Landslide boundary automatic detection method combining SBAS-InSAR and global-local short-term dense connection network

Abstract

The invention provides a landslide boundary automatic detection method combining SBAS-InSAR and a global-local short-term dense connection network, and the method comprises the following steps: 1, obtaining multi-scene SAR image data, precise orbit data and DEM data, carrying out the preprocessing of the data, and obtaining time sequence deformation, cumulative deformation and annual average deformation rate; 2, geocoding the preprocessed data, and making an SAR image landslide data set; step 3, establishing a global-local short-term dense connection network; 4, inputting the SAR image landslide data set into the global-local short-term dense connection network for training, and detecting the landslide boundary of the SAR image; 5, fusing the time sequence deformation, the accumulated deformation, the annual average deformation rate and the landslide boundary at the detection position; according to the method, global and local contexts can be captured on multiple scales, and the position and range of the landslide can be efficiently and automatically extracted with high precision.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of landslide boundary detection, and in particular to a landslide boundary automatic detection method combining SBAS-InSAR and global-local short-term dense connection network. BACKGROUND

[0002] Landslide refers to the sliding geological phenomenon of slope rock-soil mass along the through shear failure surface caused by human activities and environmental conditions. The Ministry of Natural Resources released the China Natural Resources Bulletin 2024, which shows that there were 5719 geological disasters in China in 2024, of which 3316 were landslides. Landslide hazard identification and monitoring are of great significance to prevent disasters and protect people's lives and property safety.

[0003] When monitoring landslide deformation in plateau mountainous areas, the conventional ground monitoring methods such as level, theodolite and total station are usually affected by terrain and natural environment, and have the disadvantages of high labor intensity, low work efficiency, low automation degree and high cost, and can only monitor the deformation of landslide at certain points, with low spatial resolution and cannot reflect the overall deformation of landslide. InSAR is a new spatial geodetic observation technology developed in recent decades, which has developed into D-InSAR technology and has been successfully applied to monitoring ground deformation with a theoretical accuracy of centimeter to millimeter. A series of time series InSAR technologies represented by SBAS-InSAR, PS-InSAR and coherent target method can detect centimeter or even millimeter slow deformation of the ground over a long period of time, and have the advantages of high accuracy, wide monitoring range, low cost, all-day and all-weather, etc., and have been widely applied to landslide and urban geological disaster monitoring fields.

[0004] Time series InSAR technology can detect centimeter or even millimeter slow landslide deformation over a long period of time, but it needs to manually draw the landslide boundary, which is time-consuming and laborious. The continuous emergence of deep learning methods represented by CNN and Transformer provides a rich technical method basis for intelligent landslide identification. However, compared with the field of computer vision, the overall intelligent processing of InSAR data is still in its infancy and faces the difficulty of sample scarcity, mainly due to the relative difficulty of obtaining InSAR data and the lack of high-quality labeled data sets, and there is currently a lack of landslide identification research combining deep learning and InSAR. SUMMARY

[0005] Therefore, the purpose of the present application is to provide a landslide boundary automatic detection method combining SBAS-InSAR and global-local short-term dense connection network, to at least solve the above problems.

[0006] The technical scheme adopted by the present application is as follows: A landslide boundary automatic detection method combining SBAS-InSAR and global-local short-term dense connection network, comprising the following steps: Step 1: Obtain multi-scene SAR image data, precise orbit data and DEM data, preprocess the data, and obtain time series deformation, cumulative deformation and annual average deformation rate; Step 2: Geocode the preprocessed data, make SAR image landslide data set, perform data enhancement processing, and divide into training set, validation set and test set; Step 3: Establish a global-local short-term dense connection network; Step 4: Input the SAR image landslide data set into the global-local short-term dense connection network for training, and detect the landslide boundary of the SAR image; Step 5: Fuse the time series deformation, cumulative deformation and annual average deformation rate with the detected landslide boundary to obtain an accurate landslide boundary with time series deformation, cumulative deformation and annual average deformation rate.

