A method, device, and storage medium for landslide hazard identification based on three-dimensional geometric features of slope profiles and machine learning models.

Abstract

This invention provides a method, apparatus, and storage medium for landslide hazard identification based on the three-dimensional geometric features of slope plan and profile and a machine learning model. It relates to the field of landslide hazard identification technology and includes: obtaining planar and sectional geometric features of slope units based on the main sliding surface and main profile of slope units within a target working area, using historical landslide disasters or known landslide hazard points; standardizing the high-dimensional slope geometric feature data to obtain corresponding high-dimensional geometric feature vectors; inputting the sample set into a trained autoencoder model, outputting the corresponding reconstructed feature vectors, and determining the discrimination threshold; inputting the high-dimensional geometric feature vectors corresponding to all target slope units to be identified within the target working area into the autoencoder model, calculating the corresponding reconstruction error through the autoencoder model, and determining potential landslide hazards; and eliminating isolated misjudged units among potential landslide hazards. Using the identification method provided by this invention, landslide hazards can be identified quickly and accurately.

Description

Technical Field

[0001] This invention relates to the field of landslide hazard identification technology, and in particular to a method, device and storage medium for landslide hazard identification based on the three-dimensional geometric features of slope profile and machine learning model. Background Technology

[0002] Currently, there are various methods for identifying landslide hazards over a wide area, which can be broadly classified into two categories: qualitative identification and intelligent identification.

[0003] Qualitative identification methods: Traditional human-computer interaction-based interpretation methods involve technicians comparing multiple periods of optical remote sensing imagery and InSAR (Synthetic Aperture Radar) surface deformation data. Based on the characteristics of past disaster-affected areas, they determine the locations of potential landslide hazards (landslide risks) or delineate areas in optical images that reflect surface deformation characteristics as landslide hazard zones. This method has a relatively mature technical process and does not require a large amount of landslide sample data. However, it is highly dependent on the technical experience of the interpreters and cannot accurately identify the locations of potential landslide hazards (potential landslide disaster sites) over a large area in a short time with high precision, efficiency, and accuracy, thus limiting its scalability.

[0004] Intelligent identification methods: Intelligent hazard identification techniques based on comprehensive remote sensing data (optical remote sensing, synthetic aperture radar, lidar, and multi-source data such as meteorological, basic geological, and land cover data) are constantly developing and have achieved certain results in practice. Based on a large number of landslide hazard samples with rich annotation features (optical remote sensing images or multimodal data), different types of machine learning (logistic regression, random forest, etc.) and deep learning (convolutional neural networks, recurrent neural networks, Transformer models, etc.) are used to train the intelligent model, achieving intelligent identification of landslide hazards. This method typically requires a large number of hazard identification samples to train the intelligent model. Obtaining a sufficient number of standardized sample data for training the intelligent model is usually a challenge and constraint for the widespread application of this method. At the same time, due to various reasons such as inaccurate hazard data recording, limited data volume, and certain errors in multimodal data acquisition, this identification method is still in the exploratory stage and is difficult to effectively play a role in practical work.

[0005] Specifically, existing research findings and technical methods mainly face the following two challenges and shortcomings: (1) Qualitative methods are inefficient, costly, and difficult to replicate and promote. Currently, the main methods used are based on manual visual remote sensing interpretation and field surveys to identify the spatial location and boundary distribution of landslide hazards. This method is highly subjective, requires a large amount of manpower, involves a large workload in field surveys, and takes a long time. It is difficult to promote and apply it widely, and it is difficult to identify it in a wide area. It is also difficult to quickly implement and promote the identification technology results in different regions, thus limiting the efficiency of technology use. (2) Quantitative / intelligent methods require large amounts of data, have high requirements, are difficult to develop models, are still immature, and are difficult to replicate and promote. In recent years, the application of artificial intelligence technology in landslide hazard identification research has gradually increased and made significant progress. However, this type of method has high model training costs, requires a large number of basic data, especially landslide hazard samples, which are difficult to obtain, have insufficient data quality and coverage, high cost of high-resolution remote sensing data, lack of data in remote areas, and InSAR is easily affected by vegetation and atmospheric interference, limiting the accuracy of deformation monitoring and causing large errors. Imbalanced samples and labeling difficulties, few landslide samples, and a high proportion of negative samples (non-landslide areas) lead to model overfitting; historical landslide data labeling relies on expert experience, and automated labeling tools are not yet mature. At the same time, due to the significant differences in disaster characteristics in different regions, the replication and promotion of technical methods and successful experiences are difficult.

[0006] From the perspective of the controlling characteristics of landslide occurrence, the topographic features of the slope are the most important controlling factor in whether landslide disasters are likely to occur. Moreover, topographic data is more difficult to obtain, more accurate, and more efficient, making it more conducive to replication and promotion.

[0007] Therefore, there is a need for a landslide hazard identification method, device, and storage medium based on the three-dimensional geometric features of slope profiles and machine learning models, in order to at least partially solve the above-mentioned technical problems. Summary of the Invention

[0008] In view of this, embodiments of the present invention provide a method, device and storage medium for identifying landslide hazards based on the three-dimensional geometric features of slope profiles and machine learning models, so as to at least solve one of the problems in the prior art.

