A landslide early warning method, system, device and storage medium

Abstract

The present disclosure belongs to the technical field of mountain exploration and monitoring, and particularly relates to a mountain landslide early warning method, system, device and storage medium. The method comprises: collecting a mountain image with vegetation removed, establishing a mountain model; and obtaining a crack probability map according to a pre-trained bare land digital elevation model; selecting a mountain model with the same angle as the pre-trained bare land digital elevation model to obtain a bare land mask; combining the crack probability map and the bare land mask to generate a crack skeleton map; extracting crack feature parameters in the crack skeleton map and judging a crack type; and outputting mountain landslide early warning information based on the crack type. The present disclosure mainly uses image recognition technology to identify rock-soil cracks appearing before a mountain landslide, further judges whether there is a risk of a mountain landslide, and has higher recognition accuracy.

Description

Technical Field

[0001] This disclosure belongs to the field of mountain exploration and monitoring technology, and specifically relates to a method, system, equipment and storage medium for early warning of landslides. Background Technology

[0002] Landslide development generally occurs in three stages: creep deformation, rapid sliding, and gradual stabilization. Before a landslide, a series of unusual phenomena emerge, such as soil and rock fissures, collapses, and abnormal water levels. Therefore, monitoring potentially hazardous mountains is crucial for landslide prediction. Soil and rock fissures, occurring on the mountain surface, are easier to observe than other features and can serve as a primary predictive characteristic for landslides. The shape and orientation of fissures differ at different stages and are accompanied by unusual phenomena, such as bulging fissures at their leading edges accompanied by seepage and turbidity. Therefore, image recognition methods can be used to identify fissures, thereby detecting their formation and penetration, and reducing the impact of landslides. The most important aspect of image recognition is identifying fissure features, including fissure width, fissure centerline position, fissure centerline extension direction, fissure shape, and fissure verticality. Secondary features include the smoothness of the soil surface at the fissure and the presence of seepage. Neural networks are then used to analyze these features to determine whether a cutting surface has formed. Existing technologies, such as CN104299365A, CN117854256A, and CN111006593A, disclose methods for monitoring abnormal changes in mountains. The features identified are mainly focused on identifying sudden abnormal changes in mountain features, such as the rolling of sand and gravel, but they cannot directly detect whether a landslide has occurred. Summary of the Invention

[0003] To address the above problems, this disclosure provides a landslide early warning method, the method comprising:

[0004] Collect images of the mountain after removing vegetation cover, and build a mountain model; and,

[0005] Based on the pre-trained bare ground digital elevation model, a crack probability map is obtained;

[0006] Select a mountain model with the same angle as the pre-trained bare ground digital elevation model to obtain a bare ground mask;

[0007] The crack probability map and the bare ground mask are combined to generate a crack skeleton map;

[0008] Extract crack feature parameters from the crack skeleton map and determine the crack type;

[0009] Output landslide early warning information based on crack type.

[0010] According to some embodiments of this disclosure, acquiring images of mountains with vegetation cover removed and establishing a mountain model includes:

[0011] Obtain the diffusion model; the diffusion model is a noise-adding function;

[0012] The diffusion model is trained using the U-Net network structure to obtain a pre-trained diffusion model;

[0013] The mountain image with vegetation removed is input into a pre-trained diffusion model to obtain the mountain model.

[0014] According to some embodiments of this disclosure, the U-Net network structure includes: an encoder with a downsampling path, a decoder with an upsampling path, and an output layer;

[0015] The encoder includes convolutional layers and pooling layers;

[0016] The decoder includes an upsampling layer and a convolutional layer;

[0017] The output layer includes a convolutional layer.

[0018] According to some embodiments of this disclosure, training a diffusion model using a U-Net network structure includes:

[0019] Collect sample images; the sample images include vegetation cover images and bare land images of the same mountain area;

[0020] The sample image is gradually denoised using a noise-adding function to obtain a noisy image.

[0021] The bare ground image is obtained from the noisy image using a reverse process.

[0022] According to some embodiments of this disclosure, the noise addition function expression is as follows:

[0023] ;

[0024] Where t is the number of steps, T is the length of the step; Let be a monotonically decreasing function at step t; It is pure Gaussian noise with a mean of 0 and a variance of I; Let be the weight at step t; This represents vegetation pattern residuals.

[0025] According to some embodiments of this disclosure, a bare land image is obtained from a noisy image using a reverse process, including:

[0026] Using a pre-trained diffusion model, from noisy to the vegetation cover image at step T. Initially, noise and vegetation pattern residuals added during the forward pass are predicted and removed progressively. After T iterations, the bare land image is finally reconstructed.

