A method for optimizing segmentation of geological survey remote sensing images

Abstract

The present invention relates to the field of image optimization technology, and in particular to a method for optimizing the segmentation of remote sensing images for geological exploration. The method selects dark pixel blocks in a superpixel segmentation image of a remote sensing image according to the size of pixel values, analyzes the connectivity between the dark pixel blocks and adjacent superpixel blocks, and determines seed superpixel blocks. Regional growth is performed based on the seed superpixel blocks to obtain slope pixel blocks. Adaptive difference is performed based on the corresponding positions of the slope pixel blocks in a three-dimensional grid model to obtain an optimized three-dimensional grid model. The pit slope is segmented in the optimized three-dimensional grid model. The present invention effectively fuses high-resolution remote sensing images with three-dimensional grid models to achieve effective segmentation of pit areas in abandoned mining areas.

Description

Technical Field

[0001] The invention relates to the technical field of image optimization, and in particular to a method for optimizing segmentation of geological survey remote sensing images. Background Art

[0002] Radar remote sensing images can use elevation data to restore and analyze geological areas. For abandoned mining areas, accurate geological surveys and assessment of land contiguousness are important projects in ecological restoration projects. In abandoned mining areas, the factors that affect land contiguousness are mainly large areas of pits and holes, and the locations of pits and holes can be screened out through the elevation information in remote sensing images. However, for actual abandoned mining areas, the elevation information of pits and holes is irregular. Because of the uneven surface, the depths of pits and holes vary, resulting in the consistency of the elevation information of shallower areas and pit and hole slopes, which makes it impossible to accurately segment the pits and holes in abandoned mining areas directly from the elevation information. In the prior art, data fusion can be performed based on high-resolution remote sensing images and radar remote sensing effects, and local interpolation can be performed on three-dimensional models with elevation information to improve the feature prominence of pit and hole information. However, high-resolution remote sensing image information is easily affected by light angles, geological scales, etc., which makes it difficult to determine that some areas are pit and hole areas that need to be interpolated due to perspective, texture changes, etc., which leads to the inability of the three-dimensional model after interpolation to effectively segment the pit and hole areas. Summary of the invention

[0003] In order to solve the technical problem in the prior art that when using high-resolution remote sensing images to perform local interpolation on three-dimensional models, the interpolation position is not accurately determined, resulting in the inability of the three-dimensional model after interpolation to effectively segment the pothole area, the purpose of the present invention is to provide a method for optimizing segmentation of geological survey remote sensing images, and the technical scheme adopted is as follows:

[0004] The present invention proposes a method for optimizing segmentation of geological survey remote sensing images, the method comprising:

[0005] Obtain remote sensing images and 3D grid models of abandoned mining areas;

[0006] Obtain a superpixel segmentation image of the remote sensing image; filter out dark pixel blocks according to pixel values ​​of superpixel blocks in the superpixel segmentation image; obtain a suspected slope direction according to the direction between the dark pixel block and an adjacent superpixel block; obtain a connectivity probability according to the pixel value distribution in the suspected slope direction and the shape matching degree between the dark pixel block and the adjacent superpixel block;

[0007] Among all the adjacent super pixel blocks of the dark pixel block, a seed super pixel block is selected according to the connectivity probability; the seed super pixel block is subjected to regional growth according to the preset pothole characteristic condition to obtain a slope pixel block;

[0008] In the three-dimensional grid model, according to the elevation gradient information in the grid area corresponding to the slope pixel block, an interpolation position is selected in the grid area and the elevation value is interpolated to obtain an optimized three-dimensional grid model;

[0009] The pothole slope surface is segmented in the optimized three-dimensional grid model.

[0010] Furthermore, the dark pixel block screening method includes:

[0011] The grayscale mean of the superpixel block is used as the overall grayscale feature, and Otsu threshold segmentation is performed based on the overall grayscale feature to obtain two types of superpixel blocks, and the superpixel block with the smallest overall grayscale feature is used as a suspected dark pixel block;

[0012] If the overall grayscale feature of the suspected dark pixel block is smaller than the overall grayscale features of all adjacent super pixel blocks, the corresponding suspected dark pixel block will be used as the dark pixel block.

[0013] Furthermore, the method for obtaining the connectivity probability includes:

[0014] Taking any adjacent superpixel block of the dark pixel block as the target adjacent superpixel block, obtaining a centroid line between the dark pixel block and the target adjacent superpixel block, and taking a direction perpendicular to the centroid line as the suspected slope direction;

[0015] Obtaining a first connectivity probability according to a difference in shape features between the dark pixel block and a target adjacent super pixel block, and a grayscale distribution in a direction of the suspected slope;

[0016] In the direction of the centroid line, analyzing the degree of shape perspective of the target adjacent superpixel block, and obtaining a second connectivity probability according to the degree of shape perspective and the degree of change of grayscale values ​​on the target adjacent superpixel block;

[0017] The connectivity probability between the dark pixel block and the target adjacent super pixel block is obtained according to the first connectivity probability and the second connectivity probability.

[0018] Furthermore, the method for obtaining the first connectivity probability includes:

[0019] Taking the dark pixel block as the center, the two closest superpixel blocks along the two sides of the suspected slope direction are respectively used as the top pixel block and the bottom pixel block of the dark pixel block; obtaining the top pixel block and the bottom pixel block of the target adjacent superpixel block, and obtaining the grayscale distribution consistency between the dark pixel block and the target adjacent superpixel block according to the grayscale difference of the top pixel block and the grayscale difference of the bottom pixel block between the dark pixel block and the target adjacent superpixel block;

[0020] For each superpixel block, the narrow length of the superpixel block is used as the shape feature; the shape matching degree is obtained according to the shape feature difference between the dark pixel block and the target adjacent superpixel block;

[0021] A first connectivity probability between the dark pixel block and the target adjacent super pixel block is obtained according to the grayscale distribution consistency and the shape matching degree.

