Atmospheric frontal surface center line positioning method for synthetic aperture radar image

Abstract

The invention discloses an atmospheric frontal surface center line positioning method for a synthetic aperture radar image, which adopts an improved edge detection algorithm, removes image noise and highlights the characteristics of an atmospheric frontal surface by combining adaptive thresholding and bilateral filtering technologies, thereby effectively enhancing edge information. Converting the edge image into a simplified skeleton diagram by applying a skeletonization technology, and extracting an edge path with the maximum geometric length from the simplified skeleton diagram through a geometric longest main branch screening method; and extending an edge curve to an image boundary according to geometric characteristics of the edge by adopting a self-adaptive curve extension algorithm so as to ensure continuity and integrity of the segmented boundary. According to the method, by introducing an innovative technology of curve extension and skeleton screening, the accuracy and integrity of atmospheric frontal surface edge extraction are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of radar image technology, and specifically to a method for locating the centerline of an atmospheric front in synthetic aperture radar images. Background Technology

[0002] Atmospheric fronts in synthetic aperture radar (SAR) images typically appear as bright or dark bands of a certain width and continuity. Accurate detection and segmentation of these fronts are of significant scientific value for marine meteorological analysis, weather forecasting, and climate change research. However, due to the inherent speckle noise in SAR images, complex air-sea interactions, and the multi-scale turbulent structure of the atmospheric boundary layer, traditional edge detection algorithms have significant limitations in applications involving such images. Traditional edge detection algorithms (such as Canny and Sobel) are extremely sensitive to the inherent speckle noise in SAR images, leading to unstable gradient calculations, generating numerous false edges, and masking the true weak frontal signals. In complex air-sea backgrounds, due to inappropriate threshold selection and a lack of global structural analysis, true frontal edges often exhibit breaks and discontinuities. Furthermore, traditional methods fail to fully consider the geometric characteristics of atmospheric fronts as continuous curved structures, making it difficult for extracted edges to form closed and complete segmentation boundaries.

[0003] Existing research largely focuses on improving general edge detection algorithms, while refined processing methods specifically for atmospheric fronts in SAR images are relatively lacking. Some researchers have attempted to combine multi-scale analysis and active contour models, but these methods suffer from high computational complexity and sensitivity to the initial contour location. Another approach is based on region growing and level set evolution theory, but it is prone to boundary leakage problems in weak edges and complex texture regions. Furthermore, existing methods often neglect the global geometric characteristics and topological continuity of edges when processing large-scale atmospheric front structures, resulting in segmentation results that fail to meet the requirements of high-precision meteorological analysis. Summary of the Invention

[0004] Purpose of the invention: The purpose of this invention is to provide a method for locating the centerline of atmospheric fronts in synthetic aperture radar images. It combines gradient statistical feature analysis, geometric longest path selection, and adaptive curve extension into an innovative solution. Through the synergistic optimization of multiple technical aspects, it achieves accurate extraction of the edge of atmospheric fronts in complex backgrounds, solving the problems of breakage and blurring that occur in the extraction of front edges in complex ocean-atmosphere backgrounds using traditional algorithms.

[0005] Technical solution: The present invention provides a method for locating the centerline of an atmospheric front in synthetic aperture radar images, comprising the following steps:

[0006] (1) Perform bilateral filtering on the original SAR image to achieve noise suppression and edge protection;

[0007] (2) Calculate the gradient field of the smoothed image and perform normalization and Gaussian smoothing.

[0008] (3) Adaptive thresholding is performed based on the statistical characteristics of the gradient image to generate a binary edge image;

[0009] (4) Perform morphological operations on the binary edge image to optimize edge connectivity and integrity;

[0010] (5) Skeletonize the morphologically optimized binary image and perform edge filtering;

[0011] (6) Extract the longest connected path from the filtered skeleton graph as the main branch skeleton and perform parametric curve smoothing.

[0012] (7) Based on the endpoints and local direction information of the main branch skeleton, the adaptive curve extension technique is used to extend the smooth curve to the image boundary to generate a complete atmospheric front centerline.

[0013] Furthermore, in step (1), the bilateral filtering process adopts a multi-parameter optimization configuration, and achieves a smooth edge-preserving effect through the dual weight calculation of spatial proximity and pixel value similarity.

[0014] Furthermore, in step (2), the gradient calculation uses the Sobel operator and is normalized and Gaussian smoothed to enhance gradient continuity.

[0015] Furthermore, in step (3), adaptive thresholding dynamically calculates the threshold based on the statistical characteristics of the gradient image and optimizes the threshold selection by combining the edge density feedback mechanism.

[0016] Furthermore, in step (4), the morphological operation uses elliptical structuring elements to perform closing and opening operations in sequence to connect broken edges and eliminate noise.

[0017] Furthermore, in step (6), the geometrically longest main branch selection algorithm extracts the longest connected path from the skeleton graph as the main branch skeleton through endpoint detection and path tracing, and reduces the computational complexity through candidate endpoint optimization strategy.