[0007] Further, the data preprocessing in step 1 and the obtaining of time series deformation, cumulative deformation and annual average deformation rate are as follows: Using the atmospheric correction method based on SAR data itself and the atmospheric correction method based on GACOS, using the SBAS-InSAR deformation monitoring method, the multi-scene SAR image is processed by time series InSAR to generate the unwrapped D-InSAR phase diagram and SAR intensity diagram, and the time series deformation, cumulative deformation and annual average deformation rate are obtained.

[0008] Further, step 2 specifically includes: Step 21: Use DEM to geocode the unwrapped D-InSAR phase diagram and SAR intensity diagram, and fuse them into a three-channel image; Step 22: Use ArcGIS to visually interpret the unwrapped D-InSAR phase diagram, draw the landslide boundary vector, and convert it into a raster true value label; Step 23: Use Python to crop and normalize the three-channel image and raster true value label, and use rotation, flipping, contrast transformation and other methods for data enhancement; Step 24: Divide the enhanced data into training set, validation set and test set to generate SAR image landslide data set.

[0009] Further, the global-local short-term dense connection network in step 3 includes a short-term dense connection network, a global-local attention module and a multilayer perceptron.

[0010] Further, the short-term dense connection network is composed of 5 stages, stage 1 and stage 2 each use a convolution block, the convolution block includes a convolution layer, a batch normalization layer and a ReLU activation layer; stage 3, stage 4 and stage 5 each use a short-term dense connection module with a step of 2 and a short-term dense connection module with a step of 1.

[0011] Further, the global-local attention module includes a global branch and a local branch, used to extract global and local context information, aggregate the extracted global and local context information, and output global-local context information through depth convolution, batch normalization and standard convolution; the global branch uses a window-based multi-head attention mechanism to obtain global context information; the local branch uses two parallel convolution layers to extract local context information.

[0012] Further, step 4 is specifically: Step 41: input the SAR image landslide data set and its grid true value label into the global-local short-term dense connection network, and extract four feature maps through the short-term dense connection network, the feature Figures 1-4 The output feature channel number is 64, 256, 512 and 1024 respectively, and the feature Figures 1-4 The size is 1 / 4, 1 / 8, 1 / 16 and 1 / 32 of the input image size respectively; Step 42: input the fourth feature into the convolution layer and the batch normalization layer, and then input the output feature into the global-local attention module and the multilayer perceptron output feature in turn, the channel number of the output feature is 64, and the image size of the output feature is 1 / 32 of the input image; Step 43: the output feature in step 42 and the third feature are used for feature aggregation using weighted summation operation, and the aggregated feature is input into the global-local attention module and the multilayer perceptron output feature in turn, the channel number of the output feature is 64, and the image size of the output feature is 1 / 16 of the input image; Step 44: the output feature in step 43 and the second feature are used for feature aggregation using weighted summation operation, and the aggregated feature is input into the global-local attention module and the multilayer perceptron output feature in turn, the channel number of the output feature is 64, and the image size of the output feature is 1 / 8 of the input image; Step 45: sample and decode the final feature to generate the detection result of the landslide boundary; Step 46: based on the final feature and the true value label, use the dice loss function and the cross entropy loss function to optimize the global-local short-term dense connection network.

[0013] Compared with the prior art, the beneficial effects of the present application are: The application provides a landslide boundary automatic detection method combining SBAS-InSAR and a global-local short-term dense connection network. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only preferred embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0015] Figure 1 It is a flow chart of a landslide boundary automatic detection method combining SBAS-InSAR and a global-local short-term dense connection network provided by the embodiments of the present application. Figure 2 It is a schematic diagram of a global-local short-term dense connection network provided by the embodiments of the present application. Figure 3 It is a schematic diagram of a global-local attention module provided by the embodiments of the present application. Figure 4 It is a schematic diagram of a Lu Tan No.1 strip mode SAR image provided by the embodiments of the present application. Figure 5 It is a schematic diagram of a landslide boundary recognition effect of a Lu Tan No.1 SAR image provided by the embodiments of the present application. DETAILED DESCRIPTION

[0016] The technical solutions of the present application will be further described in detail below in combination with the drawings and specific embodiments. Unless otherwise defined, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs. The terms used in the specification of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. In the following description, the expression "some embodiments" describes a subset of all possible embodiments, but it should be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments and can be combined with each other without conflict.