[0022] Furthermore, the method for acquiring the second connectivity probability includes:

[0023] Obtaining a minimum circumscribed window of the target adjacent superpixel block, wherein the minimum circumscribed window is an isosceles trapezoid, and the bottom side is parallel to the centroid line; obtaining the perspective degree of the shape according to the area ratio of the target adjacent superpixel block in the minimum circumscribed window;

[0024] Obtaining a texture feature value of each pixel in the target adjacent superpixel block, and obtaining the grayscale value variation degree according to the variance of the texture feature value;

[0025] A second connectivity probability between the dark pixel block and the target adjacent super pixel block is obtained according to the shape perspective degree and the gray value change degree.

[0026] Furthermore, the preset pothole characteristic conditions include:

[0027] During the region growing process, if the grayscale interval overlap between the superpixel block to be grown and the grown region is greater than the preset overlap threshold, the grayscale variance difference is less than the preset variance condition threshold, and the gradient information difference is less than the preset gradient information threshold, then the superpixel block to be grown is deemed to meet the pit feature conditions.

[0028] Furthermore, the method for screening the interpolation position includes:

[0029] The average elevation gradient amplitude of the boundary grid of the grid area is obtained, and the sampling coefficient is obtained after negative correlation mapping and normalization of the average elevation gradient amplitude; the number of pixels in the slope pixel block is multiplied by the sampling coefficient to obtain the sampling number; pixel points selected in the slope pixel block according to the sampling number in descending order of edge gradient amplitude are used as interpolation pixel points, and the position of the interpolation pixel point corresponding to the three-dimensional grid model is the interpolation position.

[0030] Furthermore, the method for interpolating the elevation value of the interpolation position includes:

[0031] For each interpolation position, the pixel to which the interpolation position belongs in the three-dimensional grid model is taken as the target pixel;

[0032] For a vertex of the target pixel, negatively correlate the distance between the vertex and the interpolation position and normalize it to obtain a distance weight; obtain a pixel value ratio between the interpolation position and the vertex; multiply the distance weight, the pixel value ratio and the elevation value of the vertex to obtain initial interpolation data of the vertex;

[0033] The average initial interpolation data of all vertices of the target pixel is used as the elevation value of the interpolation position for interpolation.

[0034] Further, the step of selecting a seed superpixel block according to the connectivity probability includes:

[0035] Among all the adjacent superpixel blocks of the dark pixel block, the adjacent superpixel block with the largest connectivity probability is selected as the seed superpixel block.

[0036] Furthermore, the preset pit characteristic condition also includes setting a growth area threshold. During the regional growth process, if the grown area is greater than or equal to the growth area threshold, the growth is stopped.

[0037] The present invention has the following beneficial effects:

[0038] The present invention first analyzes image information in remote sensing images. Considering that potholes will produce obvious shadow areas due to depression, the shadow areas produced by potholes are easier to identify than slope areas. Therefore, the embodiment of the present invention selects dark pixel blocks according to pixel values, that is, dark pixel blocks can be considered as shadow areas produced by potholes. Based on the dark pixel blocks, further analysis of the adjacent super pixel blocks can determine the slope pixel blocks with obvious features. Considering that the slope pixel blocks and the dark pixel blocks corresponding to the shadow areas are both the results of potholes, their shapes have obvious matching degrees, and the pixel value distribution in the extension direction of the slope will show consistent characteristics. Therefore, the seed super pixel blocks can be effectively selected through the connectivity probability, that is, the seed super pixel blocks are slope pixel blocks that can effectively characterize the slope features, and then the slope pixel blocks can be obtained by regional growth based on the seed super pixel blocks. In the three-dimensional grid model, the interpolation position can be selected and the elevation difference can be performed based on the elevation gradient information in the grid area corresponding to the slope pixel blocks, and the optimized three-dimensional grid model with obvious pothole features can be obtained, thereby realizing the effective segmentation of the pothole area. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0040] Figure 1 A flow chart of a method for optimizing segmentation of geological survey remote sensing images provided by one embodiment of the present invention;

[0041] Figure 2 A schematic diagram of a remote sensing image provided by an embodiment of the present invention;

[0042] Figure 3 A synthetic aperture radar image provided by an embodiment of the present invention;

[0043] Figure 4 A local magnified image of a remote sensing image provided by an embodiment of the present invention;

[0044] Figure 5 A superpixel segmented image provided by an embodiment of the present invention;

[0045] Figure 6 A post-processed image of a superpixel segmented image provided by an embodiment of the present invention;

[0046] Figure 7 A schematic diagram of dark pixel block distribution provided by an embodiment of the present invention;

[0047] Figure 8 A schematic diagram of a perspective result in a remote sensing image provided by an embodiment of the present invention;

[0048] Fig. 9 A schematic diagram of a minimum external window provided by an embodiment of the present invention;

[0049] Fig.10 A schematic diagram of a three-dimensional grid model conversion process provided by an embodiment of the present invention;

[0050] Fig.11 A schematic diagram of a three-dimensional rasterized model provided by an embodiment of the present invention;

