A method and system for identifying geological landslides

Abstract

The present invention discloses a method and system for identifying geological landslides, relating to the technical field of geological disaster monitoring. The method is as follows: after obtaining remote sensing images and lidar scanning data of the area where the target slope is located, first, based on the preprocessing of the remote sensing images, the formation lithology label values, vegetation coverage label values, and the distance values to the nearest water body of each target pixel point are obtained, and based on the preprocessing of the scanning data, the elevation values, slope values, and aspect values of each target pixel point are obtained. Then, these numerical values are imported into a landslide body recognition model that has been pre-trained based on an artificial intelligence algorithm to obtain the landslide body recognition results of each target pixel point, and a binarized image of the landslide body distribution of the target slope is generated by summarization. Finally, edge contour extraction processing, connected domain area calculation, and area comparison are performed on the landslide body connected domains in the binarized image, and when the comparison conditions are met, the landslide area surrounded by the edge contour is used as the geological landslide recognition result, which is beneficial to environmental protection and disaster reduction and prevention.

Description

Technical Field

[0001] The present invention belongs to the technical field of geological disaster monitoring, and particularly relates to a method and system for identifying geological landslides. Background Art

[0002] Under the influence of human activities and natural conditions, the phenomenon that the rock mass or soil mass on a slope loses its stability and then slides along the failure surface is called a landslide disaster. Landslide geological disasters are very common, with characteristics such as group occurrence, multiple occurrence, suddenness, and harmfulness. They have become a global environmental problem, seriously threatening the lives and property safety of the people. Their impact on human society is increasing and cannot be ignored. Ranking by the harm degree caused by natural disasters, landslide geological disasters have become the largest geological disaster except earthquakes. The problem of landslide disasters is closely related to people's living environment. Improving the research level of landslide disaster prevention and control has important theoretical and engineering practical significance for environmental protection and disaster reduction and prevention.

[0003] Currently, there are mainly the following three existing geological landslide identification schemes: (1) Field investigation, the advantage is that it can observe the shape of the landslide at close range and delineate each landslide element, but the disadvantage is low work efficiency and inability to complete landslide surveys in areas that are difficult to reach; (2) Satellite remote sensing technology (i.e., "digital landslide"), which is further divided into optical satellite remote sensing technology and imaging radar satellite remote sensing technology. Its disadvantage is that data acquisition is greatly restricted by weather, cannot penetrate vegetation, has limited ability to identify landslides under vegetation cover, and cannot obtain a DEM (Digital Elevation Model) model with meter-level accuracy at the same time. Therefore, it is impossible to generate fine parameters that can depict the minute deformation of the ground surface, restricting research such as landslide body identification and hazard assessment; (3) Aerial remote sensing technology, also known as aerial photogrammetric remote sensing technology. Its aerial images can reach a resolution of decimeter level and generate a DEM model with a stereoscopic image pair. It has achieved relatively remarkable results in the field of landslide disaster identification. Especially with the DEM model generated by its stereoscopic image pair, relevant parameters of the landslide can be calculated relatively accurately in a three-dimensional system. However, aerial images still cannot penetrate vegetation, and the identification effect of landslide elements under vegetation cover or in the shadow is poor, especially powerless for minute deformation under such geomorphic conditions.

[0004] Therefore, in the face of the above problems existing in the existing geological landslide identification schemes, how to provide a new type of geological landslide identification scheme to improve work efficiency and the accuracy of landslide identification results, and thus facilitate environmental protection and disaster reduction and prevention, is an urgent research topic for those skilled in the art. Summary of the Invention

[0005] The object of the present invention is to provide a method, a system, a computer device, a computer-readable storage medium and a computer program product for identifying geological landslides, so as to solve the problems of low working efficiency, limited landslide identification ability and poor identification effect existing in the existing geological landslide identification solutions.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, a method for identifying geological landslides is provided, including:

[0008] Obtain remote sensing images and lidar scanning data of the area where the target slope is located;

[0009] For each pixel point in the remote sensing image, import the corresponding pixel value and the pixel values of the surrounding pixel points into a formation lithology classification model pre-trained based on a first artificial intelligence algorithm, and output the corresponding formation lithology classification result, where the formation lithology classification result is used to indicate the corresponding pixel point and a certain formation lithology label value among multiple formation lithology label values, and the multiple formation lithology label values correspond to multiple formation types one by one;

[0010] For each pixel point, import the corresponding pixel value and the pixel values of the surrounding pixel points into a vegetation coverage identification model pre-trained based on a second artificial intelligence algorithm, and output the corresponding vegetation coverage identification result, where the vegetation coverage identification result is used to indicate the vegetation coverage label value of the corresponding pixel point, and the vegetation coverage label value is represented by the numerical value 1 for being covered by vegetation and the numerical value 0 for not being covered by vegetation;

[0011] For each pixel point, import the corresponding pixel value and the pixel values of the surrounding pixel points into a surface water body identification model pre-trained based on a third artificial intelligence algorithm, and output the corresponding surface water body identification result, where the surface water body identification result is used to indicate whether the corresponding pixel point is a water body pixel point;

[0012] For each target pixel point in the remote sensing image and located within the target slope, according to the surface water body identification result of each pixel point, determine the water body pixel point closest to the corresponding pixel point, and calculate the distance value between the water body pixel point and the corresponding pixel point;

[0013] Generate a digital elevation model of the area where the target slope is located according to the lidar scanning data;

[0014] According to the digital elevation model, obtain the elevation distribution image, slope distribution image, and aspect distribution image of the area where the target slope is located. Among them, the pixel value of each pixel point in the elevation distribution image is the elevation value, the pixel value of each pixel point in the slope distribution image is the slope value, and the pixel value of each pixel point in the aspect distribution image is the aspect value;

[0015] For each of the target pixel points, import the corresponding formation lithology label value, vegetation cover label value, distance value, elevation value, slope value, and aspect value, as well as the formation lithology label value, vegetation cover label value, distance value, elevation value, slope value, and aspect value of the corresponding surrounding target pixel points into the landslide recognition model pre-trained based on the fourth artificial intelligence algorithm, and output the corresponding landslide recognition result, where the landslide recognition result is used to indicate whether the corresponding pixel point is a landslide pixel point;

[0016] According to the landslide recognition results of each of the target pixel points, generate a binary image of the landslide distribution of the target slope, where the binary image of the landslide distribution uses a first pixel value to represent that the corresponding pixel point is a landslide pixel point, and uses a second pixel value to represent that the corresponding pixel point is not a landslide pixel point;

[0017] Perform edge contour extraction processing on the landslide connected regions in the binary image of the landslide distribution to obtain the edge contours of the landslide connected regions;

[0018] According to the edge contours of the landslide connected regions, calculate the area of the landslide connected regions;

[0019] If the area of the landslide connected region reaches a preset area threshold, then use the landslide area surrounded by the edge contours of the landslide connected region as the geological landslide recognition result.

[0020] Based on the above invention content, a new solution for geological landslide identification based on the mixture of remote sensing images and lidar scanning data is provided. That is, after obtaining the remote sensing images and lidar scanning data of the area where the target slope is located, first, the formation lithology label value, vegetation coverage label value, and the distance value to the nearest water body of each target pixel point are obtained based on the preprocessing of the remote sensing images, and the elevation value, slope value, and aspect value of each target pixel point are obtained based on the preprocessing of the scanning data. Then, these numerical values are imported into the landslide body identification model pre-trained based on the artificial intelligence algorithm to obtain the landslide body identification results of each target pixel point, and a binary image of the landslide body distribution of the target slope is generated by summarization. Finally, edge contour extraction processing, connected domain area calculation, and area comparison are performed on the landslide body connected domains in the binary image, and when the comparison conditions are met, the landslide area surrounded by the edge contour is used as the geological landslide identification result. In this way, not only can the work efficiency be improved, but also the implicit landslide influence information in the remote sensing images and lidar scanning data can be fully utilized to make the best use of their advantages and avoid their disadvantages, effectively improving the landslide identification ability, identification accuracy, and identification effect, which is beneficial to environmental protection and disaster reduction and prevention, and is convenient for practical application and popularization.

[0021] In a possible design, the first artificial intelligence algorithm, the second artificial intelligence algorithm, the third artificial intelligence algorithm, or the fourth artificial intelligence algorithm adopts a machine learning algorithm based on support vector machine, stochastic gradient descent method, multivariable linear regression, multi-layer perceptron, decision tree, backpropagation neural network, or radial basis function network.

