Mountain torrent disaster classification method, device and equipment

Abstract

The invention discloses a mountain torrent disaster classification method, device and equipment, and relates to the technical field of disaster monitoring, and the method comprises the steps: obtaining a geographic feature data set of a plurality of drainage basin nodes of a target region, and constructing a feature matrix according to the geographic feature data set; inputting the feature matrix into a mountain torrent disaster classification model, and outputting the classification probability that each drainage basin node belongs to each mountain torrent disaster classification; the mountain torrent disaster classification model is used for calculating an edge weight matrix of a graph structure through a multi-head attention mechanism according to the feature matrix and the graph structure formed by drainage basin nodes of the target area; sparse processing is carried out on the edge weight matrix, and graph convolution operation is carried out based on the feature matrix and the edge weight matrix after sparse processing to obtain aggregation features; performing global feature enhancement on the aggregated features to obtain enhanced features; and outputting the classification probability that each drainage basin node belongs to each mountain torrent disaster classification according to the enhanced features. The accuracy of mountain torrent disaster classification can be improved.

Description

Technical Field

[0001] This application relates to the technical field of disaster monitoring, and particularly to a mountain flood disaster classification method, device and equipment. Background Art

[0002] For the risk assessment of mountain flood disasters, there is currently a probability early warning of mountain flood disasters based on machine learning. However, there is an inherent contradiction between the dynamic element relationship modeling and the calculation efficiency in the existing methods. If a complex structure is used to capture the time-varying interaction features of multiple elements in a rainstorm event, it will face the bottleneck of exponential growth of computing resources; if the calculation efficiency is pursued, it is difficult to break through the modeling limitations of the static graph structure. On the other hand, the problem of balancing global feature perception and local detail preservation stems from the inherent defects of traditional model architectures. In addition, the differences in the representation scale, modal dimension and semantic space of multi-source heterogeneous data (such as meteorological time series data, geographical raster data, and disaster text data) make it impossible for existing methods to construct a unified feature mapping space, restricting the in-depth mining of the coupling relationship of multiple elements, and thus affecting the accuracy of mountain flood disaster classification. Summary of the Invention

[0003] The purpose of this application is to provide a mountain flood disaster classification method, device and equipment, which can improve the accuracy of mountain flood disaster classification.

[0004] To achieve the above purpose, this application provides the following solutions: In the first aspect, this application provides a mountain flood disaster classification method, and the mountain flood disaster classification includes: Obtain the geographical feature dataset of multiple watershed nodes in the target area, and construct a feature matrix according to the geographical feature dataset; Input the feature matrix into the mountain flood disaster classification model, and output the classification probability of each watershed node belonging to each mountain flood disaster classification; the mountain flood disaster classification model is used for: calculating the edge weight matrix of the graph structure according to the feature matrix and the graph structure composed of each watershed node in the target area through the multi-head attention mechanism; performing sparsification processing on the edge weight matrix, and performing graph convolution operation based on the feature matrix and the sparsified edge weight matrix to obtain an aggregated feature; performing global feature enhancement on the aggregated feature to obtain an enhanced feature; outputting the classification probability of each watershed node belonging to each mountain flood disaster classification according to the enhanced feature; Take the mountain flood disaster classification corresponding to the maximum value among the classification probabilities of each watershed node belonging to each mountain flood disaster classification as the mountain flood disaster classification of each watershed node.

[0005] Optionally, the feature matrix includes the elevation standard deviation, slope variation, surface coverage rate, basin area, terrain undulation degree, average slope, average elevation, shape coefficient, length of the longest confluence path, gradient of the longest confluence path, gradient of the longest confluence path 1085, stable infiltration rate, roughness coefficient, reach length, reach gradient, unit hydrograph peak discharge modulus, maximum basin confluence time, river power index, terrain humidity index, normalized difference vegetation index, disaster occurrence time, longitude, and latitude of each watershed node.

[0006] Optionally, the formula for calculating the edge weight matrix of the graph structure through the multi-head attention mechanism is expressed as: ; where, represents the edge weight matrix, and the edge weight matrix is an asymmetric edge weight matrix, is the query matrix, is the key matrix, , , is the feature matrix, and are both trainable parameters, represents a dataset of dimension, d represents the geographical feature dimension of each watershed node, is the first attention head dimension, ⊙ represents the Hadamard product, and M is the adjacency mask matrix.

