Landslide identification method, system and terminal

Abstract

The invention belongs to the technical field of landslide identification, and relates to a landslide identification method, which comprises the steps of configuring a plurality of preset point locations, and obtaining a reference image of each preset point location of a mountain; a pre-inspection image of each preset point location of the mountain is collected in sequence; an improved CLAHE algorithm is adopted to preprocess the pre-inspection image, and an inspection image is obtained; registering the inspection image and the reference image of the same preset point location, performing multi-scale feature extraction and fusion on the registered reference image and inspection image, constructing a logarithmic difference chart, performing segmentation and morphological post-processing on the logarithmic difference chart to obtain a change region, and taking the area of the change region as the landslide area; and judging whether the preset point location has landslide or not based on the proportion of the landslide area to the total area of the mountain. The invention also provides a landslide identification system and a terminal. The method can achieve autonomous inspection, does not need to depend on a large amount of labeled data in the recognition process, and has the advantages of high anti-interference capability, high recognition precision and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of landslide identification, and in particular to a landslide identification method, system and terminal. BACKGROUND

[0002] Landslide is a collapse phenomenon of rock under the action of gravity. The occurrence of landslide is usually related to many factors, including river erosion, groundwater activity, rainwater soaking, earthquake and artificial cutting slope, etc. Landslide can cause great disaster, and even in severe cases, it can destroy entire villages, kill people and livestock, destroy factories, power stations and block roads; at the same time, a large amount of stones and soil blocks falling into the river after the occurrence of landslide will also block the river and form flood disaster. In order to reduce the impact and harm of landslide, it is necessary to carry out inspection and identification of landslide, which not only saves valuable time for disaster response, but also enhances people's ability to cope with natural disasters.

[0003] The existing landslide inspection and identification methods mainly include unmanned aerial vehicle remote sensing image, satellite remote sensing and artificial inspection and identification methods. These methods have certain deficiencies: the artificial inspection and identification method is low in efficiency; the satellite remote sensing identification method is poor in real-time performance; and the unmanned aerial vehicle inspection is easily affected by weather. In satellite remote sensing identification and unmanned aerial vehicle inspection identification, landslide identification algorithms mainly include traditional image processing-based algorithms and deep learning-based algorithms. The traditional image processing-based algorithm mainly adopts edge detection combined with morphological analysis and change detection method. This algorithm has high real-time performance, can directly extract cracks and deformation areas based on image gradient / texture features, but is easily disturbed by shadows and weather. The deep learning-based algorithm includes UNet, Siamese network and Transformer network-based landslide identification method, which has good recognition effect, but needs a large amount of labeled data for training. Since it is difficult to collect landslide data, it is very difficult to train a general model with good generalization ability, and the actual implementation is difficult.

[0004] Therefore, it is urgent to propose a landslide inspection and identification method which can be self-inspected, has strong anti-interference ability, high recognition accuracy and does not need to rely on labeled data. SUMMARY

[0005] The present application aims to provide a landslide identification method, system and terminal, which adopts a robot dog to collect pre-inspection images of each preset point of a mountain, has strong anti-interference ability, processes the pre-inspection images based on an improved CLAHE algorithm, and can well suppress noise and retain the edge features of the images on the premise of enhancing the features. In addition, the present application can well represent the deep and shallow features of the images by performing multi-scale feature extraction and fusion on the reference images and the inspection images, and constructing a logarithmic difference map, and can realize high-precision identification of the landslide without a large amount of labeled data.

[0006] To achieve the above object, the present application adopts the following technical solutions: In a first aspect, the present application provides a mountain landslide identification method, comprising the following steps: S1. configuring a plurality of preset points, and obtaining a reference image of each preset point of the mountain; S2. sequentially collecting a pre-inspection image of each preset point of the mountain; S3. pre-processing the pre-inspection image by using an improved CLAHE algorithm to obtain an inspection image; S4. registering the inspection image and the reference image of the same preset point, and calculating a landslide area based on the registered images; S5. determining whether the mountain landslide occurs at the preset point based on the proportion of the landslide area to the total area of the mountain.