[0051] Fig.12 A schematic diagram of pothole area segmentation provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0052] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the method for optimizing segmentation of remote sensing images for geological exploration proposed by the present invention, its specific implementation method, structure, features and effects, in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0053] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0054] The scenario targeted by the embodiment of the present invention is geological survey of abandoned mining areas. High-resolution geological remote sensing images can be collected by remote sensing satellites, which are referred to as remote sensing images in the following description; and elevation maps containing elevation information can be collected by synthetic aperture radar. Figure 2 , which shows a remote sensing image schematic diagram provided by an embodiment of the present invention. Figure 3 , which shows a synthetic aperture radar image provided by an embodiment of the present invention, namely, Figure 3 is an elevation map containing elevation information. Figure 3 The gray value in represents the elevation value of the corresponding position. Figure 2 and Figure 3 As shown in the figure, remote sensing images and elevation images have different information expression capabilities. Remote sensing images are mainly based on surface visual information and represent the current geological structure through features such as color and shadow in the image; while the grayscale value of each position in the elevation image represents the elevation value, and the current geological results are represented by the size of the elevation value.

[0055] In order to analyze the contiguous situation of abandoned mining areas, it is necessary to convert the elevation map into a three-dimensional grid model. The conversion method of the three-dimensional grid model is a technical means well known to those skilled in the art, and will not be described in detail. Fig.10 , which shows a schematic diagram of a three-dimensional grid model conversion process provided by an embodiment of the present invention, converting the elevation map into an elevation model, and then rasterizing the elevation model to obtain the following Fig.11 A three-dimensional rasterized model is shown.

[0056] As mentioned in the background technology, if the three-dimensional rasterized model is directly used to identify the pothole area, the pothole identification error will occur due to the irregular elevation information of the potholes and the different depths of the potholes, which will affect the contiguous analysis results. Because remote sensing images are visual image information, in order to facilitate the understanding of the different depths of potholes, please refer to Figure 4 , which shows a local magnified image of a remote sensing image provided by an embodiment of the present invention. Figure 4 The two potholes circled in the figure are obviously different in depth. One pothole is deep and has a large slope, while the other is shallow and has a low slope. The edges of the potholes are easily affected by the terrain, and the shallower potholes are almost connected to the outside area. However, the elevation information represented by the two potholes in the elevation map is obviously different, which easily makes it difficult to identify the shallower potholes. Therefore, it is necessary to extract the pothole slope area in the remote sensing image and then interpolate the elevation data in the three-dimensional rasterized model.

[0057] See also Figure 1 , which shows a flow chart of a method for optimizing segmentation of geological survey remote sensing images provided by an embodiment of the present invention, the method comprising:

[0058] Step S1: Obtain remote sensing images and three-dimensional grid models of abandoned mining areas.

[0059] In an embodiment of the present invention, after obtaining the remote sensing image of the abandoned mining area, it needs to be processed by grayscale, that is, the pixel values ​​proposed in the subsequent steps are all grayscale values. The method of obtaining the three-dimensional grid model has been explained in the above content. In the three-dimensional grid model, the spacing between the row and column grids is a constant, and each grid can be regarded as a pixel, and each pixel records the elevation data of the corresponding position. After obtaining the three-dimensional grid model, it is also necessary to unify the projection coordinates of the remote sensing image and the three-dimensional grid model, and use the geo-registration algorithm to spatially align the remote sensing image and the three-dimensional grid model. After alignment, each position in the remote sensing image has a corresponding position in the three-dimensional grid model.

[0060] Step S2: Obtain a superpixel segmentation image of the remote sensing image; filter out dark pixel blocks according to the pixel values ​​of the superpixel blocks in the superpixel segmentation image; obtain the suspected slope direction according to the direction between the dark pixel blocks and the adjacent superpixel blocks; obtain the connectivity probability according to the pixel value distribution in the suspected slope direction and the shape matching degree between the dark pixel blocks and the adjacent superpixel blocks.

[0061] In order to identify the inconspicuous slope area in the complex terrain of the abandoned mining area, the embodiment of the present invention uses superpixel segmentation technology to perform superpixel segmentation on the remote sensing image to obtain a superpixel segmented image. Figure 5 , which shows a superpixel segmentation image provided by an embodiment of the present invention. The principle of superpixel segmentation is to combine similar pixels into larger area blocks, namely superpixel blocks. Through the segmentation of superpixel blocks, more hidden slope information can be efficiently and accurately identified.

[0062] It should be noted that superpixel segmentation is a technical means well known to those skilled in the art. In the superpixel segmentation process, the noise and unnecessary details in the image can be reduced by setting an appropriate number of superpixel blocks, introducing spatial regularization terms, and other operations. The specific means are well known to those skilled in the art and will not be described in detail here.

[0063] In some implementations of an embodiment of the present invention, in order to facilitate the screening of dark pixel blocks and the analysis of connectivity probability in subsequent processes, a post-processing process is further performed after obtaining the superpixel segmentation image, and the gray value of each pixel is set to the average gray value of the superpixel block to which it belongs, such as Figure 6As shown, it shows a post-processed image of a super-pixel segmented image provided by an embodiment of the present invention. Through post-processing, the image information can be simplified, because the subsequent screening of dark pixel blocks and analysis of connectivity probability are all analyzed for the entire super-pixel block, and do not involve the pixel value of each pixel point. Therefore, the analysis efficiency can be improved through post-processing. It should be noted that in the subsequent process of step S2, the analysis can be directly performed in the post-processed image of the super-pixel segmented image, and the post-processing operation may not be performed in other implementation methods of the embodiments of the present invention.