[0007] Optionally, the formula for sparsifying the edge weight matrix is expressed as: ; where, is the edge weight matrix after sparsification, represents the edge weight matrix, ⊙ represents the Hadamard product, is the indicator function, is the learnable threshold parameter.

[0008] Optionally, the formula for performing graph convolution operation based on the feature matrix and the edge weight matrix after sparsification is expressed as: ; where, is the aggregated feature, represents the rectified linear unit function, D is the degree matrix, is the edge weight matrix after sparsification, is the feature matrix, is the projection matrix, represents a dataset of dimension, d represents the geographical feature dimension of each watershed node, is the second attention head dimension.

[0009] Optionally, perform global feature enhancement on the aggregated features to obtain enhanced features, specifically including: Perform Transformer encoding on the aggregated features to obtain enhanced features; The formula for performing Transformer encoding on the aggregated features is: ; where is the aggregated feature, is the enhanced feature, represents layer normalization, represents the multi-head attention mechanism.

[0010] Optionally, the flash flood disaster classification model is obtained by training with a training set, and the joint loss function used during training is expressed as: ; where L represents the joint loss value, is the cross-entropy loss, represents the edge weight matrix, and are both weight coefficients, represents the edge weight from the i-th watershed node to the j-th watershed node, represents the edge weight from the j-th watershed node to the i-th watershed node.

[0011] Optionally, according to the enhanced features, output the classification probabilities of each watershed node belonging to each flash flood disaster classification, specifically including: Use a multi-layer perceptron to output the classification probabilities of each watershed node belonging to each flash flood disaster classification according to the enhanced features.

[0012] In a second aspect, the present application provides a flash flood disaster classification device, which applies the flash flood disaster classification method described in any one of the above, and the flash flood disaster classification device includes: A feature matrix determination module, configured to obtain a geographical feature dataset of multiple watershed nodes in a target area, and construct a feature matrix according to the geographical feature dataset; The flash flood disaster classification probability prediction module is used to input the feature matrix into the flash flood disaster classification model and output the classification probabilities of each watershed node belonging to each flash flood disaster classification; the flash flood disaster classification model is used to: calculate the edge weight matrix of the graph structure according to the feature matrix and the graph structure composed of each watershed node in the target area through the multi-head attention mechanism; sparsify the edge weight matrix, and perform graph convolution operations based on the feature matrix and the sparsified edge weight matrix to obtain aggregated features; perform global feature enhancement on the aggregated features to obtain enhanced features; output the classification probabilities of each watershed node belonging to each flash flood disaster classification according to the enhanced features. The classification determination module is used to take the flash flood disaster classification corresponding to the maximum value among the classification probabilities of each watershed node belonging to each flash flood disaster classification as the flash flood disaster classification of each watershed node.

[0013] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the steps of the flash flood disaster classification method described in any one of the above.

[0014] According to the specific embodiments provided by the present application, the present application discloses the following technical effects: The present application provides a flash flood disaster classification method, device and equipment. According to the feature matrix and the graph structure composed of each watershed node in the target area, the edge weight matrix of the graph structure is calculated through the multi-head attention mechanism, breaking through the limitation of the static graph structure, enhancing the visual interpretability of the edge weight matrix, and thus improving the prediction accuracy; sparsify the edge weight matrix, and perform graph convolution operations based on the feature matrix and the sparsified edge weight matrix to obtain aggregated features, realizing a two-stage architecture of multi-head attention and graph convolution, alleviating the conflict between global calculation efficiency and local feature preservation, and further improving the accuracy of flash flood disaster classification. Description of the Drawings

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0016] Figure 1 It is a schematic flowchart of a flash flood disaster classification method provided by an embodiment of the present application.

[0017] Figure 2 It is a schematic principle diagram of a flash flood disaster classification method provided by an embodiment of the present application.

[0018] Figure 3 Schematic diagram of the functional modules of a flash flood disaster classification device provided in an embodiment of the present application.

[0019] Figure 4 Schematic diagram of the structure of a computer device provided in an embodiment of the present application. Detailed implementation manners

[0020] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0021] To make the above objects, features, and advantages of the present application more obvious and understandable, the present invention will be described in detail below in conjunction with the drawings and specific implementation manners.