[0007] As a possible implementation manner, S3 comprises the following sub-steps: S30. dividing the pre-inspection image into a plurality of overlapping sub-blocks; S31. calculating the cumulative distribution function and the bilinear interpolation weight of each sub-block; S32. weightedly summing the cumulative distribution function and the bilinear interpolation weight of the plurality of overlapping sub-blocks to obtain the inspection image.

[0008] As a possible implementation manner, S4 comprises the following sub-steps: S40. registering the inspection image and the reference image of the same preset point; S41. extracting the multi-scale features of the registered inspection image to obtain an inspection feature map, and extracting the multi-scale features of the registered reference image to obtain a reference feature map; S42. constructing a logarithmic difference map based on the inspection feature map and the reference feature map, segmenting and morphologically post-processing the logarithmic difference map to obtain a change region, and taking the area of the change region as the landslide area.

[0009] As a possible implementation manner, S40 comprises the following sub-steps: S400. respectively obtaining reference image SIFT feature points and inspection image SIFT feature points based on the differences between Gaussian smoothed images under different scales; S401. constructing reference image SIFT descriptors based on the amplitude and direction of pixel gradients in the neighborhood of the reference image SIFT feature points, and constructing inspection image SIFT descriptors based on the amplitude and direction of pixel gradients in the neighborhood of the inspection image SIFT feature points; S402. Calculate the matching relationship of the SIFT descriptors of the reference image and the SIFT descriptors of the inspection image based on the Euclidean distance, and screen out a plurality of matching feature point pairs in the reference image and the inspection image; S403. Establish the correspondence of the reference image and the inspection image based on the plurality of matching feature point pairs, and calculate the transformation model based on the correspondence; S404. Transform the inspection image based on the transformation model to align the inspection image with the reference image, that is, complete the SIFT feature registration of the inspection image and the reference image.

[0010] As a possible implementation manner, S41 is specifically: Construct a feature pyramid for the registered inspection image based on the feature pyramid network, use ResNet50 as the backbone network to extract multi-scale features of the registered inspection image, and fuse the extracted multi-scale features to obtain an inspection feature map; Construct a feature pyramid for the registered reference image based on the feature pyramid network, use ResNet50 as the backbone network to extract multi-scale features of the registered reference image, and fuse the extracted multi-scale features to obtain a reference feature map.

[0011] As a possible implementation manner, S42 includes the following sub-steps: S420. Calculate the ratio of the inspection feature map plus 1 and the reference feature map plus 1, calculate the absolute value of the logarithm of the ratio, and obtain a logarithmic difference map; S421. Calculate the inter-class variance of two types of pixels in the logarithmic difference map, determine a segmentation threshold, perform OTSU segmentation on the logarithmic difference map based on the inter-class variance and the segmentation threshold, and obtain a segmented change area; S422. Perform morphological post-processing on the segmented change area, and calculate the area of the change area, which is the landslide area.

[0012] As a possible implementation manner, the following method is used to determine the segmentation threshold: Iterate through all pixels in the logarithmic difference map, calculate the gray value of all pixels, and calculate the inter-class variance by taking the gray value of each pixel as a threshold one by one. The threshold with the maximum inter-class variance is the segmentation threshold.

[0013] As a possible implementation manner, S5 is specifically: A landslide threshold is preset, the ratio of the landslide area to the total area of the mountain is calculated, and if the ratio is greater than the landslide threshold, it is determined that a mountain landslide has occurred; otherwise, it is determined that a mountain landslide has not occurred.

[0014] In a second aspect, the present application provides a mountain landslide identification system, comprising: The benchmark image acquisition unit is configured with a plurality of preset points, and benchmark images of each preset point of the mountain are acquired in advance; The pre-inspection image acquisition unit controls the robot dog to sequentially acquire pre-inspection images of each preset point of the mountain; The mountain landslide identification unit adopts the improved CLAHE algorithm to pre-process the pre-inspection images, obtains inspection images, registers the inspection images and the benchmark images of the same preset point, calculates the landslide area based on the registered images, and judges whether the preset point has a mountain landslide based on the proportion of the landslide area in the total area of the mountain.

