Landslide sample data set construction method based on remote sensing image

Abstract

The invention discloses a landslide sample data set construction method based on a remote sensing image. And a high-quality and information-rich landslide sample data set is constructed to directly influence the intelligent geological element information extraction capability. The method comprises the following steps: acquiring an optical satellite image and digital elevation model (DEM) data of a research area, and preprocessing; marking a landslide in the corresponding area according to the determined area range; extracting an image and mark data of the research area; and segmenting the obtained image and mark data into a data set. According to the method, the Google Earth can be used for identifying the landslide by means of the terrain height difference, and compared with a common optical image, the landslide is easier to identify and the identification precision is higher; the quadtree algorithm is adopted for landslide sample cutting, and samples obtained for landslides of different scales are more complete and higher in quality.

Description

Technical Field

[0001] The invention belongs to the technical field of remote sensing interpretation, and in particular relates to a method for constructing a landslide sample data set based on remote sensing images. Background Art

[0002] Landslides and other surface geological disasters are key geological factors that affect the entire life cycle of line selection, construction, and operation and maintenance of typical complex and dangerous mountain railways and large-scale water conservancy and hydropower projects. Such projects involve large areas, multiple types of disasters, and high levels of project impact, and therefore have a significant demand for remote sensing interpretation.

[0003] In recent years, people have used deep learning technology to significantly improve the results of image feature extraction. However, due to the narrow application scenarios and the strong professional knowledge required, it is difficult to construct landslide remote sensing sample datasets in engineering geology, and the automatic interpretation capability of multi-source remote sensing images is still insufficient. Currently, there are few landslide sample datasets, and there are problems such as inconsistent classification systems, small interpretation sample size, insufficient diversity, single sample image source, and fixed sample size, which have seriously affected the efficiency and quality of large-scale multi-source heterogeneous remote sensing image interpretation. Constructing a high-quality, information-rich landslide sample dataset directly affects the ability to intelligently extract geological element information, and is the basis and key step for the successful implementation of intelligent remote sensing landslide interpretation. Summary of the invention

[0004] In order to make up for the shortcomings of the prior art, the present invention provides a method for constructing a landslide sample data set based on remote sensing images.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A method for constructing a landslide sample data set based on remote sensing images, characterized in that it comprises the following steps:

[0007] S1: Obtain optical satellite images and digital elevation model (DEM) data of the study area and perform preprocessing;

[0008] S2: According to the area range determined in step S1, marking the landslide in the corresponding area;

[0009] S3: Extract image and label data of the study area;

[0010] S4: Divide the images and labeled data obtained in S3 into data sets.

[0011] Further, the specific method of step S1 is:

[0012] S1-1: After obtaining the optical remote sensing satellite data, pre-process it first, and then cut out the optical satellite image of the corresponding size according to the longitude and latitude of the study area;

[0013] S1-2: The DEM data produced by synthetic aperture radar is resampled using the cubic convolution interpolation method to obtain DEM data with a spatial resolution of 2.5m.

[0014] Further, the specific method of step S2 is:

[0015] S2-1: Input the longitude and latitude coordinates of the study area determined in step S1 into Google Earth, switch to the terrain data view, and then mark the landslides in the area through manual visual interpretation;

[0016] S2-2: Manual visual interpretation needs to be distinguished based on the image characteristics of the landslide;

[0017] S2-3: Marking landslides. After marking all the landslides in the study area, export them as KML files.

[0018] Further, the specific method of step S3 is:

[0019] S3-1: Convert the KML file obtained in step S2 into a shapefile;

[0020] S3-2: Import the optical satellite image and DEM data obtained in step S1 and the shapefile file obtained in step S3-1 into ENVI software respectively, take the shapefile file as the area of ​​interest, extract the optical image and DEM data, obtain the optical image and DEM data of the same size, and convert the shapefile file into binary labeled data.