[0022] In an exemplary embodiment, the present application provides a flash flood disaster classification method, as Figure 1 and Figure 2 shown, the flash flood disaster classification includes steps 101 to 102.

[0023] Step 101: Obtain a geographical feature data set of multiple watershed nodes in the target area, and construct a feature matrix according to the geographical feature data set.

[0024] Step 102: Input the feature matrix into the flash flood disaster classification model, and output the classification probability that each watershed node belongs to each type of flash flood disaster classification; the flash flood disaster classification model is used to: calculate the edge weight matrix of the graph structure according to the feature matrix and the graph structure composed of each watershed node in the target area through the multi-head attention mechanism; perform sparsification processing on the edge weight matrix, and perform graph convolution operation based on the feature matrix and the sparsified edge weight matrix to obtain aggregated features; perform global feature enhancement on the aggregated features to obtain enhanced features; output the classification probability that each watershed node belongs to each type of flash flood disaster classification according to the enhanced features.

[0025] Step 103: Take the flash flood disaster classification corresponding to the maximum value among the classification probabilities that each watershed node belongs to each type of flash flood disaster classification as the flash flood disaster classification of each watershed node.

[0026] The graph structure is a fully connected graph structure.

[0027] The flash flood disaster classification includes rainstorm type, debris flow type, breach type, and snowmelt type.

[0028] This application also includes visualizing the graph structure, mining unrecognized types of mountain flood disasters through node connectivity, and is expected to discover subclasses or subtypes under the main types of mountain flood disasters, so as to achieve accurate classification and risk zoning of mountain flood disasters.

[0029] In an exemplary embodiment, step 101 specifically includes: The computer reads the geographical feature dataset of N basin nodes in the target area from the storage device and constructs a feature matrix . Denote as the data of, is the geographical feature dimension of each basin node.

[0030] Among them, the known quantity is the measured geographical feature data of each basin node; the unknown quantity is the potential association strength between each basin node.

[0031] The feature elements in the feature matrix include the elevation standard deviation, slope variation, surface coverage rate, basin area, terrain undulation degree, average slope, average elevation, shape coefficient, longest confluence path length, longest confluence path slope (‰), longest confluence path slope 1085 (‰), stable infiltration rate, roughness coefficient, river section length, river section slope, unit hydrograph peak discharge modulus, basin maximum confluence time, river power index, terrain humidity index, normalized vegetation index, disaster occurrence time, longitude and latitude of each basin node.

[0032] In an exemplary embodiment, the edge weight matrix of the graph structure is calculated through the multi-head attention mechanism, specifically including: The computer calculates the learnable edge weight matrix through the multi-head attention mechanism . Denote as the dataset of dimension.

[0033] The formula for calculating the edge weight matrix of the graph structure through the multi-head attention mechanism is expressed as: ; Among them, denotes the edge weight matrix, and the edge weight matrix is an asymmetric edge weight matrix, is the query matrix, is the key matrix, and the superscript T represents transpose, , , is the feature matrix, are all trainable parameters, Denote as the dataset of dimension, d represents the geographical feature dimension of each basin node, is the first attention head dimension, and ⊙ represents the Hadamard product; M ∈ {0, 1}^(N×N) is the adjacency mask matrix, which is generated according to the minimum distance threshold of the watershed.

[0034] The function of calculating the edge weight matrix of the graph structure is to break through the limitations of the predefined graph structure and achieve dynamic relationship modeling.

[0035] In an exemplary embodiment, the present application realizes two-stage message passing through edge weight filtering and feature aggregation.

[0036] Edge weight filtering is to sparsify the edge weight matrix through a computer for processing, Figure 2 In the figure, nodes e, u, g, and v are all watershed nodes, represents the geographical feature of node u, represents the geographical feature of node v, represents the geographical feature of node e, represents the edge weight from node u to node g, represents the edge weight from node v to node g, represents the edge weight from node e to node g.

[0037] The formula for sparsifying the edge weight matrix is expressed as: .

[0038] Among them, is the sparsified edge weight matrix, represents the edge weight matrix, ⊙ represents the Hadamard product, is the indicator function, is the learnable threshold parameter.

[0039] The role of edge weight filtering is to break through the limitations of the predefined graph structure and achieve dynamic relationship modeling.

[0040] In an exemplary embodiment, feature aggregation includes performing graph convolution operations through a computer.

[0041] The formula for performing graph convolution operations based on the feature matrix and the sparsified edge weight matrix is expressed as: .