[0022] In a possible design, the multiple formation types include Cambrian formation, Ordovician formation, Silurian formation, Devonian formation, Carboniferous formation, Permian formation, and / or Triassic formation.

[0023] In a possible design, generating a digital elevation model of the area where the target slope is located according to the lidar scanning data includes:

[0024] Using differential GPS technology, and combining the dynamic GPS data of the airborne platform and the data of the ground GPS reference station, jointly solving to obtain the three-dimensional coordinates of the flight trajectory of the airborne platform, where the airborne platform refers to the aircraft platform used to carry the lidar scanning device to obtain the lidar scanning data;

[0025] Integrating the inertial navigation data of the airborne platform onto the three-dimensional coordinates of the flight trajectory, and then solving to obtain a new flight trajectory with the instantaneous position and attitude information of the lidar scanning device;

[0026] Combine the new flight trajectory and the lidar scan data, and incorporate system error correction parameters and coordinate projection parameters to solve for lidar point cloud data with three-dimensional coordinates in the WGS-84 coordinate system;

[0027] Perform ground point extraction processing on the lidar point cloud data to obtain ground point cloud data;

[0028] Generate a digital elevation model of the area where the target slope is located based on the ground point cloud data.

[0029] In a possible design, perform edge contour extraction processing on the landslide-connected regions in the landslide distribution binary image to obtain the edge contours of the landslide-connected regions, including:

[0030] Use the region growing method to perform hole removal processing on the landslide-connected regions in the landslide distribution binary image to obtain new landslide-connected regions;

[0031] Perform edge contour extraction processing based on the Canny algorithm on the new landslide-connected regions to obtain the edge contours of the landslide-connected regions.

[0032] In a possible design, after taking the landslide area defined by the edge contours of the landslide-connected regions as the geological landslide identification result, the method further includes:

[0033] Obtain precipitation distribution images of the area where the target slope is located in the current nearest K historical unit time periods and the current next future unit time period, where the pixel value of each pixel point in the precipitation distribution image is precipitation, and K represents a positive integer;

[0034] Based on the precipitation distribution images, statistically obtain the average precipitation of all pixel points located within the landslide area in the current nearest multiple historical unit time periods and the current next future unit time period;

[0035] Based on the formation lithology classification results of each pixel point, statistically obtain the proportions of multiple types of pixel points among all pixel points, where the multiple types of pixel points correspond one-to-one with multiple formation types;

[0036] Import the area of the landslide-connected region, the average precipitation, the proportions of the multiple types of pixel points, and the average vegetation coverage label value, average distance value, average elevation value, average slope value, and average aspect value of all pixel points into a landslide event occurrence prediction model pre-trained based on the fifth artificial intelligence algorithm, and output the probability value of a landslide event occurring in the current next future unit time period;

[0037] If the probability value reaches a preset probability threshold, a landslide event warning action is triggered for the landslide area.

[0038] In a possible design, based on the classification results of the formation lithology of each pixel, the proportions of various types of pixels among all the pixels are statistically obtained, including:

[0039] For each formation type among the various formation types, based on the classification results of the formation lithology of each pixel, the number of pixels with corresponding formation lithology label values among all the pixels is statistically obtained, and the ratio of this number to the total number of pixels of all the pixels is used as the proportion of the corresponding pixels among all the pixels.

[0040] In a possible design, the landslide event occurrence prediction model is pre-trained in the following manner:

[0041] Obtain multiple negative sample data and multiple positive sample data corresponding one by one to multiple historical landslide events. Among them, the negative sample data includes a first model input item and a first model output item with a value of 0. The first model input item includes the connected domain area of the landslide body in the first known landslide area, the average precipitation in K + 1 consecutive historical unit time periods when no landslide event occurs, the proportions of various types of pixels among all the pixels in the first known landslide area, and the average vegetation coverage label value, average distance value, average elevation value, average slope value, and average aspect value of all the pixels in the first known landslide area. The positive sample data includes a second model input item and a second model output item with a value of 1. The second model input item includes the connected domain area of the landslide body in the second known landslide area, the average precipitation in the K historical unit time periods closest to the occurrence of the corresponding landslide event and the historical unit time period when the corresponding landslide event occurs, the proportions of various types of pixels among all the pixels in the second known landslide area, and the average vegetation coverage label value, average distance value, average elevation value, average slope value, and average aspect value of all the pixels in the second known landslide area;

[0042] Apply the multiple negative sample data and the multiple positive sample data to perform calibration and verification modeling on a machine learning model based on the fifth artificial intelligence algorithm to obtain the landslide event occurrence prediction model. Among them, the confidence level with an output value of 1 of the landslide event occurrence prediction model is used as the probability value of a landslide event occurring in the next unit period.

[0043] In a second aspect, a geological landslide identification system is provided, including an image data acquisition unit, a formation lithology classification unit, a vegetation cover identification unit, a surface water body identification unit, a near-water distance calculation unit, a digital model generation unit, a distribution image acquisition unit, a landslide body identification unit, a binary image generation unit, an edge contour extraction unit, a connected domain area calculation unit, and a landslide area determination unit;

[0044] The image data acquisition unit is configured to acquire remote sensing images and lidar scanning data of the area where the target slope is located;

[0045] The formation lithology classification unit is communicatively connected to the image data acquisition unit, and is configured to import the corresponding pixel value and the pixel values of the corresponding surrounding pixel points for each pixel point in the remote sensing image into a formation lithology classification model pre-trained based on a first artificial intelligence algorithm, and output a corresponding formation lithology classification result, where the formation lithology classification result is used to indicate a certain formation lithology label value among multiple formation lithology label values corresponding to the corresponding pixel point, and the multiple formation lithology label values correspond to multiple formation types one by one;

[0046] The vegetation cover identification unit is communicatively connected to the image data acquisition unit, and is configured to import the corresponding pixel value and the pixel values of the corresponding surrounding pixel points for each pixel point into a vegetation cover identification model pre-trained based on a second artificial intelligence algorithm, and output a corresponding vegetation cover identification result, where the vegetation cover identification result is used to indicate the vegetation cover label value of the corresponding pixel point, and the vegetation cover label value is represented by the numerical value 1 for being covered by vegetation and the numerical value 0 for not being covered by vegetation;

[0047] The surface water body identification unit is communicatively connected to the image data acquisition unit, and is configured to import the corresponding pixel value and the pixel values of the corresponding surrounding pixel points for each pixel point into a surface water body identification model pre-trained based on a third artificial intelligence algorithm, and output a corresponding surface water body identification result, where the surface water body identification result is used to indicate whether the corresponding pixel point is a water body pixel point;

[0048] The near-water distance calculation unit is communicatively connected to the surface water body identification unit, and is configured to determine the water body pixel point closest to the corresponding pixel point for each target pixel point in the remote sensing image and located within the target slope according to the surface water body identification result of each pixel point, and calculate the distance value between the water body pixel point and the corresponding pixel point;

[0049] The digital model generation unit is communicatively connected to the image data acquisition unit, and is configured to generate a digital elevation model of the area where the target slope is located according to the lidar scanning data;

[0050] The distribution image acquisition unit is communicatively connected to the digital model generation unit and is configured to obtain an elevation distribution image, a slope distribution image, and an aspect distribution image of the area where the target slope is located according to the digital elevation model. Among them, the pixel value of each pixel point in the elevation distribution image is an elevation value, the pixel value of each pixel point in the slope distribution image is a slope value, and the pixel value of each pixel point in the aspect distribution image is an aspect value;

[0051] The landslide body identification unit is communicatively connected to the formation lithology classification unit, the vegetation cover identification unit, the near-water distance calculation unit, and the distribution image acquisition unit respectively. For each target pixel point, it imports the corresponding formation lithology label value, vegetation cover label value, distance value, elevation value, slope value, and aspect value, as well as the formation lithology label value, vegetation cover label value, distance value, elevation value, slope value, and aspect value of the corresponding surrounding target pixel points into a landslide body identification model pre-trained based on the fourth artificial intelligence algorithm, and outputs the corresponding landslide body identification result. Among them, the landslide body identification result is used to indicate whether the corresponding pixel point is a landslide body pixel point;

[0052] The binary image generation unit is communicatively connected to the landslide body identification unit and is configured to generate a binary landslide body distribution image of the target slope according to the landslide body identification results of each target pixel point. Among them, the binary landslide body distribution image uses a first pixel value to represent that the corresponding pixel point is a landslide body pixel point, and uses a second pixel value to represent that the corresponding pixel point is not a landslide body pixel point;

[0053] The edge contour extraction unit is communicatively connected to the binary image generation unit and is configured to perform edge contour extraction processing on the landslide body connected domain in the binary landslide body distribution image to obtain the edge contour of the landslide body connected domain;

[0054] The connected domain area calculation unit is communicatively connected to the edge contour extraction unit and is configured to calculate the area of the landslide body connected domain according to the edge contour of the landslide body connected domain;

[0055] The landslide area determination unit is communicatively connected to the edge contour extraction unit and the connected domain area calculation unit respectively. If the area of the landslide body connected domain reaches a preset area threshold, the landslide area surrounded by the edge contour of the landslide body connected domain is used as the geological landslide identification result.

