Radar image superpixel segmentation method, system and program based on polarization decomposition

Abstract

The invention provides a radar image superpixel segmentation method, system and program based on polarization decomposition, and the method comprises the steps: carrying out the RS scattering power decomposition of a polarization coherence matrix corresponding to an input polarization synthetic aperture radar image, and obtaining the surface scattering power, the secondary scattering power, the volume scattering power and the directed dipole power; rGB color coding is carried out on the four kinds of power, a color coding image is obtained, and the coding mode is that a red channel represents the power sum of secondary scattering power and directed dipole power, a green channel represents volume scattering power, and a blue channel represents surface scattering power; and applying a linear iterative clustering SLIC algorithm to the color coding image, and performing super-pixel segmentation according to a set super-pixel number to obtain a super-pixel segmentation result with a regular shape and fitted edges.

Description

Technical Field

[0001] This invention relates to the fields of remote sensing technology and computer vision technology, specifically to a method, system, and program for superpixel segmentation of radar images based on polarization decomposition. Background Technology

[0002] Polarimetric Synthetic Aperture Radar (PolSAR) employs multiple electromagnetic wave polarization transmission and reception modes, preserving more complete target electromagnetic scattering characteristics and acquiring richer target information. It has become an important tool in modern radar imaging technology and is currently widely used in military reconnaissance, topographic mapping, and maritime vessel detection, among other fields. Maritime vessel target detection is a significant application of PolSAR systems. Unfortunately, in high sea states and complex sea clutter scenarios, small-sized vessel targets are easily submerged in background clutter, making PolSAR vessel target detection difficult.

[0003] Currently, many superpixel segmentation algorithms based on simple linear iteration have been proposed for PolSAR and optical images, and are widely used in image classification, target detection, and tracking. However, when these algorithms are directly applied to PolSAR images in high sea states and with complex sea clutter, satisfactory segmentation results still cannot be obtained. Therefore, there is an urgent need to propose a novel superpixel segmentation algorithm suitable for images in high sea states and with complex sea clutter to accurately and efficiently distinguish small-sized ship targets and sea clutter. Summary of the Invention

[0004] This invention proposes a PolSAR image superpixel segmentation method based on polarization decomposition to solve the technical problem that existing superpixel segmentation methods are inaccurate when ships are present due to the scattering characteristics of ship structures.

[0005] To address the aforementioned technical problems, this invention provides a radar image superpixel segmentation method based on polarization decomposition, comprising the following steps:

[0006] Step S1: Perform RS scattering power decomposition on the polarization coherence matrix corresponding to the input polarization synthetic aperture radar image to obtain surface scattering power, secondary scattering power, volume scattering power and directed dipole power;

[0007] Step S2: Encode the four powers using RGB color to obtain a color-coded image. The encoding method is as follows: the red channel represents the sum of the secondary scattering power and the directed dipole power, the green channel represents the volume scattering power, and the blue channel represents the surface scattering power.

[0008] Step S3: Apply the linear iterative clustering (SLIC) algorithm to the color-coded image to perform superpixel segmentation according to the set number of superpixels, and obtain superpixel segmentation results with regular shapes and matching edges.

[0009] Preferably, in step S3, peak detection is performed on the color component histogram of the color-coded image to obtain the number of peaks, and the number of superpixels is set to be the product of the number of peaks and the set shrinkage ratio.

[0010] Preferably, a weighted distance function is used to replace the Euclidean distance function in the linear iterative clustering SLIC algorithm in step S3.

[0011] Preferably, the weighted distance function The expression is:

[0012] ;

[0013] ;

[0014] ;

[0015] ;

[0016] In the formula, This represents the closest distance between the center points of two adjacent superpixels; Indicates the compactness coefficient; Indicates the first The coordinates of each pixel; Indicates the first The coordinates of the cluster centers; Represents pixels The coherence matrix; Indicates the first The coherence matrix of the cluster centers; Indicates the search for the trace; Indicates the number of pixels; This indicates the set number of superpixels.

[0017] Preferably, edge recall rate and undersegmentation error are used as evaluation indicators for superpixel edge attachment capability.