[0042] Among them, is the aggregated feature, represents the rectified linear unit function, D is the degree matrix, is the sparsified edge weight matrix, is the feature matrix, is the projection matrix, represents A dataset of dimensions, where d represents the geographical feature dimensions of each watershed node. Is the second attention head dimension.

[0043] The role of the aggregated features is to achieve feature propagation that retains terrain details.

[0044] In an exemplary embodiment, global feature enhancement is performed on the aggregated features to obtain enhanced features, specifically including: encoding the aggregated features through a computer using Transformer to obtain enhanced features.

[0045] The formula for encoding the aggregated features using Transformer is expressed as: .

[0046] Where, Is the aggregated feature, Is the enhanced feature, Represents layer normalization, Represents the multi-head attention mechanism.

[0047] The role of local feature enhancement is to achieve feature propagation that retains terrain details.

[0048] In an exemplary embodiment, the flash flood disaster classification model is obtained by training with a training set, and the joint loss function used during training is expressed as: .

[0049] Where, L represents the joint loss value, Is the cross-entropy loss, Represents the edge weight matrix, And Are both weight coefficients, Represents the edge weight from the i-th watershed node to the j-th watershed node, Represents the edge weight from the j-th watershed node to the i-th watershed node. The latter two terms of the joint loss function respectively control graph sparsity and edge weight symmetry.

[0050] Regularization constraints are achieved through the joint loss function to improve the generalization ability in small-sample scenarios.

[0051] In an exemplary embodiment, according to the enhanced features, the classification probabilities of each watershed node belonging to each flash flood disaster classification are output, specifically including: using a multi-layer perceptron to output the classification probabilities of each watershed node belonging to each flash flood disaster classification according to the enhanced features.

[0052] The enhanced features are input into the multi-layer perceptron through a computer to obtain the classification probabilities of each watershed node belonging to each flash flood disaster classification, that is, the probability distribution of the flash flood disaster levels of each watershed node.

[0053] The prediction of the multi-layer perceptron is expressed as: .

[0054] Among them, represents the classification probability that each basin node belongs to each type of flash flood disaster classification, () represents the multi-layer perceptron, () represents the normalized exponential function.

[0055] The flash flood disaster classification model is obtained by training the improved network architecture of this application with the training set. The hardware deployment environment during training includes: (1) Computing nodes: equipped with a high-performance server cluster. (2) Storage devices: The geographical feature data of 53,000 basins are stored in a distributed file system. (3) Data interface: Geospatial raster data is processed through the open-source Geospatial Data Abstraction Library (GDAL). The geospatial raster data includes Digital Elevation Model (DEM) and land cover classification maps.

[0056] Training parameter settings: The number of attention heads h = 4, and the dimension of the first attention head = 64; The trainable parameter matrix , ∈ ; Edge retention ratio K: At the beginning of training, K = 20%, and it is gradually reduced to K = 5% through the cosine annealing strategy; Optimizer: Adam, initial learning rate 3e-4, weight decay 1e-6.

[0057] The improvement of the prediction model architecture in this application mainly includes the following 4 points.

[0058] (1) Co-design of the dynamic edge weight generation module and the multi-head attention mechanism.

[0059] (2) Decoupling and optimization of the graph structure learning and feature propagation stages.

[0060] (3) Trainability of the asymmetric edge weight matrix ( ).

[0061] (4) Coupling contradiction between local neighborhood aggregation and global dependence modeling in the message passing process.

[0062] Traditional graph neural network-based methods: Relying on predefined static graph structures (such as the topological connection of watershed hydrology), they are unable to dynamically capture the time-varying characteristics of the multi-factor interaction patterns during rainstorm processes. The message passing mechanism is limited to local neighborhood feature aggregation, making it difficult to model global dependencies across watersheds. The fixed edge weights result in a lack of adaptability in the feature propagation process. This application calculates the dynamic edge weight matrix of the graph structure through the multi-head attention mechanism, breaking through the limitations of the static graph structure, enhancing the visual interpretability of the edge weight matrix, and thus improving the prediction accuracy.