[0056] In a possible design, it further includes a precipitation acquisition unit, a precipitation averaging unit, a lithology proportion statistics unit, a landslide event prediction unit, and an early warning action execution unit;

[0057] The precipitation acquisition unit is communicatively connected to the landslide area determination unit, and is configured to, after taking the landslide area defined by the edge contour of the landslide body communication domain as the geological landslide identification result, acquire the precipitation distribution images of the area where the target slope is located in the current nearest K historical unit time periods and the next future unit time period, wherein the pixel value of each pixel point in the precipitation distribution image is precipitation, and K represents a positive integer;

[0058] The precipitation averaging unit is communicatively connected to the precipitation acquisition unit, and is configured to, according to the precipitation distribution images, statistically obtain the average precipitation of all pixel points located within the landslide area in the current nearest multiple historical unit time periods and the next future unit time period;

[0059] The lithology proportion statistics unit is communicatively connected to the formation lithology classification unit, and is configured to, according to the formation lithology classification results of each pixel point, statistically obtain the proportions of multiple types of pixel points among all pixel points, wherein the multiple types of pixel points correspond one-to-one to multiple formation types;

[0060] The landslide event prediction unit is communicatively connected to the communication domain area calculation unit, the precipitation averaging unit, the lithology proportion statistics unit, the vegetation coverage identification unit, the near water distance calculation unit, and the distribution image acquisition unit respectively, and is configured to import the landslide body communication domain area, the average precipitation, the proportions of the multiple types of pixel points, and the average vegetation coverage label value, average distance value, average elevation value, average slope value, and average aspect value of all pixel points into a landslide event occurrence prediction model pre-trained based on the fifth artificial intelligence algorithm, and output the probability value of a landslide event occurring in the next future unit time period;

[0061] The early warning action execution unit is communicatively connected to the landslide event prediction unit, and is configured to, if the probability value reaches a preset probability threshold, trigger and execute a landslide event early warning action for the landslide area.

[0062] In a third aspect, the present invention provides a computer device, including a memory, a processor, and a transceiver that are communicatively connected in sequence, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the geological landslide identification method as described in the first aspect or any possible design in the first aspect.

[0063] In a fourth aspect, the present invention provides a computer-readable storage medium, on which instructions are stored, and when the instructions run on a computer, the geological landslide identification method as described in the first aspect or any possible design in the first aspect is executed.

[0064] In a fifth aspect, the present invention provides a computer program product, including a computer program or instructions, which, when executed by a computer, implement the method for identifying a geological landslide as described in the first aspect or any possible design in the first aspect.

[0065] Beneficial effects of the above solution:

[0066] (1) The present invention provides a new solution for identifying geological landslides based on a mixture of remote sensing images and lidar scanning data. That is, after obtaining the remote sensing images and lidar scanning data of the area where the target slope is located, first, based on the preprocessing of the remote sensing images, the formation lithology label values, vegetation cover label values, and the distance to the nearest water body values of each target pixel point are obtained, and based on the preprocessing of the scanning data, the elevation value, slope value, and aspect value of each target pixel point are obtained. Then, these numerical values are imported into a landslide body identification model pre-trained based on an artificial intelligence algorithm to obtain the landslide body identification results of each target pixel point, and the landslide body distribution binary image of the target slope is generated by summarization. Finally, edge contour extraction processing, connected domain area calculation, and area comparison are performed on the connected domains of the landslide bodies in the binary image, and when the comparison conditions are met, the landslide area surrounded by the edge contour is used as the geological landslide identification result. In this way, not only can the work efficiency be improved, but also the implicit information of landslide influence in the remote sensing images and lidar scanning data can be fully utilized to effectively improve the landslide identification ability, identification accuracy, and identification effect, which is beneficial to environmental protection and disaster reduction and prevention;

[0067] (2) Landslide monitoring can be carried out in real time and early warnings can be triggered after obtaining the geological landslide identification results, and the accuracy of the early warning timing can also be ensured based on the artificial intelligence algorithm, which is further beneficial to environmental protection and disaster reduction and prevention and is convenient for practical application and promotion. Description of the Drawings

[0068] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0069] Figure 1 It is a schematic flowchart of the method for identifying a geological landslide provided by an embodiment of the present application.

[0070] Figure 2 It is a schematic structural diagram of the identification system for a geological landslide provided by an embodiment of the present application.

[0071] Figure 3 It is a schematic structural diagram of a computer device provided by an embodiment of the present application. Detailed implementation manners

[0072] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and the descriptions of the embodiments or the prior art. Obviously, the following descriptions of the structures of the accompanying drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other embodiments can be obtained based on these embodiments without creative efforts. It should be noted here that the descriptions of these embodiment modes are used to help understand the present invention, but do not constitute a limitation to the present invention.

[0073] It should be understood that although terms such as first and second etc. may be used herein to describe various objects, these objects should not be limited by these terms. These terms are only used to distinguish one object from another. For example, the first object can be called the second object, and similarly, the second object can be called the first object, without departing from the scope of the exemplary embodiments of the present invention.

[0074] It should be understood that for the term "and / or" that may appear in this article, it is only a description of the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, or A and B exist simultaneously, etc.; another example, A, B and / or C can mean any one of A, B and C or any combination of them; for the term " / and" that may appear in this article, it is a description of another association object relationship, indicating that two relationships can exist. For example, A / and B can mean: A exists alone or A and B exist simultaneously, etc.; in addition, for the character " / " that may appear in this article, generally it means that the front and rear associated objects are an "or" relationship.

[0075] Embodiment

[0076] As Figure 1 shown, the method for identifying geological landslides provided in the first aspect of this embodiment can be, but is not limited to, executed by a computer device with certain computing resources, such as a cloud server, an edge computer, a personal computer (Personal Computer, PC, referring to a multi-purpose computer suitable for personal use in terms of size, price and performance; desktop computers, laptops to small laptops, tablet computers and ultrabooks etc. all belong to personal computers), a smart phone, a personal digital assistant (Personal Digital Assistant, PDA) or a wearable device and other electronic devices. As Figure 1 shown, the identification method can be, but is not limited to, including the following steps S1 to S12.

[0077] S1. Obtain the remote sensing image and lidar scanning data of the area where the target slope is located.

[0078] In the step S1, the area where the target slope is located is the area including the target slope, such as a river basin area including a river and the target slopes on both sides of the river. The remote sensing image can be a satellite remote sensing image or an aerial remote sensing image obtained based on an aircraft platform, and it can be obtained conventionally. The lidar scanning data is preferably the airborne lidar scanning data obtained based on an aircraft platform to improve the subsequent modeling accuracy. That is, since the lidar signal can penetrate a certain vegetation cover to obtain pure surface information, and its data acquisition method can minimize the shadow effect caused by terrain cutting compared with aerial images, a high-precision DEM model can be generated, which is beneficial to improving the recognition accuracy of (potential) landslide bodies. In addition, the lidar scanning data can be specifically but not limited to being conventionally collected by an ALS50-II type airborne lidar system (the flight altitude range of this system is 200 meters to 6000 meters, the maximum scanning angle is 75°, the maximum laser pulse emission frequency is 150KHz, the maximum scanning frequency is 90Hz, and it can receive up to four echoes).

[0079] S2. For each pixel point in the remote sensing image, import the corresponding pixel value and the pixel values of the corresponding surrounding pixel points into the formation lithology classification model pre-trained based on the first artificial intelligence algorithm, and output the corresponding formation lithology classification result, where the formation lithology classification result is used to indicate a certain formation lithology label value among the multiple formation lithology label values corresponding to the corresponding pixel point, and the multiple formation lithology label values correspond to multiple formation types one by one.