[0017] In the following description, numerous specific details are set forth to provide a more thorough understanding of the present application. However, it will be apparent to one of skill in the art that the present application can be practiced without one or more of these specific details. In other instances, well-known features have not been described in order to avoid obscuring the present application. The features listed below are intended to be illustrative rather than exhaustive.

[0018] It is to be understood that the application can assume various alternative forms of embodiment, and it is accordingly not restricted to the examples described and illustrated herein. On the contrary, it is intended to cover any and all modifications, combinations, equivalents, and alternatives falling within the scope of the application. In addition, throughout this disclosure, the

[0019] It is also to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising", when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0020] In order that the application can be fully understood and readily put into practical effect, there shall be described in the following detailed description only details of specific embodiments of the present application. However, it should be appreciated that the present application can be carried out without one or more of these details. In other cases, in order to avoid confusion with the present application, some technical features well known in the art are not described.

[0021] With reference to Figure 1 , Figure 2 , Figure 3 The present application provides a landslide boundary automatic detection method combining SBAS-InSAR and global-local short-term dense connection network, comprising the following steps: Step 1: Obtain multi-scene SAR image data, precise orbit data and DEM data, pre-process the data, and obtain time series deformation, cumulative deformation and annual average deformation rate; The pre-processing of the data and the obtaining of the time series deformation, the cumulative deformation and the annual average deformation rate are specifically as follows: Using the atmospheric correction method based on SAR data itself and the atmospheric correction method based on GACOS, the SBAS-InSAR deformation monitoring method is used to process the time series InSAR of multi-scene SAR images, generate the unwrapped D-InSAR phase diagram and SAR intensity diagram, and obtain the time series deformation, cumulative deformation and annual average deformation rate.

[0022] Step 2: Geocoding the preprocessed data, making SAR image landslide data set, data enhancement processing, and dividing into training set, validation set and test set, including: Step 21: Geocoding the unwrapped D-InSAR phase diagram and SAR intensity diagram using DEM, and fusing into three-channel images; Step 22: Using ArcGIS to visually interpret the unwrapped D-InSAR phase diagram, draw the landslide boundary vector, and convert it into a raster true value label; Step 23: Using Python to crop and normalize the three-channel images and raster true value labels to an appropriate size, and using rotation, flipping, contrast transformation and other methods for data enhancement; Step 24: According to the ratio of 3:1:1, the enhanced data is divided into training set, validation set and test set, and the SAR image landslide data set is generated.

[0023] Step 3: Establish a global-local short-term dense connection network, including a short-term dense connection network, a global-local attention module and a multilayer perceptron; The short-term dense connection network is composed of 5 stages, stage 1 and stage 2 each use a convolution block, the convolution block includes a convolution layer with a step of 2, a batch normalization layer and a ReLU activation layer; Stage 3, stage 4 and stage 5 use 1 short-term dense connection module with a step of 2 and a short-term dense connection module with a step of 1; The global-local attention module includes a global branch and a local branch, which is used to extract global and local context information, aggregate the extracted global and local context information, and output global-local context information through depth convolution, batch normalization and standard 1*1 convolution; The global branch uses a window-based multi-head attention mechanism to obtain global context information; The local branch uses two parallel convolution layers to extract local context information, and two batch normalization operations are added before the summation operation.