[0063] Pure attention mechanism architecture methods: The fully connected attention mechanism introduces redundant calculations and faces an O(N²) complexity bottleneck when dealing with a large number of nodes; it ignores the spatial semantic associations between nodes, resulting in insufficient extraction of meteorological-geographical coupling features and a lack of the ability to handle differential features of heterogeneous data modalities. O(N²) means that the execution time of the algorithm grows quadratically with the increase in the input data volume. And for multi-model fusion methods: The cascaded architecture causes feature alignment deviations, resulting in a relatively high fusion loss rate of hydrological time series features and geographical spatial features (often reaching about 20%); inconsistent parameter optimization objectives lead to difficulties in model convergence; insufficient interpretability restricts the ability to trace the causes of disasters. The dual-stage architecture of multi-head attention and graph convolution in this application: Alleviates the conflict between global computational efficiency and local feature preservation, and the inference speed is increased by 37 times at the node scale of N = 1000 (compared with the Transformer benchmark). Through edge weight filtering and regularization constraints, it suppresses the risk of noise propagation and overfitting. By modeling with asymmetric edge weights, it overcomes the geographical semantic distortion of traditional symmetric adjacency matrices.

[0064] Based on the same inventive concept, the embodiments of this application also provide a flash flood disaster classification device for implementing the above-mentioned flash flood disaster classification method. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more of the following flash flood disaster classification device embodiments can refer to the limitations on the flash flood disaster classification method in the above text and will not be repeated here.

[0065] In an exemplary embodiment, as Figure 3 shown, a flash flood disaster classification device is provided. The flash flood disaster classification device applies the above-mentioned flash flood disaster classification method. The flash flood disaster classification device includes: A feature matrix determination module, configured to obtain a geographical feature dataset of multiple watershed nodes in a target area and construct a feature matrix according to the geographical feature dataset.

[0066] The flash flood disaster classification probability prediction module is used to input the feature matrix into the flash flood disaster classification model and output the classification probabilities of each basin node belonging to each flash flood disaster classification; the flash flood disaster classification model is used to: calculate the edge weight matrix of the graph structure according to the feature matrix and the graph structure composed of each basin node in the target area through the multi-head attention mechanism; sparsify the edge weight matrix, and perform graph convolution operations based on the feature matrix and the sparsified edge weight matrix to obtain aggregated features; perform global feature enhancement on the aggregated features to obtain enhanced features; output the classification probabilities of each basin node belonging to each flash flood disaster classification according to the enhanced features.

[0067] The classification determination module is used to take the flash flood disaster classification corresponding to the maximum value among the classification probabilities of each basin node belonging to each flash flood disaster classification as the flash flood disaster classification of each basin node.

[0068] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as Figure 4 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through the system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store flash flood disaster classification data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements a flash flood disaster classification method.

[0069] Those skilled in the art can understand that Figure 4 the structure shown in

[0070] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program which, when executed by a processor, implements the steps in the above-described method embodiments.

[0071] In an exemplary embodiment, a computer program product is provided, including a computer program which, when executed by a processor, implements the steps in the above-described method embodiments.

[0072] It should be noted that the collection, use, and processing of the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application need to comply with relevant regulations.

[0073] Those of ordinary skill in the art can understand that all or part of the processes in the above-described method embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-described method embodiments. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memories can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0074] In each of the embodiments provided in the present application, the databases involved may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on a blockchain, etc., without limitation. In each of the embodiments provided in the present application, the processor may be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a data processing logic unit of a programmable logic device, etc., without limitation.

[0075] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0076] Specific examples are used in this article to elaborate on the principles and implementation manners of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A method for classifying mountain flood disasters, characterized in that, The classification of mountain flood disasters includes: Obtain the geographical feature dataset of multiple basin nodes in the target area, and construct a feature matrix according to the geographical feature dataset; Input the feature matrix into the mountain flood disaster classification model, and output the classification probabilities of each basin node belonging to each mountain flood disaster classification; the mountain flood disaster classification model is used to: calculate the edge weight matrix of the graph structure according to the feature matrix and the graph structure composed of each basin node in the target area through the multi-head attention mechanism; perform sparsification processing on the edge weight matrix, and perform graph convolution operation based on the feature matrix and the sparsified edge weight matrix to obtain aggregated features; perform global feature enhancement on the aggregated features to obtain enhanced features; output the classification probabilities of each basin node belonging to each mountain flood disaster classification according to the enhanced features; Take the mountain flood disaster classification corresponding to the maximum value among the classification probabilities of each basin node belonging to each mountain flood disaster classification as the mountain flood disaster classification of each basin node.