[0027] According to some embodiments of this disclosure, the expression for obtaining the bare land image from the noisy image using the reverse process is as follows:

[0028] ;

[0029] in, Let be the scheduling variance at step t; The learning parameters for training the network are: At that time, the Gaussian noise map output by the training network; The learning parameters for training the network are: At that time, the residual of the vegetation pattern map output by the training network; Let z be the scheduling variance, and z be a random number. Because the signal is during the forward process Reduce, denominator This means amplifying the signal back to its original scale.

[0030] According to some embodiments of this disclosure, a crack probability map is obtained based on a pre-trained bare ground digital elevation model, including:

[0031] Obtain a digital elevation model of bare land without vegetation cover;

[0032] The bare land digital elevation model is trained by using a loss function to obtain a pre-trained bare land digital elevation model.

[0033] Obtain the actual bare land digital elevation model image and input it into the pre-trained bare land digital elevation model to obtain the crack probability map.

[0034] According to some embodiments of this disclosure, a crack probability map is combined with a bare ground mask to obtain a crack skeleton map, including:

[0035] By comparing the crack probability map with the bare ground mask pixel by pixel, a probability map showing the presence of cracks only in the bare ground is obtained.

[0036] Binarize the probability map of cracks existing only in bare land to obtain a binary crack map;

[0037] The binary crack map is refined to obtain the crack skeleton map.

[0038] According to some embodiments of this disclosure, the pre-trained bare ground digital elevation model is a full-scale jump connection network;

[0039] The full-scale skip connection network is a multi-layer lightweight U-Net network.

[0040] According to some embodiments of this disclosure, the loss function expression is as follows:

[0041]

[0042] Where LBCE is the binary cross-entropy function, and LDice is the Dice coefficient loss. , As weight.

[0043] According to some embodiments of this disclosure, the crack characteristic parameters include at least one of crack length, average width, orientation, aspect ratio, radius of curvature, extension direction, and crack density.

[0044] According to some embodiments of this disclosure, the crack types include trailing edge tensile cracks, flank shear cracks, and leading edge bulging cracks.

[0045] According to some embodiments of this disclosure, crack feature parameters are extracted from the crack skeleton diagram, and the crack type is determined, including:

[0046] The extracted crack feature parameters are compared with the pre-stored crack feature-type mapping table to obtain the crack type of the current crack.

[0047] According to some embodiments of this disclosure, landslide early warning information is output based on crack type, including:

[0048] Based on the pre-existing correspondence between crack type and landslide stage, determine the landslide development stage indicated by the crack type;

[0049] Based on the development stage of the landslide, corresponding warning information is generated and output.

[0050] According to some embodiments of this disclosure, the landslide development stage indicated by the crack type is determined based on a pre-stored correspondence between crack type and landslide stage, including:

[0051] When the crack type is a trailing edge tensile crack, the landslide stage is the creep deformation stage;

[0052] When the crack type is lateral shear crack, the landslide stage is the rapid sliding stage; and...

[0053] When the crack type is a leading-edge bulging crack, the landslide stage is a gradually stabilizing stage.

[0054] According to some embodiments of this disclosure, the levels of the warning information include:

[0055] A primary warning corresponding to the creeping deformation stage;

[0056] A medium-level warning corresponding to the gradual stabilization phase; and,

[0057] Advanced warning for the phase of rapid sliding.

[0058] This disclosure also provides a landslide early warning system, the system comprising:

[0059] The mountain model module is used to acquire images of mountains after removing vegetation cover and to build a mountain model; and,

[0060] The crack probability map module is used to obtain a crack probability map based on a pre-trained bare ground digital elevation model.

[0061] The bare ground mask module is used to select a mountain model with the same angle as the pre-trained bare ground digital elevation model to obtain a bare ground mask;

[0062] The crack skeleton map module is used to combine the crack probability map and the bare ground mask to generate a crack skeleton map.

[0063] The judgment module is used to extract crack feature parameters from the crack skeleton map and determine the crack type;

[0064] The early warning module is used to output landslide early warning information based on crack type.

[0065] According to some embodiments of this disclosure, a mountain model module is used to acquire mountain images with vegetation cover removed and to establish a mountain model, including:

[0066] Obtain the diffusion model; the diffusion model is a noise-adding function;

[0067] The diffusion model is trained using the U-Net network structure to obtain a pre-trained diffusion model;

[0068] The mountain image with vegetation removed is input into a pre-trained diffusion model to obtain the mountain model.