[0024] Step 4: Input the SAR image landslide data set into the global-local short-term dense connection network for training, and detect the landslide boundary of the SAR image, specifically: Step 41: Input the SAR image landslide dataset and its raster ground truth labels into a global-local short-term dense connection network. The short-term dense connection network extracts four feature maps. Figures 1-4 The number of output feature channels are 64, 256, 512, and 1024, respectively. Figures 1-4 The sizes are 1 / 4, 1 / 8, 1 / 16 and 1 / 32 of the input image size, respectively; Step 42: Input the fourth feature into the convolutional layer and the batch normalization layer, and then input the output feature into the global-local attention module and the multilayer perceptron output feature in sequence. The number of channels of the output feature is 64, and the image size of the output feature is 1 / 32 of the input image. For example, step 42 specifically involves: inputting the fourth feature into a convolutional layer and a batch normalization layer to output the first feature, the number of channels of the output first feature is 64, and the image size of the output first feature is 1 / 32 of the input image; The first feature is input into the batch normalization layer, and then input into the global-local attention module to output the second feature. The number of channels of the output second feature is 64, and the image size of the output second feature is 1 / 32 of the input image. The second feature is input into the batch normalization layer, and then into the multilayer perceptron to output the third feature. The number of channels of the output third feature is 64, and the image size of the output third feature is 1 / 32 of the input image. Step 43: Perform a weighted summation operation on the output features from Step 42 and the third feature to aggregate the features. Then, input the aggregated features into the global-local attention module and the multilayer perceptron output features in sequence. The number of channels in the output features is 64, and the image size of the output features is 1 / 16 of the input image. For example, step 43 specifically involves: performing a weighted summation operation on the third feature and the third feature to output a fourth feature, with 64 channels for the output fourth feature and the image size of the output fourth feature being 1 / 16 of the input image; The fourth feature is input into the batch normalization layer, and then into the global-local attention module to output the fifth feature. The number of channels of the output fifth feature is 64, and the image size of the output fifth feature is 1 / 16 of the input image. The fifth feature is then input into the batch normalization layer again, and then into the multilayer perceptron to output the sixth feature. The number of channels for the output sixth feature is 64, and the image size of the output sixth feature is 1 / 16 of the input image. The output features from step 43 are aggregated with the second feature using a weighted summation operation. The aggregated features are then sequentially input into the global-local attention module and the multilayer perceptron to output the final features. The final output features have 64 channels and the image size of the final output features is 1 / 8 of the input image. Specifically, step 44 is specifically: the sixth feature and the second feature are aggregated using a weighted summation operation to output a seventh feature, the channel of the output seventh feature is 64, and the image size of the output seventh feature is 1 / 8 of the input image; The seventh feature is input into a batch normalization layer, and then input into a global-local attention module to output an eighth feature, the number of channels of the output eighth feature is 64, and the image size of the output eighth feature is 1 / 8 of the input image. The eighth feature is input into a batch normalization layer, and then input into a multilayer perceptron to output a final feature, the number of channels of the output final feature is 64, and the image size of the output final feature is 1 / 8 of the input image.

[0025] Step 45: sampling and decoding the final feature to generate the detection result of the landslide boundary.

[0026] Step 46: based on the final feature and the true value label, using the dice loss function and the cross entropy loss function to optimize the global-local short-term dense connection network.

[0027] Specifically, the dice loss function is:

[0028] wherein, is the number of samples in the batch, is the total number of categories, is the true label of the sample , and is the prediction probability output of the model for the sample . The cross entropy loss function is:

[0029] wherein, is the number of samples in the batch, is the total number of categories, is the true label of the sample , and is the prediction probability output of the model for the sample , indicating the confidence of the sample belonging to the category .

[0030] Step 5: fuse the time series deformation, cumulative deformation, and annual average deformation rate with the detected landslide boundary to obtain an accurate landslide boundary with time series deformation, cumulative deformation, and annual average deformation rate.

[0031] Specifically, referring to Figure 4, The embodiment adopts Lu Tan No.1 SAR image to monitor a landslide, selects 20 scenes of domestic L-band Lu Tan No.1 strip mode 1HH polarization SAR image and precise orbit data, downloads ALOS World 3D 30m external DEM data through a website (https: / / www.eorc.jaxa.jp / ALOS / en / dataset / aw3d_e.htm), downloads high spatial resolution zenith troposphere delay provided by a GACOS system through a website (http: / / www.gacos.net), and uses the method to detect the landslide, and a recognition effect diagram obtained is as shown in Figure 5 Figure 5 In the table, the first column represents a true value label, the second column represents a landslide boundary schematic diagram detected by the method, and the third column represents a landslide boundary schematic diagram detected by a traditional method.

