A co-seismic landslide extraction method, system, electronic device and storage medium

Abstract

The present invention discloses a method, system, electronic device and storage medium for extracting co-seismic landslides. The method includes: extracting the spectral band principal components of a remote sensing image by using the principal component analysis method; obtaining the texture features of the spectral band principal components by using the gray-level co-occurrence matrix; constructing a landslide gain index based on the texture features; constructing a multi-feature deep learning model based on the texture features, the landslide gain index and the vegetation cover index; and extracting co-seismic landslides by using the trained multi-feature deep learning model. The present invention can eliminate the interference caused by bare land and roads to the remote sensing extraction of co-seismic landslides, quickly and accurately identify and extract co-seismic landslides from images, and achieve the purpose of quickly responding to earthquake disasters.

Description

Technical Field

[0001] The present invention relates to the technical field of geological disasters, and particularly relates to a method, a system, an electronic device and a storage medium for extracting co-seismic landslides. Background Art

[0002] Co-seismic landslides are secondary disasters caused by earthquakes, which pose significant challenges to post-disaster rescue, recovery and reconstruction during and after an earthquake. These landslides generated in tectonically active areas and earthquake-prone areas pose a serious threat to the environment, vegetation and human life, especially in mountainous areas. Therefore, timely acquisition of co-seismic landslide data is crucial for post-disaster rescue and risk assessment, especially in remote areas where investigators cannot easily access.

[0003] Currently, the methods for extracting co-seismic landslides using remote sensing images mainly include visual interpretation, pixel-based, object-oriented and deep learning. Among them, visual interpretation requires a large amount of professional knowledge and high requirements for the resolution of remote sensing images, which is time-consuming and laborious and cannot achieve a rapid response to disasters. Most of the pixel-based landslide extraction methods are based on traditional statistical analysis and machine learning techniques, and the extraction methods are relatively simple. However, they only consider the features of individual pixel points and do not consider the relevant features of other attributes, such as shape, texture, spatial structure, context, etc., resulting in the loss of correlation between pixels. The object-oriented classification technology segments homogeneous images, collects adjacent pixels as analysis objects, and uses high-resolution and multi-spectral data for high-precision classification. Object-oriented classification is based on image segmentation, and the results depend on the choice of segmentation scale. Experience and parameter adjustment are required in the process of setting segmentation parameters and classification rules, and complex and large-scale remote sensing data cannot be processed quickly, which cannot meet the needs of earthquake emergency. Deep learning can extract complex features from high-dimensional data, has the advantages of feature learning in an unsupervised or semi-supervised manner, and uses hierarchical feature extraction instead of manual recognition. The extraction speed and accuracy are greatly improved, which can meet the requirements of earthquake disaster emergency response. However, due to the similar spectral features of bare land and roads to landslides, they are often misidentified as landslides, affecting the extraction accuracy of co-seismic landslides. Summary of the Invention

[0004] The purpose of the present invention is to provide a method, a system, an electronic device and a storage medium for extracting co-seismic landslides, so as to eliminate the interference caused by bare land and roads to the remote sensing extraction of co-seismic landslides, quickly achieve accurate identification and extraction of co-seismic landslides from images, and achieve the purpose of rapid response to earthquake disasters.

[0005] To achieve the above purpose, the present invention provides the following solutions:

[0006] A method for extracting co-seismic landslides includes:

[0007] Extract the principal components of the spectral bands of the remote sensing image using the principal component analysis method;

[0008] Obtain the texture features of the principal components of the spectral bands using the gray-level co-occurrence matrix; the texture features include mean, variance, homogeneity, contrast, dissimilarity, entropy, angular second moment, and correlation;

[0009] Construct a landslide gain index based on the texture features;

[0010] Construct a multi-feature deep learning model based on the texture features, the landslide gain index, and the vegetation cover index;

[0011] Extract co-seismic landslides using the trained multi-feature deep learning model.

[0012] Optionally, constructing a landslide gain index based on the texture features specifically includes:

[0013] Use the ROI statistical analysis tool to analyze the distribution characteristics of co-seismic landslides, bare land, and roads in terms of texture features, and determine the differential texture features based on the distribution characteristics; the differential texture features include: contrast and dissimilarity;

[0014] Perform normalization calculation on the differential texture features to construct a landslide gain index.

