Method and apparatus for extracting the morphological structure of shallow small landslides

Abstract

The present invention discloses a method and device for extracting the morphological structure of shallow small-scale landslides. The method includes: generating a digital orthophoto map (DOM) and a digital elevation model (DEM) based on the original data obtained from aerial surveys of the study area of shallow small-scale landslides; selecting the optimal segmentation scale for different ground objects in the DOM to construct a multi-level model; constructing a multi-feature rule set according to the interpretation features of various ground objects, and extracting landslide candidate areas layer by layer in the multi-level model to obtain the landslide spatial range; based on this range and the DEM, using the strip profile method to conduct a detailed analysis of the terrain of the landslide area, and determining the optimal threshold to extract the morphological structure of shallow small-scale landslides. The present invention can construct a multi-level morphological structure extraction model for shallow small-scale landslides in sudden-occurring areas, accurately quantify the terrain undulation changes of shallow small-scale landslides, and then effectively extract their morphological structures, which is of great significance for disaster assessment, emergency rescue, reconstruction planning, etc.

Description

Technical Field

[0001] The present invention relates to the technical field of landslide geological disaster identification, and particularly to a method and device for extracting the morphological structure of shallow small landslides considering multi-feature information of UAV images. Background Art

[0002] This section aims to provide background or context for the embodiments of the present invention stated in the claims. The descriptions herein are not admitted to be prior art merely because they are included in this section.

[0003] A landslide is a phenomenon in which a part of a slope is affected by external factors and undergoes shear movement along one or more slip surfaces in the slope direction under the action of gravity. After a landslide disaster occurs, a series of special landslide terrains will be formed on the slope, such as landslide walls, source areas, sliding areas, accumulation areas, etc. These landslide terrain elements are also called the morphological structure of the landslide. Qualitatively, positionally, and quantitatively obtaining relevant information on the landslide morphological structure is crucial for understanding the characteristics of landslide geological disasters, quantitatively assessing disasters, and identifying unstable landslides.

[0004] A Digital Elevation Model (DEM) contains a large amount of topographic and geomorphic information and can better describe the fine geomorphic surface change characteristics in the landslide area, making it possible to extract the landslide morphological structure. Currently, the methods for extracting the landslide morphological structure using DEM mainly include visual interpretation, pixel-based statistical analysis methods, and object-based image analysis methods. Visual interpretation relies on a large amount of prior knowledge and experience. Pixel-based statistical analysis methods are suitable for large landslide areas with obvious terrain undulations. Object-based image analysis methods have become a powerful tool for extracting the morphological structure of shallow small landslides, and can divide the complex internal structure of the landslide into objects with specific spatial organizations to distinguish different landslide terrain elements. However, in the extraction process, it faces a huge challenge of threshold setting. At the same time, due to the suddenness of landslides, in most cases, only post-disaster DEM data can be obtained. Therefore, how to use post-disaster DEM to accurately quantify the terrain undulation changes of shallow small landslides and then effectively extract their morphological structure is a key issue. Summary of the Invention

[0005] Embodiments of the present invention provide a method for extracting the morphological structure of shallow small landslides, which is used to accurately quantify the terrain undulation changes of shallow small landslides in a sudden shallow small landslide area using post-disaster DEM, and then effectively extract their morphological structure. The method includes:

[0006] Generating a Digital Orthophoto Map (DOM) and a DEM according to the original data obtained by aerial survey of the study area of shallow small landslides;

[0007] Through a multi-scale segmentation method, the optimal segmentation scale is selected according to the spectral and shape features of different ground objects in the DOM, and a multi-level model is constructed based on the optimal segmentation scale; the multi-level model is used to depict objects at three different description scales of the landslide background area, the landslide itself, and its morphological structure.

[0008] A multi-feature rule set is constructed according to the interpretation features of various ground objects in the DOM, and the landslide candidate areas are extracted layer by layer in the multi-level model according to the multi-feature rule set to obtain the landslide spatial range.

[0009] Based on the landslide spatial range and the DEM, the strip profile method is used to conduct a detailed analysis of the terrain in the landslide study area, and after determining the optimal threshold, the morphological structure of shallow small landslides is extracted.

[0010] An embodiment of the present invention also provides a device for extracting the morphological structure of shallow small landslides, which is used to accurately quantify the terrain undulation changes of shallow small landslides in the sudden shallow small landslide area by using the post-disaster DEM, and then effectively extract its morphological structure. The device includes:

[0011] A generation unit for generating DOM and DEM according to the original data obtained from the aerial survey of the shallow small landslide study area.

[0012] A model construction unit for selecting the optimal segmentation scale according to the spectral and shape features of different ground objects in the DOM through a multi-scale segmentation method, and constructing a multi-level model based on the optimal segmentation scale; the multi-level model is used to depict objects at three different description scales of the landslide background area, the landslide itself, and its morphological structure.

[0013] A landslide spatial range determination unit for constructing a multi-feature rule set according to the interpretation features of various ground objects in the DOM, and extracting the landslide candidate areas layer by layer in the multi-level model according to the multi-feature rule set to obtain the landslide spatial range.

[0014] A morphological structure extraction unit for conducting a detailed analysis of the terrain in the landslide study area based on the landslide spatial range and the DEM by using the strip profile method, and extracting the morphological structure of shallow small landslides after determining the optimal threshold.

[0015] An embodiment of the present invention also provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the above method for extracting the morphological structure of shallow small landslides is implemented.

[0016] An embodiment of the present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above method for extracting the morphological structure of shallow small landslides is implemented.

[0017] An embodiment of the present invention also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the above-mentioned method for extracting the morphological structure of shallow small landslides.

[0018] In the embodiment of the present invention, for the solution for extracting the morphological structure of shallow small landslides, specifically: the method for extracting the morphological structure of shallow small landslides provided by the embodiment of the present invention, when working: based on the original data obtained from aerial surveys of the research area of shallow small landslides, DOM and DEM are generated; through the multi-scale segmentation method, the optimal segmentation scale is selected according to the spectral and shape characteristics of different ground objects in the DOM, and a multi-level model is constructed according to the optimal segmentation scale; the multi-level model is used to depict objects at three different description scales: the landslide background area, the landslide itself, and its morphological structure; a multi-feature rule set is constructed according to the interpretation features of various ground objects in the DOM, and the landslide candidate areas are extracted layer by layer in the multi-level model according to the multi-feature rule set to obtain the landslide spatial range; based on the landslide spatial range and the DEM, the strip profile method is used to conduct a detailed analysis of the terrain of the landslide research area, and after determining the optimal threshold, the morphological structure of the shallow small landslide is extracted. This solution can construct a multi-level extraction model for the morphological structure of shallow small landslides in the area of sudden shallow small landslides, accurately quantify the terrain undulation changes of shallow small landslides, and then effectively extract their morphological structure, which is of great significance for disaster assessment, emergency rescue, reconstruction planning, and other aspects. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] 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 description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings. In the drawings:

[0020] Figure 1 is the overall flowchart of the method for extracting the morphological structure of shallow small landslides considering multi-feature information of UAV images in the embodiment of the present invention;

[0021] Figure 2a is the UAV DOM of a certain survey area obtained through processing in the embodiment of the present invention;

[0022] Figure 2b is the DEM in the embodiment of the present invention;