[0032] The above merely describes preferred embodiments of the present application and is not intended to limit the present application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.​

Claims

1. A landslide boundary automatic detection method combining SBAS-InSAR and global-local short-term dense connection network, characterized in that, The method comprises the following steps: Step 1: obtaining multi-scene SAR image data, precise orbit data and DEM data, preprocessing the data, and obtaining time series deformation, cumulative deformation and annual average deformation rate; Step 2: geocoding the preprocessed data, making a SAR image landslide data set, performing data enhancement processing, and dividing into a training set, a validation set and a test set; Step 3: establishing a global-local short-term dense connection network; Step 4: inputting the SAR image landslide data set into the global-local short-term dense connection network for training to detect the landslide boundary of the SAR image; Step 5: fusing the time series deformation, cumulative deformation and annual average deformation rate with the detected landslide boundary to obtain an accurate landslide boundary with time series deformation, cumulative deformation and annual average deformation rate. 2.The landslide boundary automatic detection method of combining SBAS-InSAR and global-local short-term dense connection network according to claim 1, wherein, In step 1, the data is preprocessed, and the time series deformation, cumulative deformation and annual average deformation rate are obtained as follows: Using the atmospheric correction method based on SAR data itself and the atmospheric correction method based on GACOS, the SBAS-InSAR deformation monitoring method is used to perform time series InSAR processing on the multi-scene SAR image to generate the unwrapped D-InSAR phase diagram and SAR intensity diagram, and obtain the time series deformation, cumulative deformation and annual average deformation rate.

3. The landslide boundary automatic detection method of claim 2, wherein, Step 2 specifically includes: Step 21: geocoding the unwrapped D-InSAR phase diagram and SAR intensity diagram using DEM, and fusing them into a three-channel image; Step 22: visually interpreting the unwrapped D-InSAR phase diagram using ArcGIS, drawing a landslide boundary vector, and converting it into a raster true value label; Step 23: using Python to crop and normalize the three-channel image and the raster true value label, and using rotation, flipping, contrast transformation and other methods for data enhancement; Step 24: dividing the enhanced data into a training set, a validation set and a test set to generate a SAR image landslide data set.

4. The landslide boundary automatic detection method of claim 3, wherein, The global-local short-term dense connection network in step 3 includes a short-term dense connection network, a global-local attention module and a multilayer perceptron.

5. The landslide boundary automatic detection method of claim 4, wherein, The short-term dense connection network is composed of 5 stages, stage 1 and stage 2 each use a convolution block, the convolution block includes a convolution layer, a batch normalization layer and a ReLU activation layer; stage 3, stage 4 and stage 5 each use a short-term dense connection module with a step of 2 and a short-term dense connection module with a step of 1.

6. The landslide boundary automatic detection method of claim 5, wherein, The global-local attention module includes a global branch and a local branch, which are used to extract global and local context information, aggregate the extracted global and local context information, and output global-local context information through depth convolution, batch normalization and standard convolution; the global branch uses a window-based multi-head attention mechanism to obtain global context information; the local branch uses two parallel convolution layers to extract local context information.

7. The landslide boundary automatic detection method of claim 6, wherein, Step 4 specifically includes: Step 41: input the SAR image landslide dataset and its grid true value label into the global-local short-term dense connection network, and extract four feature maps through the short-term dense connection network, the output feature channel numbers of feature maps 1-4 are 64, 256, 512 and 1024 respectively, and the sizes of feature maps 1-4 are 1 / 4, 1 / 8, 1 / 16 and 1 / 32 of the input image size respectively; Step 42: input the fourth feature into the convolution layer and the batch normalization layer, and then input the output feature into the global-local attention module and the multilayer perceptron output feature in turn, the channel number of the output feature is 64, and the image size of the output feature is 1 / 32 of the input image; Step 43: use weighted summation operation to aggregate the output feature in step 42 and the third feature, and then input the aggregated feature into the global-local attention module and the multilayer perceptron output feature in turn, the channel number of the output feature is 64, and the image size of the output feature is 1 / 16 of the input image; Step 44: use weighted summation operation to aggregate the output feature in step 43 and the second feature, and then input the aggregated feature into the global-local attention module and the multilayer perceptron output feature in turn to output the final feature, the channel number of the final feature is 64, and the image size of the final feature is 1 / 8 of the input image; Step 45: sample and decode the final feature to generate the detection result of the landslide boundary; Step 46: based on the final feature and the true value label, use the dice loss function and the cross entropy loss function to optimize the global-local short-term dense connection network.

Images