[0015] Optionally, the construction formula of the landslide gain index is as follows:

[0016]

[0017] Where LGI represents the landslide gain index, Contrast represents contrast, and Dissimilarity represents dissimilarity.

[0018] Optionally, constructing a multi-feature deep learning model based on the texture features, the landslide gain index, and the vegetation cover index specifically includes:

[0019] Construct a landslide recognition feature raster set based on the texture features, the landslide gain index, and the vegetation cover index;

[0020] Manually interpret co-seismic landslides in the remote sensing image and perform annotation to obtain a label raster;

[0021] Construct a multi-feature deep learning model through the landslide recognition feature raster set and the label raster.

[0022] Optionally, it further includes verifying the accuracy of the trained multi-feature deep learning model; the indicators for the accuracy verification include: overall accuracy, kappa coefficient, accuracy, recall rate, and F1 score.

[0023] The present invention also provides a co-seismic landslide extraction system, including:

[0024] A spectral band principal component extraction module, configured to extract the spectral band principal components of a remote sensing image by using the principal component analysis method;

[0025] A texture feature acquisition module, configured to acquire the texture features of the spectral band principal components by using a gray-level co-occurrence matrix; the texture features include mean, variance, homogeneity, contrast, dissimilarity, entropy, angular second moment, and correlation;

[0026] A landslide gain index construction module, configured to construct a landslide gain index based on the texture features;

[0027] A multi-feature deep learning model construction module, configured to construct a multi-feature deep learning model based on the texture features, the landslide gain index, and the vegetation coverage index;

[0028] A co-seismic landslide extraction module, configured to extract co-seismic landslides by using the trained multi-feature deep learning model.

[0029] Optionally, the landslide gain index construction module specifically includes:

[0030] A differential texture feature determination unit, configured to analyze the distribution characteristics of co-seismic landslides, bare land, and roads in terms of texture features by using an ROI statistical analysis tool, and determine differential texture features based on the distribution characteristics; the differential texture features include: contrast and dissimilarity;

[0031] A landslide gain index construction unit, configured to perform normalization calculation on the differential texture features to construct a landslide gain index.

[0032] Optionally, the multi-feature deep learning model construction module specifically includes:

[0033] A landslide recognition feature raster set construction unit, configured to construct a landslide recognition feature raster set based on the texture features, the landslide gain index, and the vegetation coverage index;

[0034] A label raster construction unit, configured to manually interpret and label co-seismic landslides in the remote sensing image to obtain a label raster;

[0035] A multi-feature deep learning model construction unit, configured to construct a multi-feature deep learning model through the landslide recognition feature raster set and the label raster.

[0036] The present invention also provides an electronic device, including a memory and a processor, where the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above-mentioned co-seismic landslide extraction method.

[0037] The present invention also provides a computer-readable storage medium storing a computer program, which when executed by a processor implements the above-mentioned co-seismic landslide extraction method.

[0038] According to the specific embodiments provided by the present invention, the following technical effects are disclosed:

[0039] The present invention extracts the texture features of remote sensing images through the principal component analysis method (PCA) and the gray-level co-occurrence matrix (GLCM); statistically analyzes the main feature differences between co-seismic landslides, bare land, and roads in texture features. The statistical results show that co-seismic landslides exhibit obvious feature differences from bare land and roads in terms of contrast and dissimilarity; normalizes the calculation of contrast and dissimilarity to generate a landslide enhancement index (LGI); uses LGI, texture features, and the normalized difference vegetation index (NDVI) as the remote sensing identification feature bands for co-seismic landslides to construct a multi-feature deep learning model to identify and extract co-seismic landslides. The multi-feature deep learning model not only performs non-linear mapping on the internal features of co-seismic landslide data but also fully learns the complex features of co-seismic landslides from high-dimensional data. The co-seismic landslide extraction results are closer to the ground truth landslides in terms of shape and area; under the advantages of the original deep learning algorithm, the multi-feature deep learning model incorporates NDVI and band texture features, enabling more comprehensive learning of co-seismic landslide features; in particular, LGI can further expand the feature differences between co-seismic landslides and bare land and roads, effectively eliminating the interference caused by bare land and roads to the remote sensing extraction of co-seismic landslides, making the final extraction results more accurate. Therefore, the multi-feature deep learning model constructed in the present invention has high feasibility and can accurately extract co-seismic landslides. The present invention provides a new method for future landslide extraction research, which is fast and efficient, and is beneficial to the prevention and control of landslide geological disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] 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 embodiments. Obviously, the following described 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.

