prediction model based on the fusion data cube of the target mountain, quantify the hidden danger type risk probability map and deformation state map of the target mountain, and evaluate the hidden danger type risk classification warning map of the target mountain;

[0108] The hidden danger type status analysis module is electrically connected to the hidden danger type early warning map module. The hidden danger type status analysis module is used to use a clustering algorithm to perform clustering division on the hidden danger type risk classification early warning map of the target mountain, verify the multi-source change trend of each high hidden danger type risk area of ​​the target mountain, and obtain the hidden danger type risk status of the target mountain;

[0109] The early warning module is electrically connected to the hidden danger type status analysis module. The early warning module is used to determine whether the hidden danger type risk status of the target mountain exceeds the hidden danger type risk safety threshold. If not, it is determined that there is no abnormality. If so, it is determined to be abnormal and a hidden danger type risk warning is issued.

[0110] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions merely illustrate the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent matching and fusion of multi-source geographic information data is disclosed, which is characterized by: include: S1. Based on heterogeneous sensors, obtain multi-source geographic information data of the target mountain; S2. Perform data alignment and registration based on the multi-source geographic information data of the target mountain to generate a fused data cube of the target mountain; S3. Based on the fusion data cube of the target mountain, a spatiotemporal graph neural network prediction model is established to quantify the risk probability map and deformation state map of the target mountain’s hidden danger types, and to evaluate the risk classification warning map of the target mountain’s hidden danger types; S4. Using a clustering algorithm, cluster the target mountain’s hidden danger type risk classification warning map, verify the multi-source change trend of each high hidden danger type risk area of ​​the target mountain, and obtain the hidden danger type risk status of the target mountain; S5. Determine whether the hidden danger type risk status of the target mountain exceeds the hidden danger type risk safety threshold. If not, determine that there is no abnormality. If so, determine that there is an abnormality and issue a hidden danger type risk warning.

2. According to claim 1, the method for intelligent matching and fusion of multi-source geographic information data is disclosed, characterized in that The S2 includes: Based on the satellite data of the target mountain, the deformation rate of the target mountain in the satellite data is extracted by using phase unwrapping and deformation calculation; Based on the UAV aerial photography data, the DEM grid of the target mountain is generated by rasterizing the point cloud according to the ICP registration; According to the volume change rate of the target mountain and the DEM grid of the target mountain, a deformation registration error function is established to generate the geometric consistency error between the volume change rate of the target mountain and the DEM grid of the target mountain; The singular value decomposition (SVD) is used to solve the geometric consistency error between the target mountain's body shape change rate and the target mountain's DEM grid, and the optimal rotation matrix and translation matrix are obtained. Using the optimal rotation matrix and translation matrix, the original point cloud data of the UAV aerial photography data is transformed into the coordinate system of the target point cloud, the transformed point cloud of the UAV aerial photography data is calculated, and the transformed DEM grid of the target mountain is determined; The coordinate difference between the transformed point cloud of the UAV aerial photography data and the original point cloud data of the human-machine aerial photography data is compared to calculate the transformation deformation rate of the target mountain.

3. The method for intelligent matching and fusion of multi-source geographic information data disclosed in claim 2 is characterized in that Said S2 further comprises: A grid joint field is established based on the transformed DEM grid of the target mountain and the transformed deformation rate of the target mountain to generate a deformation-DEM joint field of the target mountain; Step 202: Gaussian kernel mapping is performed based on the position of the sensor in the multi-source geographic information data of the target mountain to obtain a spatial code based on the multi-source geographic information data of the target mountain; Use sine and cosine functions to map the timestamps in the target mountain multi-source geographic information data to the same unit time range, extract the trend characteristics of the target mountain multi-source geographic information data, and obtain the time code of the target mountain multi-source geographic information data; the trend characteristics include: climate change, temperature change Based on the spatial coding of the multi-source geographic information data of the target mountain and the temporal coding of the multi-source geographic information data of the target mountain, a multi-dimensional information spatiotemporal state matrix of the target mountain is constructed.

4. The method for intelligent matching and fusion of multi-source geographic information data disclosed in claim 3 is characterized in that Said S2 further comprises: Based on the geological data in the multi-source geographic information data of the target mountain, a high-resolution lithologic distribution map of the target mountain is established; Based on the rainfall observation data of the target mountain, the rainfall vector of the target mountain is marked according to the time series regression analysis; The rainfall vector of the target mountain is weighted according to the spatial distance between the sensor position of the multi-source geographic information data of the target mountain and the corresponding grid point of the two-dimensional geographic area of ​​the target mountain, and the rainfall field of the target mountain is calculated according to the weighted formula; Based on the high-resolution lithologic distribution map of the target mountain, the hydraulic conductivity and rainfall infiltration of the target mountain are determined, a dynamic geological and meteorological field function is established, and the stability index field and pore water pressure field of the target mountain are generated.

5. The method for intelligent matching and fusion of multi-source geographic information data disclosed in claim 4 is characterized in that Said S2 further comprises: Using bicubic interpolation, the data sampling rate of the high-resolution lithologic distribution map of the target mountain and the rainfall vector of the target mountain are synchronized; Based on the multi-head attention of the Transformer model, the target mountain's deformation-DEM joint field is used as a query, and the target mountain's rainfall field, stability index field, and pore water pressure field are used as key values ​​to generate multi-dimensional features of the target mountain. Based on the fusion of the multi-dimensional characteristics of the target mountain and the multi-dimensional information spatiotemporal state matrix of the target mountain, the data is divided according to the corresponding grid points of the two-dimensional geographical area of ​​the target mountain to form a fusion data cube of the target mountain.