[0015] In a third aspect, the present application provides a terminal, comprising a processor and a communication interface coupled with the processor, and the processor is used to run a computer program or instruction to realize the mountain landslide identification method provided in the first aspect.

[0016] Compared with the prior art, the present application has the following beneficial effects: 1. The mountain landslide identification method provided by the present application adopts the improved CLAHE algorithm to pre-process the pre-inspection images, divides the pre-inspection images into a plurality of overlapping sub-blocks based on the block overlap strategy, so that the sub-blocks are more smooth, can well suppress noise and retain the edge features of the image on the premise of enhancing the features, and effectively improves the chessboard effect.

[0017] 2. The mountain landslide identification method provided by the present application extracts and fuses the multi-scale features of the registered inspection images, constructs a logarithmic difference map, which can better reflect the difference between the change region and the non-change region, realizes accurate identification and calculation of the change region, does not need to rely on a large amount of labeled data, and has high identification accuracy.

[0018] 3. The mountain landslide identification method provided by the present application adopts the inspection robot dog to acquire the pre-inspection images, can autonomously move to complete the inspection task, can adapt to various complex inspection scenes, has high flexibility and strong anti-interference ability. DETAILED DESCRIPTION

[0019] The accompanying drawings described herein are used to provide further understanding of the present application, and form a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation on the present application. In the drawings: Figure 1 The mountain landslide identification method flowchart in the embodiment of the present application; Figure 2 The registration flowchart of the inspection images and the benchmark images of the same preset point in the embodiment of the present application; Figure 3 The benchmark image and the pre-inspection image of a certain point acquired in advance in the embodiment of the present application; Figure 4The reference image and the inspection image obtained after the reference image and the pre-inspection image are enhanced by using the improved CLAHE algorithm in the embodiment of the present application; Figure 5 The reference image and the inspection image after registration in the embodiment of the present application; Figure 6 The change region diagram segmented by using the OTSU segmentation algorithm in the embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to clearly describe the technical scheme of the embodiment of the present application, in the embodiment of the present application, the same items or similar items with basically the same function and role are distinguished by using the words of 'first','second', etc. For example, the first threshold and the second threshold are only used to distinguish different thresholds, and the order is not limited. Those skilled in the art can understand that the words of 'first','second', etc. do not limit the number and execution order, and the words of 'first','second', etc. also do not limit the difference.

[0021] It should be noted that in the present application, the words of 'exemplary' or 'for example' are used to represent an example, illustration or description. Any embodiment or design scheme described as 'exemplary' or 'for example' in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. In fact, the words of 'exemplary' or 'for example' are intended to present the relevant concept in a specific way.

[0022] In the present application, 'at least one' means one or more, and 'a plurality of' means two or more. The association relationship of the associated objects is described by 'and / or', which means that there can be three kinds of relationships, for example, A and / or B can represent the following three cases: A exists alone, A and B exist together, and B exists alone, wherein A and B can be singular or plural. The character ' / ' generally represents an 'or' relationship between the associated objects before and after it. The following at least one (or similar expressions) means any combination of these items, including any combination of single item (or plural items). For example, at least one of a, b or c can represent: a, b, c, the combination of a and b, the combination of a and c, the combination of b and c, or the combination of a, b and c, wherein a, b and c can be single or plural.

[0023] The embodiment of the present application aims to provide a landslide identification method, system and terminal, adopts a robot dog to collect pre-inspection images of each preset point of a mountain, has strong anti-interference capability, processes the pre-inspection images based on an improved CLAHE algorithm, and under the premise of enhancing features, well suppresses noise and retains the edge features of the images. In addition, the present application fuses multi-scale feature extraction of the reference images and the inspection images, constructs a logarithmic difference map, can well represent the deep and shallow features of the images, and can realize high-precision identification of the mountain landslide without a large amount of labeled data.