[0018] Preferably, the edge recall rate The expression is:

[0019] ;

[0020] In the formula, This indicates the number of superpixel edge pixels that overlap with the real edge; This represents the number of pixels at the actual edge.

[0021] Preferably, the undersegmentation error The expression is:

[0022] ;

[0023] In the formula, Indicates the number of pixels; Represents a set of superpixels; Represents the true set of partitions.

[0024] This invention also provides a PolSAR image superpixel segmentation system based on polarization decomposition, applicable to the above-mentioned methods, comprising:

[0025] The decomposition module is used to perform RS scattering power decomposition on the polarization coherence matrix corresponding to the input polarization synthetic aperture radar image to obtain surface scattering power, secondary scattering power, volume scattering power and directed dipole power.

[0026] The encoding module is used to perform RGB color encoding on the scattering power to generate a color-coded image, where the red channel represents the sum of the secondary scattering power and the directed dipole power, the green channel represents the volume scattering power, and the blue channel represents the surface scattering power.

[0027] The segmentation module is used to perform superpixel segmentation on the color-coded image based on a linear iterative clustering algorithm to generate a segmentation result with edge-fitting.

[0028] Preferably, the system further includes a distance metric module, used to replace the Euclidean distance function in the linear iterative clustering algorithm with a weighted distance function during the linear iterative clustering process.

[0029] The present invention also provides a computer program comprising instructions for performing the method described above, wherein the computer program, when executed by a computing device, is used to implement a radar image superpixel segmentation method based on polarization decomposition.

[0030] The beneficial effects of the present invention include at least the following: The present invention uses the RS scattering power decomposition method to decompose the polarization coherence matrix corresponding to radar high sea state and complex sea clutter images, and uses RGB color encoding to synthesize color images from the decomposed scattering power, thereby fully exploring the differences in polarization scattering characteristics between the target and the environment.

[0031] By analyzing the scattering characteristics of ship structures, it is found that there are many upright metal components on the ship deck, similar to a steel frame structure. These components and the deck form a complex scattering structure. In this invention, these complex scatterings are represented by a directional dipole component scattering matrix, which has obvious advantages for the identification of ship targets. It is particularly suitable for accurately and efficiently distinguishing small-sized ship targets and sea clutter in high sea states and complex sea clutter scenarios.

[0032] Subsequently, the RS scattering power decomposition was combined with the SLIC algorithm to perform superpixel segmentation on PolSAR images of high sea state and complex sea clutter, which greatly improved the similarity representation capability and generated superpixels with regular shapes, smooth edges, and good fit. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention;

[0034] Figure 2 This is a PolSAR image of a certain sea area input for an embodiment of the present invention;

[0035] Figure 3 A schematic diagram of a color-coded image commonly used in the industry;

[0036] Figure 4 This is a schematic diagram of a color-coded image generated according to an embodiment of the present invention;

[0037] Figure 5 This is a schematic diagram comparing the edge recall rates of various algorithms in embodiments of the present invention;

[0038] Figure 6 This is a schematic diagram comparing the undersegmentation errors of various algorithms in embodiments of the present invention;

[0039] Figure 7 This is a schematic diagram of the superpixel segmentation result structure according to an embodiment of the present invention;

[0040] Figure 8 This is a schematic diagram of the segmentation results using the SLIC algorithm based on RS scattering power decomposition and universal color coding.

[0041] Figure 9 This is a schematic diagram of the segmentation results using the SLIC algorithm based on Y4R scattering power decomposition and universal color coding.

[0042] Figure 10 This is a schematic diagram of the segmentation results based on the SLIC algorithm and universal color coding using G4U scattering power decomposition.

[0043] Figure 11 This is a schematic diagram of the segmentation results of the original SLIC algorithm and general color coding. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0045] Example 1

[0046] This invention provides a radar image superpixel segmentation method based on polarization decomposition, comprising the following steps:

[0047] Step S1: Perform RS scattering power decomposition on the polarization coherence matrix corresponding to the input polarization synthetic aperture radar image to obtain surface scattering power, secondary scattering power, volume scattering power and directed dipole power.