[0080] In the step S2, the surrounding pixel points can be conventionally determined by a coverage radius R, that is, if the second pixel point is within a circular area centered at the first pixel point with a radius of R, then the second pixel point is a surrounding pixel point of the first pixel point; otherwise, the second pixel point is not a surrounding pixel point of the first pixel point. In addition, when the aforementioned first pixel point is an edge pixel point, the pixel value of the first pixel point can be, but is not limited to, used as the pixel value of the missing surrounding pixel points. The aforementioned artificial intelligence algorithm is a core artificial intelligence algorithm that specifically studies how a computer simulates or implements human learning behaviors to acquire new knowledge or skills and reorganizes the existing knowledge structure to continuously improve its own performance, and is the fundamental way to make a computer intelligent. Specifically, the first artificial intelligence algorithm can be, but is not limited to, a machine learning algorithm based on support vector machines, stochastic gradient descent, multivariate linear regression, multi-layer perceptrons, decision trees, backpropagation neural networks, or radial basis function networks, etc., so as to quickly and accurately find the patterns in the data. Therefore, based on a certain amount of sample data (such as the remote sensing pixel values of pixel points located in the exposed area of a certain formation type and the remote sensing pixel values of the surrounding pixel points of the pixel points) and label data (such as the formation lithology label values of the certain formation type), through a conventional calibration and verification modeling process (specifically including a model calibration process and a verification process, that is, first comparing the model simulation results with the measured data, and then adjusting the model parameters according to the comparison results to make the simulation results coincide with the actual situation), the formation lithology classification model can be trained. Preferably, in the calibration and verification modeling process of the formation lithology classification model, a Bayesian optimization algorithm based on a tree structure is used to optimize the model parameters. In addition, specifically, the multiple formation types include, but are not limited to, Cambrian formations, Ordovician formations, Silurian formations, Devonian formations, Carboniferous formations, Permian formations, and / or Triassic formations, etc.

[0081] S3. For each of the pixel points, import the corresponding pixel value and the pixel values of the corresponding surrounding pixel points into a vegetation coverage recognition model pre-trained based on a second artificial intelligence algorithm, and output the corresponding vegetation coverage recognition result, where the vegetation coverage recognition result is used to indicate the vegetation coverage label value of the corresponding pixel point, and the vegetation coverage label value is represented by the numerical value 1 for being covered by vegetation and the numerical value 0 for not being covered by vegetation.

[0082] In the step S3, specifically, the second artificial intelligence algorithm can also but is not limited to specifically adopting machine learning algorithms based on support vector machines, stochastic gradient descent methods, multivariate linear regression, multi-layer perceptrons, decision trees, backpropagation neural networks, or radial basis function networks, etc. And it can also be pre-trained based on a certain amount of positive sample data (such as the remote sensing pixel values of pixel points within the vegetation coverage area and the remote sensing pixel values of the surrounding pixel points of the pixel point) and positive label data (such as the vegetation coverage label value of 1) and negative sample data (such as the remote sensing pixel values of pixel points within the non-vegetation coverage area and the remote sensing pixel values of the surrounding pixel points of the pixel point) and negative label data (such as the vegetation coverage label value of 0), and the vegetation coverage recognition model can be obtained through the conventional calibration and verification modeling process. In addition, preferably, in the calibration and verification modeling process of the vegetation coverage recognition model, the Bayesian optimization algorithm based on the tree structure can also be used to optimize the model parameters.

[0083] S4. For each of the pixel points, import the corresponding pixel value and the pixel values of the corresponding surrounding pixel points into the surface water body recognition model pre-trained based on the third artificial intelligence algorithm, and output the corresponding surface water body recognition result, where the surface water body recognition result is used to indicate whether the corresponding pixel point is a water body pixel point.

[0084] In the step S4, specifically, the third artificial intelligence algorithm can also but is not limited to specifically adopting machine learning algorithms based on support vector machines, stochastic gradient descent methods, multivariate linear regression, multi-layer perceptrons, decision trees, backpropagation neural networks, or radial basis function networks, etc. And it can also be pre-trained based on a certain amount of positive sample data (such as the remote sensing pixel values of pixel points within water body areas such as river surfaces or lake surfaces and the remote sensing pixel values of the surrounding pixel points of the pixel point) and positive label data (such as the water body label value of 1) and negative sample data (such as the remote sensing pixel values of pixel points within non-water body areas and the remote sensing pixel values of the surrounding pixel points of the pixel point) and negative label data (such as the water body label value of 0), and the surface water body recognition model can be obtained through the conventional calibration and verification modeling process. In addition, preferably, in the calibration and verification modeling process of the surface water body recognition model, the Bayesian optimization algorithm based on the tree structure can also be used to optimize the model parameters.

[0085] S5. For each target pixel point in the remote sensing image and within the target slope, determine the water body pixel point closest to the corresponding pixel point according to the surface water body recognition result of each pixel point, and calculate the distance value between the water body pixel point and the corresponding pixel point.

[0086] In the step S5, the target slope, for example, is the mountain slope on both sides of the river, which can be pre-defined manually. In addition, the specific determination method of the nearest water body pixel points can be conventionally realized by traversing and calculating the distances between pixel points and comparing the distances.

[0087] S6. Generate a digital elevation model of the area where the target slope is located according to the lidar scanning data.

[0088] In the step S6, the digital elevation model is a digital simulation of the ground terrain through limited terrain elevation data (i.e., the digital expression of the terrain surface form). It is a kind of solid ground model representing the ground elevation in the form of an ordered numerical array, and can describe the spatial distribution of various geomorphic factors including elevation, such as linear and non-linear combinations of factors such as slope, aspect, and slope change rate. When the lidar scanning data is airborne lidar scanning data obtained based on an aircraft platform, specifically, generating the digital elevation model of the area where the target slope is located according to the lidar scanning data includes, but is not limited to, the following steps S601 to S606.

[0089] S601. Adopt differential GPS (Global Positioning System) technology, and combine the dynamic GPS data of the airborne platform and the data of the ground GPS reference station to jointly solve for the three-dimensional coordinates of the flight trajectory of the airborne platform, where the airborne platform refers to the aircraft platform used to carry the lidar scanning device to obtain the lidar scanning data.

[0090] In the step S601, the differential GPS technology (Differential Global Positioning System, abbreviated as DGPS, that is, differential global positioning system) is an existing technology that installs a GPS monitoring receiver at an accurate known position (such as a reference station) and calculates the distance correction number between the reference station and the GPS satellite. Therefore, the three-dimensional coordinates of the flight trajectory of the airborne platform can be conventionally jointly solved.

[0091] S602. Integrate the inertial navigation data of the airborne platform onto the three-dimensional coordinates of the flight trajectory, and then solve for a new flight trajectory with the instantaneous position and attitude information of the lidar scanning device.

[0092] S603. Combine the new flight trajectory and the lidar scanning data, and add system error correction parameters and coordinate projection parameters to solve for the lidar point cloud data with three-dimensional coordinates in the WGS-84 coordinate system.

[0093] S604. Perform ground point extraction processing on the laser point cloud data to obtain ground point cloud data.

[0094] S605. Generate a digital elevation model of the area where the target slope is located based on the ground point cloud data.

[0095] S7. Based on the digital elevation model, obtain the elevation distribution image, slope distribution image, and aspect distribution image of the area where the target slope is located. Among them, the pixel value of each pixel point in the elevation distribution image is the elevation value, the pixel value of each pixel point in the slope distribution image is the slope value, and the pixel value of each pixel point in the aspect distribution image is the aspect value.

[0096] In step S7, since the digital elevation model can describe the spatial distribution of various geomorphic factors including elevation, such as linear and non-linear combinations of factors such as slope, aspect, and slope change rate, etc., the elevation distribution image, the slope distribution image, and the aspect distribution image can be routinely obtained based on this. In addition, to ensure that their pixel points have a one-to-one correspondence with the pixel points in the remote sensing image, the elevation distribution image, the slope distribution image, and the aspect distribution image need to have the same size as the remote sensing image respectively.

[0097] S8. For each target pixel point, import the corresponding formation lithology label value, vegetation coverage label value, distance value, elevation value, slope value, and aspect value, as well as the formation lithology label value, vegetation coverage label value, distance value, elevation value, slope value, and aspect value of the corresponding surrounding target pixel points into the landslide recognition model pre-trained based on the fourth artificial intelligence algorithm, and output the corresponding landslide recognition result. Among them, the landslide recognition result is used to indicate whether the corresponding pixel point is a landslide pixel point.