[0018] Furthermore, in step (7), the adaptive curve extension technique calculates the extension direction based on the endpoints and local direction information of the main branch skeleton using the multi-point averaging method, and extends the curve to the image boundary in unit steps.

[0019] Furthermore, step (7) also includes outputting the geometric parameters of the centerline, including the total length, average curvature, endpoint coordinates, and orientation angle.

[0020] The present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the methods described herein.

[0021] An electronic device according to the present invention includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement any of the methods described herein.

[0022] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0023] 1. Enhanced robustness of edge detection: By employing bilateral filtering combined with adaptive thresholding, the algorithm can effectively suppress the inherent speckle noise in SAR images while preserving details of atmospheric front edges. This overcomes the problem of false edges and weak signal loss that traditional edge detection algorithms are prone to in complex noise environments, significantly improving the robustness of the algorithm and the accuracy of edge detection.

[0024] 2. Enhancing Edge Continuity and Integrity: Utilizing the longest geometrically defined main branch selection algorithm, the algorithm extracts the continuous path with the maximum geometric length from the skeleton image as the main branch skeleton, effectively avoiding edge breaks and discontinuities. Simultaneously, through adaptive curve extension technology, the edges are smoothly extended to the image boundary, forming a complete and continuous atmospheric front centerline, ensuring global consistency of the segmentation results.

[0025] 3. Adaptive parameter adjustment mechanism: Based on the statistical characteristics of gradient images, an adaptive threshold is dynamically calculated, and the threshold selection is optimized by combining an edge density feedback mechanism. This enables the algorithm to adapt to different imaging conditions and noisy environments, reducing manual intervention and improving automation and applicability.

[0026] 4. Efficient geometric feature extraction: Through skeletonization and longest geometric path filtering, representative atmospheric frontal branches can be quickly identified from complex edge structures. Combined with spline interpolation and direction estimation, smooth geometric curves are generated, providing high-quality and high-precision geometric parameter support for subsequent frontal dynamics analysis.

[0027] 5. Optimized computational efficiency: When there are a large number of endpoints, a boundary point selection strategy (such as selecting coordinate extreme points as candidate endpoints) is adopted, which improves the running efficiency of the algorithm and is suitable for processing large-scale, high-resolution SAR images.

[0028] 6. Comprehensive morphological optimization: By adopting a morphological processing sequence of closing operations followed by opening operations, combined with elliptical structural elements, edge gaps are effectively filled, broken fragments are connected, and isolated noise is removed, further improving edge connectivity and structural integrity, laying a solid foundation for skeletonization processing. Attached Figure Description

[0029] Figure 1 This is a flowchart of the present invention;

[0030] Figure 2 This is the atmospheric front edge detection process of the present invention;

[0031] Figure 3 This invention describes the process for extracting the edge lines of atmospheric fronts.

[0032] Figure 4 This is the final result diagram of the atmospheric front centerline of the present invention. Detailed Implementation

[0033] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0034] like Figure 1 As shown, this embodiment of the invention provides a method for locating the centerline of an atmospheric front in a synthetic aperture radar image, comprising the following steps:

[0035] Step 1: As Figure 2 As shown in b, the filter diameter is set to 100 pixels, the color space standard deviation to 750, and the coordinate space standard deviation to 100. This parameter combination effectively smooths image noise while preserving edge information well, providing high-quality input data for subsequent edge detection. The mathematical expression for bilateral filtering is:

[0036] The spatial weight function And gray-scale similarity weight function They are defined as follows:

[0037]

[0038]

[0039] Step 2: As Figure 2 As shown in Figure c, the Sobel operator is used to calculate the gradient components in the x and y directions respectively, and then the complete gradient image is obtained by calculating the gradient magnitude. The gradient calculation is as follows:

[0040]

[0041] G represents the gradient magnitude, Gx represents the horizontal gradient component, and Gy represents the vertical gradient component. The gradient image is then normalized.

[0042]

[0043] The gradient image is then transformed to the range of 0-255 and then blurred using a 9×9 Gaussian kernel to enhance the continuity of the gradient.

[0044] Step 3: Calculate a dynamic threshold based on the statistical characteristics of the gradient image, extract all pixels with a gradient value greater than 10, and calculate the mean and standard deviation of these pixels.

[0045]

[0046]

[0047] The formula for calculating the basic threshold is: The threshold coefficient k is dynamically adjusted based on the gradient standard deviation σ.

[0048]

[0049] Threshold adjustment is performed based on the density of edge pixels after binarization:

[0050] ,in ,

[0051] W and H represent the image width and height. This multi-level adjustment mechanism ensures the stability of the binarization effect. Final threshold calculation:

[0052]

[0053] Step 4: As Figure 2 As shown in e, morphological operations are performed using an elliptical structuring element to achieve binarization. The structuring element size is set to 17×17 pixels. First, a morphological closing operation is performed:

[0054]

[0055] in' ' represents erosion, '㊉' represents dilation, and then an opening operation is performed:

[0056]

[0057] Opening operations are used to remove isolated noise points. This sequential morphological operation can effectively improve the connectivity and integrity of edges.