[0041] Figure 1 It is a flowchart of the co-seismic landslide extraction method provided by the present invention;

[0042] Figure 2 It is a specific flowchart of the co-seismic landslide extraction method provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0044] The purpose of the present invention is to provide a co-seismic landslide extraction method, system, electronic device and storage medium, so as to eliminate the interference caused by bare land and roads to the remote sensing extraction of co-seismic landslides, quickly and accurately identify and extract co-seismic landslides from images, and achieve the purpose of quickly responding to earthquake disasters.

[0045] In order to make the above objects, features and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0046] Embodiment 1

[0047] The present invention provides a co-seismic landslide extraction method, as Figure 1 - Figure 2 shown, the method includes the following steps:

[0048] Step 101: Extract the main components of the spectral bands of the remote sensing image by using the principal component analysis method.

[0049] In practical applications, obtain the remote sensing image from the remote sensing data platform, and perform corresponding preprocessing on the image, including radiometric calibration and atmospheric correction. Use the principal component analysis method (PCA) to extract the main components of the image spectral bands. After PCA analysis, the first principal component (PC band1) contains more than 80% of the image information. In order to improve efficiency, prevent information intersection, and save computing space, PC band1 of the image is retained. Use PC band1 as the basic data to carry out the following work.

[0050] Step 102: Obtain the texture features of the main components of the spectral bands by using the gray-level co-occurrence matrix; the texture features include mean, variance, homogeneity, contrast, dissimilarity, entropy, angular second moment and correlation.

[0051] In practical applications, affected by human engineering activities, the remote sensing identification of coseismic landslides is prone to confusion with exposed land in terms of morphology and geomorphic location, and is prone to confusion with roads in terms of tone indication. Therefore, in addition to considering spectral features and index features, the remote sensing identification of coseismic landslides should also fully consider the image texture features. Utilizing the image texture features that can reflect the microscopic structure of the image can improve the accuracy of the remote sensing identification of coseismic landslides. The image texture features of the PC band1 are extracted using the Gray Level Co-occurrence Matrix (GLCM). GLCM is defined by the joint probability density of two-position pixels. It not only reflects the distribution characteristics of brightness, but also reflects the position distribution characteristics between pixels with the same or similar brightness. It is a second-order statistical feature related to the brightness change of the image and is the basis for defining a set of texture features. The GLCM of an image can reflect the comprehensive information of the image gray level regarding direction, adjacent interval, and change amplitude. In ENVI, the parameter settings for the window size of the pixel statistics for GLCM image texture feature extraction are 5×5, the moving step is set to 2, the moving direction is set to 2, and the gray level quantization level is set to 64. The extracted texture features include Mean, Variance, Homogeneity, Contrast, Dissimilarity, Entropy, SecondMoment, and Correlation.

[0052] Step 103: Construct a landslide gain index based on texture features. Specifically, it includes: using the ROI statistical analysis tool to analyze the distribution characteristics of coseismic landslides, bare land, and roads in terms of texture features, and determining the differential texture features based on the distribution characteristics; the differential texture features include: Contrast and Dissimilarity; performing normalization calculations on the differential texture features to construct a landslide gain index.