[0023] Figure 3a is the result of the first segmentation layer of multi-scale and multi-level in the embodiment of the present invention;

[0024] Figure 3bIt is the result of the second segmentation layer with multiple scales and levels in the embodiment of the present invention;

[0025] Figure 3c It is the result of the third segmentation layer with multiple scales and levels in the embodiment of the present invention;

[0026] Figure 4 It is the sub - flow chart for hierarchically extracting the landslide spatial range by a multi - feature rule set in the embodiment of the present invention;

[0027] Figure 5 It is the result map of the landslide spatial range extraction in the embodiment of the present invention;

[0028] Figure 6 It is the sub - flow chart for extracting the landslide morphological structure by the strip profile method in the embodiment of the present invention;

[0029] Figure 7 It is the result map of the landslide morphological structure extraction in the embodiment of the present invention;

[0030] Figure 8 It is the schematic flow chart of the method for extracting the morphological structure of shallow small - scale landslides in the embodiment of the present invention;

[0031] Figure 9 It is the schematic structural diagram of the device for extracting the morphological structure of shallow small - scale landslides in the embodiment of the present invention. Detailed implementation manners

[0032] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer and more understandable, the following further elaborates on the embodiments of the present invention with reference to the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but not to limit the present invention.

[0033] The embodiments of the present invention disclose a shallow small - scale landslide (shallow: depth less than 10m; volume less than 10×10 4Scheme for extracting the morphological structure of small landslides, i.e., this scheme is a scheme for extracting the morphological structure of shallow small landslides based on UAV images, including the following steps: (1) Conduct UAV aerial survey and photography on the demarcated landslide area, and the obtained original data is preprocessed to generate a Digital OrthophotoMap (DOM) and a Digital Elevation Model (DEM); (2) Select the optimal segmentation scale through a multi-scale segmentation method and construct a top-down multi-level model to facilitate the characterization of objects at different description scales; (3) Construct a multi-feature rule set to hierarchically extract the landslide spatial range; (4) On the basis of effectively identifying the landslide spatial range, use the strip profile method to conduct a detailed analysis of the terrain in the landslide area and determine the optimal threshold to achieve the extraction of the landslide morphological structure. The embodiments of the present invention can construct a multi-level extraction model for the morphological structure of shallow small landslides by integrating the multi-feature information of UAV DOM and post-disaster DEM data in the area of sudden shallow small landslides, effectively identify the internal terrain elements of the landslide, and are of great significance for disaster assessment, emergency rescue, reconstruction planning, etc. The following is a detailed introduction to the scheme for extracting the morphological structure of shallow small landslides.

[0034] Figure 8 It is a schematic flow chart of the method for extracting the morphological structure of shallow small landslides in the embodiments of the present invention, as Figure 8 described, this method includes the following steps:

[0035] Step 101: Generate DOM and DEM based on the original data obtained from the aerial survey of the research area of shallow small landslides;

[0036] Step 102: Through the multi-scale segmentation method, select the optimal segmentation scale for the spectral and shape features of different ground objects in the DOM, and construct a multi-level model according to the optimal segmentation scale; the multi-level model is used to characterize objects at three different description scales: the landslide background area, the landslide itself, and its morphological structure;

[0037] Step 103: Construct a multi-feature rule set according to the interpretation features of various ground objects in the DOM, and layer by layer extract the landslide candidate areas and eliminate the landslide false alarm areas in the multi-level model to obtain the landslide spatial range;

[0038] Step 104: Based on the landslide spatial range and DEM, use the strip profile method to conduct a detailed analysis of the terrain in the landslide research area, determine the optimal threshold, and then extract the morphological structure of shallow small landslides.

[0039] The extraction method of the shallow small landslide morphological structure provided by the embodiment of the present invention, when working: generate DOM and DEM according to the original data obtained by aerial survey of the shallow small landslide research area; through the multi-scale segmentation method, select the optimal segmentation scale according to the spectral and shape characteristics of different ground objects in the DOM, and construct a multi-level model according to the optimal segmentation scale; the multi-level model is used to depict objects at three different description scales of the landslide background area, the landslide itself and its morphological structure; construct a multi-feature rule set according to the interpretation characteristics of various ground objects in the DOM, and layer by layer extract the landslide candidate area and eliminate the landslide false alarm area in the multi-level model to obtain the landslide spatial range; based on the landslide spatial range and DEM, use the strip profile method to finely analyze the terrain of the landslide research area, determine the optimal threshold and then extract the shallow small landslide morphological structure. This method can construct a multi-level shallow small landslide morphological structure extraction model in the sudden shallow small landslide area, accurately quantify the terrain undulation change of the shallow small landslide, and then effectively extract its morphological structure, which is of great significance to disaster assessment, emergency rescue, reconstruction planning and other aspects. The following is a detailed introduction to the extraction method of the shallow small landslide morphological structure.

[0040] The purpose of the embodiment of the present invention is to provide a method for extracting the shallow small landslide morphological structure considering multi-feature information of UAV images, so as to accurately quantify the terrain undulation change of the shallow small landslide by using the post-disaster DEM in the sudden shallow small landslide area, and then effectively extract its morphological structure.

[0041] To achieve the above invention purpose, the method for extracting the shallow small landslide morphological structure considering multi-feature information of UAV images provided by the present invention specifically includes the following steps:

[0042] Step S1: Conduct UAV aerial survey shooting on the demarcated landslide area, and generate DOM and DEM after preprocessing the obtained original data; that is, after preprocessing the obtained original data, generate DOM and post-disaster DEM;

[0043] Step S2: Through a multi-scale and multi-level segmentation method, select the optimal segmentation scale for the spectral and shape features of different ground objects in the DOM (select the optimal segmentation scale for the target ground objects at each level), and construct a top-down multi-level model to facilitate the description of objects at three different scales of the landslide background area, the landslide itself, and its morphological structure (the object contains ground objects, that is, the same type of adjacent ground objects with similar spectral features are segmented and merged into the same object, and these objects are not merged with other types of ground objects and have clear boundaries, taking into account to a large extent the spectral, texture, geometric, spatial and other feature information and local details of the ground objects shown in the DOM): The higher the level of the segmentation scale, the more detailed the segmentation object. There is a topological relationship between high-level objects and low-level objects, that is, the landslide background area is characterized at a larger scale (the first scale), mainly by segmenting the vegetation with the largest coverage area in the study area. The medium scale (the second scale) level segments out ground objects such as landslide bodies, roads, bare soil, and buildings. The smaller scale (the third scale) level mainly segments the internal morphological structure of the landslide;

[0044] Step S3: Combine the interpretation features of various ground objects in the DOM of the study area, construct a multi-feature rule set, and layer by layer extract landslide candidate areas and eliminate false alarms of landslides in the multi-level model in Step S2, so as to obtain the landslide spatial range;

[0045] Step S4: On the basis of effectively identifying the landslide spatial range, use the strip profile method to finely analyze the terrain of the landslide area, determine the optimal threshold, and then extract the landslide morphological structure.

[0046] As an optimal implementation method, the specific steps of the UAV data preprocessing in Step S1 are as follows:

[0047] Step S1.1: First, perform aerial triangulation on the UAV visible light RGB images (the pixel coordinates can be directly measured from the photos, but the corresponding ground coordinates are required in the end. Therefore, through this step of aerial triangulation, the ground coordinates of these points are calculated). Combine the UAV positioning POS data with the ground control point coordinates measured in actuality for joint regional network adjustment, solve the true spatial position and attitude of the images, obtain the ground coordinates of the key connection points (the key connection points refer to the common feature points in multiple photos, and some key points can form a sparse point cloud), and generate a sparse point cloud.