[0048] The RS scattering power decomposition method is easy to implement, computationally simple and fast, and highly applicable. By interpreting the physical scattering mechanism through color, it generates clear, vivid, and aesthetically pleasing RGB color-coded images. In this embodiment, RS scattering power decomposition includes, but is not limited to, decomposition methods based on physical models, decomposition methods based on mathematical models, hybrid decomposition methods, and other convolutional neural network-assisted scattering decomposition methods.

[0049] For example, the scattering power decomposition method of this embodiment is shown below.

[0050] The calculation method is as follows: Input the polarization coherence matrix corresponding to the PolSAR high sea state and complex sea clutter images. , and then decompose it.

[0051] ;

[0052] in Represents a surface scattering substrate; The shape parameter for surface scattering; This represents the surface scattering power.

[0053] Represents a secondary scattering substrate; The shape parameter for secondary scattering; This represents the secondary scattering power.

[0054] Represents a directional dipole scattering substrate; It is the power of a directional dipole.

[0055] Representative scattering substrate, according to and The power ratio is selected using a volume scattering model, with the boundary set to ±2dB according to industry standards. When hour, ;when hour, ;when hour, Re denotes taking the real part; For volume scattering power.

[0056] Surface scattering power can be calculated using RS scattering power decomposition. Secondary scattering power Directed dipole power Volume scattering power .

[0057] Step S2: Encode the four powers using RGB color to obtain a color-coded image. The encoding method is as follows: the red channel represents the sum of the secondary scattering power and the directed dipole power, the green channel represents the volume scattering power, and the blue channel represents the surface scattering power.

[0058] Figure 2 This is a PolSAR image of a certain sea area. In high sea states, small ship targets are submerged in background clutter. Figure 3 For general color-coded images, Figure 4 This is the color-coded image of the present invention, where red represents secondary scattering power. With directed dipole power The power and the green color represent the volume scattering power. Blue indicates surface scattering power. .

[0059] from Figure 3 As can be seen, in the color-coded image of this invention, the ship portion appears pink, which is more consistent with the actual scattering mechanism. Because there are many upright metal components on the ship's deck, similar to a steel frame structure, these components and the deck form a complex scattering structure. These complex scatterings are represented by a directional dipole component scattering matrix. This color coding is particularly suitable for ship target image recognition under PolSAR conditions of high sea states and complex sea clutter.

[0060] Step S3: Apply the linear iterative clustering (SLIC) algorithm to the color-coded image to perform superpixel segmentation according to the set number of superpixels, and obtain superpixel segmentation results with regular shapes and matching edges.

[0061] The process of generating superpixels using the SLIC algorithm is as follows:

[0062] Assuming the image has There are 1 pixel, and the number of superpixels is 1. The average size of each superpixel is The nearest distance between the centers of two adjacent superpixels is represented as... , With step size as Initialize the cluster centers in the grid; then at each cluster center of Within the neighborhood, the similarity between each pixel in the image and the center of its nearest superpixel is calculated. The label of the most similar superpixel center is assigned to the pixel. This process is iterated until convergence, and finally each pixel is assigned the best matching class label.

[0063] The original SLIC algorithm uses Euclidean distance in the CIELAB color space as the distance metric, as shown below:

[0064] (1)

[0065] (2)

[0066] (3)

[0067] in, It is the spatial distance between pixels and cluster centers, subscript and Representing the first The pixel and the Cluster centers; It refers to the color distance in the CIELAB color space. , , The CIELAB color space value of the pixel; It's the tightness coefficient. The larger the size, the greater the proportion of spatial distance, and the more compact the resulting superpixel will be, which is likely to be a rectangle or a regular hexagon; This indicates the initial grid side length.

[0068] Example 2

[0069] The SLIC algorithm was proposed for optical images. However, since the PolSAR images used in this embodiment have speckle noise, the superpixel segmentation effect is poor when SLIC is directly applied to PolSAR images.