[0069] According to some embodiments of this disclosure, a crack probability map module is used to obtain a crack probability map based on a pre-trained bare ground digital elevation model, including:

[0070] The crack probability map module is used to obtain a digital elevation model of bare land without vegetation cover;

[0071] The bare land digital elevation model is trained by using a loss function to obtain a pre-trained bare land digital elevation model.

[0072] Obtain the actual bare land digital elevation model image and input it into the pre-trained bare land digital elevation model to obtain the crack probability map.

[0073] According to some embodiments of this disclosure, a crack skeleton map module is used to combine a crack probability map and a bare ground mask to generate a crack skeleton map, including:

[0074] The crack skeleton map module is used to compare the crack probability map with the bare ground mask pixel by pixel to obtain the probability map that cracks exist only in the bare ground.

[0075] Binarize the probability map of cracks existing only in bare land to obtain a binary crack map;

[0076] The binary crack map is refined to obtain the crack skeleton map.

[0077] This disclosure also provides an electronic device, including:

[0078] Processor and memory;

[0079] The processor invokes the computer program stored in the memory to execute the landslide early warning method.

[0080] This disclosure also provides a computer-readable storage medium,

[0081] The computer-readable storage medium stores a computer program that, when executed by a processor, enables the processor to perform the landslide early warning method.

[0082] This disclosure has the following beneficial effects:

[0083] This disclosure establishes a network architecture for identifying cracks before landslides. By obtaining features such as crack length, average width, orientation, aspect ratio, radius of curvature, extension direction, and crack density, the crack type is determined, and the stage of the landslide is further identified, providing technical support for subsequent prevention work. The accuracy of landslide identification through cracks is relatively high.

[0084] Other features and advantages of this disclosure will be set forth in the following description and will be apparent in part from the description or may be learned by practicing the disclosure. The objects and other advantages of this disclosure may be realized and obtained by means of the structures pointed out in the description and the accompanying drawings. Attached Figure Description

[0085] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0086] Figure 1 A diagram illustrating the landslide early warning method in an embodiment of this disclosure is shown.

[0087] Figure 2 A diagram of a landslide early warning system in an embodiment of this disclosure is shown;

[0088] Figure 3 A diagram of an electronic device according to an embodiment of this disclosure is shown. Detailed Implementation

[0089] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this disclosure more comprehensive and complete, and to fully convey the concept of the example embodiments to those skilled in the art. The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced with one or more of the specific details omitted, or other methods, components, apparatus, steps, etc., can be employed. In other instances, well-known technical solutions are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0090] Furthermore, the accompanying drawings are merely illustrative of this disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware units or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0091] The flowchart shown in the attached diagram is merely an illustrative example and does not necessarily include all steps. For example, some steps may be broken down, while others may be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0092] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein.

[0093] Furthermore, the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion, such that a process, method, system, product, or device that includes a series of steps or sub-modules is not necessarily limited to those steps or sub-modules that are explicitly listed, but may include other steps or sub-modules that are not explicitly listed or that are inherent to such process, method, product, or device.

[0094] This disclosure mainly utilizes image recognition technology to identify rock and soil cracks that appear before a landslide, further determining whether there is a risk of a landslide, with higher recognition accuracy.

[0095] like Figure 1 As shown, this disclosure provides a method for early warning of landslides, the method comprising:

[0096] Collect images of the mountain after removing vegetation cover, and build a mountain model; and,

[0097] Based on the pre-trained bare ground digital elevation model, a crack probability map is obtained;

[0098] Select a mountain model with the same angle as the pre-trained bare ground digital elevation model to obtain a bare ground mask;

[0099] The crack probability map and the bare ground mask are combined to generate a crack skeleton map;

[0100] Extract crack feature parameters from the crack skeleton map and determine the crack type;

[0101] Output landslide early warning information based on crack type.

[0102] Specifically, the process involves acquiring images of the mountain after vegetation cover has been removed, and then building a mountain model, including:

[0103] Obtain the diffusion model; the diffusion model is a noise-adding function;

[0104] The diffusion model is trained using the U-Net network structure to obtain a pre-trained diffusion model;

[0105] The mountain image with vegetation removed is input into a pre-trained diffusion model to obtain the mountain model.

[0106] Specifically, the U-Net network structure includes: an encoder with a downsampling path, a decoder with an upsampling path, and an output layer;

[0107] The encoder includes convolutional layers and pooling layers;

[0108] The decoder includes an upsampling layer and a convolutional layer;

[0109] The output layer includes a convolutional layer.

