A method for extracting road intersections from SAR images based on omnidirectional triangulation model

Abstract

The present invention provides a method for extracting road intersections from SAR images based on an omnidirectional triangulation model, comprising the following steps: S1. performing high-quality speckle suppression processing on a high-resolution SAR image to obtain a processed high-resolution SAR image; S2. performing rough extraction of road intersections on the processed high-resolution SAR image; S3. confirming the roughly extracted road intersections based on the omnidirectional triangulation model to determine the final road intersection. The present invention utilizes the Laplace operator to optimize the edge detection model of the AD filter, effectively achieving speckle noise suppression in the road intersection area. The use of the omnidirectional triangulation model can further resist strong speckle noise that is difficult to suppress, effectively achieving accurate extraction of common two-dimensional road intersections. Compared with current SAR image road intersection extraction methods, this method has significantly improved resistance to speckle noise.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image data processing, and in particular to a SAR image road intersection extraction method based on an omnidirectional triangle model. Background Art

[0002] As a core technology in the field of active microwave remote sensing, Synthetic Aperture Radar (SAR) has the unique advantage of all-weather and all-day Earth observation and strong penetration ability of certain land features (such as vegetation canopies, deserts, shallow water, etc.). It has become a very important supporting means in the fields of environmental monitoring, global change, urban planning, resource exploration, disaster assessment and planetary exploration, and is currently one of the most cutting-edge new Earth observation technologies in the world.

[0003] Road intersections are important hubs where roads converge. They are of great significance in many areas, including computer vision, image processing, stereo matching, and aerial road detection. For example:

[0004] (1) In the semi-automatic road network construction method, road intersections can be used as the initial seed points of the algorithm to guide the extraction of roads and the establishment of the road network;

[0005] (2) As a stable and reliable feature in the road network, road intersections can be used as effective ground control points (GCPs), which play a key role in the establishment of geometric correction models for remote sensing images;

[0006] (3) As a basic element of the road network, road intersections can provide accurate location information, direction information, and connection information for roads, and are widely used in areas such as automatic matching of road networks, assisted navigation, and change monitoring and updating of geographic information data;

[0007] (4) In low- and medium-resolution remote sensing images, road intersections can be viewed as point targets (usually the intersection of curves or straight lines). Relying on their reliable and stable characteristics, they can play an important role in the image registration process.

[0008] (5) For modern high-tech warfare, accurate identification of military targets (such as large road intersections) is crucial.

[0009] However, the recognition of road intersections in remote sensing images has long been neglected, especially in SAR images, where research is extremely scarce, which is not commensurate with its importance. Therefore, effectively extracting road intersections from SAR images has become one of the important tasks in remote sensing image interpretation.

[0010] With the advancement of SAR imaging technology, image resolution has greatly improved. Civilian spaceborne SAR satellites currently achieve resolutions up to 1 meter (e.g., in TerraSAR-X spotlight mode). High-resolution remote sensing imagery provides a vast amount of complex information for image interpretation, but it also presents new challenges for road intersection extraction. For example, in high-resolution imagery, the more complex terrain details and texture information in the area surrounding the road intersection further interfere with intersection extraction. Furthermore, road intersections exhibit different characteristics in SAR imagery of different resolutions. In low- and medium-resolution images, road intersections can be considered point features (the intersection of lines or curves); in higher-resolution SAR imagery, road intersections appear as regional features with a certain area. Road intersection extraction methods designed for low- and medium-resolution SAR imagery are clearly difficult to apply to medium- and high-resolution SAR imagery. Among various types of high-resolution remote sensing imagery, significant differences exist between SAR satellite imagery and optical remote sensing imagery. Imaging characteristics of images from different sensors also vary. Due to the unique coherent imaging mechanism of SAR, SAR images are saturated with a large amount of typical speckled multiplicative noise, which is more difficult to suppress than additive noise in optical imagery. Existing road intersection extraction models for SAR images are often more susceptible to interference from coherent speckle noise, resulting in limited road intersection extraction accuracy. This is the technical problem that this application mainly solves. Summary of the Invention

[0011] In response to the above problems existing in the prior art, the present invention provides a method for extracting road intersections from SAR images based on an omnidirectional triangulation model, which can effectively overcome the problem of limited noise resistance of existing high-resolution SAR image road intersection extraction methods.