[0098] In step S8, according to existing research, it is found that formation lithology, vegetation coverage, distance to the nearest water body, elevation, slope, and aspect are important factors affecting geological landslides. Therefore, the formation lithology label value, vegetation coverage label value, distance value, elevation value, slope value, and aspect value of the target pixel and its surrounding pixels are used as model input items, so as to determine whether the target pixel is a landslide pixel through the landslide body identification model. Specifically, the fourth artificial intelligence algorithm can also but is not limited to specifically using machine learning algorithms based on support vector machines, stochastic gradient descent methods, multivariable linear regression, multi-layer perceptrons, decision trees, backpropagation neural networks, or radial basis function networks, etc. And it can also be pre-trained based on a certain amount of positive sample data (such as the remote sensing pixel values of pixels located in known landslide areas and the remote sensing pixel values of the surrounding pixels of the pixel) and positive label data (such as landslide body label values with a value of 1), and negative sample data (such as the remote sensing pixel values of pixels located in known non-landslide areas and the remote sensing pixel values of the surrounding pixels of the pixel) and negative label data (such as landslide body label values with a value of 0) to obtain the landslide body identification model through the conventional calibration and verification modeling process. In addition, preferably, in the calibration and verification modeling process of the landslide body identification model, the Bayesian optimization algorithm based on the tree structure can also be used to optimize the model parameters.

[0099] S9. Generate a binarized landslide body distribution image of the target slope according to the landslide body identification results of the respective target pixels, where the binarized landslide body distribution image uses a first pixel value to represent that the corresponding pixel is a landslide pixel, and uses a second pixel value to represent that the corresponding pixel is not a landslide pixel.

[0100] In step S9, for example, the first pixel value is 255 and the second pixel value is 0.

[0101] S10. Perform edge contour extraction processing on the landslide body connected regions in the binarized landslide body distribution image to obtain the edge contours of the landslide body connected regions.

[0102] In the step S10, in order to eliminate the holes in the connected domain and improve the accuracy of edge contour extraction, preferably, edge contour extraction processing is performed on the landslide connected domain in the landslide distribution binary image to obtain the edge contour of the landslide connected domain, including but not limited to the following steps: S101. The hole removal processing is performed on the landslide connected domain in the landslide distribution binary image by using the region growing method to obtain a new landslide connected domain; S102. Edge contour extraction processing based on the Canny algorithm is performed on the new landslide connected domain to obtain the edge contour of the landslide connected domain. The aforementioned region growing method and Canny algorithm are both existing technical means. In addition, the number of landslide connected domains in the landslide distribution binary image can be one or more. When there are multiple ones, the edge contour extraction processing needs to be performed for each landslide connected domain respectively to obtain the corresponding edge contour of the landslide connected domain.

[0103] S11. According to the edge contour of the landslide connected domain, the area of the landslide connected domain is calculated.

[0104] S12. If the area of the landslide connected domain reaches the preset area threshold, the landslide area surrounded by the edge contour of the landslide connected domain is used as the geological landslide recognition result.

[0105] In the step S12, the preset area threshold can be determined in advance according to historical experience. In addition, if the area of the landslide connected domain does not reach the preset area threshold, the corresponding landslide connected domain can be regarded as noise and excluded.

[0106] Based on the recognition method described in the foregoing steps S1 to S12, a new scheme for geological landslide recognition based on a mixture of remote sensing images and lidar scanning data is provided. That is, after obtaining the remote sensing image and lidar scanning data of the area where the target slope is located, first, the formation lithology label value, vegetation coverage label value, and nearest water body distance value of each target pixel point are obtained based on the preprocessing of the remote sensing image, and the elevation value, slope value, and aspect value of each target pixel point are obtained based on the preprocessing of the scanning data. Then, these numerical values are imported into the landslide recognition model pre-trained based on the artificial intelligence algorithm to obtain the landslide recognition results of each target pixel point, and the landslide distribution binary image of the target slope is generated by summarization. Finally, edge contour extraction processing, connected domain area calculation, and area comparison are performed on the landslide connected domain in the binary image, and when the comparison conditions are met, the landslide area surrounded by the edge contour is used as the geological landslide recognition result. In this way, not only can the work efficiency be improved, but also the implicit information of landslide influence in the remote sensing image and lidar scanning data can be fully utilized to make the best use of the advantages and avoid the disadvantages, effectively improving the landslide recognition ability, recognition accuracy, and recognition effect, which is beneficial to environmental protection and disaster reduction and prevention, and is convenient for practical application and promotion.

[0107] Based on the technical solution in the first aspect of this embodiment, a possible design 1 for real-time landslide monitoring and triggering early warning after obtaining the geological landslide identification result is also provided. That is, after taking the landslide area defined by the edge contour of the landslide body connected domain as the geological landslide identification result, the method further includes but is not limited to the following steps S131 to S135.

[0108] S131. Obtain the precipitation distribution images of the area where the target slope is located in the current nearest K historical unit time periods and the current next future unit time period. Among them, the pixel value of each pixel point in the precipitation distribution image is the precipitation, and K represents a positive integer.

[0109] In the step S131, the unit time period can be but is not limited to examples such as 1 hour, 1 shichen, half a day, or 1 day, etc. The current nearest K historical unit time periods need to include the unit time period currently being experienced. For example, if the current unit time period is 12:00 on the 5th day and K is 12, then it is necessary to obtain the precipitation distribution image of the area where the target slope is located from 1:00 on the 5th day to 13:00 on the 5th day. Among them, from 1:00 on the 5th day to 12:00 on the 5th day is the current nearest K historical unit time periods, 13:00 on the 5th day is the current next future unit time period, and the precipitation distribution image at 13:00 on the 5th day is conventional weather forecast data. In addition, the precipitation distribution image can be specifically obtained by conventional query from the weather forecast server.

[0110] S132. According to the precipitation distribution image, statistically obtain the average precipitation of all pixel points located in the landslide area in the current nearest multiple historical unit time periods and the current next future unit time period.

[0111] In the step S132, it is necessary to respectively statistically obtain the average precipitation of all pixel points located in the landslide area in the corresponding time period for each unit time period in the current nearest multiple historical unit time periods and the current next future unit time period, so as to form an average precipitation time series data. In addition, the specific statistical method is an existing conventional technical means.

[0112] S133. According to the classification results of the formation lithology of each pixel point, statistically obtain the proportions of multiple types of pixel points among all pixel points, where the multiple types of pixel points correspond one-to-one to multiple formation types.

[0113] In the step S133, specifically, according to the classification results of the formation lithology of each pixel, the proportions of multiple types of pixels among all the pixels are statistically obtained, including but not limited to the following steps: For each formation type among the multiple formation types, according to the classification results of the formation lithology of each pixel, the number of pixels with the corresponding formation lithology label value among all the pixels is statistically obtained, and the ratio of this number to the total number of pixels of all the pixels is used as the proportion of the corresponding pixels among all the pixels. For example, if the multiple formation types include Cambrian formation, Ordovician formation, Silurian formation, Devonian formation, Carboniferous formation, Permian formation, Triassic formation, etc., and the total number of pixels of all the pixels is 216, among which the number of pixels with the formation lithology label value corresponding to the Cambrian formation is 10, the number of pixels with the formation lithology label value corresponding to the Ordovician formation is 33, the number of pixels with the formation lithology label value corresponding to the Silurian formation is 4, the number of pixels with the formation lithology label value corresponding to the Devonian formation is 0, the number of pixels with the formation lithology label value corresponding to the Carboniferous formation is 27, the number of pixels with the formation lithology label value corresponding to the Permian formation is 54, and the number of pixels with the formation lithology label value corresponding to the Triassic formation is 88, then the proportions of multiple types of pixels among all the pixels can be statistically obtained as 0.046, 0.153, 0.019, 0, 0.125, 0.25, and 0.407 in sequence.

[0114] S134. Import the landslide body connected domain area, the average precipitation, the proportions of the multiple types of pixels, and the average vegetation cover label value, average distance value, average elevation value, average slope value, and average aspect value of all the pixels into a landslide event occurrence prediction model pre-trained based on the fifth artificial intelligence algorithm, and output the probability value of a landslide event occurring in the next future unit time period.

[0115] In step S134, the average vegetation coverage label value, average distance value, average elevation value, average slope value, and average aspect value of all the pixel points can be conventionally statistically obtained based on the vegetation coverage recognition results of the respective pixel points, the distance values corresponding to the respective target pixel points, the elevation distribution image, the slope distribution image, and the aspect distribution image. Since factors such as the area size of the landslide area, precipitation, formation lithology, vegetation coverage, distance to the nearest water body, elevation, slope, and aspect are important factors triggering geological landslide events, the landslide body connected domain area, the average precipitation, the proportion of various pixel points, and the average vegetation coverage label value, average distance value, average elevation value, average slope value, and average aspect value of all the pixel points are used as model input items, so as to predict the probability of a landslide event occurring in the current next future unit time period through the landslide event occurrence prediction model.