[0110] Specifically, the diffusion model is trained using the U-Net network structure, including:

[0111] Collect sample images; the sample images include vegetation cover images and bare land images of the same mountain area;

[0112] The sample image is gradually denoised using a noise-adding function to obtain a noisy image.

[0113] The bare ground image is obtained from the noisy image using a reverse process.

[0114] Specifically, the expression for the noise-adding function is:

[0115] ;

[0116] Where t is the number of steps, T is the length of the step; Let be a monotonically decreasing function at step t; It is pure Gaussian noise with a mean of 0 and a variance of I; Let be the weight at step t; This represents vegetation pattern residuals.

[0117] Specifically, the process of obtaining a bare ground image from a noisy image using a reverse process includes:

[0118] Using a pre-trained diffusion model, from noisy to the vegetation cover image at step T. Initially, noise and vegetation pattern residuals added during the forward pass are predicted and removed progressively. After T iterations, the bare land image is finally reconstructed.

[0119] Specifically, the expression for obtaining the bare land image from the noisy image using the reverse process is as follows:

[0120] ;

[0121] in, Let be the scheduling variance at step t; The learning parameters for training the network are: At that time, the Gaussian noise map output by the training network; The learning parameters for training the network are: At that time, the residual of the vegetation pattern map output by the training network; Let z be the scheduling variance, and z be a random number. Because the signal is during the forward process Reduce, denominator This means amplifying the signal back to its original scale.

[0122] Specifically, based on the pre-trained bare ground digital elevation model, a crack probability map is obtained, including:

[0123] Obtain a digital elevation model of bare land without vegetation cover;

[0124] The bare land digital elevation model is trained by using a loss function to obtain a pre-trained bare land digital elevation model.

[0125] Obtain the actual bare land digital elevation model image and input it into the pre-trained bare land digital elevation model to obtain the crack probability map.

[0126] Specifically, the crack probability map and the bare ground mask are combined to obtain a crack skeleton map, including:

[0127] By comparing the crack probability map with the bare ground mask pixel by pixel, a probability map showing the presence of cracks only in the bare ground is obtained.

[0128] Binarize the probability map of cracks existing only in bare land to obtain a binary crack map;

[0129] The binary crack map is refined to obtain the crack skeleton map.

[0130] Specifically, the pre-trained bare ground digital elevation model is a full-scale jump connection network;

[0131] The full-scale skip connection network is a multi-layer lightweight U-Net network.

[0132] Specifically, the loss function expression is as follows:

[0133]

[0134] in, is the loss function; LBCE is the binary cross-entropy function; LDice is the Dice coefficient loss; , As weight.

[0135] Specifically, the crack characteristic parameters include at least one of the following: crack length, average width, orientation, aspect ratio, radius of curvature, extension direction, and crack density.

[0136] Specifically, the crack types include trailing edge tensile cracks, flank shear cracks, and leading edge bulging cracks.

[0137] Specifically, crack feature parameters are extracted from the crack skeleton map, and the crack type is determined, including:

[0138] The extracted crack feature parameters are compared with the pre-stored crack feature-type mapping table to obtain the crack type of the current crack.

[0139] Specifically, landslide early warning information is output based on crack type, including:

[0140] Based on the pre-existing correspondence between crack type and landslide stage, determine the landslide development stage indicated by the crack type;

[0141] Based on the development stage of the landslide, corresponding warning information is generated and output.

[0142] Specifically, based on the pre-existing correspondence between crack types and landslide stages, the landslide development stage indicated by the crack type is determined, including:

[0143] When the crack type is a trailing edge tensile crack, the landslide stage is the creep deformation stage;

[0144] When the crack type is lateral shear crack, the landslide stage is the rapid sliding stage; and...

[0145] When the crack type is a leading-edge bulging crack, the landslide stage is a gradually stabilizing stage.

[0146] Specifically, the levels of the warning information include:

[0147] A primary warning corresponding to the creeping deformation stage;

[0148] A medium-level warning corresponding to the gradual stabilization phase; and,

[0149] Advanced warning for the phase of rapid sliding.

[0150] The detailed process of this method is as follows:

[0151] Step S100: Remove mountain vegetation using a diffusion model (i.e., a noise-adding function) to establish a multi-angle mountain model;

[0152] In step S100, a mountain model is established using a U-Net network structure, which includes an encoder (downsampling path), a decoder (upsampling path), and an output layer.

[0153] The encoder includes convolutional layers and pooling layers; the decoder includes upsampling layers and convolutional layers; the output layer is... Convolutional layer.