[0048] Step S1.2: For the images whose true spatial position and attitude have been restored in Step S1.1, use the multi-view image dense matching technology to identify the homologous points in multiple images and establish a high-density point cloud of the survey area.

[0049] Step S1.3: Use the dense point cloud generated in Step S1.2 to construct an irregular triangular network, thereby establishing a digital surface model (DSM).

[0050] Step S1.4: Process the irregular triangular grid generated in Step S1.3 with an encryption filtering algorithm to separate ground points from non-ground points, remove non-ground point information such as vegetation and buildings, retain only ground point information, and construct an elevation grid for the ground points to generate the DEM of the study area.

[0051] Step S1.5: According to the DEM model generated in Step S1.4, orthorectify the single-frame UAV images, and perform mosaicking, cropping, etc. on the overlapping areas to finally obtain the complete DOM of the survey area.

[0052] As can be seen from the above, in one embodiment, the original data of the aerial survey is UAV visible light RGB images and UAV POS data; according to the original data of the aerial survey of the shallow small landslide study area, generating DOM and DEM may include:

[0053] Perform aerial triangulation on the UAV visible light RGB images, and perform joint block adjustment by combining the UAV positioning POS data with the ground control point coordinates measured actually to solve the true spatial position and attitude of the images, obtain the ground coordinates of the key connection points, and generate a sparse point cloud;

[0054] Adopt the multi-view image dense matching technology to identify the corresponding points in multiple images from the images with true spatial position and attitude, and establish the dense point cloud of the survey area according to the corresponding points in multiple images and the sparse point cloud;

[0055] Use the dense point cloud to construct an irregular triangular grid;

[0056] Process the irregular triangular grid with an encryption filtering algorithm to separate the ground points from the non-ground points to obtain the ground point information, construct an elevation grid for the ground points, and generate the DEM of the study area;

[0057] According to the DEM, orthorectify the single-frame UAV images, and perform mosaicking and cropping on the overlapping areas to finally obtain the complete DOM of the study area.

[0058] As a preferred embodiment, the specific steps of constructing a top-down multi-level model by the multi-scale segmentation method in Step S2 are as follows:

[0059] Step S2.1: Load the UAV DOM, and set the four main parameters of layer weight, scale factor, shape factor, and compactness factor. Among them, the layers participating in the segmentation are the R, G, and B layers, and the weight of each layer is 1.

[0060] Step S2.2: Set both the shape factor (the shape factor affects the geometric feature differences of the segmented object) and the compactness factor (the compactness factor focuses on showing the overall tightness of the image segmentation object) to 0.5. The scale parameter ranges from [50, 300] and increases in units of 25 for experiments. Then, according to the effects, select a more suitable scale parameter. This step is to use the method of controlling variables. After fixing the shape factor and the compactness factor, conduct multiple experiments on the scale factor and select a suitable value (preferred scale parameter value) according to the segmentation effect;

[0061] Step S2.3: Set the scale factor to the preferred scale parameter value obtained in Step S2.2, and adjust the shape factor and the compactness factor respectively. According to the segmentation effect of the actual ground objects in the UAV image, determine the final parameter values of the shape factor and the compactness factor;

[0062] Step S2.4: Under the condition of determining the parameter values of the shape factor and the compactness factor in Step S2.3, use the ESP algorithm to adjust the starting scale and the number of cycles to obtain the local variance change rate (ROC-LV) curve:

[0063]

[0064] Among them, L is the local variance of the target level, and L-1 is the local variance of the next level. Select the point where the peak value is located as the potential optimal segmentation scale. Conduct experiments on each potential optimal scale parameter respectively, find the more suitable parameters for the target ground object, and determine it as the final scale factor (see the introduction of the following example. The final scale factor of the test experiment in the embodiment of the present invention is 215), thus completing the segmentation of the first level. The operations of the segmentation of the first level are the operations executed in the above Steps S2.1 to S2.4.

[0065] Step S2.5: Inherit the segmentation result of Step S2.4, and repeat Steps S2.2 to 2.4 for the target ground object (the first-level target ground object: vegetation; the second-level target ground objects: landslides, bare lands, roads, buildings, etc.; the third-level target ground object: the internal morphological structure of the landslide) to further refine the segmented object and obtain the segmentation result of the second level.

[0066] Step S2.6: Inherit the segmentation result of Step S2.5, and repeat Steps S2.2 to 2.4 for the target ground object to refine the segmented object again and obtain the segmentation result of the third level.

[0067] As can be seen from the above, in one embodiment, through the multi-scale segmentation method, the optimal segmentation scale is selected according to the spectral and shape features of different ground objects in the DOM, and a multi-level model can be constructed according to the optimal segmentation scale, including:

[0068] Load the DOM; the layers participating in the segmentation in the DOM are the three layers of R, G, and B, and the weight of each layer is 1;

[0069] Set both the shape factor and the compactness factor to 0.5, and the scale parameter ranges from [50, 300] and increases in increments of 25 for experiments. Then, select the optimal scale parameter value according to the effect;

[0070] Set the scale factor to the optimal scale parameter value, and adjust the shape factor and the compactness factor respectively. Determine the final parameter values of the shape factor and the compactness factor according to the segmentation effect of the actual ground objects in the DOM;

[0071] When the final parameter values of the shape factor and the compactness factor are determined, use the ESP algorithm to adjust the initial scale and the number of loops to obtain the local variance change rate curve; select the point where the peak of the curve is located as the potential optimal segmentation scale, and conduct experiments on the parameters of each potential optimal segmentation scale respectively. Find the parameters most suitable for the target ground object as the final optimal segmentation scale, and complete the first-level segmentation according to the final optimal segmentation scale to obtain the first-level segmentation result;

[0072] Inherit the first-level segmentation result, and perform the second-level segmentation operation for the target ground object with reference to the operation of the first-level segmentation, further refining the segmented object to obtain the second-level segmentation result;

[0073] Inherit the second-level segmentation result, and perform the third-level segmentation operation for the target ground object with reference to the operation of the first-level segmentation, refining the segmented object again to obtain the third-level segmentation result, and obtain the multi-level model.

[0074] As a preferred implementation, in step S3, combine the interpretation features of various ground objects in the combined area to construct a multi-feature rule set, and extract the landslide candidate areas and eliminate the landslide false alarm areas layer by layer in the multi-level model of step S2. The specific steps to obtain the landslide spatial range are as follows:

[0075] Step S3.1: Conduct feature analysis on landslides and non-landslides, and establish direct or indirect interpretation signs such as the tone, shape, and position of ground objects.

[0076] Step S3.2: For the segmented objects obtained in step 2 (the objects obtained after segmenting various ground objects in the DOM), calculate the feature values of each object (including spectral mean, aspect ratio, mean of gray-level co-occurrence matrix, etc.) to obtain the multi-feature attributes such as spectrum, texture, geometry, and space of the object.

[0077] Step S3.3: Selection of sample vector points, establishment of sample point files for ground features such as vegetation, bare land, buildings, roads, and landslides, and collection of sample points respectively.