[0012] The present invention provides a method for extracting road intersections from SAR images based on an omnidirectional triangulation model, comprising the following steps:

[0013] S1. performing high-quality speckle reduction processing on the high-resolution SAR image to obtain a processed high-resolution SAR image;

[0014] S2. performing rough extraction of road intersections on the processed high-resolution SAR image;

[0015] S3. Confirm the roughly extracted road intersection based on the omnidirectional triangulation model to determine the final road intersection.

[0016] Preferably, in step S1, a speckle reduction model based on partial differential equations is used to perform high-quality speckle reduction processing on the high-resolution SAR image.

[0017] Preferably, the speckle suppression model based on partial differential equations is as follows:

[0018]

[0019] Among them, div represents the divergence operator, represents the gradient operator, I0 is the original image, I(t) represents the SAR image at time t, and c(x) represents the diffusion function.

[0020] Preferably, the specific model of c(x) is as follows:

[0021]

[0022] Among them, E T is the edge detection threshold, which is obtained by averaging the edge detection values ​​at all pixels of the high-resolution SAR image.

[0023] Preferably, the edge detection model is as follows:

[0024]

[0025] Among them, E x,y Represents the edge detection value at the pixel position x, y, Max, Min and Mean represent the maximum value, minimum value and mean value respectively, Ω represents the neighborhood rectangular window corresponding to the pixel position, k is a constant, L x,y Represents the Laplace result at pixel position x,y.

[0026] Preferably, the Jacobi iteration method is used to solve the problem, and the model is as follows:

[0027]

[0028] in, It represents the pixel value at the pixel position x,y after the nth iteration, △t is the preset time step, Indicates the pixel change factor at position x,y after the n-1th iteration, Represents the diffusion coefficient at position x,y after the n-1th iteration.

[0029] Preferably, step S2 specifically includes the following sub-steps:

[0030] S21. Sharpening the high-resolution SAR image after filtering in step S1;

[0031] S22. Use the Bottom-Hat algorithm to detect the sharpened high-resolution SAR image, remove obvious non-road intersection areas, and detect candidate road intersection areas;

[0032] S23. Binarize the high-resolution SAR image after the Bottom-Hat operation to form a binarized high-resolution SAR image, and then use the seed filling method to mark the connected areas of the binarized high-resolution SAR image, and use the center point coordinates of the connected areas as the center point coordinates of the candidate road intersection area.

[0033] Preferably, in step S21, the specific model of the sharpening process is as follows:

[0034]

[0035] Among them, X ij is the pixel value of the high-resolution SAR image, △d2 is the pixel value X ij The difference between the mean of the pixels in the 8 surrounding neighborhoods, T2 is the preset threshold, and C is a constant value.

[0036] Preferably, in step S22, the Bottom-Hat operation is specifically as follows:

[0037] B=f·bf

[0038] Where f represents the high-resolution SAR image after filtering and sharpening, b represents the structural element, and f·b represents the closing operation. The closing operation is a combination of the basic morphological operations of dilation and erosion, which can be expressed as follows:

[0039]

[0040] in, and denote the dilation and erosion operators respectively.

[0041] The present invention also provides a storage medium containing computer-executable instructions, wherein the computer-executable instructions are used to execute the above-mentioned method when executed by a computer processor.

[0042] The accuracy of road intersection extraction from existing high-resolution SAR images is often affected by speckle noise. This invention focuses on this problem and designs a SAR image road intersection extraction method based on an omnidirectional triangulation model, based on the speckle noise characteristics of SAR images and the characteristics of road intersection models. This method first uses the Laplace operator to optimize the edge detection model of the AD filter, effectively achieving speckle noise suppression in the road intersection area and providing good SAR quality for subsequent road intersection extraction. Secondly, the designed omnidirectional triangulation model can further resist strong speckle noise that is difficult to suppress, effectively achieving accurate extraction of common two-dimensional road intersections. Overall, compared with current SAR image road intersection extraction methods, this method has significantly improved its resistance to speckle noise. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 4 is a flowchart of a method for extracting road intersections from SAR images based on an omnidirectional triangulation model according to an embodiment of the present invention.

[0044] Figure 2 Schematic diagram of an omnidirectional triangle model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The present invention will be further described in detail below with reference to the accompanying drawings and examples. It will be understood that the specific embodiments described herein are intended only to illustrate the present invention and are not intended to limit the present invention. It should also be noted that, for ease of description, the accompanying drawings only illustrate portions relevant to the present invention, not all structures.