[0009] In a first aspect, embodiments of the present invention provide a method for identifying landslide hazards based on three-dimensional geometric features of slope profiles and machine learning models, the identification method comprising: Based on the main sliding surface and main profile of the slope unit with the same or similar geological environmental conditions in the target work area, obtain the planar geometric features and cross-sectional geometric features of the slope unit; The high-dimensional slope geometric feature data, which consists of planar geometric features and cross-sectional geometric features, is standardized and preprocessed to obtain the corresponding high-dimensional geometric feature vectors. A sample set is then constructed based on all the high-dimensional geometric feature vectors. The sample set is input into the trained autoencoder model, the autoencoder model outputs the corresponding reconstructed feature vector and calculates the reconstruction error, and the discrimination threshold is determined based on the reconstruction error distribution of all reconstructed feature vectors. The high-dimensional geometric feature vectors corresponding to all target slope units to be identified extracted in the target working area are input into the corresponding autoencoder model. The corresponding reconstruction error is calculated by the autoencoder model. Target slope units whose corresponding reconstruction error does not exceed the discrimination threshold are judged as potential landslide hazards. The density clustering algorithm is used to remove isolated misclassified units from potential landslide hazards, thus obtaining the landslide hazard identification results.

[0010] Secondly, embodiments of the present invention also provide a landslide hazard identification device based on the three-dimensional geometric features of a slope profile and a machine learning model, the identification device comprising: Memory is used to store executable instructions for a computer; A processor, used to implement the identification method of the above-mentioned technical solution when executing computer-executable instructions stored in the memory.

[0011] Thirdly, embodiments of the present invention also provide a storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute the identification method of the above-described technical solution.

[0012] According to the identification method of this invention, the three-dimensional geometric features (planar morphology: the vertical projection features of the slope unit in the plane; lateral morphology: the directional projection features of the slope unit in the profile) of the slope unit in the target work area are utilized and mined for quantification. A series of quantitative discrimination models (i.e., trained autoencoder models) based on the planar and profile geometric features of the slope unit are established. Then, based on the model, all slope units in the work area are quantitatively identified, thereby completing the rapid identification and extraction of landslide hazards in the slope units in the area, improving the efficiency and accuracy of landslide hazard investigation at the regional scale.

[0013] The identification method of this invention significantly reduces the requirements for relevant basic data, lowers the technical threshold, and improves the feasibility of technology application compared to current common research and development approaches and technical routes. The technology proposed in this invention reduces the high demands on sample data volume, the complexity and completeness of sample feature annotations, and the precision required in previous intelligent hazard identification models, as well as the complexity of model training and construction. This allows for the accurate identification of landslide hazard locations (potential landslide disaster sites) over a wide area with lower requirements for basic data, sample data input, and model construction difficulty. This provides more precise and accurate target area and point information for further on-site verification and assessment.

[0014] Additional advantages, objects, and features of the invention will be set forth in part in the description which follows, and will also become apparent in part to those skilled in the art upon studying the description, or may be learned by practice of the invention. The objects and other advantages of the invention can be realized and obtained by means of the structures specifically pointed out in the description and drawings.

[0015] Those skilled in the art will understand that the objectives and advantages achievable with the present invention are not limited to those specifically described above, and that the above and other objectives achievable with the present invention will become clearer from the following detailed description. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, are not intended to limit the scope of the invention. The components in the drawings are not drawn to scale but are merely illustrative of the principles of the invention. For ease of illustration and description of certain parts of the invention, corresponding portions in the drawings may be enlarged, i.e., may appear larger relative to other components in an exemplary device actually manufactured according to the invention. In the drawings: Figure 1 This is a flowchart of an identification method according to an embodiment of the present invention; Figure 2 This is a schematic main cross-section of a historical landslide disaster or known landslide hazard point identification method according to an embodiment of the present invention. Figure 3 This is a schematic diagram of the main cross-section of historical landslide disasters or known landslide hazard points used to calculate the geometric features of a slope unit profile in an identification method according to an embodiment of the present invention. Figure 4 This is a schematic diagram of the main cross-section of another historical landslide disaster or known landslide hazard point used to calculate the geometric features of the slope unit profile in the identification method according to an embodiment of the invention. Figure 5This is a schematic diagram of the target slope unit boundary to be identified and its main sliding surface / main profile extraction results in the identification method according to an embodiment of the present invention; Figure 6 This is a schematic diagram of an identification device according to an embodiment of the present invention; Figure 7 This is a schematic diagram of an identification system according to an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.

[0018] It should also be noted that, in order to avoid obscuring the invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the invention are shown in the accompanying drawings, while other details that are not closely related to the invention are omitted.

[0019] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0020] It should also be noted that, unless otherwise specified, the term "connection" in this article can refer not only to a direct connection, but also to an indirect connection involving an intermediary.

[0021] In the following description, embodiments of the invention will be illustrated with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar parts, or the same or similar steps.

[0022] First, refer to Figure 1 This application describes a landslide hazard identification method 100 based on three-dimensional geometric features of slope profiles and a machine learning model, according to embodiments of this application. Figure 1 As shown, the identification method 100 may include steps S110 to S150, as detailed below: In step S110, the planar geometric features and profile geometric features of the slope unit are obtained based on the main sliding surface and main profile of the slope unit with the same or similar geological environment conditions of historical landslide disasters or known landslide hazard points in the target working area.