[0058] Step 5: Skeletonize the binary edge image to obtain a skeleton structure with a width of one pixel. Then, perform edge filtering based on connected component analysis: calculate the quartiles and maximum values ​​of the areas of all connected components, and set the filtering threshold to the larger of the 75th quartile and 50% of the maximum value, while ensuring that it is not lower than a preset minimum length threshold. Only retain edge segments with a length exceeding this threshold to form the main edge skeleton.

[0059] Step 6: Identify all endpoints from the skeleton, i.e., pixels with a neighbor count of 1. When the number of endpoints exceeds 8, a boundary point selection strategy is adopted: the leftmost and rightmost endpoints in x-coordinate, and the topmost and bottommost endpoints in y-coordinate are selected as candidate endpoints. By calculating the paths between all candidate endpoint pairs, the path with the longest Euclidean geometric length is selected as the main branch skeleton.

[0060] The path tracing uses a breadth-first search algorithm to find a connected path between two endpoints, and then calculates the cumulative Euclidean distance of the path as a geometric length evaluation criterion.

[0061]

[0062] x i y i Let be the coordinates of the i-th point on the path.

[0063] Step 7: Sort the main branch skeleton points by x-coordinate and smooth the curve using univariate spline interpolation. Create spline functions for both x and y coordinates, dynamically adjusting the smoothing parameters based on the number of points. Generate at least 200 interpolation points to construct the smooth curve, ensuring all curve points are within the valid range of the image.

[0064] Step 8: Based on the leftmost and rightmost endpoints of the original skeleton, and combined with the local direction information of the smooth curve, extend the curve towards the image boundary. The direction calculation uses a multi-point averaging method: the starting direction is calculated using the first three points of the curve, and the ending direction is calculated using the last three points. After normalizing the direction vector, the curve is gradually extended along the direction vector from the original endpoint until it reaches the image boundary.

[0065] During the extension process, it extends in the opposite direction from the leftmost endpoint to the left boundary, and in the forward direction from the rightmost endpoint to the right boundary, eventually forming a complete center line that runs through the entire width of the image.

Claims

1. A method for locating the centerline of an atmospheric front in synthetic aperture radar images, characterized in that, Includes the following steps: (1) Perform bilateral filtering on the original SAR image to achieve noise suppression and edge protection; (2) Calculate the gradient field of the smoothed image and perform normalization and Gaussian smoothing. (3) Adaptive thresholding is performed based on the statistical characteristics of the gradient image to generate a binary edge image; (4) Perform morphological operations on the binary edge image to optimize edge connectivity and integrity; (5) Skeletonize the morphologically optimized binary image and perform edge filtering; (6) Extract the longest connected path from the filtered skeleton graph as the main branch skeleton and perform parametric curve smoothing. (7) Based on the endpoints and local direction information of the main branch skeleton, the adaptive curve extension technique is used to extend the smooth curve to the image boundary to generate a complete atmospheric front centerline.

2. The method for locating the centerline of an atmospheric front in a synthetic aperture radar image according to claim 1, characterized in that, In step (1), the bilateral filtering process adopts a multi-parameter optimization configuration and achieves a smooth edge-preserving effect through the dual weight calculation of spatial proximity and pixel value similarity.

3. The method for locating the centerline of an atmospheric front in a synthetic aperture radar image according to claim 1, characterized in that, In step (2), the gradient calculation uses the Sobel operator and is normalized and Gaussian smoothed to enhance gradient continuity.

4. The method for locating the centerline of an atmospheric front in a synthetic aperture radar image according to claim 1, characterized in that, In step (3), adaptive thresholding dynamically calculates the threshold based on the statistical characteristics of the gradient image and optimizes the threshold selection by combining the edge density feedback mechanism.

5. The method for locating the centerline of an atmospheric front in a synthetic aperture radar image according to claim 1, characterized in that, In step (4), morphological operations use elliptical structuring elements to perform closing and opening operations in sequence to connect broken edges and eliminate noise.

6. The method for locating the centerline of an atmospheric front in a synthetic aperture radar image according to claim 1, characterized in that, In step (6), the longest geometric main branch selection algorithm extracts the longest connected path in the skeleton graph as the main branch skeleton through endpoint detection and path tracing, and reduces the computational complexity through candidate endpoint optimization strategy.

7. The method for locating the centerline of an atmospheric front in a synthetic aperture radar image according to claim 1, characterized in that, In step (7), the adaptive curve extension technique calculates the extension direction based on the endpoints and local direction information of the main branch skeleton, and extends the curve to the image boundary in unit steps.

8. The method for locating the centerline of an atmospheric front in a synthetic aperture radar image according to claim 1, characterized in that, Step (7) also includes outputting the geometric parameters of the centerline, including the total length, average curvature, endpoint coordinates and orientation angle.

9. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1-7.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor executes the program to implement the method as described in any one of claims 1-7.

Images