2. The flash flood disaster classification method according to claim 1, wherein, The feature matrix includes the elevation standard deviation, slope variation, surface coverage rate, basin area, terrain undulation degree, average slope, average elevation, shape coefficient, longest confluence path length, longest confluence path slope, longest confluence path slope 1085, stable infiltration rate, roughness coefficient, river section length, river section slope, unit hydrograph peak modulus, basin maximum confluence time, river power index, terrain humidity index, normalized vegetation index, disaster occurrence time, longitude and latitude of each basin node; The mountain flood disaster classification includes rainstorm type, debris flow type, breach type and snowmelt type.

3. The flash flood disaster classification method according to claim 1, wherein The formula for calculating the edge weight matrix of the graph structure through the multi-head attention mechanism is expressed as: ； Among them, represents the edge weight matrix, and the edge weight matrix is an asymmetric edge weight matrix. is the query matrix. is the key matrix. , , is the feature matrix. and are both trainable parameters. represents a dataset of dimension, where d represents the geographical feature dimension of each basin node. is the first attention head dimension, ⊙ represents the Hadamard product, and M is the adjacency mask matrix.

4. The flash flood disaster classification method according to claim 1, characterized in that, The formula for performing sparsification processing on the edge weight matrix is expressed as: ； Among them, is the edge weight matrix after sparsification processing, represents the edge weight matrix, ⊙ represents the Hadamard product, is the indicator function, is the learnable threshold parameter.

5. The flash flood disaster classification method according to claim 1, wherein The formula for performing graph convolution operation based on the feature matrix and the sparsified edge weight matrix is expressed as: ； Among them, is the aggregation feature, represents the rectified linear unit function, D is the degree matrix, is the edge weight matrix after sparsification processing, is the feature matrix, is the projection matrix, represents a dataset of dimension, d represents the geographical feature dimension of each basin node, is the second attention head dimension.

6. The flash flood disaster classification method according to claim 1, characterized in that Performing global feature enhancement on the aggregated features to obtain enhanced features specifically includes: Perform Transformer encoding on the aggregated features to obtain enhanced features; The formula for performing Transformer encoding on the aggregated features is expressed as: ； Among them, is the aggregation feature, is the enhanced feature, represents layer normalization, represents the multi-head attention mechanism.

7. The flash flood disaster classification method according to claim 1, wherein The mountain flood disaster classification model is obtained by training with a training set, and the joint loss function used during training is expressed as: ； where L represents the combined loss value, is the cross-entropy loss, represents the edge weight matrix, and are both weight coefficients, represents the edge weight from the i-th watershed node to the j-th watershed node, represents the edge weight from the j-th watershed node to the i-th watershed node.

8. The flash flood disaster classification method according to claim 1, characterized in that, Outputting the classification probabilities of each basin node belonging to each mountain flood disaster classification according to the enhanced features specifically includes: Adopt a multi-layer perceptron to output the classification probabilities of each basin node belonging to each mountain flood disaster classification according to the enhanced features.

9. A flash flood disaster classification device, characterized in that, The mountain flood disaster classification device applies the mountain flood disaster classification method described in any one of claims 1-8, and the mountain flood disaster classification device includes: A feature matrix determination module, configured to obtain the geographical feature dataset of multiple basin nodes in the target area, and construct a feature matrix according to the geographical feature dataset; The mountain flood disaster classification probability prediction module is used to input the feature matrix into the mountain flood disaster classification model and output the classification probabilities of each basin node belonging to each mountain flood disaster classification; the mountain flood disaster classification model is used for: calculating the edge weight matrix of the graph structure according to the feature matrix and the graph structure composed of each basin node in the target area through the multi-head attention mechanism; sparsifying the edge weight matrix, and performing graph convolution operations based on the feature matrix and the sparsified edge weight matrix to obtain aggregated features; performing global feature enhancement on the aggregated features to obtain enhanced features; and outputting the classification probabilities of each basin node belonging to each mountain flood disaster classification according to the enhanced features. The classification determination module is used to take the mountain flood disaster classification corresponding to the maximum value among the classification probabilities of each basin node belonging to each mountain flood disaster classification as the mountain flood disaster classification of each basin node.

10. A computer device, comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the mountain flood disaster classification method according to any one of claims 1-8.

Images