[0023] In step S120, the high-dimensional slope geometric feature data composed of planar geometric features and cross-sectional geometric features are standardized to obtain the corresponding high-dimensional geometric feature vectors, and a sample set is constructed based on all high-dimensional geometric feature vectors.

[0024] In step S130, the sample set is input into the trained autoencoder model, the autoencoder model outputs the corresponding reconstructed feature vector and calculates the reconstruction error, and the discrimination threshold is determined based on the reconstruction error distribution of all reconstructed feature vectors.

[0025] In step S140, the high-dimensional geometric feature vectors corresponding to all target slope units to be identified extracted in the target working area are input into the corresponding autoencoder model. The corresponding reconstruction error is calculated by the autoencoder model. Target slope units whose corresponding reconstruction error does not exceed the discrimination threshold are judged as potential landslide hazards.

[0026] In step S150, isolated misjudged units in potential landslide hazards are removed based on density clustering algorithm to obtain landslide hazard identification results.

[0027] In the embodiments of this application, firstly, planar geometric features and cross-sectional geometric features of slope units are obtained based on the main sliding surface and main profile of slope units with historical landslide disasters or known landslide hazard points having the same or similar geological environmental conditions within the target working area; the high-dimensional slope geometric feature data composed of planar geometric features and cross-sectional geometric features are standardized and preprocessed to obtain corresponding high-dimensional geometric feature vectors, and a sample set is constructed based on all high-dimensional geometric feature vectors; then, the sample set is input into a trained autoencoder model, which outputs the corresponding reconstructed feature vector and calculates the reconstruction error, and a discrimination threshold is determined based on the reconstruction error distribution of all reconstructed feature vectors; then, the high-dimensional geometric feature vectors corresponding to all target slope units to be identified extracted within the target working area are input into the corresponding autoencoder model, and the corresponding reconstruction error is calculated by the autoencoder model; target slope units whose corresponding reconstruction error does not exceed the discrimination threshold are judged as potential landslide hazards; finally, isolated misjudged units in potential landslide hazards are removed based on a density clustering algorithm to obtain the landslide hazard identification result.

[0028] As described above, the identification method 100 of this application extracts, quantifies, statistically analyzes, and fits model construction and feature analysis of the plan and profile geometric morphology of landslide disasters that have occurred in the work area (or areas with similar disaster-prone conditions) before or after investigation of landslide hazard points. It then constructs a regional landslide hazard discrimination model system based on the three-dimensional geometric morphological features of the landslide hazard plan and profile (vertical projection features of slope unit planes and directional projection features of slope unit profiles). Based on this, by extracting slope units and their main sliding profiles within the work area and comparing them with the constructed series of hazard discrimination models, it achieves rapid identification and extraction of slope units that conform to the geometric morphological feature model, improving the efficiency and accuracy of landslide hazard investigation at the regional scale, and providing more accurate target area and target point basis for further on-site verification and assessment.

[0029] The following will combine Figure 1 The specific description covers the above-described steps of the identification method 100 according to embodiments of this application.

[0030] In the embodiments of this application, in step S110, the planar geometric features and cross-sectional geometric features of the slope unit are obtained based on the main sliding surface and main profile of the slope unit with the same or similar geological environmental conditions of historical landslide disasters or known landslide hazard points in the target working area.

[0031] Specifically, before proceeding to step S110, it is necessary to obtain data on historical landslide disasters or known landslide hazard points within the target work area that have the same or similar geological environmental conditions as the target slope unit to be identified. For example, this can be obtained by consulting relevant historical data and combining it with field exploration. Then, the main sliding surface and main profile of the slope unit of the historical landslide disaster or known landslide hazard point are extracted.

[0032] It should be noted that the target working area in this embodiment refers to a working area of ​​a certain size, such as the area of ​​a township or county. Historical landslide disasters or known landslide hazard points with the same or similar geological environmental conditions refer to those with the same or similar geological environmental conditions as the target slope unit to be identified. Therefore, depending on the different geological environmental conditions of the target slope unit to be identified, the historical landslide disasters or known landslide hazard points obtained within the target working area will also be different, and the trained autoencoder models appearing in steps S130 and S140 below will also be different accordingly.

[0033] Regarding the methods for extracting the main sliding surface and main profile of slope units with historical landslide disasters or known landslide hazard points having the same or similar geological environmental conditions within the target work area, existing technologies can be used, and will not be detailed here. A schematic diagram of the main profile of the slope unit can be found here. Figure 2 .

[0034] Based on the main sliding surface and main profile of slope units from historical landslide disasters or known landslide hazard points, the planar geometric features and cross-sectional geometric features of the slope units are obtained. The planar geometric features include the length, width, and aspect ratio of the slope unit. The cross-sectional geometric features include the height and aspect ratio of the slope unit; the angle and ratio between the line segments connecting the horizontal midpoint of the slope curve corresponding to the transverse coordinate center point to the start and end points of the slope curve and the horizontal plane; the average curvature of the two curve segments divided by the horizontal midpoint of the slope curve corresponding to the transverse coordinate center point; the angle and ratio between the line segments connecting the vertical midpoint of the slope curve corresponding to the longitudinal coordinate center point to the start and end points of the slope curve and the horizontal plane; the average curvature of the two curve segments divided by the vertical midpoint of the slope curve corresponding to the longitudinal coordinate center point; and the area ratio of the area below the cross-sectional curve to the area below the line segments connecting the start and end points of the cross-sectional curve (see Table 1).