[0024] In a first aspect, the embodiment of the present application provides a landslide identification method, referring to Figure 1 , comprising the following steps: S1. Configuring a plurality of preset points, acquiring reference images of each preset point of a mountain; S2. Collecting pre-inspection images of each preset point of the mountain in turn; As an example, a plurality of preset points are configured , the reference images of each preset point are collected in advance . Based on the actual inspection requirements, the robot dog is controlled to inspect according to the planned inspection frequency, and the pre-inspection images of each preset point are collected in turn each time , until all the preset points are completed.

[0025] S3. Preprocessing the pre-inspection images by using an improved CLAHE algorithm to obtain inspection images; As a possible implementation manner, S3 comprises the following sub-steps: S30. Dividing the pre-inspection images into a plurality of overlapping sub-blocks; S31. Calculating the cumulative distribution function and bilinear interpolation weight of each sub-block; S32. Weighted sum of the cumulative distribution function and bilinear interpolation weight of the plurality of overlapping sub-blocks to obtain the inspection images.

[0026] The traditional CLAHE algorithm divides the image into a plurality of non-overlapping sub-blocks, if the parameters are not properly selected, the image details are easily lost, and the chessboard effect appears. The embodiment improves the traditional CLAHE algorithm, adopts a block overlap strategy, divides the pre-inspection images into a plurality of overlapping sub-blocks, makes the sub-blocks more smooth between each other, and effectively improves the chessboard effect.

[0027] As an example, the pre-inspection images are preprocessed by using the following method: Wherein, represents the pre-inspection images, represents the inspection images, represents the sub-block​ cumulative distribution function of the sum of the absolute values of the differences between the pixel values of the sub-blocks, denote bilinear interpolation weights of the sub-blocks, denote pixel points in the sub-blocks.

[0028] S4. Register the inspection image and the reference image at the same preset point, and calculate the landslide area based on the registered images; As a possible implementation manner, S4 includes the following sub-steps: S40. Register the inspection image and the reference image at the same preset point; As a possible implementation manner, refer to Figure 2 , S40 includes the following sub-steps: S400. Obtain the reference image SIFT feature points and the inspection image SIFT feature points based on the differences between the Gaussian smoothing images at different scales; S401. Construct the reference image SIFT descriptor based on the amplitude and direction of the pixel gradient in the neighborhood of the reference image SIFT feature points, and construct the inspection image SIFT descriptor based on the amplitude and direction of the pixel gradient in the neighborhood of the inspection image SIFT feature points; S402. Calculate the matching relationship of the reference image SIFT descriptor and the inspection image SIFT descriptor based on the Euclidean distance, and filter out a plurality of matching feature point pairs in the reference image and the inspection image; S403. Establish the correspondence between the reference image and the inspection image based on the plurality of matching feature point pairs, and calculate the transformation model based on the correspondence; S404. Transform the inspection image based on the transformation model to align the inspection image with the reference image, that is, complete the SIFT feature registration of the inspection image and the reference image.

[0029] As an example, denote the reference image as , and the inspection image as , first extract the SIFT features of the reference image and the inspection image . Specifically, detect feature points in the Difference of Gauss (DoG) scale space, and select neighborhood feature points with a radius of around the feature points by the following method, calculate the amplitude and direction of the pixel gradient in the neighborhood of the feature points, and construct the reference image SIFT descriptor: wherein, denotes the Gaussian difference scale space image where the feature point is located, denotes the feature point​ The magnitude of the pixel gradient in the neighborhood, Representing feature points The direction of pixel gradients within the neighborhood.

[0030] Next, the matching relationship between the SIFT descriptor of the reference image and the SIFT descriptor of the inspection image is calculated based on the Euclidean distance, and multiple matching feature point pairs in the reference image and the inspection image are screened. The following method is used to establish the corresponding relationship between the reference image and the inspection image based on multiple matching feature point pairs: in, represents the feature points extracted from the reference image, represents the feature points extracted from the inspection image, Represents the transformation model. Solve the transformation model in the formula , perform inspection images Transform to align the inspection image with the reference image.