[0064] In remote sensing images, when the lighting angle of the annular slope of a pothole changes, an obvious shadow area will be generated, that is, a dark pixel block is connected to a bright pixel block in the superpixel segmentation image, and the dark pixel block and the connected bright pixel block belong to the pothole information, and there is an obvious connectivity feature between these pixel blocks. Therefore, the embodiment of the present invention firstly filters out the dark pixel block based on the pixel value of the superpixel block in the superpixel block segmentation image, that is, the lower the pixel value and the dark pixel block is connected to the bright pixel block.

[0065] Further analyze the distribution relationship and shape relationship between the dark pixel block and the adjacent super pixel block. If the adjacent super pixel block and the dark pixel block are the pixel blocks of the same pit, the dark pixel block is a shadow area, and the direction between it and the adjacent super pixel block has a great correlation with the direction of the slope of the pit. Therefore, the suspected slope direction can be obtained first according to the direction between the dark pixel block and the adjacent super pixel block. It should be noted that, because it is unclear whether a certain adjacent super pixel block is a real pixel block belonging to the same pit as the dark pixel block in the image analysis stage, it is necessary to traverse each adjacent super pixel block for analysis, so the result obtained is a suspected slope direction, and the corresponding adjacent super pixel block needs to be further analyzed based on the suspected slope direction.

[0066] Because the dark pixel block is a pixel block generated when the illumination angle of the annular slope of the pit changes, that is, the dark pixel block is a slope, and the image information of the top of the slope and the slope can be obtained based on the extension direction of the slope. Because the dark pixel block has a pixel value distribution feature in the direction of the suspected slope, the analyzed adjacent super pixel block also has a pixel value distribution feature. If the adjacent super pixel block and the dark pixel block are image information of the same slope, that is, the connection probability between the adjacent super pixel block and the dark pixel block is large, then the corresponding pixel value distribution feature should have obvious consistency. Therefore, the embodiment of the present invention analyzes the pixel value distribution in the direction of the suspected slope to determine the connection probability.

[0067] Further considering that if the dark pixel block and the adjacent super pixel block have obvious connectivity features, that is, they belong to the same slope, then the shapes of the two have obvious matching degree, so the embodiment of the present invention further analyzes the shape matching degree between the dark pixel block and the adjacent super pixel block to obtain the connectivity probability. The greater the connectivity probability, the more the corresponding adjacent super pixel block belongs to the same slope area connected to the dark pixel block.

[0068] Step S3: among all the adjacent super pixel blocks of the dark pixel block, a seed super pixel block is selected according to the connectivity probability; and the seed super pixel block is subjected to regional growth according to the preset pothole characteristic conditions to obtain a slope pixel block.

[0069] The greater the connectivity probability, the more likely the adjacent super-pixel block belongs to the same slope pixel block of the dark pixel block. Therefore, in the embodiment of the present invention, the seed super-pixel block is screened out according to the connectivity probability among all the adjacent super-pixel blocks of the dark pixel block. Because the seed super-pixel block is not a special pixel block such as the dark pixel block, it can be used as a reference data to obtain other slope information. Therefore, in the embodiment of the present invention, the seed super-pixel block is subjected to regional growth according to the preset pit characteristic conditions to obtain the slope pixel block. It should be noted that the slope pixel block is the result of regional growth, so it is a combination of super-pixel blocks, while the dark pixel block, the adjacent super-pixel block and the seed super-pixel block are all super-pixel blocks.

[0070] Step S4: In the three-dimensional grid model, according to the elevation gradient information in the grid area corresponding to the slope pixel block, the interpolation position is selected in the grid area and the elevation value is interpolated to obtain the optimized three-dimensional grid model.

[0071] Because there are differences in slope features at different scales, different geological structures, and different angles, the embodiment of the present invention uses dark pixel blocks to locate seed superpixel blocks. Each seed superpixel block can specifically extract the slope features of the pit where it is located, and multiple slope pixel blocks can be obtained. Different slope pixel blocks may represent different pit slope features. The slope pixel blocks in the remote sensing image correspond to a grid area composed of multiple pixels in the three-dimensional grid model. Therefore, the elevation data in the three-dimensional grid model can be adaptively interpolated based on the slope pixel blocks, thereby improving the flexibility and accuracy of pit slope recognition in complex terrain.

[0072] In the interpolation process, the interpolation position needs to be screened out first. In the grid area, the weaker the elevation gradient information, the less obvious the pothole characteristics reflected by the elevation data in the corresponding position area, and the more interpolation should be performed. Therefore, based on the elevation gradient information, the interpolation position can be screened out in the grid area and the elevation value difference can be performed to obtain an optimized three-dimensional grid model.

[0073] It should be noted that, since the embodiment of the present invention aims to determine the slope pixel blocks and then interpolate the three-dimensional grid model based on the rough positioning results of the slope pixel blocks, the interpolation process can be implemented by existing technologies and will not be described in detail here.

[0074] It should be noted that, although there is almost no blind spot in the remote sensing perspective, the dark pixel blocks obtained in step S2 are not necessarily all slope areas. The slope pixel blocks obtained only from the remote sensing image are a rough segmentation result, not a refined segmentation result. The dark pixel blocks can cover most of the pothole slope areas. Analyzing them can divide the slope areas hidden from the elevation data based on their high coverage. Therefore, compared with directly using the slope pixel blocks as the segmentation results, the three-dimensional grid model is adaptively interpolated based on the obtained slope pixel blocks, and then the pothole slope segmentation result is more accurate based on the optimized three-dimensional grid model.

[0075] Step S5: Segment the pothole slope surface in the optimized three-dimensional grid model.