[0035] Table 1 The planar and cross-sectional geometric features of the slope element are obtained as follows: Basic data needs to be prepared in advance, along with auxiliary data.

[0036] Basic data: High-precision DEM (≤5m resolution), historical landslide boundary vector data, and geological maps. If a high-precision DEM is unavailable, 3D point cloud data can be generated using UAV oblique photogrammetry, and slope units can be extracted through point cloud segmentation.

[0037] Supporting data: remote sensing images (to verify surface deformation), rainfall / earthquake-triggered event records.

[0038] Specific features extracted include: 1. Planar geometric characteristics of slope elements The length of the main sliding surface along the main profile direction is taken as the length of the slope element, and the width is taken as the perpendicular direction. Thus, the length, width, and aspect ratio of the slope element are obtained.

[0039] 2. Geometric features of slope unit profile First, interpolate along the main profile using DEM precision to obtain the elevation curve of the main profile. Based on this, obtain the height and length-to-height ratio of the slope element.

[0040] See Figure 3 The process involves obtaining the point corresponding to the center point in the X direction (horizontal) (referred to as HCenter), obtaining two line segments: one from the starting point to HCenter, and the other from HCenter to the ending point. The angles between these two line segments and the horizontal plane are then calculated, along with the ratio of these angles. HCenter divides the profile curve into two segments, and the average curvature of each segment is calculated.

[0041] See Figure 4 Obtain the point corresponding to the center point in the Y direction (longitudinal direction) (called VCenter). Similarly, obtain the two line segments from the starting point to VCenter and from VCenter to the ending point, and obtain the angles between the two line segments and the horizontal plane, as well as the ratio of the angles. VCenter divides the profile curve into two segments, and calculates the average curvature of each segment. Obtain the area under the profile curve, and the area under the line segment directly connecting the beginning and end points of the profile curve. Calculate the ratio of the two areas.

[0042] In the embodiments of this application, step S120 involves standardizing and preprocessing the high-dimensional slope geometric feature data composed of planar geometric features and cross-sectional geometric features to obtain the corresponding high-dimensional geometric feature vectors, and constructing a sample set based on all high-dimensional geometric feature vectors.

[0043] Specifically, 16-dimensional high-dimensional slope geometric feature data were obtained through step S110. This 16-dimensional feature data was then standardized to eliminate the influence of dimensions, resulting in corresponding high-dimensional geometric feature vectors. All these high-dimensional geometric feature vectors were then used to construct a sample set.

[0044] Among them, data standardization to eliminate the influence of dimensions is a well-known existing data preprocessing method, such as range standardization, which will not be described in detail here.

[0045] In the embodiments of this application, in step S130, the sample set is input into the trained autoencoder model, the autoencoder model outputs the corresponding reconstructed feature vector and calculates the reconstruction error, and the discrimination threshold is determined based on the reconstruction error distribution of all reconstructed feature vectors.

[0046] Specifically, before proceeding to step S130, a trained autoencoder model needs to be constructed. Constructing the trained autoencoder model includes the following steps: The training set is a sample set based on all high-dimensional geometric feature vectors, which is then input into the initial autoencoder model. The initial autoencoder model employs a deep autoencoder network structure with fully connected layers. An autoencoder is an unsupervised neural network model that achieves feature extraction and dimensionality reduction by learning the low-dimensional latent space (low-dimensional latent representation) of the input data. Its core structure consists of an encoder and a decoder. The encoder progressively reduces the dimensionality to the low-dimensional latent space, while the decoder symmetrically restores the original dimensionality.

[0047] The encoder and decoder of the initial autoencoder model transform each high-dimensional geometric feature vector in the training set into a reconstructed feature vector of the same dimension.

[0048] The initial autoencoder model is trained in an unsupervised manner, and the reconstruction error can be minimized using the Adam optimizer. Dropout layers and L2 regularization can be added during training to prevent overfitting.

[0049] Model training can be terminated using early stopping and the maximum number of training epochs. Specifically, during training, the model monitors the reconstruction error on an independent validation set. If the validation set error stops decreasing or even starts to increase over several consecutive epochs, training is stopped to prevent overfitting.

[0050] That is, training is stopped when the reconstruction error stops decreasing or starts to increase over several consecutive epochs to prevent overfitting. Alternatively, to prevent excessively long training times, the model is trained until the preset maximum number of training epochs is reached (e.g., Epochs=100), at which point the autoencoder model is considered well-trained. The ultimate goal is to find the weights that best perform on the validation set to ensure generalization ability.

[0051] The reconstruction error can be expressed as mean squared error (MSE). The reconstruction error (e.g., mean squared error) is calculated by comparing the difference between the input and output. The larger the error, the more likely the sample is to be an anomaly (potential landslide risk).

[0052] Further, in step S130, the autoencoder model outputs the corresponding reconstructed feature vector, and a discrimination threshold is determined based on the reconstruction error distribution of all reconstructed feature vectors, specifically including: First, the encoder of the trained autoencoder model transforms the high-dimensional geometric feature vectors of the sample set into a low-dimensional latent representation. The decoder of the autoencoder model then transforms the low-dimensional latent representation into a reconstructed feature vector with the same dimension as the high-dimensional geometric feature vectors of the sample set. Specifically, as follows... in, It is a high-dimensional geometric eigenvector. It is a low-dimensional latent representation. It is a weight matrix. It is the activation function ReLU. It is a bias term.