[0154] The encoder consists of multiple convolutional blocks, each typically containing two convolutional layers. Each convolutional layer is followed by an activation function (such as ReLU) and batch normalization. The pooling layers use max pooling to halve the spatial dimension of the downsampled feature map while increasing the number of channels in the feature map.

[0155] The decoder gradually restores the spatial resolution of the feature map through upsampling layers, which is achieved through transpose convolution. Each upsampling concatenates the feature map of the corresponding encoder layer with the feature map of the decoder. After each upsampling step, a convolutional layer is set to further process the concatenated feature map.

[0156] Output layer The convolutional layer generates an output feature map of the same size as the input image.

[0157] Within the framework of the diffusion model (i.e., the noise function), U-Net is used as the denoising network for training to improve the modeling accuracy of the mountain model. During training, vegetation is treated as noise, and its impact on terrain data is simulated through a forward diffusion process to construct a diffusion model. Using the trained diffusion model, the real terrain without vegetation interference is inversely recovered from the vegetation-contaminated terrain data. The training process specifically includes:

[0158] Step S101: Collect vegetation cover images and bare land images of the same mountain area to obtain a sample library;

[0159] Step S102: Establish a vegetation-bare ground elevation coupling noise function of length T. Noise addition is performed step by step, noise addition function The expression is:

[0160] ;

[0161] Where t is the number of steps, T is the length of the step; Let be a monotonically decreasing function at step t; It is pure Gaussian noise with a mean of 0 and a variance of I; Let be the weight at step t; The vegetation pattern residual is a fixed sample pre-extracted from an external vegetation bank. , This is a bare ground image. Image of vegetation cover;

[0162] Step S103: Given any newly measured vegetation cover image ,from Starting point, gradually generating This led to the final vegetation weight While terrain details are preserved. Stopping at t=1 yields... Theoretically That is, vegetation height deviation It has been reduced at each level.

[0163] The specific process of step S103 is as follows:

[0164] Step S1031, initial pure noise, take ;

[0165] Step S1032: Perform noise reduction and update gradually.

[0166] ;

[0167] in, For scheduling variance, The learning parameters for training the network are: At that time, the Gaussian noise map output by the training network, The learning parameters for training the network are: At that time, the residual of the vegetation pattern map output by the training network; Let z be the scheduling variance, and z be a random number. Because the signal is during the forward process Reduce, denominator This means amplifying the signal back to its original scale.

[0168] The loss function L in the training phase of step S100 is:

[0169] ;

[0170] in, for gradient plot; for gradient plot; , , These are the weights; This represents the Gaussian noise map added during the forward pass; Image of bare ground; This is the image with added noise at t=0.

[0171] The neural network is trained in step S100. An image of the mountain model to be built is input, and the mountain model is constructed. A multi-angle mountain model is created for later use in determining the type of cracks.

[0172] Rock and soil fissures generally form cutting surfaces in mountains and subsequently cause landslides when the following three conditions are met. Each type of fissure has unique characteristics:

[0173] (1) Tensile cracks at the rear edge: an arc-shaped tensile crack appears at the top of the slope, and the crack gradually widens and deepens, possibly accompanied by misalignment (vertical displacement), indicating that the cutting surface at the rear edge is forming.

[0174] (2) Lateral shear cracks: Equestrian or feather-like shear cracks appear on both sides of the slope, extending in a direction parallel to the landslide direction, indicating the development of lateral cutting surfaces.

[0175] (3) Bulging cracks at the leading edge: Bulging cracks or radial cracks appear at the toe of the slope, which may be due to the leading edge of the sliding body being squeezed and bulging, indicating that the sliding surface has been sheared out to the leading edge.

[0176] By identifying cracks through image recognition and comparing them with a mountain model, characteristics such as the direction of the cracks can be obtained, thus determining what type of crack it is.

[0177] Step S200: Obtain the crack features in the crack image, compare them with the mountain model, and determine the crack features.

[0178] In step S200, the method for obtaining crack features in the crack image is as follows:

[0179] Step S201: Obtain the DEM of bare land without vegetation cover;

[0180] The bare land DEM mentioned above is a special form of Digital Elevation Model (DEM), referring to raster data that reflects only the elevation of exposed ground after removing all surface vegetation, buildings, and other artificial or biological cover. It is the core foundational layer for tasks such as landslide crack identification and terrain analysis. In step S201, ground point clouds are obtained by recording the multiple echoes from a lidar penetrating the tree canopy to generate a bare land DEM without vegetation obstruction.