[0076] Terrain features such as slope, curvature and other features can be extracted from the optimized 3D grid model. For example, artificially excavated slopes usually have a more regular shape and structure, with exposed rocks, low weathering and oxidation, and obvious signs of artificial excavation. Naturally formed slopes usually have a more natural shape and structure, may have irregular boundaries and undulations, and are different from the surrounding surface. Therefore, the pit slope features can be used to identify significant or hidden pit slopes. The recognition model in the prior art can be used to accurately segment the pit slope, which will not be elaborated or limited here. Please refer to Fig.12 , which shows a schematic diagram of a pothole area segmentation provided by an embodiment of the present invention. It should be noted that Fig.12 In order to optimize the two-dimensional schematic diagram of the pothole recognition results in the three-dimensional grid model, a false color map is used for plane presentation. The false color map can highlight specific information. Fig.12 The dark area is the pit area. After obtaining the pit area, the remaining land area can be divided according to the natural terrain to obtain the contiguous situation of the abandoned mining area and complete the geological survey.

[0077] In summary, the embodiment of the present invention first analyzes image information in the remote sensing image. Considering that the potholes will produce obvious shadow areas due to the depression, the shadow areas produced by the potholes are easier to identify than the slope areas. Therefore, the embodiment of the present invention selects dark pixel blocks according to pixel values, that is, dark pixel blocks can be considered as shadow areas produced by potholes. Based on the dark pixel blocks, further analysis of the adjacent super pixel blocks can determine the slope pixel blocks with obvious features. Considering that the slope pixel blocks and the dark pixel blocks corresponding to the shadow areas are both the results of the potholes, their shapes have obvious matching degrees, and the pixel value distribution in the extension direction of the slope will show consistent characteristics. Therefore, the seed super pixel blocks can be effectively selected through the connectivity probability, that is, the seed super pixel blocks are slope pixel blocks that can effectively characterize the slope features, and then the slope pixel blocks can be obtained by regional growth based on the seed super pixel blocks. In the three-dimensional grid model, the interpolation position can be selected and the elevation difference can be performed based on the elevation gradient information in the grid area corresponding to the slope pixel blocks, and the optimized three-dimensional grid model with obvious pothole features can be obtained, thereby realizing the effective segmentation of the pothole area.

[0078] Preferably, in some implementations of the embodiments of the present invention, the method for screening dark pixel blocks includes:

[0079] The grayscale mean of the superpixel block is used as the overall grayscale feature, and Otsu threshold segmentation is performed based on the overall grayscale feature to obtain two types of superpixel blocks. The superpixel block with the smallest overall grayscale feature is used as the suspected dark pixel block. If the overall grayscale feature of the suspected dark pixel block is smaller than the overall grayscale features of all adjacent superpixel blocks, the corresponding suspected dark pixel block is used as the dark pixel block. That is, the dark pixel block satisfies the requirement of having a small average grayscale value and being surrounded by superpixels with relatively large grayscale values, that is, it presents the characteristics of light and dark surrounding distribution.

[0080] Preferably, in some implementations of the embodiments of the present invention, the method for acquiring the connectivity probability includes:

[0081] Take any adjacent superpixel block of the dark pixel block as the target adjacent superpixel block, obtain the centroid line between the dark pixel block and the target adjacent superpixel block, and take the direction perpendicular to the centroid line as the suspected slope direction. Figure 7 , which shows a schematic diagram of dark pixel block distribution provided by an embodiment of the present invention, Figure 7 In the figure, point O is the centroid of the dark pixel block, point a is the centroid of the target adjacent superpixel block, and the direction of the centroid line Oa is the suspected slope extension direction, that is, first assume that the dark pixel block and the target adjacent superpixel block belong to the slope pixel block, and the target adjacent superpixel block is the slope pixel block extension result of the dark pixel block, then the vertical direction on the centroid line is the suspected slope direction, that is, one side of the suspected slope direction corresponds to the top or bottom of the slope, and the other side corresponds to the bottom or top of the slope.

[0082] The first connection probability is obtained based on the shape feature difference between the dark pixel block and the target adjacent superpixel block, and the grayscale distribution in the suspected slope direction. The smaller the shape feature difference and the more consistent the grayscale distribution in the suspected slope direction, the more likely the target adjacent superpixel block is to be a superpixel block extending from the slope to which the dark pixel block belongs, and the greater the first connection probability.

[0083] Further considering that in remote sensing images, the adjacent pixel blocks connected to the slope shadow are at the turning point of the light angle, which will have an obvious perspective effect, such as Figure 8 As shown, under the remote sensing perspective, the turning arc at the top of the slope is longer and brighter, while the turning arc at the bottom of the slope is shorter and darker. In addition, on the curved slope at the turning position of the light angle, there are more light-facing surfaces, so there are richer light and shadow changes inside the super-pixel block. The perspective effect will make the slope area appear deep and three-dimensional, and will also affect the perception of the distance and size of the slope in the remote sensing image. Therefore, the embodiment of the present invention further analyzes the degree of shape perspective of the target adjacent super-pixel blocks in the direction of the centroid line, and obtains the second connectivity probability according to the degree of shape perspective and the degree of change of grayscale values ​​on the target adjacent super-pixel blocks. The greater the degree of shape perspective and the greater the degree of change of grayscale values, the more the target adjacent super-pixel blocks belong to the slope pixel blocks generated by the normal perspective effect.

[0084] The connectivity probability between the dark pixel block and the target adjacent super pixel block is obtained according to the first connectivity probability and the second connectivity probability.

[0085] Furthermore, in some implementations of the embodiments of the present invention, the method for obtaining the first connectivity probability includes:

[0086] Taking the dark pixel block as the center, the two closest super pixel blocks on both sides of the suspected slope direction are respectively used as the top pixel block and the bottom pixel block of the dark pixel block. The top pixel block and the bottom pixel block of the target adjacent super pixel block are obtained.