[0053] in, It is to reconstruct the feature vector. It is a weight matrix. It is a bias term.

[0054] Then, the reconstruction error between each high-dimensional geometric feature vector and its corresponding reconstructed feature vector in the sample set is calculated. Based on the calculation results, the reconstruction error distribution of all reconstructed feature vectors is determined.

[0055] The reconstruction error can be represented by mean squared error (MSE). The method for calculating MSE is existing technology and will not be elaborated upon here.

[0056] The discrimination threshold is determined by setting the reconstruction error corresponding to a set quantile based on the reconstruction error distribution of all reconstructed feature vectors.

[0057] Based on the calculated reconstruction error distribution (e.g., statistical quantiles of the error) of all training samples (sample set), a discrimination threshold (e.g., 30th percentile) is determined to distinguish between "normal" and "abnormal" data. This threshold reflects the distribution characteristics of the reconstruction error in the training data and is used for subsequent anomaly judgment of new samples. The training samples represent the geometric feature distribution of "disaster / hazardous slopes," and their reconstruction errors are relatively small. Non-landslide hazards are considered "abnormal" samples, and their reconstruction errors will deviate significantly from the normal distribution.

[0058] In landslide hazard identification tasks, autoencoders automatically learn the essential features of slope geometry by minimizing the mean square error (MSE) between the input data (16-dimensional geometric features) and the reconstructed data. The advantage of this method is that it can capture the nonlinear structure of the data without requiring labeled data, making it particularly suitable for processing high-dimensional geological feature data.

[0059] In addition, the results of field investigations (such as false positives / false negatives) can be fed back into the model to optimize feature weights and discrimination thresholds through active learning.

[0060] In the embodiments of this application, in step S140, the high-dimensional geometric feature vectors corresponding to all target slope units to be identified extracted in the target working area are input to the corresponding autoencoder model, and the corresponding reconstruction error is calculated by the autoencoder model. Target slope units whose corresponding reconstruction error does not exceed the discrimination threshold are judged as potential landslide hazards.

[0061] Specifically, before proceeding to step S140, it is necessary to pre-obtain the high-dimensional geometric feature vectors corresponding to all target slope units to be identified within the target working area. This includes the following steps: Step S141: Extract the target slope units to be identified within the target working area in advance.

[0062] Step S142: Extract the main slip surface and main profile of the target slope element to be identified. (See below) Figure 5 The extraction method can also utilize existing technologies.

[0063] Step S143: Based on the extracted main sliding surface and main profile of the target slope unit to be identified, obtain the planar geometric features and cross-sectional geometric features of the target slope unit to be identified. For details, please refer to the relevant content of step S110, which will not be repeated here.

[0064] Step S141 may include: Hydrological analysis methods (such as the D8 algorithm) can be used to initially delineate slope units. Ensure that each unit represents an independent topographic catchment and stress distribution zone.

[0065] By using curvature thresholding (e.g., through curvature maps) to identify slope inflection lines, hydrological analysis can be optimized, and slope units that better reflect actual terrain boundaries can be segmented, thus refining the unit boundaries.

[0066] Introducing terrain roughness indices (such as TRI) to verify unit boundaries, assessing the rationality of segmentation results, ensuring terrain uniformity within units, and identifying target slope units to be identified within the target working area.

[0067] Then, the high-dimensional slope geometric feature data of the target slope unit to be identified, which consists of planar geometric features and cross-sectional geometric features, is also standardized to obtain the corresponding high-dimensional geometric feature vector.

[0068] As mentioned earlier, depending on the geological environment conditions of the target slope unit to be identified, the trained autoencoder model used to calculate the corresponding reconstruction error will also be different. The high-dimensional geometric feature vectors corresponding to all the target slope units to be identified extracted within the target working area are input into the corresponding autoencoder model (or a series of autoencoder models). The corresponding reconstruction error is calculated by the autoencoder model. Target slope units whose corresponding reconstruction error does not exceed the discrimination threshold are judged as potential landslide hazards and can be marked as high-risk units.

[0069] In the embodiments of this application, in step S150, isolated misjudged units in potential landslide hazards are eliminated based on density clustering algorithm to obtain landslide hazard identification results.

[0070] Specifically, the density clustering algorithm can use the DBSCAN algorithm.

[0071] First, the spatial coordinates of potential landslide hazards (such as latitude and longitude or grid number) can be used as clustering input.

[0072] Taking any potential landslide hazard as the starting point, if the number of potential landslide hazard corresponding to the starting point within the set neighborhood radius (e.g., neighborhood radius of 50-100 meters) is greater than the set minimum number of points (which can be set to 3-5), then the potential landslide hazard is the core point.

[0073] Then, a new cluster is created for each core point, and the potential landslide hazards within the core point and its defined neighborhood radius are added to the current new cluster.

[0074] These newly added potential landslide hazards are processed recursively until no new core points or boundary points are added to the current cluster. Specifically, for each core point, a potential landslide hazard within a defined neighborhood radius is considered a boundary point if it is not a core point but has already been added to a cluster.

[0075] Next, potential landslide hazards that are neither core points nor boundary points are considered isolated misjudged units and marked as noise points. Noise points may be due to model misjudgments or false positives caused by irrelevant terrain fluctuations. These noise points are then directly deleted.