[0078] Step S3.4: Select corresponding segmentation objects as sample objects based on the positions of the sample points generated in Step S3.3.

[0079] Step S3.5: Extract landslide candidate areas (i.e., non-vegetation) in the first segmentation layer (segmentation result of the first level) obtained in Step S2.4, with the aim of removing ground features (vegetation) with large areas, concentrated distributions, and relatively regular shapes in the area.

[0080] Specifically, the visible light vegetation index VDVI can be used to distinguish vegetation and non-vegetation in the area, remove large areas of vegetation with high VDVI values, and thus obtain landslide candidate areas:

[0081]

[0082] Among them, R, G, and B respectively represent the red, green, and blue bands in the UAV DOM.

[0083] Step S3.6: After performing class filtering on the extraction results (vegetation and non-vegetation) of Step S3.5 to obtain landslide candidate areas, inherit them in the second segmentation layer (segmentation result of the second level) of Step S2.5, and use eigenvalue such as band mean, brightness value, mean or standard deviation of gray-level co-occurrence matrix, aspect ratio, shape index, boundary index, coordinates, etc., to obtain multi-feature information such as spectrum, texture, geometry, and space of ground features such as bare land, buildings, roads, and landslides, and remove false alarm landslide areas (non-landslide areas such as bare land, buildings, roads, etc.) one by one to obtain the landslide spatial range.

[0084] As can be seen from the above, in one embodiment, constructing a multi-feature rule set according to the interpretation features of various ground features in the DOM, and extracting landslide candidate areas and removing false alarm landslide areas layer by layer in the multi-level model to obtain the landslide spatial range may include:

[0085] Constructing a multi-feature rule set according to the interpretation features of various ground features in the DOM, and extracting landslide candidate areas and removing false alarm landslide areas layer by layer in the multi-level model to obtain the landslide spatial range, including:

[0086] Conduct feature analysis on landslides and non-landslides to establish direct or indirect interpretation marks;

[0087] Based on the direct or indirect interpretation marks, calculate multiple feature values of each object for segmentation objects at different description scales to obtain the multi-feature attributes of the objects;

[0088] Based on the multi-feature attributes of objects, sample point files of different types of ground objects are established, and the acquisition of sample vector points is carried out separately.

[0089] Taking the positions of the acquired sample vector points as the reference, the corresponding segmentation objects are selected as sample objects.

[0090] Based on the sample objects, the extraction of landslide candidate areas is carried out in the segmentation results of the first level to eliminate the ground objects in the area with an area larger than the preset area value or concentrated distribution and regular shape, and the extraction result is obtained:

[0091] After performing class filtering on the extraction result to obtain the landslide candidate area, inheriting it in the segmentation results of the second level, using the band mean, brightness value, mean or standard deviation of the gray-level co-occurrence matrix, aspect ratio, shape index, boundary index, and coordinate eigenvalue, the spectral, texture, geometric, and spatial multi-feature information of bare land, buildings, roads, and landslide ground objects is obtained, and the false alarm landslide areas are eliminated one by one to obtain the landslide spatial range.

[0092] As a preferred implementation manner, on the basis of the effective identification of the landslide spatial range in step S4, the strip profile method is used to finely analyze the terrain of the landslide area, and the specific steps for extracting the landslide morphological structure after determining the optimal threshold are as follows:

[0093] Step S4.1: After exporting the landslide spatial range extracted in step 3.6, the DEM of the landslide area is cropped.

[0094] Step S4.2: Draw the strip profile of the landslide area.

[0095] As a preferred implementation manner, the specific steps of the strip profile in step S4.2 are as follows:

[0096] Step S4.2.1: Using the DEM of the above-mentioned landslide area as the base map, a new surface layer is created, and a strip-shaped rectangle with a certain range (set according to the length and width of the landslide) is drawn with the main sliding axis of the landslide body as the basic orientation.

[0097] Step S4.2.2: Using the strip-shaped rectangle obtained in step S4.2.1 as a template, it is equally spaced divided to generate grid strips in SHP format and the centroid points corresponding to each grid.

[0098] Step S4.2.3: Rotate and slightly move the grid strips generated in step S4.2.2 to make them consistent with the landslide body movement direction.

[0099] Step S4.2.4: Using the equally spaced grid as the data set area and the DEM of the landslide area as the data layer, the DEM within the strip area is statistically analyzed by partition.

[0100] Step S4.2.5: Statistically analyze the maximum elevation value, minimum elevation value, average elevation value, and undulation degree (elevation difference) obtained from the zonal statistics in Step S4.2.4 along with their corresponding distances from the starting point, establish the mapping relationship between the centroid point of each grid and the relevant elevation values of the corresponding grid, and generate the strip profile elevation curve of the landslide area.

[0101] Step S4.3: Calculate the first derivative of the elevation curve, which serves to enhance the elevation change rate of equally spaced sample areas within the strip profile area:

[0102]

[0103] where (x i , y i ) are the coordinates of the centroid point of the i-th grid, and L' is the elevation curve after derivation.

[0104] Step S4.4: Divide the elevation curve after derivation in Step S4.3 into a topographic interval according to a undulation range. The topographic interval represents the topographic structure elements of the landslide, and the elevation demarcation value of the landslide topographic elements is the curve undulation turning value.

[0105] Step S4.5: Perform class filtering on the extraction results (roads, bare land, buildings, landslide spatial range, etc.) in Step S3.6 to obtain the landslide spatial range, inherit it in the third segmentation layer (the segmentation result of the third level) in Step S2.6, and use the elevation demarcation value obtained by the above method as the optimal threshold for landslide topographic structure element extraction to perform landslide morphology structure extraction.

[0106] Step S4.6: Further correct the local details in the extraction results of Step S4.5 according to actual situations using feature information such as slope, spectrum, and space.

[0107] As can be seen from the above, in one embodiment, based on the landslide spatial range and DEM, the strip profile method is used to perform a detailed analysis of the topography of the landslide study area, and after determining the optimal threshold, the extraction of the shallow small landslide morphology structure can include:

[0108] Export the extracted landslide spatial range and then clip the DEM of the landslide area;

[0109] Draw the strip profile of the landslide area;

[0110] Calculate the first derivative of the elevation curve of the strip profile in the landslide area to enhance the elevation change rate of equally spaced sample areas within the strip profile area:

[0111] Divide the elevation curve after derivation into a topographic interval according to a preset undulation range. The topographic interval represents the topographic structure elements of the landslide, and the elevation demarcation value of the landslide topographic elements is the curve undulation turning value;

[0112] Perform a classification filtering process on the landslide spatial range. Inherit it from the segmentation results of the third level. Use the elevation demarcation value as the optimal threshold for extracting landslide terrain structure elements to extract the landslide morphological structure.

[0113] As can be seen from the above, in one embodiment, drawing the strip profile of the landslide area may include:

[0114] Taking the DEM of the landslide study area as the base map, create a new polygon layer, and draw a strip-shaped rectangle with a preset range based on the main sliding axis of the landslide body;

[0115] Taking the strip-shaped rectangle as a template, perform equally spaced division to generate grid strips in SHP format and the centroid points corresponding to each grid;

[0116] Rotate and slightly move the grid strips to make them consistent with the landslide movement direction;

[0117] Taking the equally spaced grids as the dataset area and the DEM of the landslide area as the data layer, perform zonal statistics on the DEM within the strip area;

[0118] Statistically analyze the maximum elevation value, minimum elevation value, average elevation value, relief degree obtained from the zonal statistics and their corresponding distances from the starting point, establish the mapping relationship between the centroid points of each grid and the relevant elevation values of the corresponding grids, and generate the strip profile elevation curve of the landslide area.