[0031] S41 extracts multi-scale features of the inspection image after registration to obtain an inspection feature map, extracts multi-scale features of the reference image after registration to obtain a reference feature map; As a possible implementation method, S41 specifically includes: constructing a feature pyramid for the registered inspection image based on the feature pyramid network FPN, using ResNet50 as the backbone network to extract multi-scale features of the registered inspection image, and fusing the extracted multi-scale features to obtain an inspection feature map; A feature pyramid is constructed for the registered reference image based on the feature pyramid network. ResNet50 is used as the backbone network to extract multi-scale features of the registered reference image, and the extracted multi-scale features are fused to obtain the reference feature map.

[0032] S42. Construct a logarithmic difference map based on the inspection feature map and the baseline feature map. Segment and morphologically post-process the logarithmic difference map to obtain the change region, and use the area of ​​the change region as the landslide area.

[0033] As a possible implementation, S42 includes the following sub-steps: S420 solves the ratio of the inspection feature map plus 1 to the reference feature map plus 1, calculates the absolute value of the logarithm of the ratio, and obtains a logarithmic difference map; As an example, the following method is used to construct a logarithmic difference plot: in, represents a logarithmic difference plot, Represents the inspection feature diagram, Represents the benchmark feature map.

[0034] S421. Calculate the inter-class variance of two types of pixels in the log difference map, determine a segmentation threshold, perform OTSU segmentation on the log difference map based on the inter-class variance and the segmentation threshold, and obtain a segmented change region; As a possible implementation manner, the following method is used to determine the segmentation threshold: Traverse all pixels in the log difference map, calculate the gray value of all pixels, and take the gray value of each pixel as a threshold to calculate the inter-class variance one by one. The threshold that maximizes the inter-class variance is the segmentation threshold.

[0035] As an example, the following method is used for OTSU segmentation: Wherein: , , denotes the inter-class variance, denotes the probability of the inspection feature pixel appearing in the log difference map, denotes the probability of the reference feature pixel appearing in the log difference map, denotes the gray value of the inspection feature pixel, denotes the gray value of the reference feature pixel, denotes the global gray mean value of the image, denotes the number of pixels with a gray level of , denotes the total number of pixels of the image.

[0036] In this embodiment, the gray value that maximizes the inter-class variance is taken as the segmentation threshold.

[0037] S422. Perform morphological post-processing on the segmented change region, and calculate the area of the change region, that is, the landslide area.

[0038] S5. Determine whether the landslide occurs at the preset point based on the proportion of the landslide area to the total area of the mountain.

[0039] As a possible implementation manner, S5 is specifically: A preset landslide threshold is calculated. If the ratio of the landslide area to the total area of the mountain is greater than the landslide threshold, it is determined that the landslide occurs; otherwise, it is determined that the landslide does not occur.

[0040] As an example, the landslide threshold is 10%, which is obtained based on an empirical value.

[0041] In a second aspect, the embodiments of the present application provide a mountain landslide identification system, comprising: The benchmark image acquisition unit is configured with a plurality of preset points, and benchmark images of each preset point of the mountain are acquired in advance; The pre-inspection image acquisition unit sequentially acquires pre-inspection images of each preset point of the mountain; The mountain landslide identification unit adopts the improved CLAHE algorithm to pre-process the pre-inspection images, obtains inspection images, registers the inspection images of the same preset point with the benchmark images, calculates the landslide area based on the registered images, and judges whether the landslide occurs at the preset point based on the proportion of the landslide area in the total area of the mountain.

[0042] In a third aspect, an embodiment of the present application provides a terminal, which comprises a processor and a communication interface coupled with the processor, and the processor is configured to run a computer program or instructions to implement the mountain landslide identification method provided in the first aspect.

[0043] The implementation process of the present application will be further described below in combination with specific embodiments.