[0053] In practical applications, the ROI statistical analysis tool is used to statistically analyze all the characteristics of typical co-seismic landslide areas, bare land, and road areas. According to the average value curve graph of the texture characteristics of landslides, bare land, and roads, there are significant differences in texture characteristics between landslides and bare land / roads. The statistical results show that the average value differences in their texture characteristics are mainly manifested between two features: Contrast and Dissimilarity. Contrast reflects the clarity of the image and the depth of the texture grooves, while Dissimilarity reflects the similarity of the image grayscale. After a landslide occurs, surface gravel accumulates, and the landslide scarp presents various irregular shapes. At the same time, an obvious landslide surface will be generated after the landslide. The interior of the landslide surface is rougher compared to bare land and roads, and the depth of the texture grooves is more prominent, while bare land and roads are relatively smooth. In addition, there is a certain slope of the landslide surface relative to the flat ground, making the boundary line more obvious and the similarity of the image grayscale relatively low. Therefore, landslides, bare land, and roads can be effectively distinguished in terms of Contrast and Dissimilarity.

[0054] Normalize Contrast and Dissimilarity for calculation to construct the Landslide Gain Index LGI. This index can expand the texture feature differences between landslides and bare land / roads, effectively eliminate the interference caused by bare land and roads to the remote sensing extraction of co-seismic landslides, and ensure the accuracy of the co-seismic landslide extraction results. The construction formula of the Landslide Gain Index LGI is as follows:

[0055]

[0056] Contrast and Dissimilarity represent Contrast and Dissimilarity respectively.

[0057] Step 104: Construct a multi-feature deep learning model based on texture features, the Landslide Gain Index, and the vegetation cover index.

[0058] In practical applications, NDVI is a simple and effective measurement parameter in the field of remote sensing that reflects the surface vegetation cover and growth conditions. After a landslide occurs, it will cause a certain degree of impact and damage to the surface vegetation. The surface of the landslide development area is bare, with a low vegetation coverage rate, showing a relatively high brightness on the remote sensing image, and the vegetation cover tone is relatively light. Therefore, using NDVI as an index feature for landslide identification can effectively improve the accuracy of landslide identification. Taking NDVI, the Landslide Gain Index (LGI), and texture features as the feature bands for the remote sensing identification of co-seismic landslides, through band fusion, a landslide identification feature raster set is formed, enriching the image information.

[0059] Manually interpret co-seismic landslides from remote sensing images and divide the interpretation results into two parts, including training samples and validation samples. The training samples are used to construct a deep learning model, and the validation samples are used to validate the deep learning model.

[0060] Training the model to recognize features requires one or more images containing labeled pixel data, i.e., label rasters. Convert the training samples and validation samples of vector data into polygon regions of interest (ROIs), and create an image containing labeled pixel data from the ROIs, i.e., the label raster.

[0061] Use the landslide recognition feature raster set and the sample Label Raster to construct a multi-feature deep learning model. The model is the ENVINet5 deep learning model under the TensorFlow framework.

[0062] Before starting training, initialize an empty mask-based ENVINet5 model using the landslide recognition feature raster set and the landslide label raster. Here, the patch size of the model needs to be defined. A patch is a small image passed into the model for training. The larger the value, the higher the efficiency, but the higher the requirement for video memory. Considering the computer performance and prediction effect, after multiple calculations, the patch size of the model is defined as 256.

[0063] Training the multi-feature deep learning model requires inputting the initialized ENVINet5 model and setting the training rasters and validation rasters in the label raster. This step also requires setting the key input parameters recognized by the model. The setting of the parameters directly affects the accuracy of the landslide recognition result. Determine the parameter settings through randomized parameter experiments. The number of epochs for model training is set to 25; the number of patches per iteration is set to 300; the solid distance is set to 13.75; the blur distance is set to 2.23; the class weight size is set to 1.37; the loss weight for training is set to 0.063.

[0064] After the multi-feature deep learning model is trained, classify the landslide recognition feature raster based on the multi-feature deep learning model (TensorFlow Mask Classification). The classification result is the class activation map of the landslide class. The class activation map is a grayscale image that shows the probability of pixels belonging to the feature of interest, i.e., the probability that the raster in the grayscale image is predicted as the landslide class, with a value range of 0 to 1. Use threshold segmentation to determine the probability threshold to complete the extraction of co-seismic landslides.