[0119] The beneficial effects of the embodiments of the present invention are as follows: The present invention provides a scheme for extracting the morphological structure of shallow small landslides considering multi-feature information of UAV images. Based on UAV DOM and DEM, a top-down hierarchical model is obtained by performing multi-scale and multi-level segmentation on the DOM; taking the segmentation objects as basic units, a multi-feature hierarchical extraction rule set is constructed by integrating multi-feature information such as spectrum, texture, geometry, and space in the DOM, and landslide candidate areas are identified and landslide false alarm areas are eliminated layer by layer, and then the spatial range of the landslide is obtained; the strip profile method is used to finely analyze the terrain undulation of the landslide area, obtain the optimal threshold for dividing the terrain structure and extract the landslide morphological structure, and the DEM after the disaster can be used to accurately quantify the terrain undulation changes of shallow small landslides, and then effectively extract their morphological structures, providing important information for the drawing of landslide inventories, and providing certain data support for emergency rescue, risk assessment, reconstruction and recovery work in landslide affected areas, which is of great significance for remote sensing identification of landslide disasters.

[0120] To facilitate understanding of how the present invention is implemented, the following will be combined with the attached Figures 2a to 7The steps of a method for extracting the morphological structure of a shallow small landslide taking into account the multi-feature information of UAV images provided by the present invention are described in detail.

[0121] In step S1, a UAV is used to take aerial photos of the sudden shallow small landslide area, and the original UAV data is processed by aerial triangulation, dense matching, point cloud filtering, etc. to obtain the DOM and DEM of the survey area. Figure 2a and Figure 2b The DOM and DEM of a certain survey area drone provided by the present invention are respectively, and the specific steps are as follows:

[0122] Step S1.1: Perform aerial triangulation on the original UAV visible light RGB image, combine the UAV POS data with the actual measured ground control point coordinates for adjustment, calculate the real spatial position and posture of the image, obtain the ground coordinates of the key connection points, and generate a sparse point cloud.

[0123] Step S1.2: The images whose real spatial position and posture have been restored in step S1.1 are used to identify the points with the same name in multiple images by using the multi-view image dense matching technology, and a high-density point cloud of the survey area is established.

[0124] Step S1.3: Use the dense point cloud generated in step S1.2 to construct an irregular triangulated network to establish a digital surface model (DSM).

[0125] Step S1.4: The irregular triangular mesh generated in step S1.3 is processed by encryption filtering algorithm to separate ground points from non-ground points, remove non-ground information such as vegetation and buildings, retain only ground information, and construct elevation grids for ground points to generate DEM of the study area ( Figure 2a ).

[0126] Step S1.5: Based on the DEM model generated in step S1.4, a certain mathematical model is used to perform orthorectification on the single UAV image, and mosaicking and cropping are performed on the overlapping areas to finally obtain the complete DOM of the survey area ( Figure 2b ).

[0127] In step S2, a top-down hierarchical model is constructed through a multi-scale and multi-level segmentation method. The segmentation results are as follows: Figures 3a to 3c As shown, the specific steps are as follows:

[0128] Step S2.1: Load the drone DOM, set the R, G, and B layers as the layers involved in segmentation, and set the weight of each layer to 1.

[0129] Step S2.2: fix the shape factor and compactness factor, and set the parameter values to 0.5, and temporarily determine a more suitable scale parameter;

[0130] Step S2.3: Set the parameter values calculated in Step S2.2 as the scale parameters, adjust the shape factor and compactness factor respectively, and determine the final shape factor as 0.1 and the compactness factor as 0.5 according to the segmentation effect of the actual ground objects in the UAV images.

[0131] Step S2.4: Set the shape factor as 0.1 and the compactness factor as 0.5, adopt the ESP algorithm, adjust the starting scale as 10 and the number of loops as 150, and obtain the local variance change rate (ROC-LV) curve:

[0132]

[0133] Among them, L is the local variance of the target level, and L-1 is the local variance of the next level. Select the point where the peak value is located as the potential optimal segmentation scale (including: 82, 98, 147, 197, 215, 271), conduct experiments on each potential optimal scale parameter respectively, obtain the parameter suitable for the target ground object (vegetation) as 215, and determine it as the final scale factor parameter, thus completing the segmentation of the first level ( Figure 3a ).

[0134] Step S2.5: Inherit the segmentation result of Step S2.4. For the target ground object categories, which include non-vegetation ground objects such as bare land, roads, buildings, landslides, etc., repeat Steps S2.2 - 2.4 to further refine the segmented objects, determine the shape factor as 0.3, the compactness factor as 0.3, and the scale factor as 197, and obtain the segmentation result of the second level ( Figure 3b ).

[0135] Step S2.6: Inherit the segmentation result of Step S2.5. For the target ground object categories, which include landslide structure elements such as landslide walls, source areas, sliding areas, accumulation areas, etc., repeat Steps S2.2 - 2.4 to refine the segmented objects again, determine the shape factor as 0.3, the compactness factor as 0.6, and the scale factor as 147, and obtain the segmentation result of the third level ( Figure 3c ).

[0136] As Figure 4 shown, it is the sub - flow chart for hierarchical extraction of the landslide spatial range by the multi - feature rule set, that is, the detailed steps of the above Step S3, and the specific content is as follows:

[0137] Step S3.1: Analyze the multi - feature information such as spectrum, texture, geometry, and space of the landslides and non - landslides in the surveyed area, and establish the direct or indirect remote sensing interpretation marks of various ground objects.

[0138] Step S3.2: Calculate the eigenvalue of the object corresponding to each type of ground feature (including spectral mean, aspect ratio, mean of gray-level co-occurrence matrix, etc.), and obtain the multi-feature attributes of the object such as spectrum, texture, geometry, and space.

[0139] Step S3.3: Select sample vector points, establish sample point files for ground features such as vegetation, bare land, buildings, roads, and landslides in the format of SHP. Sample points for each type of ground feature are collected separately, and the number of samples is 100 for each type. When collecting samples, make the samples evenly and randomly distributed in the area.

[0140] Step S3.4: Establish a mapping relationship between the vector sample points generated in Step S3.3 and the segmentation objects, and use the objects corresponding to the vector sample points as sample objects.

[0141] Step S3.5: Extract potential landslide areas in the first segmentation layer obtained in Step S2.4. Landslides mostly occur in mountainous areas, and the ground features with a large coverage area or wide distribution in the area are generally vegetation. Therefore, the target ground feature in the first segmentation layer is selected as vegetation.

[0142] Use the visible light vegetation index VDVI to distinguish vegetation from non-vegetation in the area. The range of VDVI is between [-1, 1], and the threshold of the vegetation area is mostly above 0. Therefore, large-area vegetation areas are excluded, and then potential landslide areas are obtained:

[0143]

[0144] Among them, R, G, and B respectively represent the red, green, and blue bands in the UAV DOM.