[0076] Finally, the remaining potential landslide hazards after eliminating noise points are the identified landslide hazards.

[0077] Furthermore, the identified landslide hazards can be further verified by combining geological data, ultimately generating a landslide hazard distribution map.

[0078] Based on the above description, the identification method according to the embodiments of this application can achieve accurate identification of landslide hazard locations (potential landslide disaster sites) over a wide area with high precision, high efficiency, and high accuracy, requiring less data resources (reducing data costs and solving the problem of small sample data), utilizing more lightweight model algorithms (reducing training and model calculation costs), and more efficient calculation processes (improving calculation efficiency). This establishes a rapid identification technology for wide-area landslide hazard locations based on the three-dimensional geometric features of slopes.

[0079] The main innovative aspects of this application include: Innovation Point 1: Providing a complete technical approach and implementation path. By comparing the 3D geometric features, indicator system, and standards of landslide hazard profiles with wide-area measured topographic data, rapid identification and extraction of potential landslide hazards across the entire region is achieved. This reduces data costs and model training computation costs, improves computational efficiency, and enables precise identification of landslide hazard locations over a large area with high accuracy, efficiency, and precision. This application, through quantitative geometric modeling and regional adaptive framework design, provides a new paradigm for landslide hazard identification that combines efficiency and accuracy, further promoting the transformation of geological disaster investigation and hazard identification from "experience-driven" to "data-mechanism synergistic driving" approaches.

[0080] Innovation Point Two: Providing a set of algorithms for slope element extraction and 3D feature quantization. The quantization algorithms for the planar vertical projection features and the directional projection features of slope element profiles enable rapid extraction and geometric feature quantization of slope elements and their main profile lines based on wide-area high-precision terrain (DEM) data.

[0081] Innovation Point 3: Providing a set of identification rules and models. By automatically extracting and quantifying the plan and profile topographic features of historical landslide disasters, a set of three-dimensional geometric feature (length / width; length / height; width / height, etc.) indicator system and identification standards for regional landslide hazards have been established.

[0082] This application mainly utilizes readily available high-precision terrain data, effectively reducing the current identification model's need for multi-source basic data, improving the applicability of this application, enhancing the efficiency and accuracy of landslide hazard investigation at the regional scale, and providing more accurate target area and target point basis for further field verification and assessment.

[0083] refer to Figure 6 An identification device 200 for implementing the identification method according to an embodiment of this application includes a processor 210 and a memory 220. The identification device 200 may include one or more processors 210 and one or more memories 220. The memory 220 stores an executable program executed by the processor 210, which, when executed by the processor 210, causes the processor 210 to execute the identification method 100 described above according to an embodiment of this application.

[0084] The processor 210 may be a central processing unit (CPU) or other processing units with data processing capabilities and / or instruction execution capabilities.

[0085] The memory 220 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 210 may execute the program instructions to implement the client functions (implemented by the processor) in the embodiments of this application described herein, and / or other desired functions. Various applications and various data may also be stored in the computer-readable storage medium, such as various data used and / or generated by the applications.

[0086] The identification device 200 may also include input and output devices, which are interconnected via a bus system and / or other forms of connection mechanisms. It should be noted that... Figure 6 The components and structure of the identification device 200 shown are merely exemplary and not limiting; the identification device 200 may also have other components and structures as needed.

[0087] The input device can be a device used by a user to input commands, and can include one or more of a keyboard, mouse, microphone, and touchscreen. Furthermore, the input device can also be any interface for receiving information.

[0088] The output device can output various information (e.g., images or sounds) to the outside (e.g., a user), and may include one or more of a display, speaker, etc. Furthermore, the output device can also be any other device with output functionality.

[0089] For example, the example identification device 200 for implementing the identification method 100 according to the embodiments of this application can be applied to terminal devices (such as mobile phones), tablet computers, laptop computers, ultra-mobile personal computers (UMPCs), handheld computers, netbooks, personal digital assistants (PDAs), wearable devices (such as smartwatches, smart glasses, or smart helmets), augmented reality (AR), virtual reality (VR) devices, smart home devices, in-vehicle computers, and other electronic devices. The embodiments of this application do not impose any limitations on this.

[0090] Those skilled in the art can understand the specific operation of the identification device 200 for implementing the identification method 100 according to the embodiments of this application in conjunction with the content described above. For the sake of brevity, the specific details will not be repeated here, but only some main operations of the processor 210 will be described.

[0091] In one embodiment of this application, when the executable program is run by the processor 210, the processor 210 performs the following steps: Based on the main sliding surface and main profile of slope units with similar or identical geological conditions in the target working area, planar and profile geometric features of the slope units are obtained. The high-dimensional slope geometric feature data, composed of planar and profile geometric features, are standardized to obtain corresponding high-dimensional geometric feature vectors. A sample set is constructed based on all high-dimensional geometric feature vectors. The sample set is input into a trained autoencoder model, which outputs corresponding reconstructed feature vectors and calculates the reconstruction error. A discrimination threshold is determined based on the distribution of reconstruction errors across all reconstructed feature vectors. The high-dimensional geometric feature vectors corresponding to all target slope units extracted from the target working area are input into the corresponding autoencoder model. The autoencoder model calculates the corresponding reconstruction error, and target slope units whose reconstruction errors do not exceed the discrimination threshold are identified as potential landslide hazards. Isolated misclassified units among potential landslide hazards are removed using a density clustering algorithm to obtain the landslide hazard identification results.