[0087] like Figure 8 As shown, the top pixel block corresponding to the dark pixel block is Io1 pixel block, and the bottom pixel block is Io2 pixel block; the top pixel block of the target adjacent super pixel block is Ia1 pixel block, and the bottom pixel block is Ia2 pixel block. Figure 8 In the figure, since the Ia1 pixel block does not belong to the adjacent super pixel block of the dark pixel block, it is not shown. It should be noted that the Io1 pixel block can also be used as the bottom pixel block, and the Io2 pixel block can be used as the top pixel block, and the target adjacent super pixel block is the same. That is, it is only necessary to ensure that the top pixel block and the bottom pixel block of the dark pixel block and the target adjacent super pixel block are selected in the same direction.

[0088] According to the grayscale difference of the top pixel block and the grayscale difference of the bottom pixel block between the dark pixel block and the target adjacent superpixel block, the grayscale distribution consistency between the dark pixel block and the target adjacent superpixel block is obtained.

[0089] In a specific implementation method of an embodiment of the present invention, the average grayscale value of the slope top pixel block and the slope ground pixel block is used as the grayscale feature. The dark pixel block and the target pixel block each have a set of grayscale values. The dark pixel block and the target pixel block can be regarded as each corresponding to a two-dimensional coordinate. The grayscale distribution consistency can be obtained by calculating the Euclidean distance between the two-dimensional coordinates and negatively correlating the Euclidean distance and normalizing it. That is, the larger the Euclidean distance, the more inconsistent the grayscale distribution of the two, and the smaller the grayscale distribution consistency.

[0090] It should be noted that negative correlation mapping and normalization can be implemented by conventional mathematical means, such as taking the opposite number of the data as the power of an exponential function with a natural constant as the base; or first normalizing by using range standardization, and then obtaining the result of negative correlation mapping and normalization by subtracting a positive integer 1. The specific means will not be limited or elaborated.

[0091] Because the slope should be a long and narrow area, for each super pixel block, the narrow length of the super pixel block is used as the shape feature. The shape matching degree is obtained according to the shape feature difference between the dark pixel block and the target adjacent super pixel block.

[0092] In a specific implementation of the embodiment of the present invention, the maximum inner diameter and the minimum inner diameter of the super pixel block passing through the centroid are obtained, and the ratio of the maximum inner diameter to the minimum inner diameter is used as the slit length. The absolute value of the slit length difference between the dark pixel block and the target adjacent super pixel block is obtained to obtain the shape feature difference. The shape feature difference is also subjected to negative correlation mapping and normalization to obtain the shape matching degree.

[0093] In a specific implementation method of an embodiment of the present invention, after obtaining the shape feature difference between the dark pixel block and the target adjacent super-pixel block, in order to further reflect the difference characteristics, the shape feature difference between the dark pixel block and each adjacent super-pixel block is obtained and averaged, and the average value is used as the denominator, and the shape feature difference of the target adjacent super-pixel block is used as the numerator. The smaller the numerator and the larger the denominator of the fraction, the more it can reflect the connectivity between the target adjacent super-pixel block and the dark pixel block. Therefore, the fraction can be normalized by negative correlation mapping to obtain the shape matching degree.

[0094] The first connectivity probability between the dark pixel block and the target adjacent superpixel block is obtained according to the grayscale distribution consistency and shape matching.

[0095] In a specific implementation of the embodiment of the present invention, because the grayscale distribution consistency and the shape matching degree are both data with a value range between 0 and 1, the average of the two can be used as the first connection probability. In another specific implementation of the embodiment of the present invention, the two can also be averaged and then further normalized to obtain the first connection probability.

[0096] Preferably, in one embodiment of the present invention, the method for acquiring the second connectivity probability includes:

[0097] Get the minimum bounding window of the target adjacent superpixel block. The minimum bounding window is an isosceles trapezoid, and the bottom edge is parallel to the centroid line. Get the shape perspective degree according to the area ratio of the target adjacent superpixel block in the minimum bounding window. Fig. 9 , which shows a schematic diagram of a minimum external window provided by an embodiment of the present invention, Fig. 9 o is the centroid of the dark pixel block, a is the centroid of the target adjacent superpixel block, and the long side of the constructed minimum bounding window faces upward and is parallel to the centroid line. The larger the area of ​​the target adjacent superpixel block in the minimum bounding window, the more likely the target adjacent superpixel block is a trapezoid, and the more likely the target adjacent superpixel block is to have a perspective effect in the direction of slope extension.

[0098] The texture feature value of each pixel in the target adjacent superpixel block is obtained, and the grayscale value change degree is obtained according to the variance of the texture feature value. In some implementations of the embodiments of the present invention, the texture feature value is obtained by processing the superpixel block with a Laplace operator, that is, each pixel in the superpixel block corresponds to a texture feature value obtained by a Laplace operator. The obtained variance is normalized to obtain the grayscale value change degree. The greater the grayscale value change degree, the richer the light and shadow changes of the target adjacent superpixel block, and the more it belongs to the slope extension pixel block of the dark pixel block.

[0099] The second connectivity probability between the dark pixel block and the target adjacent super pixel block is obtained according to the shape perspective degree and the gray value change degree. In the embodiment of the present invention, the shape perspective degree and the gray value change degree are directly multiplied to obtain the second connectivity probability.

[0100] In a specific implementation of the embodiment of the present invention, the first connectivity probability characterizes the slope extension feature between the dark pixel block and the target adjacent super pixel block, and the second connectivity probability describes the angle change feature of light, so after obtaining the first connectivity probability and the second connectivity probability, the two are multiplied to obtain the connectivity probability. Because each adjacent super pixel block corresponds to a connectivity probability, the adjacent super pixel block with the largest connectivity probability is selected as the seed super pixel block.