[0046] Existing road intersection extraction models for high-resolution SAR imagery are often susceptible to interference from speckle noise, resulting in limited road intersection extraction accuracy. To address this issue, this method proposes a new edge detection model based on the Laplace operator and optimizes the AD filter, effectively suppressing the anisotropic effect of speckle in SAR images. This reduces the susceptibility of intersection edges to the noise reduction model, while also adopting a hierarchical strategy for coarse and fine intersection extraction. In the coarse extraction process, the morphological features of intersections in SAR images are sharpened and fully integrated to better identify candidate intersection regions. In the fine intersection extraction process, an omnidirectional triangulation model is designed, and based on prior knowledge, accurate and effective intersection extraction is achieved.

[0047] like Figure 1 As shown, the present invention provides a method for extracting road intersections from SAR images based on an omnidirectional triangulation model, which mainly includes the following steps:

[0048] S1. Perform high-quality speckle suppression processing on the high-resolution SAR image to obtain a processed high-resolution SAR image.

[0049] Coherent speckle is an inevitable inherent property of high-resolution SAR images. It seriously affects image quality and causes unclear contrast between road intersections and surrounding objects, thus interfering with the accurate extraction of road intersections. It is urgent to suppress it.

[0050] According to a preferred embodiment of the present invention, a partial differential equation-based speckle reduction model is used to perform high-quality speckle reduction processing on high-resolution SAR images. Specifically, a diffusion function is constructed to calculate the diffusion coefficient at each pixel in the high-resolution SAR image. The diffusion coefficient ranges from 0 to 1. Based on this diffusion coefficient, speckle reduction at different scales (0-1) can be achieved for each pixel in the high-resolution SAR image.

[0051] The speckle suppression model based on partial differential equations is:

[0052]

[0053] Among them, div represents the divergence operator, represents the gradient operator, I0 is the original image, I(t) represents the SAR image at time t, and c(x) represents the diffusion function. The specific model is as follows:

[0054]

[0055] Among them, q represents the edge detection model, q T Represents the edge detection threshold.

[0056] The above model effectively achieves diffusion at different scales for each pixel in high-resolution SAR images. However, determining the specific diffusion coefficient value for each pixel within the range of 0-1 is even more crucial. It should be noted that high-resolution SAR images require speckle reduction at different scales because noise filtering often results in the loss of edge points, which are crucial for subsequent road intersection extraction. Therefore, a high-quality speckle reduction model often requires maintaining high-scale suppression in homogeneous regions and low-scale suppression in edge regions. Therefore, determining edge regions is key to more accurately achieving different-scale filtering for each pixel in SAR images.

[0057] The edge detection operator of traditional AD filters has poor performance. The SAR filter model constructs an instantaneous coefficient of variation, but its interference ability against strong noise is insufficient. To address this, this application introduces Laplace and proposes a new edge detection model as follows:

[0058]

[0059] Among them, E x,y Represents the edge detection value at the pixel position x, y, Max, Min and Mean represent the maximum value, minimum value and mean value respectively, Ω represents the neighborhood rectangular window corresponding to the pixel position, k is a constant, L x,yRepresents the Laplace result at pixel position x,y.

[0060] Based on the proposed new edge detection model, it can be introduced into the diffusion function to form a new diffusion function model as follows:

[0061]

[0062] Among them, E T is the edge detection threshold, which is obtained by averaging the edge detection values ​​at all pixels of the high-resolution SAR image.

[0063] Therefore, based on the proposed new edge detection model and diffusion function, the coherent speckle suppression model based on partial differential equations can be further updated to form an improved AD filter model. That is, the coherent speckle suppression model based on partial differential equations adopted in the present invention is as follows:

[0064]

[0065] Among them, div represents the divergence operator, represents the gradient operator, I0 is the original image, I(t) represents the SAR image at time t, and c(x) represents the diffusion function.

[0066] The solution of the improved AD filter model is consistent with the solution of the traditional AD filter model, using the Jacobi iteration method. The model is as follows:

[0067]

[0068] in, It represents the pixel value at the pixel position x,y after the nth iteration, △t is the preset time step, Indicates the pixel change factor at position x,y after the n-1th iteration,

[0069] Represents the diffusion coefficient at position x,y after the n-1th iteration.

[0070] S2. performing rough extraction of road intersections on the processed high-resolution SAR image.

[0071] Step S2 specifically includes the following sub-steps:

[0072] S21. Sharpening the high-resolution SAR image after filtering in step S1.