6. The method for intelligent matching and fusion of multi-source geographic information data disclosed in claim 5 is characterized in that The S3 includes: Based on the grid cells corresponding to the two-dimensional geographical area of ​​the target mountain as nodes, the fused data cube of the corresponding target mountain as attributes, and the deformation Euclidean distance, elevation gradient and lithology consistency of the deformation-DEM joint field between each node as physical constraint edges; Based on the ST-GNN spatiotemporal graph neural network model, the deformation Euclidean distance, elevation gradient and lithology consistency of the deformation-DEM joint field between each node are used as physical constraint edges as input. The topological mask is used to mark the corresponding grid cells of the two-dimensional geographical area of ​​the target mountain connected by the physical constraint edges as the original attention scores corresponding to the nodes. The softmax function is input to generate the weights of the original attention scores as dynamic spatiotemporal constraint edges to generate a dynamic graph of the target mountain.

7. The method for intelligent matching and fusion of multi-source geographic information data disclosed in claim 6 is characterized in that Said S3 further comprises: Based on the dynamic map of the target mountain, input the first layer of risk probability prediction of hidden danger type and deformation prediction; Based on the hidden danger type risk probability prediction head layer, the convolution layer is used to compress the number of channels of the dynamic image of the target mountain, and then input into the batch normalization layer. The LeakyReLU activation function is applied to generate a compressed feature map of the target mountain. The depthwise separable convolution is input to extract local spatial features and obtain the spatial feature map of the target mountain. Based on the jump connection between the compression feature map of the target mountain and the spatial feature map of the target mountain, the Sigmoid activation function is input to output the risk probability map of the hidden danger type of the target mountain; Based on the deformation prediction head layer, transposed convolution is used to perform spatial upsampling on the dynamic image of the target mountain to generate an upsampled feature map of the target mountain. This map is then substituted into the temporal convolution layer to mark the temporal change trend of the upsampled feature map of the target mountain and obtain the temporal feature map of the target mountain. The multi-channel convolutional layer is used to project the temporal feature map of the target mountain into the Tanh activation function, and the deformation prediction value of the target mountain is output to obtain the deformation state map of the target mountain.

8. The method for intelligent matching and fusion of multi-source geographic information data disclosed in claim 7 is characterized in that The S4 includes: Based on the risk probability map of the hidden danger type of the target mountain and the deformation state map of the target mountain, grid resampling is performed, and the grid cells corresponding to the two-dimensional geographical area of ​​the target mountain are unified into the UTM coordinate system; Based on the hidden danger type risk probability map and the target mountain deformation state map, randomly mark the pixels in the area with hidden danger type risk probability and deformation rate as the initial seeds; According to the known risk type and deformation rate of the target mountain, the propagation radius of the initial seed in the risk probability map of the target mountain's hidden danger type and the deformation state map of the target mountain is determined; According to the hidden danger type risk probability map of the target mountain and the propagation radius of the initial seed in the deformation state map of the target mountain, density reachable expansion iteration is performed, with 2 times the standard deviation of the elevation gradient as the first stopping iteration condition and the graphic boundary as the second stopping iteration condition to obtain the hidden danger type risk-deformation area set of the target mountain.

9. The method for intelligent matching and fusion of multi-source geographic information data disclosed in claim 8 is characterized in that Said S4 further comprises: Based on the hidden danger type risk-deformation area set of the target mountain, determine the correlation factors that affect the hidden danger type-deformation; Based on random forest, the hidden danger type of the target mountain is used as the leaf node, the correlation factors affecting the hidden danger type-deformation in the fused data cube of the target mountain is used as the branch threshold, the fused data cube of the hidden danger type risk-deformation area set of the target mountain is used as input, and the hidden danger type risk index of the target mountain is used as output.

10. A multi-source geographic information data intelligent matching and fusion system is disclosed, which is characterized by: include: Data acquisition module, data fusion module, hidden danger type early warning map module, hidden danger type status analysis module, early warning module; The data acquisition module is used to obtain multi-source geographic information data of the target mountain based on heterogeneous sensors; The data fusion module is wirelessly connected to the data acquisition module, and is used to perform data alignment and registration based on multi-source geographic information data of the target mountain to generate a fused data cube of the target mountain; The hidden danger type warning map module is electrically connected to the data fusion module. The hidden danger type warning map module is used to establish a spatiotemporal graph neural network prediction model based on the fusion data cube of the target mountain, quantify the hidden danger type risk probability map and deformation state map of the target mountain, and evaluate the hidden danger type risk classification warning map of the target mountain; The hidden danger type status analysis module is electrically connected to the hidden danger type early warning map module. The hidden danger type status analysis module is used to use a clustering algorithm to perform clustering division on the hidden danger type risk classification early warning map of the target mountain, verify the multi-source change trend of each high hidden danger type risk area of ​​the target mountain, and obtain the hidden danger type risk status of the target mountain; The early warning module is electrically connected to the hidden danger type status analysis module. The early warning module is used to determine whether the hidden danger type risk status of the target mountain exceeds the hidden danger type risk safety threshold. If not, it is determined that there is no abnormality. If so, it is determined to be abnormal and a hidden danger type risk warning is issued.

Images