[0181] Step S202: Obtain a crack probability map using crack information from the bare land DEM image;

[0182] Obtaining the crack probability map requires a full-scale skip connection network based on a 3-layer lightweight U-Net. The 3-layer U-Net network structure includes three encoders (level-0, level-1, level-2, etc.) and three decoders (level-2', level-1', level-0', etc.). The network structure is trained by skipping decoder level-0', encoder level-1 skipping decoder level-1', and encoder level-2 skipping decoder level-2'. The training process includes: running four tools (Hillshade, Slope, Positive Openness, and Sky View Factor) on bare land DEM images in QGIS (Quantum Geographic Information System) to output bare land shadow maps, slope maps, positive openness maps, and sky view maps at the same resolution, and then stitching them together; loss function... for:

[0183] ;

[0184] Where LBCE is the binary cross-entropy function; LDice is the Dice coefficient loss; , As weight;

[0185] ;

[0186] ;

[0187] Where y is the true label pixel value, taking either 0 (background) or 1 (crack); p is the crack probability output by the model, ranging from... .

[0188] In step S202, the crack probability map can be obtained by substituting the trained model with the actual bare land DEM image.

[0189] Step S203: Select a mountain model with the same angle as the bare ground DEM to obtain a bare ground mask;

[0190] The bare ground mask is a binary raster layer;

[0191] Step S204: Combine the crack probability map and the bare ground mask to zero out the non-bare ground area and skeletonize it into single-pixel lines to obtain the skeleton map; the specific method is as follows:

[0192] Step S2041: Compare the crack probability map with the bare ground mask pixel by pixel. Set the probability map to 0 for positions where the bare ground mask is 0; keep the original value for positions where the bare ground mask is 1; thus obtaining the probability map of cracks existing only in bare ground.

[0193] Step S2042: Set a threshold (e.g., 0.5) for the probability map of cracks only in bare land. Set the value to 1 if it is greater than the threshold, and set the value to 0 otherwise to form a binary crack map.

[0194] Step S2043: The binary image is processed using a thinning algorithm to retain the crack centerline and compress the width to a single pixel to obtain the skeleton image.

[0195] Step S205: Perform connected component segmentation on the skeleton graph and mark each crack;

[0196] Step S206: Obtain the length, average width, orientation, aspect ratio, radius of curvature, extension direction, and crack density characteristics of each crack.

[0197] Step S207: Determine the crack type using Table 1.

[0198] Table 1 Crack Judgment Table

[0199]

[0200] By correlating the identified crack types with landslide stages, the development stage of the landslide is determined, and corresponding early warning information is output based on the development stage, for example:

[0201] When the crack type is a trailing edge tensile crack and the landslide stage is a creep deformation stage, a primary early warning message is output.

[0202] When the crack type is a lateral shear crack and the landslide stage is a rapid sliding stage, output advanced early warning information;

[0203] When the crack type is a leading-edge bulging crack and the landslide stage is a gradually stabilizing stage, a medium-level early warning information is output.

[0204] like Figure 2 The present disclosure also provides a landslide early warning system, the system comprising:

[0205] The mountain model module is used to acquire images of mountains after removing vegetation cover and to build a mountain model; and,

[0206] The crack probability map module is used to obtain a crack probability map based on a pre-trained bare ground digital elevation model.

[0207] The bare ground mask module is used to select a mountain model with the same angle as the pre-trained bare ground digital elevation model to obtain a bare ground mask;

[0208] The crack skeleton map module is used to combine the crack probability map and the bare ground mask to generate a crack skeleton map.

[0209] The judgment module is used to extract crack feature parameters from the crack skeleton map and determine the crack type;

[0210] The early warning module is used to output landslide early warning information based on crack type.

[0211] Specifically, the mountain model module is used to acquire mountain images after vegetation cover has been removed and to build a mountain model, including:

[0212] Obtain the diffusion model; the diffusion model is a noise-adding function;

[0213] The diffusion model is trained using the U-Net network structure to obtain a pre-trained diffusion model;

[0214] The mountain image with vegetation removed is input into a pre-trained diffusion model to obtain the mountain model.

[0215] Specifically, the crack probability map module is used to obtain a crack probability map based on a pre-trained bare ground digital elevation model, including:

[0216] The crack probability map module is used to obtain a digital elevation model of bare land without vegetation cover;

[0217] The bare land digital elevation model is trained by using a loss function to obtain a pre-trained bare land digital elevation model.

[0218] Obtain the actual bare land digital elevation model image and input it into the pre-trained bare land digital elevation model to obtain the crack probability map.