[0065] After the multi-feature deep learning model is trained, it is necessary to verify the accuracy of the multi-feature deep learning model. The accuracy verification metrics include overall accuracy (OA), kappa coefficient (KC), precision, recall, and F1-score. Among them, OA is the ratio between the number of correct predictions of the model on all test sets and the total number; precision represents the proportion of the number of correctly classified pixels in the classification result of this category of pixels; recall represents the proportion of the correctly classified pixels in the actual number of pixels in this category. If a classifier has good performance, it should have the following performance: while the recall value increases, the precision value remains at a high level. And a classifier with poor performance may sacrifice a lot of precision values to improve the recall value. Usually, the F1-score is used to show the trade-off between precision and recall of the classifier. The F1-score is the harmonic mean of precision and recall and is a comprehensive evaluation index.

[0066] The calculation formula of OA is:

[0067]

[0068] The calculation formula of KC is:

[0069]

[0070] The calculation formula of precision is:

[0071]

[0072] The calculation formula of recall is:

[0073]

[0074] The calculation formula of F1-score is:

[0075]

[0076] Among them, TP is true positive; FP is false positive; TN is true negative; FN is false negative.

[0077] The larger the values of OA, KC, precision, recall, and F1-score, the higher the model accuracy and the better the coseismic landslide extraction effect. After verification, OA, KC, precision, recall, and F1-score reached 97%, 0.83, 94%, 94%, and 0.94 respectively, achieving satisfactory evaluation results.

[0078] Therefore, the multi-feature deep learning model constructed by the present invention can quickly and accurately extract coseismic landslides, achieving the purpose of quickly responding to earthquake disasters.

[0079] Step 105: Use the trained multi-feature deep learning model to extract coseismic landslides.

[0080] Embodiment 2

[0081] To execute the method corresponding to the above Embodiment 1 to achieve the corresponding functions and technical effects, a coseismic landslide extraction system is provided below.

[0082] The system includes:

[0083] A spectral band principal component extraction module, which is used to extract the spectral band principal components of remote sensing images using the principal component analysis method.

[0084] A texture feature acquisition module, which is used to obtain the texture features of the spectral band principal components using the gray-level co-occurrence matrix; the texture features include mean, variance, homogeneity, contrast, dissimilarity, entropy, angular second moment, and correlation.

[0085] A landslide gain index construction module, which is used to construct a landslide gain index based on the texture features.

[0086] A multi-feature deep learning model construction module, which is used to construct a multi-feature deep learning model based on the texture features, landslide gain index, and vegetation coverage index.

[0087] A coseismic landslide extraction module, which is used to extract coseismic landslides using the trained multi-feature deep learning model.

[0088] Among them, the landslide gain index construction module specifically includes:

[0089] A differential texture feature determination unit, which is used to analyze the distribution characteristics of coseismic landslides, bare land, and roads in texture features using the ROI statistical analysis tool, and determine the differential texture features based on the distribution characteristics; the differential texture features include: contrast and dissimilarity;

[0090] A landslide gain index construction unit, which is used to perform normalization calculation on the differential texture features to construct a landslide gain index.

[0091] Among them, the multi-feature deep learning model construction module specifically includes:

[0092] A landslide recognition feature raster set construction unit, which is used to construct a landslide recognition feature raster set based on texture features, landslide gain index, and vegetation coverage index;

[0093] A label raster construction unit, which is used to manually interpret co-seismic landslides in remote sensing images and perform annotation to obtain a label raster;

[0094] A multi-feature deep learning model construction unit, which is used to construct a multi-feature deep learning model through the landslide recognition feature raster set and the label raster.

[0095] Embodiment III

[0096] Embodiment III of the present invention provides an electronic device, including a memory and a processor. The memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the co-seismic landslide extraction method of Embodiment I.

[0097] The above-mentioned electronic device can be a server.

[0098] Embodiment IV

[0099] Embodiment IV of the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the co-seismic landslide extraction method of Embodiment I.