[0101] Preferably, in one embodiment of the present invention, the preset pothole characteristic conditions include:

[0102] In the process of regional growth, if the grayscale interval overlap between the superpixel block to be grown and the grown region is greater than the preset overlap threshold, the difference in grayscale variance is less than the preset variance condition threshold, and the gradient information difference is less than the preset gradient difference threshold, then the superpixel block to be grown is determined to meet the pit feature condition. That is, the superpixel block to be grown that meets the three conditions at the same time during the growth process is determined to meet the pit feature condition, and it is merged with the grown region until the grown region can no longer be merged to obtain a slope pixel block.

[0103] In the embodiment of the present invention, the grayscale interval overlap can be obtained by the intersection and union ratio of two grayscale intervals, and the overlap threshold is set to 60%. The variance condition threshold is set to 25. The gradient information includes the gradient amplitude and the gradient direction, and the gradient difference threshold also includes the gradient amplitude difference threshold and the gradient direction difference threshold. The gradient amplitude difference threshold is set to 10, and the gradient direction difference threshold is set to 30°.

[0104] It should be noted that the pothole characteristic conditions can also be determined through multiple experiments to determine the final condition type, such as information entropy, texture characteristics, etc., which will not be limited or elaborated here.

[0105] Furthermore, in some implementations of the embodiments of the present invention, in order to avoid the slope pixel block formed in the grown area being too large, the pit feature condition also sets a growth area threshold, and if the grown area is greater than or equal to the growth area threshold during the regional growth process, the growth is stopped. In the embodiments of the present invention, considering that the pit is a ring structure, the shadow area corresponding to the dark pixel block is the pit slope at one of the backlight angles, and the ring pit slope can be considered as a pit slope including four angles, so the embodiments of the present invention set a growth area threshold for the growth process of each seed super pixel block, add the area of ​​the seed super pixel block and the corresponding dark pixel block, and then multiply by 4 to obtain the growth area threshold corresponding to the seed super pixel block.

[0106] Preferably, in an embodiment of the present invention, the method for screening the interpolation position includes:

[0107] The average elevation gradient amplitude of the boundary grid of the grid area is obtained, and the sampling coefficient is obtained after negative correlation mapping and normalization of the average elevation gradient amplitude. In an embodiment of the present invention, the gradient amplitude of the boundary grid is the difference in elevation value between each edge pixel and the external adjacent pixel. The method of negative correlation mapping and normalization has been described in the aforementioned content and will not be repeated here. That is, the smaller the average elevation gradient amplitude is, the less obvious the difference in elevation value between the grid area and the adjacent area is, and the more likely it is hidden pothole information. Therefore, the corresponding sampling coefficient is larger, and more interpolation positions need to be selected.

[0108] The number of pixels in the slope pixel block is multiplied by the sampling coefficient to obtain the number of samples; the pixels selected in the slope pixel block according to the number of samples in descending order of edge gradient amplitude are used as interpolation pixels, and the positions corresponding to the interpolation pixels in the three-dimensional grid model are the interpolation positions. That is, the pixels are sorted in descending order of edge gradient amplitude, and the first number of pixels in the sorting result are used as interpolation pixels, that is, the pixels with larger edge gradient amplitude are used as interpolation pixels, and interpolation is performed at the corresponding positions of the three-dimensional grid model to further highlight the characteristics of potholes.

[0109] Preferably, in some implementations of the embodiments of the present invention, the method for interpolating the elevation value of the interpolation position includes:

[0110] For each interpolation position, the pixel to which the interpolation position belongs in the three-dimensional grid model is used as the target pixel.

[0111] Because a pixel includes four vertices, each vertex also has elevation information, so the embodiment of the present invention uses the elevation value of each vertex as reference data for analysis. For a vertex of the target pixel, the distance between the vertex and the interpolation position is negatively correlated and normalized to obtain a distance weight. That is, the smaller the distance, the closer the interpolation position is to the vertex, the greater the reference of the data of the vertex to the interpolation position, and the greater the distance weight.

[0112] Similarly, a vertex also corresponds to a pixel value in the remote sensing image, and the pixel value ratio of the interpolated position and the vertex is further obtained. The pixel value ratio is used as a coefficient to further adjust the value of the vertex elevation value, that is, by comparing the information in the remote sensing image, the reference degree of the vertex elevation value is further determined.

[0113] The distance weight, pixel value ratio and vertex elevation value are multiplied to obtain the initial interpolation data of the vertex. The elevation values ​​of all vertices are further combined, and the average initial interpolation data of all vertices of the target pixel is used as the elevation value of the interpolation position for interpolation.

[0114] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0115] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A method for optimizing segmentation of geological survey remote sensing images, characterized in that: The method comprises: Obtain remote sensing images and 3D grid models of abandoned mining areas; Obtain a superpixel segmentation image of the remote sensing image; filter out dark pixel blocks according to pixel values ​​of superpixel blocks in the superpixel segmentation image; obtain a suspected slope direction according to the direction between the dark pixel block and an adjacent superpixel block; obtain a connectivity probability according to the pixel value distribution in the suspected slope direction and the shape matching degree between the dark pixel block and the adjacent superpixel block; Among all the adjacent super pixel blocks of the dark pixel block, a seed super pixel block is selected according to the connectivity probability; the seed super pixel block is subjected to regional growth according to the preset pothole characteristic condition to obtain a slope pixel block; In the three-dimensional grid model, according to the elevation gradient information in the grid area corresponding to the slope pixel block, an interpolation position is selected in the grid area and the elevation value is interpolated to obtain an optimized three-dimensional grid model; The pothole slope surface is segmented in the optimized three-dimensional grid model.