[0073] In the rough extraction of road intersections, the high-resolution SAR image after filtering in step S1 is first sharpened to further highlight the edges of the road intersections. Sharpening requires calculating the pixel value X of the high-resolution SAR image after filtering.ij If the absolute value of the difference △d2 between the pixel value and the mean of the eight neighboring pixels is greater than the preset threshold T2, the pixel value is subtracted from the given constant C value as the new pixel value, otherwise the original pixel value is retained. The specific model is as follows:

[0074]

[0075] S22. Use the Bottom-Hat operation to detect the sharpened high-resolution SAR image, eliminate obvious non-road intersection areas, and detect candidate road intersection areas.

[0076] Based on the road intersection model characteristics of SAR images, the shape and size of the structural elements are set, and the Bottom-Hat method is used to directly detect the entire image, eliminate obvious non-road intersection areas, and detect candidate road intersection areas.

[0077] In high-resolution SAR images, road intersections appear as darker areas. The morphological Bottom-Hat operation can effectively highlight darker areas and separate dark spots. The Bottom-Hat operation is as follows:

[0078] B=f·bf

[0079] Where f represents the high-resolution SAR image after filtering and sharpening, b represents the structural element, and f·b represents the closing operation. The closing operation is a combination of the basic morphological operations of dilation and erosion, which can be expressed as follows:

[0080]

[0081] in, and denote the dilation and erosion operators respectively.

[0082] It should be pointed out that the key to morphological operations is the selection of the structural element b. Based on the morphological characteristics, the structural elements can be divided into linear, square, circular, prismatic, etc. The shape and size of the selected structural element must be similar to the geometric properties of the target image to be processed. Although the shapes of the road intersection area are diverse, the intersection and the road are connected as one, and the connection center presents a circular area feature with relatively uniform grayscale. Therefore, the present invention sets the structural element to be nearly circular. In addition, the diameter of the selected nearly circular structural element cannot be larger than the inner diameter of the road intersection, otherwise there will be non-road intersection areas in the area covered by the structural element, which will easily blur the road intersection. At the same time, in order to remove other areas of the road and retain only the road intersection, the diameter of the structural element also needs to be larger than the road width.

[0083] Even after the Bottom-Hat operation, the image still contains a significant amount of stray noise. Therefore, according to a preferred embodiment of the present invention, a closing operation can be further performed on the image. This closing operation effectively removes noise with smaller structures in the SAR image and effectively fills any small gaps in the image.

[0084] S23. Binarize the high-resolution SAR image after the Bottom-Hat operation to form a binarized high-resolution SAR image, and then use the seed filling method to mark the connected areas of the binarized high-resolution SAR image, and use the center point coordinates of the connected areas as the center point coordinates of the candidate road intersection area.

[0085] After the closing operation, the features such as road intersections in the SAR image still appear as darker areas. g And compare each pixel value of the closed image with the threshold T one by one g Comparison, the pixel value is less than the threshold T g Some of the images are marked as 1 and the others are marked as 0, thus forming a binary image.

[0086] After binarization, the parts with pixel values ​​of 1 are mostly part of the road intersection, and there are many of them. There is a road intersection area with many small binary pixel values ​​of 1. The seed filling method is used to mark the connected areas of the binary image, and any foreground pixel is regarded as a seed point. Following the two basic conditions of adjacent positions and identical pixel values, the foreground pixels in the neighborhood (the present invention uses 8 neighborhoods) are merged into a pixel set to finally obtain a connected area. The connected area can be considered as the integration of the areas with more small binary pixel values ​​of 1 in the local area. Since it is a coarse extraction, we select the centers of these connected areas as the candidate road intersection judgment reference positions for further confirmation of the road intersection.

[0087] S3. Confirm the roughly extracted road intersection based on the omnidirectional triangulation model to determine the final road intersection.

[0088] The confirmation of road intersections uses an omnidirectional triangulation model. The steps to confirm the road intersection model using the omnidirectional triangulation model are as follows: ① Construct an omnidirectional triangulation model. The model is designed as 16 identical isosceles triangles, each with a vertex angle of 22.5°, covering a total of 360° in all directions. Figure 2As shown in the figure below; ② Calculate the grayscale mean of the triangular area within the model. Taking each candidate intersection reference position as the starting point, 0° as the starting direction, and 360° as the ending direction, calculate the grayscale mean of the triangular coverage area in 22.5° steps. ③ Draw a line graph. Statistically calculate the 16 grayscale means and draw a line graph. ④ Confirm the road intersection. Based on the number of trough points in the line graph and the angles of adjacent trough points, determine the road intersection type according to Table 1 and complete the road intersection confirmation.