[0219] Specifically, the crack skeleton map module is used to combine the crack probability map and the bare ground mask to generate a crack skeleton map, including:

[0220] The crack skeleton map module is used to compare the crack probability map with the bare ground mask pixel by pixel to obtain the probability map that cracks exist only in the bare ground.

[0221] Binarize the probability map of cracks existing only in bare land to obtain a binary crack map;

[0222] The binary crack map is refined to obtain the crack skeleton map.

[0223] like Figure 3 As shown, corresponding to the landslide early warning method provided above, this disclosure also provides an electronic device. Since the embodiment of this device is similar to the embodiment of the method described above, the description is relatively simple. For relevant details, please refer to the description in the method embodiment section above. The device described below is merely illustrative. This device may include: a processor 1, a memory 2, a communication bus (i.e., the aforementioned device bus), and a lookup engine. The processor 1 and the memory 2 communicate with each other through the communication bus and communicate with external systems through a communication interface. The processor 1 can call logical instructions in the memory 2 to execute the landslide early warning method.

[0224] Furthermore, the logical instructions in the aforementioned memory 2 can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this disclosure. The aforementioned storage medium includes various media capable of storing program code, such as memory chips, USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0225] On the other hand, this disclosure also provides a processor-readable storage medium storing a computer program 3, which, when executed by a processor 1, is implemented to perform the landslide early warning method provided in the above embodiments.

[0226] The processor-readable storage medium can be any available medium or data storage device that the processor 1 can access, including but not limited to magnetic memory (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO)), optical memory (e.g., CD, DVD, BD, HVD), and semiconductor memory (e.g., ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)).

[0227] Embodiments of this disclosure also provide a computer program product comprising a computer program that includes computer program code means stored on a computer-readable medium or carrier wave, the computer program code means being configured to cause a computer or processor to control the execution of steps of a method according to any embodiment of this disclosure.

[0228] Those skilled in the art should understand that, despite the detailed description of this disclosure with reference to the foregoing embodiments, modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure.

Claims

1. A method for early warning of landslides, characterized in that, The method includes: Acquiring images of mountains with vegetation removed and establishing a mountain model includes: obtaining a diffusion model; the diffusion model is a noise-adding function; training the diffusion model using a U-Net network structure to obtain a pre-trained diffusion model; and inputting the images of mountains with vegetation removed into the pre-trained diffusion model to obtain the mountain model. The process of training a diffusion model using the U-Net network structure includes: collecting sample images; the sample images include vegetation cover images and bare land images of the same mountain area; progressively adding noise to the sample images using a noise-adding function to obtain noisy images; and retrieving bare land images from the noisy images using a reverse process, including: using the pre-trained diffusion model to process the vegetation cover images from the noisy images up to the T-th step. Initially, noise and vegetation pattern residuals added during the forward pass are predicted and removed progressively. After T iterations, the bare land image is finally reconstructed; and, Based on the pre-trained bare ground digital elevation model, a crack probability map is obtained; Select a mountain model with the same angle as the pre-trained bare ground digital elevation model to obtain a bare ground mask; The crack probability map and the bare ground mask are combined to generate a crack skeleton map, including: comparing the crack probability map and the bare ground mask pixel by pixel to obtain a crack probability map where only bare ground exists; performing binarization on the crack probability map where only bare ground exists to obtain a binary crack map; and refining the binary crack map to obtain the crack skeleton map. Extract crack feature parameters from the crack skeleton map and determine the crack type; The method for outputting landslide early warning information based on crack type includes: determining the landslide development stage indicated by the crack type according to the pre-stored crack type-landslide stage correspondence; and generating and outputting early warning information of the corresponding level according to the landslide development stage.

2. The landslide early warning method according to claim 1, characterized in that, The U-Net network structure includes: an encoder with a downsampling path, a decoder with an upsampling path, and an output layer; The encoder includes convolutional layers and pooling layers; The decoder includes an upsampling layer and a convolutional layer; The output layer includes a convolutional layer.

3. The landslide early warning method according to claim 1, characterized in that, The expression for the noise-adding function is: ； Where t is the number of steps, T is the length of the step; Let be a monotonically decreasing function at step t; It is pure Gaussian noise with a mean of 0 and a variance of I; Let be the weight at step t; This represents vegetation pattern residuals.

4. The landslide early warning method according to claim 1, characterized in that, The expression for obtaining the bare land image from the noisy image using the reverse process is as follows: ； in, Let be the scheduling variance at step t; The learning parameters for training the network are: At that time, the Gaussian noise map output by the training network; The learning parameters for training the network are: At that time, the residual of the vegetation pattern map output by the training network; Let z be the scheduling variance, and z be a random number. Because the signal is during the forward process Reduce, denominator This means amplifying the signal back to its original scale.