[0145] Step S3.6: Inherit the extraction result of the vegetation in Step S3.5 into the second segmentation layer in Step S2.5. For the objects that have not been extracted, calculate their eigenvalue such as band mean, brightness value, mean or standard deviation of gray-level co-occurrence matrix, aspect ratio, shape index, boundary index, coordinates, etc., and obtain the multi-feature information of ground features such as bare land, buildings, roads, and landslides in terms of spectrum, texture, geometry, and space. Eliminate the false alarm landslide areas (non-landslide areas such as bare land, buildings, roads, etc.) one by one to obtain the landslide spatial range ( Figure 5 ).

[0146] As Figure 6 shown, it is the sub-flowchart for extracting the landslide morphology and structure by the strip profile method, that is, the detailed steps of the above Step S3. The specific content is as follows:

[0147] Step S4.1: Export the landslide spatial range extracted in Step 3.6 and then crop the DEM of the landslide area.

[0148] Step S4.2: Draw the strip profile of the landslide area.

[0149] As a preferred embodiment, the specific steps of the strip profile in step S4.2 are as follows:

[0150] Step S4.2.1: Using the DEM of the above-mentioned landslide area as the base map, create a new polygon feature layer in SHP format. Draw a strip-shaped rectangle with a certain range along the main sliding axis of the landslide body. In this example, the length of the strip is 160 m and the width is 10 m.

[0151] Step S4.2.2: Using the strip-shaped rectangle obtained in step S4.2.1 as a template, with an equal interval of 10×10 m, a total of 16 grids, the length and width being the same as the above strip rectangle, generate a grid strip in SHP format and the centroid point corresponding to each grid.

[0152] Step S4.2.3: Rotate and slightly shift the grid strip generated in step S4.2.2 to make it consistent with the moving direction of the landslide body.

[0153] Step S4.2.4: Using the equally spaced grids as the dataset area and the DEM of the landslide area as the data layer, conduct zonal statistics on the DEM within the strip area.

[0154] Step S4.2.5: Statistically analyze the maximum elevation value, minimum elevation value, average elevation value, and undulation (elevation difference) obtained from the zonal statistics in step S4.2.4 and their corresponding distances from the starting point, establish the mapping relationship between the centroid point of each grid and the relevant elevation value of the corresponding grid, and generate the strip profile elevation curve of the landslide area.

[0155] Step S4.3: Calculate the first derivative of the elevation curve to enhance the elevation change rate:

[0156]

[0157] where (x i , y i ) is the coordinate of the centroid point of the i-th grid, and L′ is the elevation curve after derivation.

[0158] Step S4.4: Divide the elevation curve after derivation in step S4.4 into a terrain interval according to a undulation range. The terrain interval represents the terrain structure elements of the landslide. Using the curve undulation turning value as the elevation demarcation value of the landslide terrain elements, the demarcation values in this example are 2261.16 m and 2227.86 m, and thus extract the landslide source area, sliding area, and accumulation area.

[0159] Step S4.5: Inherit the previous extraction results in the third segmentation layer of Step S2.6. The elevation demarcation values obtained by the above method are 2261.16 m and 2227.86 m respectively as the optimal thresholds for landslide topographic structure element extraction, and perform landslide morphological structure extraction. The extraction results are as Figure 7 shown.

[0160] Step S4.6: For the local details in the extraction results of Step S4.5, further correct them according to the actual situation using feature information such as slope, spectrum, and space.

[0161] The present invention objectively evaluates the extraction results using a confusion matrix, compares the test data with the extraction results, and obtains an overall accuracy of 94% and a Kappa coefficient of 0.92. The method for extracting the morphological structure of shallow small landslides considering multi-feature information of UAV images can comprehensively utilize multi-feature information such as the spectrum, texture, geometry, and space of UAV DOM in sudden shallow small landslide disasters, and use the post-disaster DEM to make strip profiles to finely analyze the terrain with small undulations to determine the optimal threshold, and then effectively extract the landslide morphological structure, which has certain application value. In addition, this method is also applicable to the extraction of landslide morphological structures in similar areas.

[0162] In an embodiment of the present invention, an extraction device for the morphological structure of shallow small landslides is also provided as described in the following embodiments. Since the principle of the device for solving problems is similar to the method for extracting the morphological structure of shallow small landslides, the implementation of the device can refer to the implementation of the method for extracting the morphological structure of shallow small landslides, and the repeated parts will not be described again.

[0163] Figure 9 is a schematic structural diagram of the extraction device for the morphological structure of shallow small landslides in an embodiment of the present invention, as Figure 9 shown, and the device includes:

[0164] A generation unit 01 for generating DOM and DEM according to the original data obtained by aerial survey of the study area of shallow small landslides;

[0165] A model construction unit 02 for selecting the optimal segmentation scale for different ground objects in DOM through a multi-scale segmentation method and constructing a multi-level model according to the optimal segmentation scale; the multi-level model is used to depict objects at three different description scales of the landslide background area, the landslide itself, and its morphological structure;

[0166] A landslide spatial range determination unit 03 for constructing a multi-feature rule set according to the interpretation features of various ground objects in DOM, and extracting landslide candidate areas layer by layer in the multi-level model according to the multi-feature rule set to obtain the landslide spatial range;

[0167] The morphological structure extraction unit 04 is used to finely analyze the terrain of the landslide study area based on the landslide spatial range and DEM by using the strip profile method, and extract the morphological structure of shallow small landslides after determining the optimal threshold.

[0168] In one embodiment, the original data captured by aerial survey is the visible light RGB image of the drone and the POS data of the drone; the generating unit is specifically used for:

[0169] Perform aerial triangulation on the visible light RGB image of the drone, combine the drone POS data with the coordinates of the ground control points measured actually for joint regional network adjustment, solve the true spatial position and attitude of the image, obtain the ground coordinates of the key connection points, and generate a sparse point cloud;

[0170] Adopt the multi-view image dense matching technology to identify the homologous points in multiple images from the images with true spatial position and attitude, and establish a dense point cloud of the survey area according to the homologous points in multiple images and the sparse point cloud;

[0171] Use the dense point cloud to construct an irregular triangular grid;

[0172] Perform encryption filtering algorithm processing on the irregular triangular grid, separate the ground points from the non-ground points to obtain ground point information, construct an elevation grid for the ground points, and generate the DEM of the study area;

[0173] According to the DEM, perform orthorectification on a single drone image, and perform mosaicking and cropping processing on the overlapping areas to finally obtain the complete DOM of the study area.

[0174] In one embodiment, the model construction unit is specifically used for:

[0175] Load the DOM; the layers participating in the segmentation in the DOM are the three layers of R, G, and B, and the weights of each layer are all 1;

[0176] Set both the shape factor and the compactness factor to 0.5, the scale parameter is in the range of [50, 300], and experiments are carried out with an increment of 25 as a unit, and the optimal scale parameter value is selected according to the effect;

[0177] Set the scale factor to the optimal scale parameter value, adjust the shape factor and the compactness factor respectively, and determine the final parameter values of the shape factor and the compactness factor according to the segmentation effect of the actual ground objects in the DOM;

[0178] When the final shape factor and compactness factor parameter values are determined, the ESP algorithm is used to adjust the initial scale and the number of cycles to obtain the local variance change rate curve; the point where the peak value of the curve is located is selected as the potential optimal segmentation scale, and the parameters of each potential optimal segmentation scale are experimentally tested to find the most suitable parameters for the target ground object as the final optimal segmentation scale, and the first-level segmentation is completed according to the final optimal segmentation scale to obtain the first-level segmentation result;

[0179] Inheriting the first-level segmentation result, for the target ground object, referring to the operation of the first-level segmentation, the second-level segmentation operation is performed to further refine the segmented object to obtain the second-level segmentation result;

[0180] Inheriting the second-level segmentation result, for the target ground object, referring to the operation of the first-level segmentation, the third-level segmentation operation is performed to refine the segmented object again to obtain the third-level segmentation result, and the multi-level model is obtained.