[0089] Table 1 Road intersection model characteristics

[0090]

[0091] The present invention also provides a storage medium containing computer-executable instructions, characterized in that the computer-executable instructions are used to execute the method described in the above embodiment when executed by a computer processor.

[0092] Through the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented with the help of software and necessary general-purpose hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0093] Although the present invention has been described in detail above using general explanations, specific embodiments, and experiments, it will be apparent to those skilled in the art that modifications and improvements may be made based on the present invention. Therefore, such modifications and improvements, which do not depart from the spirit of the present invention, are intended to be within the scope of protection claimed herein.

Claims

1. A method for extracting road intersections from SAR images based on an omnidirectional triangulation model, comprising the following steps: S1. performing high-quality speckle reduction processing on the high-resolution SAR image to obtain a processed high-resolution SAR image; S2. performing rough extraction of road intersections on the processed high-resolution SAR image; S3. Confirm the roughly extracted road intersection based on the omnidirectional triangulation model to determine the final road intersection; In step S1, the edge detection model used in the speckle reduction process is as follows: Among them, E x,y Represents the edge detection value at the pixel position x, y, Max, Min and Mean represent the maximum value, minimum value and mean value respectively, Ω represents the neighborhood rectangular window corresponding to the pixel position, k is a constant, L x,y Represents the Laplace result at pixel position x,y; The steps to identify road intersections using the omnidirectional triangulation model are as follows: Construct an omnidirectional triangle model. The model is designed as 16 identical isosceles triangles, each with a vertex angle of 22.5°, covering a total of 360° in all directions. Calculate the grayscale mean of the triangular area within the model. Starting from each candidate intersection reference position, with 0° as the starting direction and 360° as the ending direction, the grayscale mean of the triangular coverage area is calculated in steps of 22.5°. Draw a line graph, and calculate the 16 grayscale means and draw them into a line graph; The confirmation of road intersections is completed through the number of trough points in the line graph and the angles between adjacent trough points.

2. The method according to claim 1, characterized in that In step S1 , high-quality speckle reduction processing is performed on high-resolution SAR images using a speckle reduction model based on partial differential equations.

3. The method according to claim 2, characterized in that The speckle suppression model based on partial differential equations is as follows: Where div represents the divergence operator, ▽ represents the gradient operator, I0 is the original image, I(t) represents the SAR image at time t, and c(x) represents the diffusion function.

4. The method according to claim 3, characterized in that The specific model of c(x) is as follows: Among them, E T is the edge detection threshold, which is obtained by averaging the edge detection values ​​at all pixels of the high-resolution SAR image.

5. The method according to claim 1, wherein The Jacobi iteration method is used to solve the problem. The model is as follows: in, represents the pixel value at the pixel position x,y after the nth iteration, Δt is the preset time step, Indicates the pixel change factor at position x,y after the n-1th iteration, Represents the diffusion coefficient at position x,y after the n-1th iteration.

6. The method according to claim 1, wherein Step S2 specifically includes the following sub-steps: S21. Sharpening the high-resolution SAR image after filtering in step S1; S22. Use the Bottom-Hat algorithm to detect the sharpened high-resolution SAR image, remove obvious non-road intersection areas, and detect candidate road intersection areas; S23. Binarize the high-resolution SAR image after the Bottom-Hat operation to form a binarized high-resolution SAR image, and then use the seed filling method to mark the connected areas of the binarized high-resolution SAR image, and use the center point coordinates of the connected areas as the center point coordinates of the candidate road intersection area.

7. The method according to claim 6, characterized in that In step S21, the specific model of the sharpening process is as follows: Among them, X ij is the pixel value of the high-resolution SAR image, Δd2 is the pixel value X ij The difference between the mean of the pixels in the 8 surrounding neighborhoods, T2 is the preset threshold, and C is a constant value.

8. The method according to claim 6, characterized in that In step S22, the Bottom-Hat operation is specifically as follows: B=f·bf Where f represents the high-resolution SAR image after filtering and sharpening, b represents the structural element, and f·b represents the closing operation. The closing operation is a combination of the basic morphological operations of dilation and erosion, which can be expressed as follows: in, and denote the dilation and erosion operators respectively.

9. A storage medium containing computer-executable instructions, characterized in that: The computer executable instructions are used to perform the method according to any one of claims 1 to 8 when executed by a computer processor.

Images