5. The landslide early warning method according to claim 1, characterized in that, Based on the pre-trained bare ground digital elevation model, a crack probability map is obtained, including: Obtain a digital elevation model of bare land without vegetation cover; The bare land digital elevation model is trained by using a loss function to obtain a pre-trained bare land digital elevation model. Obtain the actual bare land digital elevation model image and input it into the pre-trained bare land digital elevation model to obtain the crack probability map.

6. The landslide early warning method according to claim 1, characterized in that, The pre-trained bare ground digital elevation model is a full-scale jump connection network; The full-scale skip connection network is a multi-layer lightweight U-Net network.

7. The landslide early warning method according to claim 5, characterized in that, The loss function expression is as follows: in, is the loss function; LBCE is the binary cross-entropy function; LDice is the Dice coefficient loss; , As weight.

8. The landslide early warning method according to claim 1, characterized in that, The crack characteristic parameters include at least one of the following: crack length, average width, orientation, aspect ratio, radius of curvature, extension direction, and crack density.

9. The landslide early warning method according to claim 1, characterized in that, The crack types include trailing edge tensile cracks, flank shear cracks, and leading edge bulging cracks.

10. The landslide early warning method according to claim 1, characterized in that, Extract crack feature parameters from the crack skeleton map and determine the crack type, including: The extracted crack feature parameters are compared with the pre-stored crack feature-type mapping table to obtain the crack type of the current crack.

11. The landslide early warning method according to claim 1, characterized in that, Based on the pre-existing correspondence between crack types and landslide stages, the landslide development stage indicated by the crack type is determined, including: When the crack type is a trailing edge tensile crack, the landslide stage is the creep deformation stage; When the crack type is lateral shear crack, the landslide stage is the rapid sliding stage; and... When the crack type is a leading-edge bulging crack, the landslide stage is a gradually stabilizing stage.

12. The landslide early warning method according to claim 11, characterized in that, The levels of the warning information include: A primary warning corresponding to the creeping deformation stage; A medium-level warning corresponding to the gradual stabilization phase; and, Advanced warning for the phase of rapid sliding.

13. A landslide early warning system, characterized in that, The system includes: The mountain model module is used to acquire mountain images with vegetation cover removed and to build a mountain model, including: obtaining a diffusion model; the diffusion model is a noise-adding function; training the diffusion model using a U-Net network structure to obtain a pre-trained diffusion model; and inputting the mountain images with vegetation cover removed into the pre-trained diffusion model to obtain the mountain model. The process of training a diffusion model using the U-Net network structure includes: collecting sample images; the sample images include vegetation cover images and bare land images of the same mountain area; progressively adding noise to the sample images using a noise-adding function to obtain noisy images; and retrieving bare land images from the noisy images using a reverse process, including: using the pre-trained diffusion model to process the vegetation cover images from the noisy images up to the T-th step. Initially, noise and vegetation pattern residuals added during the forward pass are predicted and removed progressively. After T iterations, the bare land image is finally reconstructed; and, The crack probability map module is used to obtain a crack probability map based on a pre-trained bare ground digital elevation model. The bare ground mask module is used to select a mountain model with the same angle as the pre-trained bare ground digital elevation model to obtain a bare ground mask; The crack skeleton map module is used to combine the crack probability map and the bare ground mask to generate a crack skeleton map. It includes: comparing the crack probability map with the bare ground mask pixel by pixel to obtain a crack probability map where cracks exist only in the bare ground; performing binarization on the crack probability map where cracks exist only in the bare ground to obtain a binary crack map; and refining the binary crack map to obtain the crack skeleton map. The judgment module is used to extract crack feature parameters from the crack skeleton map and determine the crack type; The early warning module is used to output landslide early warning information based on crack type.

14. The landslide early warning system according to claim 13, characterized in that, The crack probability map module is used to obtain a crack probability map based on a pre-trained bare ground digital elevation model, including: The crack probability map module is used to obtain a digital elevation model of bare land without vegetation cover; The bare land digital elevation model is trained by using a loss function to obtain a pre-trained bare land digital elevation model. Obtain the actual bare land digital elevation model image and input it into the pre-trained bare land digital elevation model to obtain the crack probability map.

15. An electronic device, characterized in that, include: Processor and memory; The processor invokes the computer program stored in the memory to execute the landslide early warning method according to any one of claims 1 to 12.

16. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the processor to perform the landslide early warning method according to any one of claims 1 to 12.

Images