[0021] Further, the specific method of step S4 is:

[0022] S4-1: according to the labeled data obtained in step S3, the quadtree algorithm is used to extract image blocks of different scales containing the target of interest from the entire remote sensing image, and used as model training samples;

[0023] S4-2: Extract image blocks that do not contain landslide areas as negative samples in the sample set;

[0024] S4-3: The final sample set contains 3 folders.

[0025] Furthermore, in S1-1, after obtaining the optical remote sensing satellite data, noise reduction, shadow removal, and geometric correction preprocessing are first performed.

[0026] Furthermore, in: S2-3, the landslide was marked using the polygon drawing tool of Google Earth.

[0027] Further, in S4-2, image blocks that do not include the landslide area are extracted at a 1:1 ratio.

[0028] Further, in S3-1, the KML file obtained in S2 is imported into the Aowei interactive map software for format conversion.

[0029] Furthermore, the three folders are optical images, DEM data, and marker data.

[0030] Beneficial effects of the present invention:

[0031] 1) The present invention uses Google Earth to identify landslides with the help of terrain height differences, which is easier to identify and has higher identification accuracy than using general optical images;

[0032] 2) The present invention adopts a quadtree algorithm to cut landslide samples, and the samples obtained for landslides of different scales are more complete and of higher quality;

[0033] 3) The present invention fills the gap in the railway survey process, has high production efficiency, and obtains samples of good quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is the operation flow chart of the present invention;

[0035] Figure 2 It is a cropped and pre-processed optical image;

[0036] Figure 3 It is a resampled DEM image;

[0037] Figure 4 is the marked landslide data;

[0038] Figure 5 The visualization results of quadtree segmentation are shown;

[0039] Figure 6 This is the final landslide sample dataset. DETAILED DESCRIPTION

[0040] The present invention is described in detail below in conjunction with specific implementation modes.

[0041] The present invention is suitable for use as a training data set for intelligent interpretation algorithms during landslide surveys and monitoring in engineering geological surveys. Figure 1 As shown, the landslide sample data set construction method of the present invention comprises the following steps:

[0042] S1: Obtain optical satellite images and digital elevation model (DEM) data of the study area and perform preprocessing;

[0043] S2: According to the area range determined in step S1, marking the landslide in the corresponding area;

[0044] S3: Extract image and label data of the study area;

[0045] S4: Divide the images and labeled data obtained in S3 into data sets.

[0046] 1) The specific method of step S1 is:

[0047] S1-1: After obtaining the optical remote sensing satellite data, first perform noise reduction, shadow removal, and geometric correction preprocessing, and then crop the optical satellite image of the corresponding size according to the longitude and latitude of the study area, such as Figure 2 As shown;

[0048] S1-2: The DEM data produced by the synthetic aperture radar (SAR) is resampled using the cubic convolution interpolation method because its spatial resolution is relatively large at 12.5m, and the DEM data with a spatial resolution of 2.5m is obtained, as shown in Figure 3 shown.

[0049] 2) The specific method of step S2 is:

[0050] S2-1: Input the longitude and latitude coordinates of the study area determined in step S1 into Google Earth, switch to the terrain data, and then mark the landslides in the area through manual visual interpretation;

[0051] S2-2: Manual visual interpretation needs to be distinguished based on the image characteristics of the landslide, which are characterized by a slope of 14-25°, an obvious landslide wall on the rear edge, double groove homologous features on the side edge, and a dustpan-shaped or arc-shaped slope foot;

[0052] S2-3: Use the polygon drawing tool of Google Earth to mark the landslides. After marking all the landslides in the study area, export them into a KML file. Figure 4 are the labeled landslide data.

[0053] 3) The specific method of step S3 is:

[0054] S3-1: Import the KML file obtained in step S2 into the Aowei interactive map software for format conversion and export the shapefile file;

[0055] S3-2: Import the optical satellite image and DEM data obtained in step S1 and the shapefile obtained in step S3-1 into ENVI software respectively, take the shapefile as the area of ​​interest, extract the optical image and DEM data, obtain the optical image and DEM data of the same size, and convert the shapefile into binary labeled data, such as Figure 4 shown.