[0116] In step S134, specifically, the fifth artificial intelligence algorithm can also but is not limited to specifically adopting machine learning algorithms based on support vector machines, stochastic gradient descent methods, multivariate linear regression, multi-layer perceptrons, decision trees, backpropagation neural networks, or radial basis function networks, etc. And in detail, the landslide event occurrence prediction model is pre-trained according to the following steps S1341 to S1342.

[0117] S1341. Obtain multiple negative sample data and multiple positive sample data corresponding one by one to multiple historical landslide events. Among them, the negative sample data includes a first model input item and a first model output item with a value of 0. The first model input item includes the landslide body connected domain area of a first known landslide area, the average precipitation in K + 1 consecutive historical unit time periods when no landslide event occurs, the proportion of various pixel points among all pixel points located in the first known landslide area, and the average vegetation coverage label value, average distance value, average elevation value, average slope value, and average aspect value of all pixel points located in the first known landslide area. The positive sample data includes a second model input item and a second model output item with a value of 1. The second model input item includes the landslide body connected domain area of a second known landslide area, the average precipitation in the K historical unit time periods closest to the occurrence of the corresponding landslide event and the historical unit time period when the corresponding landslide event occurs, the proportion of various pixel points among all pixel points located in the second known landslide area, and the average vegetation coverage label value, average distance value, average elevation value, average slope value, and average aspect value of all pixel points located in the second known landslide area.

[0118] In the step S1341, the specific obtaining methods of the above average precipitation, the proportions of various pixel points, the average vegetation coverage label value, the average distance value, the average elevation value, the average slope value, the average aspect value, etc. can be obtained by conventional derivation according to the foregoing steps, and will not be elaborated here.

[0119] S1342. Apply the multiple negative sample data and the multiple positive sample data to calibrate and verify the machine learning model based on the fifth artificial intelligence algorithm, and obtain the landslide event occurrence prediction model. Among them, the confidence level with an output value of 1 of the landslide event occurrence prediction model is used as the probability value of the landslide event occurring in the next unit period.

[0120] S135. If the probability value reaches the preset probability threshold, trigger the execution of the landslide event warning action for the landslide area.

[0121] In the step S135, the preset probability threshold can also be determined in advance according to historical experience. In addition, if the probability value does not reach the preset probability threshold, the execution of the landslide event warning action is not triggered, and the landslide event warning action is specifically but not limited to the sound and light alarm action.

[0122] Based on the foregoing possible design one, it is possible to perform real-time landslide monitoring and trigger an early warning after obtaining the geological landslide identification result, and it is also possible to ensure the accuracy of the early warning timing based on the artificial intelligence algorithm, which is further beneficial to environmental protection and disaster reduction and prevention.

[0123] As Figure 2 shown, the second aspect of this embodiment provides a virtual system for implementing the recognition method described in the first aspect or any possible design in the first aspect, including an image data acquisition unit, a formation lithology classification unit, a vegetation coverage recognition unit, a surface water body recognition unit, a near-water distance calculation unit, a digital model generation unit, a distribution image acquisition unit, a landslide body recognition unit, a binary image generation unit, an edge contour extraction unit, a connected domain area calculation unit, and a landslide area determination unit;

[0124] The image data acquisition unit is used to acquire the remote sensing image and lidar scan data of the area where the target slope is located;

[0125] The formation lithology classification unit is communicatively connected to the image data acquisition unit, and is used to import the corresponding pixel value and the pixel values of the corresponding surrounding pixel points together into a formation lithology classification model pre-trained based on a first artificial intelligence algorithm for each pixel point in the remote sensing image, and output a corresponding formation lithology classification result. Among them, the formation lithology classification result is used to indicate the corresponding pixel point and a certain formation lithology label value among multiple formation lithology label values, and the multiple formation lithology label values correspond to multiple formation types one by one;

[0126] The vegetation coverage recognition unit is communicatively connected to the image data acquisition unit, and is used to import the corresponding pixel value and the pixel values of the corresponding surrounding pixel points together into a vegetation coverage recognition model pre-trained based on a second artificial intelligence algorithm for each pixel point, and output a corresponding vegetation coverage recognition result. Among them, the vegetation coverage recognition result is used to indicate the vegetation coverage label value of the corresponding pixel point, and the vegetation coverage label value uses the numerical value 1 to represent being covered by vegetation and the numerical value 0 to represent not being covered by vegetation;

[0127] The surface water body recognition unit is communicatively connected to the image data acquisition unit, and is used to import the corresponding pixel value and the pixel values of the corresponding surrounding pixel points together into a surface water body recognition model pre-trained based on a third artificial intelligence algorithm for each pixel point, and output a corresponding surface water body recognition result. Among them, the surface water body recognition result is used to indicate whether the corresponding pixel point is a water body pixel point;

[0128] The near-water distance calculation unit is communicatively connected to the surface water body recognition unit, and is used to determine the water body pixel point closest to the corresponding pixel point according to the surface water body recognition result of each pixel point for each target pixel point in the remote sensing image and located within the target slope, and calculate the distance value between the water body pixel point and the corresponding pixel point;

[0129] The digital model generation unit is communicatively connected to the image data acquisition unit, and is used to generate a digital elevation model of the area where the target slope is located according to the lidar scan data;

[0130] The distribution image acquisition unit is communicatively connected to the digital model generation unit, and is used to obtain an elevation distribution image, a slope distribution image, and an aspect distribution image of the area where the target slope is located according to the digital elevation model. Among them, the pixel value of each pixel point in the elevation distribution image is an elevation value, the pixel value of each pixel point in the slope distribution image is a slope value, and the pixel value of each pixel point in the aspect distribution image is an aspect value;

[0131] The landslide body identification unit is communicatively connected to the formation lithology classification unit, the vegetation coverage identification unit, the near-water distance calculation unit, and the distribution image acquisition unit respectively. For each target pixel point, it imports the corresponding formation lithology label value, vegetation coverage label value, distance value, elevation value, slope value, and aspect value, as well as the formation lithology label value, vegetation coverage label value, distance value, elevation value, slope value, and aspect value of the corresponding surrounding target pixel points into a landslide body identification model pre-trained based on the fourth artificial intelligence algorithm, and outputs the corresponding landslide body identification result. Among them, the landslide body identification result is used to indicate whether the corresponding pixel point is a landslide body pixel point;

[0132] The binary image generation unit is communicatively connected to the landslide body identification unit. It is used to generate a binary image of the landslide body distribution of the target slope according to the landslide body identification results of each target pixel point. Among them, the binary image of the landslide body distribution uses a first pixel value to represent that the corresponding pixel point is a landslide body pixel point, and uses a second pixel value to represent that the corresponding pixel point is not a landslide body pixel point;

[0133] The edge contour extraction unit is communicatively connected to the binary image generation unit. It is used to perform edge contour extraction processing on the landslide body connected domain in the binary image of the landslide body distribution to obtain the edge contour of the landslide body connected domain;

[0134] The connected domain area calculation unit is communicatively connected to the edge contour extraction unit. It is used to calculate the area of the landslide body connected domain according to the edge contour of the landslide body connected domain;

[0135] The landslide area determination unit is communicatively connected to the edge contour extraction unit and the connected domain area calculation unit respectively. If the area of the landslide body connected domain reaches a preset area threshold, it takes the landslide area surrounded by the edge contour of the landslide body connected domain as the geological landslide identification result.

[0136] In a possible design, it further includes a precipitation acquisition unit, a precipitation averaging unit, a lithology proportion statistics unit, a landslide event prediction unit, and an early warning action execution unit;

[0137] The precipitation acquisition unit is communicatively connected to the landslide area determination unit. After taking the landslide area defined by the edge contour of the landslide body connected domain as the geological landslide identification result, it acquires the precipitation distribution images of the area where the target slope is located in the current nearest K historical unit time periods and the current next future unit time period. Among them, the pixel value of each pixel point in the precipitation distribution image is precipitation, and K represents a positive integer;

[0138] The precipitation averaging unit is communicatively connected to the precipitation acquisition unit, and is configured to statistically obtain the average precipitation of all pixel points located within the landslide area during the current recent multiple historical unit time periods and the current next future unit time period according to the precipitation distribution image;

[0139] The lithology proportion statistics unit is communicatively connected to the formation lithology classification unit, and is configured to statistically obtain the proportions of multiple types of pixel points among all the pixel points according to the formation lithology classification results of the respective pixel points, wherein the multiple types of pixel points correspond one-to-one to multiple formation types;

[0140] The landslide event prediction unit is respectively communicatively connected to the connected domain area calculation unit, the precipitation averaging unit, the lithology proportion statistics unit, the vegetation coverage identification unit, the distance to water calculation unit, and the distribution image acquisition unit, and is configured to import the landslide body connected domain area, the average precipitation, the proportions of the multiple types of pixel points, and the average vegetation coverage label value, average distance value, average elevation value, average slope value, and average aspect value of all the pixel points into a landslide event occurrence prediction model pre-trained based on the fifth artificial intelligence algorithm, and output a probability value of a landslide event occurring during the current next future unit time period;

[0141] The warning action execution unit is communicatively connected to the landslide event prediction unit, and is configured to trigger the execution of a landslide event warning action for the landslide area if the probability value reaches a preset probability threshold.