2. The method for optimizing segmentation of geological survey remote sensing images according to claim 1, characterized in that: The dark pixel block screening method comprises: The grayscale mean of the superpixel block is used as the overall grayscale feature, and Otsu threshold segmentation is performed based on the overall grayscale feature to obtain two types of superpixel blocks, and the superpixel block with the smallest overall grayscale feature is used as a suspected dark pixel block; If the overall grayscale feature of the suspected dark pixel block is smaller than the overall grayscale features of all adjacent super pixel blocks, the corresponding suspected dark pixel block will be used as the dark pixel block.

3. The method for optimizing segmentation of geological survey remote sensing images according to claim 1, characterized in that: The method for obtaining the connectivity probability includes: Taking any adjacent superpixel block of the dark pixel block as the target adjacent superpixel block, obtaining a centroid line between the dark pixel block and the target adjacent superpixel block, and taking a direction perpendicular to the centroid line as the suspected slope direction; Obtaining a first connectivity probability according to a difference in shape features between the dark pixel block and a target adjacent super pixel block, and a grayscale distribution in a direction of the suspected slope; In the direction of the centroid line, analyzing the degree of shape perspective of the target adjacent superpixel block, and obtaining a second connectivity probability according to the degree of shape perspective and the degree of change of grayscale values ​​on the target adjacent superpixel block; The connectivity probability between the dark pixel block and the target adjacent super pixel block is obtained according to the first connectivity probability and the second connectivity probability.

4. The method for optimizing segmentation of geological survey remote sensing images according to claim 3, characterized in that: The method for obtaining the first connectivity probability includes: Taking the dark pixel block as the center, the two closest superpixel blocks along the two sides of the suspected slope direction are respectively used as the top pixel block and the bottom pixel block of the dark pixel block; obtaining the top pixel block and the bottom pixel block of the target adjacent superpixel block, and obtaining the grayscale distribution consistency between the dark pixel block and the target adjacent superpixel block according to the grayscale difference of the top pixel block and the grayscale difference of the bottom pixel block between the dark pixel block and the target adjacent superpixel block; For each superpixel block, the narrow length of the superpixel block is used as the shape feature; the shape matching degree is obtained according to the shape feature difference between the dark pixel block and the target adjacent superpixel block; A first connectivity probability between the dark pixel block and the target adjacent super pixel block is obtained according to the grayscale distribution consistency and the shape matching degree.

5. The method for optimizing segmentation of geological survey remote sensing images according to claim 3, characterized in that: The method for acquiring the second connectivity probability includes: Obtaining a minimum circumscribed window of the target adjacent superpixel block, wherein the minimum circumscribed window is an isosceles trapezoid, and the bottom side is parallel to the centroid line; obtaining the perspective degree of the shape according to the area ratio of the target adjacent superpixel block in the minimum circumscribed window; Obtaining a texture feature value of each pixel in the target adjacent superpixel block, and obtaining the grayscale value variation degree according to the variance of the texture feature value; A second connectivity probability between the dark pixel block and the target adjacent super pixel block is obtained according to the shape perspective degree and the gray value change degree.

6. The method for optimizing segmentation of geological survey remote sensing images according to claim 1, characterized in that: The preset pothole characteristic conditions include: During the region growing process, if the grayscale interval overlap between the superpixel block to be grown and the grown region is greater than the preset overlap threshold, and the grayscale variance difference is less than the preset variance condition threshold, and the gradient information difference is less than the preset gradient information threshold, then it is determined that the superpixel block to be grown meets the pit feature condition.

7. The method for optimizing segmentation of geological survey remote sensing images according to claim 1, characterized in that: The method for screening the interpolation position includes: The average elevation gradient amplitude of the boundary grid of the grid area is obtained, and the sampling coefficient is obtained after negative correlation mapping and normalization of the average elevation gradient amplitude; the number of pixels in the slope pixel block is multiplied by the sampling coefficient to obtain the sampling number; pixel points selected in the slope pixel block according to the sampling number in descending order of edge gradient amplitude are used as interpolation pixel points, and the position of the interpolation pixel point corresponding to the three-dimensional grid model is the interpolation position.

8. The method for optimizing segmentation of geological survey remote sensing images according to claim 1, characterized in that: The method for interpolating the elevation value of the interpolation position includes: For each interpolation position, the pixel to which the interpolation position belongs in the three-dimensional grid model is taken as the target pixel; For a vertex of the target pixel, negatively correlate the distance between the vertex and the interpolation position and normalize it to obtain a distance weight; obtain a pixel value ratio between the interpolation position and the vertex; multiply the distance weight, the pixel value ratio and the elevation value of the vertex to obtain initial interpolation data of the vertex; The average initial interpolation data of all vertices of the target pixel is used as the elevation value of the interpolation position for interpolation.

9. The method for optimizing segmentation of geological survey remote sensing images according to claim 1, characterized in that: The step of selecting a seed super pixel block according to the connectivity probability comprises: Among all the adjacent superpixel blocks of the dark pixel block, the adjacent superpixel block with the largest connectivity probability is selected as the seed superpixel block.

10. The method for optimizing segmentation of geological survey remote sensing images according to claim 6, characterized in that: The preset pit characteristic condition also includes setting a growth area threshold. During the regional growth process, if the grown area is greater than or equal to the growth area threshold, the growth is stopped.

Images