[0100] In this specification, each embodiment is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0101] Specific examples are used in this article to elaborate on the principle and implementation manner of the invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention. The described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

Claims

1. A method for extracting coseismic landslides, characterized in that, Including: Extracting the principal components of the spectral bands of the remote sensing image using the principal component analysis method; Obtaining the texture features of the principal components of the spectral bands using the gray-level co-occurrence matrix; the texture features include mean, variance, homogeneity, contrast, dissimilarity, entropy, angular second moment, and correlation; Constructing a landslide gain index based on the texture features, specifically including: Analyzing the distribution characteristics of co-seismic landslides, bare land, and roads on the texture features using the ROI statistical analysis tool, and determining the differential texture features based on the distribution characteristics; the differential texture features include: contrast and dissimilarity; Performing normalization calculation on the differential texture features to construct a landslide gain index, and the construction formula of the landslide gain index is as follows: Where LGI represents the landslide gain index, Contrast represents the contrast, and Dissimilarity represents the dissimilarity; Constructing a multi-feature deep learning model based on the texture features, the landslide gain index, and the vegetation cover index; Extracting co-seismic landslides using the trained multi-feature deep learning model.

2. The coseismic landslide extraction method according to claim 1, wherein Constructing a multi-feature deep learning model based on the texture features, the landslide gain index, and the vegetation cover index, specifically including: Constructing a landslide recognition feature raster set based on the texture features, the landslide gain index, and the vegetation cover index; Manually interpreting and annotating co-seismic landslides in the remote sensing image to obtain a label raster; Constructing a multi-feature deep learning model through the landslide recognition feature raster set and the label raster.

3. The coseismic landslide extraction method according to claim 1, wherein It also includes verifying the accuracy of the trained multi-feature deep learning model; the indicators for the accuracy verification include: overall accuracy, kappa coefficient, accuracy, recall rate, and F1 score.

4. A co-seismic landslide extraction system, characterized in that, Including: A spectral band principal component extraction module for extracting the principal components of the spectral bands of the remote sensing image using the principal component analysis method; A texture feature acquisition module for obtaining the texture features of the principal components of the spectral bands using the gray-level co-occurrence matrix; the texture features include mean, variance, homogeneity, contrast, dissimilarity, entropy, angular second moment, and correlation; A landslide gain index construction module for constructing a landslide gain index based on the texture features, specifically including: Analyzing the distribution characteristics of co-seismic landslides, bare land, and roads on the texture features using the ROI statistical analysis tool, and determining the differential texture features based on the distribution characteristics; the differential texture features include: contrast and dissimilarity; Performing normalization calculation on the differential texture features to construct a landslide gain index, and the construction formula of the landslide gain index is as follows: Where LGI represents the landslide gain index, Contrast represents the contrast, and Dissimilarity represents the dissimilarity; A multi-feature deep learning model construction module for constructing a multi-feature deep learning model based on the texture features, the landslide gain index, and the vegetation cover index; A co-seismic landslide extraction module for extracting co-seismic landslides using the trained multi-feature deep learning model.

5. The coseismic landslide extraction system according to claim 4, wherein The landslide gain index construction module specifically includes: A differential texture feature determination unit, configured to analyze the distribution characteristics of co-seismic landslides, bare land, and roads in terms of texture features by using an ROI statistical analysis tool, and determine differential texture features based on the distribution characteristics; the differential texture features include: contrast and dissimilarity; A landslide gain index construction unit, configured to perform normalization calculation on the differential texture features to construct a landslide gain index.

6. The coseismic landslide extraction system according to claim 4, characterized in that The multi-feature deep learning model construction module specifically includes: A landslide recognition feature raster set construction unit, configured to construct a landslide recognition feature raster set based on the texture features, the landslide gain index, and the vegetation coverage index; A label raster construction unit, configured to manually interpret co-seismic landslides in the remote sensing image and perform annotation to obtain a label raster; A multi-feature deep learning model construction unit, configured to construct a multi-feature deep learning model through the landslide recognition feature raster set and the label raster.

7. An electronic device, characterized in that, Comprising a memory and a processor, the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the co-seismic landslide extraction method according to any one of claims 1-3.

8. A computer-readable storage medium, characterized in that, It stores a computer program, and when the computer program is executed by a processor, it implements the co-seismic landslide extraction method according to any one of claims 1-3.

Images