[0044] Referring to Figure 3 Fig. (a), which is a benchmark image of a collected point , Figure 3 Fig. (b) is a pre-inspection image of the point , and the image resolution is 800x600, and the color space is RGB. The color space of the benchmark image and the pre-inspection image is converted from RGB to LAB, and the improved CLAHE algorithm is used to enhance the image of the L channel, wherein the grid size is set to 8x8, and the adjacent grids intersect by 2 pixels. The enhanced L single-channel image A and the B channel are combined and converted to the RGB color space. The enhanced benchmark image is shown in Figure 4 Fig. (a), and the enhanced pre-inspection image, i.e., the inspection image is shown in Figure 4 Fig. (b).

[0045] The improved method is used to perform registration operation on the enhanced benchmark image and the inspection image. First, the SIFT feature points of the benchmark image and the SIFT feature points of the inspection image are acquired respectively, and the SIFT descriptors of the benchmark image and the SIFT descriptors of the inspection image are constructed respectively. Then, the feature point matching is performed using the FLANN matcher, and the homography matrix is estimated using the RANSAC algorithm based on the matching relationship of the feature points. Finally, the inspection image is perspective transformed according to the homography matrix to be registered to the benchmark image. The registered benchmark image is shown in Figure 5 Fig. (a), and the registered inspection image is shown in Figure 5 Fig. (b).

[0046] The Gaussian pyramid of the registered reference image and the inspection image is constructed respectively, including the original size, 1 / 2 size and 1 / 4 size images. The image sizes of different scales are changed to 224*224*3, and then input into the ResNet50 network to obtain the corresponding multi-scale feature maps. The 1 / 2 size and 1 / 4 size feature maps are up-sampled to restore to the normal size, and finally the three features are weighted and summed to obtain the feature map after the fusion of the reference image and the inspection image. The feature log difference map is constructed based on the fusion feature map of the reference image and the field acquisition image. The OTSU segmentation algorithm is used for segmentation on the feature log difference map. Finally, the morphological opening operation with a 7*7 rectangle as the template is used. The processing result is shown in Figure 6 The calculation can obtain that the area ratio of the changed area in the scene to the total area is 20.9%, and it is judged that the landslide occurs in the scene.

[0047] Although the present application is described herein in conjunction with various embodiments, other variations to the disclosed embodiments can be understood and effected by those skilled in the art in the field of the art, upon viewing the drawings, the disclosure and the appended claims. In the specification, the word “comprising” does not exclude other components or steps, and the indefinite article “a” or “an” does not exclude a plurality. A single processor or other unit can implement several of the functions recited in the specification. Some measures are recited in mutually different embodiments, but this does not mean that these measures cannot be combined to produce a good result.

[0048] Although the present application is described herein in conjunction with specific features and embodiments thereof, it is obvious that modifications and combinations can be made thereto without departing from the spirit and scope of the application. Accordingly, the specification and drawings are to be regarded in an illustrative rather than a restrictive sense, and it is intended that all modifications and combinations included within the scope of the present application are to be embraced, and that the scope of the present application is to be interpreted only in conformance with the appended claims. Obviously, modifications and changes can be made thereto without departing from the spirit and scope of the present application. Thus, it is intended that the present application cover modifications and variations of this application provided they come within the scope of the appended claims and their equivalents.

Claims

1. A landslide identification method, characterized in that: The steps include: S1. Configure multiple preset points and obtain a reference image of each preset point on the mountain; S2. Sequentially collect pre-inspection images of each preset point on the mountain; S3. Preprocess the pre-inspection image using the improved CLAHE algorithm to obtain the inspection image; S4. Registering the inspection image of the same preset point with the reference image, and calculating the landslide area based on the registered image; S5. Determine whether a landslide has occurred at the preset location based on the ratio of the landslide area to the total mountain area.

2. The landslide identification method according to claim 1, characterized in that: The S3 includes the following sub-steps: S30. Dividing the pre-inspection image into multiple overlapping sub-blocks; S31. Calculate the cumulative distribution function and bilinear interpolation weight of each sub-block; S32. Perform a weighted summation of the cumulative distribution functions and bilinear interpolation weights of the multiple overlapping sub-blocks to obtain an inspection image.