[0142] For the working process, working details, and technical effects of the foregoing system provided in the second aspect of this embodiment, reference may be made to the recognition method described in the first aspect or any possible design in the first aspect, and details are not described herein again.

[0143] Such as Figure 3As shown in the figure, in the third aspect of this embodiment, a computer device for executing the recognition method described in the first aspect or any possible design in the first aspect is provided, including a memory, a processor, and a transceiver that are communicatively connected in sequence. Among them, the memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer programs and execute the recognition method described in the first aspect or any possible design in the first aspect. Specifically, for example, the memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first input first output (FIFO), and / or first input last output (FILO), etc.; the processor may be, but is not limited to, a microprocessor of the STM32F105 series. In addition, the computer device may also include, but is not limited to, a power module, a display screen, and other necessary components.

[0144] For the working process, working details, and technical effects of the foregoing computer device provided in the third aspect of this embodiment, reference may be made to the recognition method described in the first aspect or any possible design in the first aspect, which will not be elaborated here.

[0145] In the fourth aspect of this embodiment, a computer-readable storage medium storing instructions including the recognition method described in the first aspect or any possible design in the first aspect is provided, that is, instructions are stored on the computer-readable storage medium, and when the instructions are run on a computer, the recognition method described in the first aspect or any possible design in the first aspect is executed. Among them, the computer-readable storage medium refers to a carrier for storing data, and may include, but is not limited to, computer-readable storage media such as floppy disks, optical discs, hard disks, flash memories, USB flash drives, and / or memory sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0146] For the working process, working details, and technical effects of the foregoing computer-readable storage medium provided in the fourth aspect of this embodiment, reference may be made to the recognition method described in the first aspect or any possible design in the first aspect, which will not be elaborated here.

[0147] In the fifth aspect of this embodiment, a computer program product is provided, including a computer program or instructions, and when the computer program or the instructions are executed by a computer, the recognition method described in the first aspect or any possible design in the first aspect is implemented. Among them, the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0148] Finally, it should be noted that the above are only preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for identifying geological landslides, characterized in that, Including: Obtaining remote sensing images and lidar scanning data of the area where the target slope is located; For each pixel point in the remote sensing image, importing the corresponding pixel value and the pixel values of the surrounding pixel points into a formation lithology classification model pre-trained based on a first artificial intelligence algorithm, and outputting the corresponding formation lithology classification result, where the formation lithology classification result is used to indicate the corresponding pixel point and a certain formation lithology label value among multiple formation lithology label values, and the multiple formation lithology label values correspond one-to-one with multiple formation types; For each pixel point, importing the corresponding pixel value and the pixel values of the surrounding pixel points into a vegetation cover recognition model pre-trained based on a second artificial intelligence algorithm, and outputting the corresponding vegetation cover recognition result, where the vegetation cover recognition result is used to indicate the vegetation cover label value of the corresponding pixel point, and the vegetation cover label value uses the numerical value 1 to represent being covered by vegetation and the numerical value 0 to represent not being covered by vegetation; For each pixel point, importing the corresponding pixel value and the pixel values of the surrounding pixel points into a surface water body recognition model pre-trained based on a third artificial intelligence algorithm, and outputting the corresponding surface water body recognition result, where the surface water body recognition result is used to indicate whether the corresponding pixel point is a water body pixel point; For each target pixel point in the remote sensing image and within the target slope, determining the water body pixel point closest to the corresponding pixel point according to the surface water body recognition result of each pixel point, and calculating the distance value between the water body pixel point and the corresponding pixel point; Generating a digital elevation model of the area where the target slope is located according to the lidar scanning data; Obtaining an elevation distribution image, a slope distribution image, and an aspect distribution image of the area where the target slope is located according to the digital elevation model, where the pixel value of each pixel point in the elevation distribution image is an elevation value, the pixel value of each pixel point in the slope distribution image is a slope value, and the pixel value of each pixel point in the aspect distribution image is an aspect value; For each target pixel point, importing the corresponding formation lithology label value, vegetation cover label value, distance value, elevation value, slope value, and aspect value, as well as the formation lithology label value, vegetation cover label value, distance value, elevation value, slope value, and aspect value of the surrounding target pixel points into a landslide body recognition model pre-trained based on a fourth artificial intelligence algorithm, and outputting the corresponding landslide body recognition result, where the landslide body recognition result is used to indicate whether the corresponding pixel point is a landslide body pixel point; Generating a landslide body distribution binary image according to the landslide body recognition results of each target pixel point, where the landslide body distribution binary image uses a first pixel value to represent that the corresponding pixel point is a landslide body pixel point and a second pixel value to represent that the corresponding pixel point is not a landslide body pixel point; Performing edge contour extraction processing on the landslide body connected regions in the landslide body distribution binary image to obtain the edge contour of the landslide body connected regions; Calculating the area of the landslide body connected regions according to the edge contour of the landslide body connected regions; If the area of the landslide body connected domain reaches a preset area threshold, the landslide area surrounded by the edge contour of the landslide body connected domain is used as the geological landslide identification result.

2. The recognition method according to claim 1, characterized in that, The first artificial intelligence algorithm, the second artificial intelligence algorithm, the third artificial intelligence algorithm or the fourth artificial intelligence algorithm adopts a machine learning algorithm based on support vector machine, stochastic gradient descent method, multivariate linear regression, multi-layer perceptron, decision tree, backpropagation neural network or radial basis function network.

3. The recognition method according to claim 1, wherein The multiple formation types include Cambrian formation, Ordovician formation, Silurian formation, Devonian formation, Carboniferous formation, Permian formation and / or Triassic formation.

4. The recognition method according to claim 1, wherein Generating a digital elevation model of the area where the target slope is located according to the lidar scan data, including: Using differential GPS technology, and combining the dynamic GPS data of the airborne platform and the data of the ground GPS reference station, jointly solving to obtain the three-dimensional coordinates of the flight trajectory of the airborne platform, where the airborne platform refers to an aircraft platform used to carry lidar scanning equipment to obtain the lidar scan data; Integrating the inertial navigation data of the airborne platform onto the three-dimensional coordinates of the flight trajectory, and then solving to obtain a new flight trajectory with the instantaneous position and attitude information of the lidar scanning equipment; Combining the new flight trajectory and the lidar scan data, and adding system error correction parameters and coordinate projection parameters, solving to obtain lidar point cloud data with three-dimensional coordinates in the WGS-84 coordinate system; Performing ground point extraction processing on the lidar point cloud data to obtain ground point cloud data; Generating a digital elevation model of the area where the target slope is located according to the ground point cloud data.

5. The recognition method according to claim 1, characterized in that Performing edge contour extraction processing on the landslide body connected domain in the landslide body distribution binary image to obtain the edge contour of the landslide body connected domain, including: Using the region growing method to perform hole removal processing on the landslide body connected domain in the landslide body distribution binary image to obtain a new landslide body connected domain; Performing edge contour extraction processing based on the Canny algorithm on the new landslide body connected domain to obtain the edge contour of the landslide body connected domain.

6. The recognition method according to claim 1, characterized in that, After using the landslide area defined by the edge contour of the landslide body connected domain as the geological landslide identification result, the method further includes: Obtaining precipitation distribution images of the area where the target slope is located in the current nearest K historical unit time periods and the current next future unit time period, where the pixel value of each pixel point in the precipitation distribution image is precipitation, and K represents a positive integer; According to the precipitation distribution images, statistically obtaining the average precipitation of all pixel points located in the landslide area in the current nearest multiple historical unit time periods and the current next future unit time period; According to the formation lithology classification results of each pixel point, statistically obtaining the proportions of multiple pixel points among all pixel points, where the multiple pixel points correspond one-to-one to multiple formation types; Import the landslide body connected domain area, the average precipitation, the proportions of the multiple types of pixel points, and the average vegetation coverage label value, average distance value, average elevation value, average slope value, and average aspect value of all the pixel points into a landslide event occurrence prediction model pre-trained based on the fifth artificial intelligence algorithm, and output the probability value of a landslide event occurring in the next future unit time period; If the probability value reaches a preset probability threshold, trigger a landslide event warning action for the landslide area.