[0092] The above exemplarily illustrates an identification method 100 according to an embodiment of this application. The following, in conjunction with... Figure 7 The following describes an identification system 300 provided in another aspect of an embodiment of this application.

[0093] Reference Figure 7 This document describes an example recognition system 300 for implementing the recognition method of the embodiments of this application. The recognition system 300 may include a geometric feature acquisition module 310, a feature vector acquisition module 320, a discrimination threshold determination module 330, a judgment module 340, and a recognition result module 350. Wherein: The geometric feature acquisition module 310 is used to: acquire the planar geometric features and cross-sectional geometric features of the slope unit based on the main sliding surface and main profile of the slope unit with the same or similar geological environmental conditions of historical landslide disasters or known landslide hazard points in the target working area.

[0094] The feature vector acquisition module 320 is used to: standardize the high-dimensional slope geometric feature data composed of planar geometric features and cross-sectional geometric features to obtain the corresponding high-dimensional geometric feature vectors, and construct a sample set based on all high-dimensional geometric feature vectors.

[0095] The discrimination threshold determination module 330 is used to: input the sample set into the trained autoencoder model, the autoencoder model outputs the corresponding reconstructed feature vector and calculates the reconstruction error, and determine the discrimination threshold based on the reconstruction error distribution of all reconstructed feature vectors.

[0096] The determination module 340 is used to: input the high-dimensional geometric feature vectors corresponding to all target slope units to be identified extracted in the target working area into the corresponding autoencoder model, calculate the corresponding reconstruction error through the autoencoder model, and determine the target slope unit whose corresponding reconstruction error does not exceed the discrimination threshold as a potential landslide hazard.

[0097] The identification result module 350 is used to: eliminate isolated misjudged units in potential landslide hazards based on density clustering algorithm, and obtain landslide hazard identification results.

[0098] The identification system 300 proposed in this embodiment of the invention can quickly and accurately identify potential landslide hazards.

[0099] Furthermore, according to embodiments of this application, this application also provides a storage medium on which a computer program is stored. When the computer program is run by a processor, it is used to execute corresponding steps of the identification method 100 of this application. The storage medium may, for example, include a memory card of a smartphone, a storage component of a tablet computer, a hard disk of a personal computer, a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a portable compact disc read-only memory (CD-ROM), a USB memory, or any combination of the above storage media. The computer-readable storage medium may be any combination of one or more computer-readable storage media.

[0100] Furthermore, according to embodiments of this application, this application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the determination method of embodiments of this application.

[0101] Although exemplary embodiments have been described herein with reference to the accompanying drawings, it should be understood that the above exemplary embodiments are merely illustrative and are not intended to limit the scope of this application. Various changes and modifications can be made therein by those skilled in the art without departing from the scope and spirit of this application. All such changes and modifications are intended to be included within the scope of this application as claimed in the appended claims.

[0102] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0103] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed.

[0104] Furthermore, those skilled in the art will understand that although some embodiments described herein include certain features but not others included in other embodiments, combinations of features from different embodiments are intended to be within the scope of this application and form different embodiments. For example, in the claims, any one of the claimed embodiments can be used in any combination.

[0105] It should be noted that the above embodiments are illustrative of this application and not restrictive, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. This application can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.

[0106] The above description is merely a specific embodiment or illustration of the embodiments of this application. The scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. The scope of protection of this application shall be determined by the scope of the claims.

Claims

1. A method for identifying landslide hazards based on the three-dimensional geometric features of slope profiles and a machine learning model, characterized in that, The identification method includes: Based on the main sliding surface and main profile of the slope unit with the same or similar geological environmental conditions in the target work area, obtain the planar geometric features and cross-sectional geometric features of the slope unit; The high-dimensional slope geometric feature data, which consists of planar geometric features and cross-sectional geometric features, is standardized and preprocessed to obtain the corresponding high-dimensional geometric feature vectors. A sample set is then constructed based on all the high-dimensional geometric feature vectors. The sample set is input into the trained autoencoder model, the autoencoder model outputs the corresponding reconstructed feature vector and calculates the reconstruction error, and the discrimination threshold is determined based on the reconstruction error distribution of all reconstructed feature vectors. The high-dimensional geometric feature vectors corresponding to all target slope units to be identified extracted in the target working area are input into the corresponding autoencoder model. The corresponding reconstruction error is calculated by the autoencoder model. Target slope units whose corresponding reconstruction error does not exceed the discrimination threshold are judged as potential landslide hazards. The density clustering algorithm is used to remove isolated misclassified units from potential landslide hazards, and the landslide hazard identification results are obtained. The planar geometric features include the length, width, and length-to-width ratio of the slope element; the profile geometric features include the height and length-to-height ratio of the slope element; the angle and angle ratio between the line segments connecting the horizontal midpoint of the slope curve corresponding to the transverse coordinate center point of the slope element to the start and end points of the slope curve and the horizontal plane; the average curvature of the two curve segments into which the slope curve is divided by the horizontal midpoint of the slope curve corresponding to the transverse coordinate center point of the slope element; the angle and angle ratio between the line segments connecting the vertical midpoint of the slope curve corresponding to the longitudinal coordinate center point of the slope element to the start and end points of the slope curve and the horizontal plane; the average curvature of the two curve segments into which the slope curve is divided by the vertical midpoint of the slope curve corresponding to the longitudinal coordinate center point of the slope element; and the area ratio of the area below the profile curve to the area below the line segments connecting the start and end points of the profile curve.