3. The landslide identification method according to claim 1, characterized in that: The S4 includes the following sub-steps: S40 will be the same preset point inspection image and the reference image registration; S41 extracts multi-scale features of the inspection image after registration to obtain an inspection feature map, extracts multi-scale features of the reference image after registration to obtain a reference feature map; S42. Construct a logarithmic difference map based on the inspection feature map and the baseline feature map. Segment and morphologically post-process the logarithmic difference map to obtain the change region, and use the area of ​​the change region as the landslide area.

4. The landslide identification method according to claim 3, characterized in that: The S40 includes the following sub-steps: S400 based on the difference between the Gaussian smoothed images at different scales to obtain the reference image SIFT feature points and inspection image SIFT feature points; S401 constructs a reference image SIFT descriptor based on the magnitude and direction of the pixel gradient within the neighborhood of the reference image SIFT feature point, and constructs a SIFT descriptor based on the magnitude and direction of the pixel gradient within the neighborhood of the inspection image SIFT feature point; S402 is based on the Euclidean distance calculation reference image SIFT descriptor and inspection image SIFT descriptor matching relationship, filter out the reference image and inspection image multiple matching feature point pairs; S403 establishes a correspondence between the reference image and the inspection image based on a plurality of matching feature points, and calculates a transformation model based on the correspondence; S404. Transform the inspection image based on the transformation model to align the inspection image with the reference image, ie, complete SIFT feature registration of the inspection image with the reference image.

5. The landslide identification method according to claim 3, characterized in that: The S41 is specifically as follows: A feature pyramid is constructed for the registered inspection image based on the feature pyramid network. ResNet50 is used as the backbone network to extract multi-scale features from the registered inspection image, and the extracted multi-scale features are fused to obtain the inspection feature map. A feature pyramid is constructed for the registered reference image based on the feature pyramid network. ResNet50 is used as the backbone network to extract multi-scale features of the registered reference image, and the extracted multi-scale features are fused to obtain the reference feature map.

6. The landslide identification method according to claim 3, characterized in that: The S42 includes the following sub-steps: S420 solves the ratio of the inspection feature map plus 1 to the reference feature map plus 1, calculates the absolute value of the logarithm of the ratio, and obtains a logarithmic difference map; S421 calculates the inter-class variance of the two types of pixels in the logarithmic difference map, determines the segmentation threshold, performs OTSU segmentation on the logarithmic difference map based on the inter-class variance and the segmentation threshold, and obtains the change area after segmentation; S422. Perform morphological post-processing on the segmented changed region and calculate the area of ​​the changed region, which is the landslide area.

7. The landslide identification method according to claim 6, characterized in that: The segmentation threshold is determined as follows: Traverse all pixels in the logarithmic difference map, calculate the grayscale values ​​of all pixels, and use the grayscale value of each pixel as the threshold to calculate the inter-class variance. The threshold that maximizes the inter-class variance is the segmentation threshold.

8. The landslide identification method according to claim 1, characterized in that: The S5 is specifically: A landslide threshold is preset, and the ratio of the landslide area to the total mountain area is calculated. If the ratio is greater than the landslide threshold, it is determined that a landslide has occurred; otherwise, it is determined that no landslide has occurred.

9. A landslide identification system, characterized in that: include: The reference image acquisition unit is configured with multiple preset points and pre-acquires the reference image of each preset point on the mountain; The pre-inspection image acquisition unit controls the robot dog to sequentially collect pre-inspection images of each preset point on the mountain; The landslide identification unit uses the improved CLAHE algorithm to preprocess the pre-inspection image to obtain the inspection image, aligns the inspection image of the same preset point with the reference image, calculates the landslide area based on the registered image, and determines whether a landslide has occurred at the preset point based on the ratio of the landslide area to the total mountain area.

10. A terminal comprising a processor and a communication interface coupled to the processor, wherein the processor is configured to run a computer program or instruction to implement the landslide identification method according to any one of claims 1 to 8.

Images