7. The recognition method according to claim 6, characterized in that, According to the stratigraphic lithology classification results of the respective pixel points, count the proportions of multiple types of pixel points among all the pixel points, including: For each stratigraphic type among the multiple stratigraphic types, according to the stratigraphic lithology classification results of the respective pixel points, count the number of pixel points with corresponding stratigraphic lithology label values among all the pixel points, and use the ratio of this number to the total number of pixel points of all the pixel points as the proportion of the corresponding pixel points among all the pixel points.

8. The recognition method according to claim 6, characterized in that, The landslide event occurrence prediction model is pre-trained in the following manner: Obtain multiple negative sample data and multiple positive sample data corresponding one by one to multiple historical landslide events. Among them, the negative sample data includes a first model input item and a first model output item with a value of 0. The first model input item includes the landslide body connected domain area of a first known landslide area, the average precipitation in K + 1 consecutive historical unit time periods when no landslide event occurs, the proportions of multiple types of pixel points among all the pixel points located in the first known landslide area, and the average vegetation coverage label value, average distance value, average elevation value, average slope value, and average aspect value of all the pixel points located in the first known landslide area. The positive sample data includes a second model input item and a second model output item with a value of 1. The second model input item includes the landslide body connected domain area of a second known landslide area, the average precipitation in the K historical unit time periods closest to the occurrence of the corresponding landslide event and the historical unit time period when the corresponding landslide event occurs, the proportions of multiple types of pixel points among all the pixel points located in the second known landslide area, and the average vegetation coverage label value, average distance value, average elevation value, average slope value, and average aspect value of all the pixel points located in the second known landslide area; Apply the multiple negative sample data and the multiple positive sample data to perform calibration verification modeling on a machine learning model based on the fifth artificial intelligence algorithm to obtain the landslide event occurrence prediction model, where the confidence level of the landslide event occurrence prediction model with an output value of 1 is used as the probability value of a landslide event occurring in the next unit period.

9. A geological landslide identification system, characterized in that, It includes an image data acquisition unit, a stratigraphic lithology classification unit, a vegetation coverage recognition unit, a surface water body recognition unit, a near-water distance calculation unit, a digital model generation unit, a distribution image acquisition unit, a landslide body recognition unit, a binary image generation unit, an edge contour extraction unit, a connected domain area calculation unit, and a landslide area determination unit; The image data acquisition unit is configured to acquire remote sensing images and lidar scan data of the area where the target slope is located; The formation lithology classification unit is communicatively connected to the image data acquisition unit, and is configured to, for each pixel point in the remote sensing image, import the corresponding pixel value and the pixel values of the surrounding pixel points together into a formation lithology classification model pre-trained based on a first artificial intelligence algorithm, and output a corresponding formation lithology classification result, where the formation lithology classification result is used to indicate a certain formation lithology label value among a plurality of formation lithology label values corresponding to the corresponding pixel point, and the plurality of formation lithology label values correspond to multiple formation types one by one; The vegetation coverage identification unit is communicatively connected to the image data acquisition unit, and is configured to, for each pixel point, import the corresponding pixel value and the pixel values of the surrounding pixel points together into a vegetation coverage identification model pre-trained based on a second artificial intelligence algorithm, and output a corresponding vegetation coverage identification result, where the vegetation coverage identification result is used to indicate the vegetation coverage label value of the corresponding pixel point, and the vegetation coverage label value uses the numerical value 1 to represent being covered by vegetation and the numerical value 0 to represent not being covered by vegetation; The surface water body identification unit is communicatively connected to the image data acquisition unit, and is configured to, for each pixel point, import the corresponding pixel value and the pixel values of the surrounding pixel points together into a surface water body identification model pre-trained based on a third artificial intelligence algorithm, and output a corresponding surface water body identification result, where the surface water body identification result is used to indicate whether the corresponding pixel point is a water body pixel point; The near-water distance calculation unit is communicatively connected to the surface water body identification unit, and is configured to, for each target pixel point in the remote sensing image and located within the target slope, determine the water body pixel point closest to the corresponding pixel point according to the surface water body identification result of each pixel point, and calculate the distance value between the water body pixel point and the corresponding pixel point; The digital model generation unit is communicatively connected to the image data acquisition unit, and is configured to generate a digital elevation model of the area where the target slope is located according to the lidar scan data; The distribution image acquisition unit is communicatively connected to the digital model generation unit, and is configured to acquire an elevation distribution image, a slope distribution image, and an aspect distribution image of the area where the target slope is located according to the digital elevation model, where the pixel value of each pixel point in the elevation distribution image is an elevation value, the pixel value of each pixel point in the slope distribution image is a slope value, and the pixel value of each pixel point in the aspect distribution image is an aspect value; The landslide body identification unit is respectively communicatively connected to the formation lithology classification unit, the vegetation coverage identification unit, the near-water distance calculation unit, and the distribution image acquisition unit. For each target pixel point, it imports the corresponding formation lithology label value, vegetation coverage label value, distance value, elevation value, slope value, and aspect value, as well as the formation lithology label value, vegetation coverage label value, distance value, elevation value, slope value, and aspect value of the corresponding surrounding target pixel points into a landslide body identification model pre-trained based on the fourth artificial intelligence algorithm, and outputs the corresponding landslide body identification result. Among them, the landslide body identification result is used to indicate whether the corresponding pixel point is a landslide body pixel point; The binary image generation unit is communicatively connected to the landslide body identification unit, and is used to generate a binary image of the landslide body distribution of the target slope according to the landslide body identification results of the respective target pixel points. Among them, the binary image of the landslide body distribution uses a first pixel value to represent that the corresponding pixel point is a landslide body pixel point, and uses a second pixel value to represent that the corresponding pixel point is not a landslide body pixel point; The edge contour extraction unit is communicatively connected to the binary image generation unit, and is used to perform edge contour extraction processing on the landslide body connected domain in the binary image of the landslide body distribution to obtain the edge contour of the landslide body connected domain; The connected domain area calculation unit is communicatively connected to the edge contour extraction unit, and is used to calculate the area of the landslide body connected domain according to the edge contour of the landslide body connected domain; The landslide area determination unit is respectively communicatively connected to the edge contour extraction unit and the connected domain area calculation unit. If the area of the landslide body connected domain reaches a preset area threshold, it takes the landslide area surrounded by the edge contour of the landslide body connected domain as the geological landslide identification result.

10. The recognition system according to claim 9, characterized in that, It further includes a precipitation acquisition unit, a precipitation averaging unit, a lithology proportion statistics unit, a landslide event prediction unit, and an early warning action execution unit; The precipitation acquisition unit is communicatively connected to the landslide area determination unit, and is used to obtain the precipitation distribution images of the area where the target slope is located in the current nearest K historical unit time periods and the current next future unit time period after taking the landslide area defined by the edge contour of the landslide body connected domain as the geological landslide identification result. Among them, the pixel value of each pixel point in the precipitation distribution image is precipitation, and K represents a positive integer; The precipitation averaging unit is communicatively connected to the precipitation acquisition unit, and is used to statistically obtain the average precipitation of all pixel points located in the landslide area in the current nearest multiple historical unit time periods and the current next future unit time period according to the precipitation distribution image; The lithology proportion statistics unit is communicatively connected to the formation lithology classification unit, and is used to statistically obtain the proportion of multiple pixel points among all pixel points according to the formation lithology classification results of the respective pixel points. Among them, the multiple pixel points correspond one by one to multiple formation types; The landslide event prediction unit is respectively communicatively connected to the connected domain area calculation unit, the average precipitation equalization unit, the lithology proportion statistics unit, the vegetation coverage identification unit, the distance to water calculation unit, and the distribution image acquisition unit, and is used to import the landslide body connected domain area, the average precipitation, the proportion of various pixel points, and the average vegetation coverage label value, average distance value, average elevation value, average slope value, and average aspect value of all pixel points into a landslide event occurrence prediction model pre-trained based on the fifth artificial intelligence algorithm, and output the probability value of a landslide event occurring in the next future unit time period at the current time; The warning action execution unit is communicatively connected to the landslide event prediction unit, and is used to trigger and execute a landslide event warning action for the landslide area if the probability value reaches a preset probability threshold.

Images