[0181] In one embodiment, the landslide spatial range determination unit is specifically configured to:

[0182] Analyze the characteristics of landslides and non-landslides to establish direct or indirect interpretation signs;

[0183] Based on the direct or indirect interpretation signs, for the segmented objects at different description scales, by calculating multiple eigenvalue of each object, the multi-feature attributes of the object are obtained;

[0184] Based on the multi-feature attributes of the object, a sample point file of different types of ground objects is established, and the acquisition of sample vector points is performed respectively;

[0185] Taking the positions of the acquired sample vector points as a reference, the corresponding segmented objects are selected as sample objects;

[0186] Based on the sample objects, the landslide candidate areas are extracted from the first-level segmentation result to remove the ground objects in the area with an area larger than the preset area value or concentrated distribution and regular shape, and the extraction result is obtained:

[0187] After performing class filtering processing on the extraction result to obtain the landslide candidate area, inheriting it in the second-level segmentation result, using the band mean, brightness value, mean or standard deviation of the gray-level co-occurrence matrix, aspect ratio, shape index, boundary index, and coordinate eigenvalue, the spectral, texture, geometric, and spatial multi-feature information of bare land, buildings, roads, and landslide ground objects are obtained, and the false alarm landslide areas are removed one by one to obtain the landslide spatial range.

[0188] In one embodiment, the morphological structure extraction unit is specifically configured to:

[0189] After exporting the extracted landslide spatial range, the DEM of the landslide area is cropped;

[0190] Draw the strip profile of the landslide area;

[0191] Calculate the first derivative of the elevation curve of the strip profile in the landslide area to enhance the elevation change rate of equally spaced sample areas within the strip profile area:

[0192] Divide the differentiated elevation curve into a terrain interval according to a preset undulation range. The terrain interval represents the terrain structure elements of the landslide, and the elevation demarcation value of the landslide terrain elements is the curve undulation turning value;

[0193] Perform a class filtering process on the landslide spatial range. In the segmentation result inherited from the third level, use the elevation demarcation value as the optimal threshold for extracting the landslide terrain structure elements to extract the landslide morphological structure.

[0194] In one embodiment, drawing the strip profile of the landslide area may include:

[0195] Using the DEM of the landslide study area as the base map, create a new polygon layer, and draw a strip-shaped rectangle with a preset range based on the main sliding axis of the landslide body;

[0196] Taking the strip-shaped rectangle as a template, perform equally spaced division to generate grid strips in SHP format and the centroid points corresponding to each grid;

[0197] Rotate and slightly shift the grid strips to make them consistent with the landslide movement direction;

[0198] Taking the equally spaced grids as the dataset area and the DEM of the landslide area as the data layer, perform zonal statistics on the DEM within the strip area;

[0199] Statistically analyze the maximum elevation value, minimum elevation value, average elevation value, undulation degree obtained from the zonal statistics and their distances from the starting point, establish the mapping relationship between the centroid points of each grid and the corresponding elevation values of the grid, and generate the elevation curve of the strip profile in the landslide area.

[0200] An embodiment of the present invention also provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the above method for extracting the morphological structure of shallow small landslides is implemented.

[0201] An embodiment of the present invention also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above method for extracting the morphological structure of shallow small landslides is implemented.

[0202] An embodiment of the present invention also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the above-mentioned method for extracting the morphological structure of shallow small landslides is implemented.

[0203] In the embodiment of the present invention, for the solution for extracting the morphological structure of shallow small landslides, specifically: the method for extracting the morphological structure of shallow small landslides provided by the embodiment of the present invention, during operation: based on the original data obtained by aerial survey of the research area of shallow small landslides, DOM and DEM are generated; through the multi-scale segmentation method, the optimal segmentation scale is selected according to the spectral and shape features of different ground objects in the DOM, and a multi-level model is constructed according to the optimal segmentation scale; the multi-level model is used to depict objects at three different description scales of the landslide background area, the landslide itself, and its morphological structure; a multi-feature rule set is constructed according to the interpretation features of various ground objects in the DOM, and the landslide candidate areas are extracted layer by layer in the multi-level model according to the multi-feature rule set to obtain the landslide spatial range; based on the landslide spatial range and the digital elevation model, the strip profile method is used to finely analyze the terrain of the landslide research area, and after determining the optimal threshold, the morphological structure of the shallow small landslide is extracted. This solution can construct a multi-level morphological structure extraction model for shallow small landslides in sudden shallow small landslide areas, accurately quantify the terrain undulation changes of shallow small landslides, and then effectively extract their morphological structures, which is of great significance for disaster assessment, emergency rescue, reconstruction planning, etc.

[0204] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0205] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one or more of the processes or multiple processes and / or blocks Figure 1 one or more of the blocks or multiple blocks.

[0206] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to work in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction device that implements the functions specified in one or more of the flows Figure 1 one or more flows and / or blocks Figure 1 specified in the block or blocks.

[0207] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one or more of the flows Figure 1 one or more flows and / or blocks Figure 1 specified in the block or blocks.

[0208] The specific embodiments described above further elaborate on the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A method for extracting the morphological structure of a shallow small landslide, characterized in that Including: Generate a Digital Orthophoto Map (DOM) and a Digital Elevation Model (DEM) based on the original data obtained from aerial surveys of the shallow small landslide study area. Through the multi-scale segmentation method, select the optimal segmentation scale according to the spectral and shape characteristics of different ground objects in the DOM. Construct a multi-level model based on the optimal segmentation scale, which includes: loading the DOM; the layers participating in the segmentation in the DOM are the three layers of R, G, and B, and the weight of each layer is 1; set both the shape factor and the compactness factor to 0.5, and the scale parameter ranges from [50, 300] and increases in units of 25 for experiments, and select the optimal scale parameter value according to the effect; set the scale factor to the optimal scale parameter value, adjust the shape factor and the compactness factor respectively, and determine the final shape factor and compactness factor parameter values according to the segmentation effect of the actual ground objects in the DOM; in the case of determining the final shape factor and compactness factor parameter values, use the ESP algorithm to adjust the initial scale and the number of cycles to obtain the local variance change rate curve; select the point where the peak value of the curve is located as the potential optimal segmentation scale, conduct experiments on the parameters of each potential optimal segmentation scale respectively, find the parameters most suitable for the target ground object as the final optimal segmentation scale, and complete the first-level segmentation according to the final optimal segmentation scale to obtain the first-level segmentation result; inherit the first-level segmentation result, and perform the second-level segmentation operation on the target ground object with reference to the operation of the first-level segmentation to further refine the segmented object and obtain the second-level segmentation result; inherit the second-level segmentation result, and perform the third-level segmentation operation on the target ground object with reference to the operation of the first-level segmentation to refine the segmented object again and obtain the third-level segmentation result, thus obtaining the multi-level model; the multi-level model is used to depict objects at three different description scales of the landslide background area, the landslide itself, and its morphological structure. Construct a multi-feature rule set according to the interpretation characteristics of various ground objects in the DOM, and extract the landslide candidate areas layer by layer in the multi-level model to obtain the landslide spatial range. Based on the landslide spatial range and the DEM, use the strip section method to conduct a detailed analysis of the terrain of the landslide study area, and extract the morphological structure of the shallow small landslide after determining the optimal threshold.