2. The identification method according to claim 1, characterized in that, The recognition method further includes pre-training an initial autoencoder model to obtain a trained autoencoder model, specifically including: The sample set based on all high-dimensional geometric feature vectors is used as the training set, and the training set is input into the initial autoencoder model; The encoder and decoder of the initial autoencoder model transform each high-dimensional geometric feature vector in the training set into a reconstructed feature vector of the same dimension; The initial autoencoder model was trained in an unsupervised manner, and the reconstruction error was minimized using the Adam optimizer. When the reconstruction error no longer decreases or begins to increase over multiple consecutive rounds, or when the training model reaches the preset maximum number of training rounds, the autoencoder model is determined to be a well-trained autoencoder model.

3. The identification method according to claim 1, characterized in that, The autoencoder model outputs the corresponding reconstructed feature vector. A discrimination threshold is determined based on the reconstruction error distribution of all reconstructed feature vectors, specifically including: The encoder of the autoencoder model transforms the high-dimensional geometric feature vectors of the sample set into a low-dimensional latent representation, and the decoder of the autoencoder model transforms the low-dimensional latent representation into a reconstructed feature vector with the same dimension as the high-dimensional geometric feature vectors of the sample set. Calculate the reconstruction error between each high-dimensional geometric feature vector and the corresponding reconstructed feature vector in the sample set, and determine the reconstruction error distribution of all reconstructed feature vectors based on the calculation results; The discrimination threshold is determined by setting the reconstruction error corresponding to a set quantile based on the reconstruction error distribution of all reconstructed feature vectors.

4. The identification method according to claim 3, characterized in that, The encoder of the autoencoder model transforms the high-dimensional geometric feature vectors of the sample set into a low-dimensional latent representation, and the decoder of the autoencoder model then transforms the low-dimensional latent representation into a reconstructed feature vector with the same dimension as the high-dimensional geometric feature vectors of the sample set. Specifically, this means: in, It is a high-dimensional geometric eigenvector. It is a low-dimensional latent representation. It is a weight matrix. It is the activation function ReLU. It is a bias term; in, It is to reconstruct the feature vector. It is a weight matrix. It is a bias term.

5. The identification method according to claim 1, characterized in that, The identification method further includes pre-acquiring high-dimensional geometric feature vectors corresponding to all target slope units to be identified within the target working area, specifically including: Pre-extract the target slope units to be identified within the target working area, including: The slope units were initially delineated using hydrological analysis methods; The curvature threshold method is used to identify slope turning lines and segment slope units that better match the actual terrain boundaries. A terrain roughness index is introduced to verify the unit boundary and determine the target slope units to be identified within the target working area. Extract the main sliding surface and main profile of the target slope element to be identified; Based on the extracted main sliding surface and main profile of the target slope unit to be identified, the planar geometric features and cross-sectional geometric features of the target slope unit to be identified are obtained. The high-dimensional slope geometric feature data of the target slope unit to be identified, which consists of planar geometric features and cross-sectional geometric features, is standardized to obtain the corresponding high-dimensional geometric feature vector.

6. The identification method according to claim 1, characterized in that, The acquisition of planar and cross-sectional geometric features of slope units based on the main sliding surface and main profile of slope units with similar or identical geological conditions within the target work area specifically refers to: Data on historical landslide disasters or known landslide hazard points within the target work area that have the same or similar geological environmental conditions as the target slope unit to be identified are obtained in advance. Extract the main sliding surface and main profile of slope units from historical landslide disasters or known landslide hazard points; Based on the extracted main sliding surface and main profile of the slope unit from historical landslide disasters or known landslide hazard points, the planar geometric features and cross-sectional geometric features of the slope unit are obtained.

7. The identification method according to claim 1, characterized in that, The process of eliminating isolated misclassified units from potential landslide hazards using density clustering algorithms to obtain landslide hazard identification results specifically refers to: Use the spatial coordinates of potential landslide hazards as clustering input; Taking any potential landslide hazard as the starting point, if the number of potential landslide hazard within a set neighborhood radius corresponding to the starting point is greater than the set minimum number of points, then the potential landslide hazard is the core point. For each core point, create a new cluster and add the core point and its defined neighborhood radius of potential landslide hazards to the current new cluster; These newly added potential landslide hazards are processed recursively until no new core points or boundary points are added to the current cluster; where, for each potential landslide hazard within a set neighborhood radius of a core point, if it is not a core point but has been previously added to a cluster, it is a boundary point. Potential landslide hazards that are neither core points nor boundary points are considered isolated misjudged units, marked as noise points, and directly deleted. The remaining potential landslide hazards after eliminating noise points are the identified landslide hazards.

8. A landslide hazard identification device based on the three-dimensional geometric features of a slope profile and a machine learning model, characterized in that, The identification device includes: Memory is used to store executable instructions for a computer; A processor, when executing computer-executable instructions stored in the memory, implements the identification method according to any one of claims 1 to 7.

9. A storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the identification method according to any one of claims 1 to 7.

Images