[0056] 4) The specific method of step S4 is:

[0057] S4-1: Based on the labeled data obtained in step S3, the quadtree algorithm is used to extract image blocks of different scales containing the target of interest from the entire remote sensing image, so that the extracted image blocks contain as many landslide areas as possible, which are used as model training samples, such as Figure 5 As shown;

[0058] S4-2: Extract image blocks that do not contain landslide areas according to a 1:1 ratio and use them as negative samples in the sample set;

[0059] S4-3: The final sample set contains three folders, namely optical images (images), DEM data (DEMs) and marked data (masks), such as Figure 6 shown.

[0060] The content of the present invention is not limited to the embodiments listed, and any equivalent changes made to the technical solution of the present invention by ordinary technicians in this field after reading the description of the present invention are covered by the claims of the present invention.

Claims

1. A method for constructing a landslide sample dataset based on remote sensing images, characterized in that: The steps include: S1: Obtain optical satellite images and digital elevation model (DEM) data of the study area and perform preprocessing; S2: According to the area range determined in step S1, marking the landslide in the corresponding area; S3: Extract image and label data of the study area; S4: Divide the images and labeled data obtained in S3 into data sets.

2. The method for constructing a landslide sample dataset based on remote sensing images according to claim 1, characterized in that: The specific method of step S1 is: S1-1: After obtaining the optical remote sensing satellite data, pre-process it first, and then cut out the optical satellite image of the corresponding size according to the longitude and latitude of the study area; S1-2: The DEM data produced by synthetic aperture radar is resampled using the cubic convolution interpolation method to obtain DEM data with a spatial resolution of 2.5m.

3. The method for constructing a landslide sample data set based on remote sensing images according to claim 2, characterized in that: The specific method of step S2 is: S2-1: Input the longitude and latitude coordinates of the study area determined in step S1 into Google Earth, switch to the terrain data view, and then mark the landslides in the area through manual visual interpretation; S2-2: Manual visual interpretation needs to be distinguished based on the image characteristics of the landslide; S2-3: Marking landslides. After marking all the landslides in the study area, export them as KML files.

4. The method for constructing a landslide sample data set based on remote sensing images according to claim 3 is characterized by: The specific method of step S3 is: S3-1: Convert the KML file obtained in step S2 into a shapefile; S3-2: Import the optical satellite image and DEM data obtained in step S1 and the shapefile file obtained in step S3-1 into ENVI software respectively, take the shapefile file as the area of ​​interest, extract the optical image and DEM data, obtain the optical image and DEM data of the same size, and convert the shapefile file into binary labeled data.

5. The method for constructing a landslide sample data set based on remote sensing images according to claim 4, characterized in that: The specific method of step S4 is: S4-1: according to the labeled data obtained in step S3, the quadtree algorithm is used to extract image blocks of different scales containing the target of interest from the entire remote sensing image, and used as model training samples; S4-2: Extract image blocks that do not contain landslide areas as negative samples in the sample set; S4-3: The final sample set contains 3 folders.

6. The method for constructing a landslide sample data set based on remote sensing images according to claim 5, characterized in that: In S1-1, after obtaining the optical remote sensing satellite data, noise reduction, shadow removal, and geometric correction preprocessing are performed first.

7. The method for constructing a landslide sample data set based on remote sensing images according to claim 6, characterized in that: In S2-3, the landslide was marked using the polygon drawing tool of Google Earth.

8. The method for constructing a landslide sample data set based on remote sensing images according to claim 7, characterized in that: In S4-2, the image blocks excluding the landslide area are extracted at a 1:1 ratio.

9. The method for constructing a landslide sample data set based on remote sensing images according to claim 8, characterized in that: In S3-1, the KML file obtained in S2 is imported into the Aowei interactive map software for format conversion.

10. The method for constructing a landslide sample data set based on remote sensing images according to claim 9, characterized in that: The three folders are optical images, DEM data and marker data.

Images