2. The method for extracting the morphological structure of a shallow small landslide according to claim 1, wherein, The original data obtained from the aerial surveys is the visible light RGB image of the unmanned aerial vehicle and the POS data of the unmanned aerial vehicle; generating the DOM and the DEM based on the original data obtained from the aerial surveys of the shallow small landslide study area includes: Perform aerial triangulation on the visible light RGB image of the unmanned aerial vehicle, and conduct joint regional network adjustment by combining the POS data of the unmanned aerial vehicle with the ground control point coordinates measured actually to solve the true spatial position and attitude of the image, obtain the ground coordinates of the key connection points, and generate a sparse point cloud. Adopt the multi-view image dense matching technology to identify the homologous points in multiple images from the images with the true spatial position and attitude, and establish a dense point cloud of the survey area according to the homologous points in multiple images and the sparse point cloud. Use the dense point cloud to construct an irregular triangular grid. Perform an encryption filtering algorithm on the irregular triangular grid, separate the ground points from the non-ground points to obtain ground point information, construct an elevation grid for the ground points, and generate a DEM of the study area; According to the DEM, correct a single UAV image for DOM, and perform mosaicking and cropping on the overlapping areas to finally obtain a complete DOM of the study area.

3. The method for extracting the morphological structure of a shallow small landslide according to claim 1, characterized in that Construct a multi-feature rule set according to the interpretation features of various ground objects in the DOM, and extract landslide candidate areas layer by layer in the multi-level model to obtain the landslide spatial range, including: Conduct feature analysis on landslides and non-landslides to establish direct or indirect interpretation signs; Based on the direct or indirect interpretation signs, calculate multiple feature values of each object for the segmentation objects at different description scales to obtain the multi-feature attributes of the objects; Based on the multi-feature attributes of the objects, establish sample point files for different types of ground objects and collect sample vector points respectively; Select the corresponding segmentation objects as sample objects based on the positions of the collected sample vector points; Based on the sample objects, extract landslide candidate areas in the segmentation results of the first level to eliminate the ground objects in the area with an area larger than the preset area value or concentrated distribution and regular shape, and obtain the extraction result: Perform class filtering on the extraction result to obtain landslide candidate areas, inherit them in the segmentation results of the second level, and use the band mean, brightness value, mean or standard deviation of the gray-level co-occurrence matrix, aspect ratio, shape index, boundary index, and coordinate feature values to obtain the spectral, texture, geometric, and spatial multi-feature information of bare land, buildings, roads, and landslide ground objects, and eliminate the false alarm landslide areas one by one to obtain the landslide spatial range.

4. The method for extracting the morphological structure of a shallow small landslide according to claim 1, characterized in that, Based on the landslide spatial range and DEM, use the strip profile method to conduct a detailed analysis of the terrain of the landslide study area, determine the optimal threshold, and then extract the morphological structure of shallow small landslides, including: Export the extracted landslide spatial range and then crop the DEM of the landslide area; Draw the strip profile of the landslide area; Calculate the first derivative of the elevation curve of the strip profile of the landslide area to enhance the elevation change rate of the equally spaced sample areas within the strip profile area: Divide the differentiated elevation curve into a terrain interval according to the preset undulation range. The terrain interval represents the terrain structure elements of the landslide, and the curve undulation turning value is used as the elevation boundary value of the landslide terrain elements; Perform class filtering on the landslide spatial range, inherit it in the segmentation results of the third level, and use the elevation boundary value as the best threshold for extracting the landslide terrain structure elements to perform landslide morphological structure extraction.

5. The method for extracting the morphological structure of a shallow small landslide according to claim 4, characterized in that, Draw the strip profile of the landslide area, including: Using the DEM of the landslide study area as the base map, create a new polygon layer, and draw a strip-shaped rectangle with a preset range based on the main sliding axis of the landslide body; Using the strip-shaped rectangle as a template, perform equally spaced division to generate grid strips in SHP format and the centroid points corresponding to each grid; Rotate and slightly move the grid strips to make them consistent with the landslide body movement direction; Using the equally spaced grids as the dataset area and the DEM of the landslide area as the data layer, conduct zonal statistics on the DEM within the strip area; Statistically analyze the maximum elevation value, minimum elevation value, average elevation value, and relief degree obtained from the partition statistics along with their corresponding distances from the starting point, establish the mapping relationship between the centroid point of each grid cell and the relevant elevation values of the corresponding grid cell, and generate the strip profile elevation curve of the landslide area.

6. An extraction device for the morphological structure of a shallow small landslide, characterized in that, Including: A generation unit for generating DOM and DEM based on the original data obtained from the aerial survey of the shallow small landslide study area. A model construction unit for selecting the optimal segmentation scale for different ground objects in the DOM according to their spectral and shape characteristics through a multi-scale segmentation method, and constructing a multi-level model based on the optimal segmentation scale. It includes: loading the DOM; the layers participating in the segmentation in the DOM are the three layers of R, G, and B, and the weight of each layer is 1; setting both the shape factor and the compactness factor to 0.5, with the scale parameter ranging from [50, 300] and incrementing in units of 25 for experiments, and selecting the optimal scale parameter value according to the effect; setting the scale factor to the optimal scale parameter value, adjusting the shape factor and the compactness factor respectively, and determining the final parameter values of the shape factor and the compactness factor according to the segmentation effect of the actual ground objects in the DOM; in the case of determining the final parameter values of the shape factor and the compactness factor, using the ESP algorithm to adjust the initial scale and the number of cycles to obtain the local variance change rate curve; selecting the point where the peak of the curve is located as the potential optimal segmentation scale, conducting experiments on the parameters of each potential optimal segmentation scale respectively, finding the most suitable parameters for the target ground object as the final optimal segmentation scale, and completing the first-level segmentation according to the final optimal segmentation scale to obtain the first-level segmentation result; inheriting the first-level segmentation result, performing the second-level segmentation operation on the target ground object with reference to the operation of the first-level segmentation to further refine the segmented object and obtain the second-level segmentation result; inheriting the second-level segmentation result, performing the third-level segmentation operation on the target ground object with reference to the operation of the first-level segmentation to refine the segmented object again and obtain the third-level segmentation result, thus obtaining the multi-level model; the multi-level model is used to depict objects at three different description scales of the landslide background area, the landslide itself, and its morphological structure. A landslide spatial range determination unit for constructing a multi-feature rule set according to the interpretation characteristics of various ground objects in the DOM, and extracting the landslide candidate area layer by layer in the multi-level model to obtain the landslide spatial range. A morphological structure extraction unit for finely analyzing the terrain of the landslide study area using the strip profile method based on the landslide spatial range and the DEM, and extracting the morphological structure of the shallow small landslide after determining the optimal threshold.

7. A computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, it implements the method according to any one of claims 1 to 5.

9. A computer program product, characterized in that, The computer program product includes a computer program, and when the computer program is executed by the processor, it implements the method according to any one of claims 1 to 5.

Images