[0070] Therefore, this embodiment improves the SLIC algorithm based on Embodiment 1 and proposes a new distance metric, defined as follows:

[0071] pixel The coherence matrix is , No. The coherence matrix of the cluster centers is , pixel and clustering The distance between them is:

[0072] (4)

[0073] in, This indicates finding the trace. Because... and All A complex matrix, therefore Too Complex matrix, This represents the total number of cluster centers.

[0074] For calculation The traces, only need to be obtained The diagonal elements of the complex matrix. , to complex matrix Expanding by rows into a column vector, that is:

[0075] ;

[0076] Then the complex matrix Expand and rearrange the columns into a single row vector, i.e.:

[0077] ;

[0078] therefore, Represented as:

[0079] (5)

[0080] Substituting equation (5) into equation (4), we get:

[0081] (6)

[0082] The weighted distance used in the improved SLIC algorithm of this invention is ultimately expressed as:

[0083] (7)

[0084] in subscript and Representing the first The pixel and the Cluster centers; It is the tightness coefficient; , The total number of pixels in the image. Indicates the number of superpixels.

[0085] This invention increases the number of superpixels Set to a multiple of the number of peaks in the histogram of color components of an RGB color-coded image, that is:

[0086] ;

[0087] in Shrinkage ratio, The number of peaks in the histogram of color components of an RGB color-coded image.

[0088] Combining RS scattering power decomposition with the improved SLIC algorithm for superpixel segmentation of PolSAR high sea state and complex sea clutter images greatly improves the similarity representation capability. The generated superpixel shapes are more regular and the edges are smoother and more closely aligned, which solves the problem that existing superpixel segmentation algorithms based on simple linear iteration cannot perform good superpixel segmentation of PolSAR high sea state and complex sea clutter images.

[0089] Furthermore, this invention determines the number of superpixels by calculating the number of peaks in the histogram of color components of an RGB color-coded image, thus overcoming the limitation that the number of superpixels in the existing simple linear iterative clustering SLIC algorithm needs to be manually set multiple times.

[0090] Example 3

[0091] This invention selects edge recall (BR) and undersegmentation error (UE) as evaluation indicators for superpixel edge attachment capability.

[0092] make , This represents the set of superpixel blocks generated from an RGB color-coded image. Indicates the first Superpixel blocks Indicates the number of superpixels. As a reference standard, representing the true set of superpixels, Indicates the first One superpixel.

[0093] Marginal recall (BR) is defined as:

[0094] ;

[0095] This indicates the number of superpixel edge pixels that overlap with the real edge; This represents the number of pixels at the true edge. A higher edge recall rate means fewer true edges are lost, resulting in better segmentation.

[0096] The undersegmentation error (UE) is defined as:

[0097] ;

[0098] This formula represents the ratio of the number of pixels outside the intersection of the superpixel and the ground truth segment (G) to the number of pixels in the ground truth segment. For any superpixel... The segmentation region with the largest overlapping area is denoted as , This represents the total number of pixels in the image. The smaller the undersegmentation error, the more accurate the segmentation result.

[0099] The improved algorithm of Example 2 was verified using the evaluation function described above. The edge recall rate and undersegmentation error of the improved SLIC algorithm based on RS scattering power decomposition, the original simple linear iterative clustering SLIC algorithm, the SLIC algorithm based on Y4R scattering power decomposition, and the SLIC algorithm based on G4U scattering power decomposition were calculated respectively. Performance curves for each algorithm were plotted. Figure 5 and Figure 6 As shown.

[0100] Depend on Figure 5 It can be seen that the edge recall rate of various algorithms increases with the number of superpixels, and the edge recall rate of each algorithm is the highest when the number of superpixels is 531. Compared with other algorithms, the edge recall rate of the method of this invention is improved from 0.706 for SLIC, 0.781 for G4U-SLIC, and 0.798 for Y4R-SLIC to 0.915.

[0101] Depend on Figure 6 It can be seen that the undersegmentation error of various algorithms decreases with the increase of the number of superpixels. When the number of superpixels is 531, the undersegmentation error of each algorithm is the smallest. Compared with other algorithms, the undersegmentation error of the algorithm of this invention is reduced from 0.283 for SLIC, 0.212 for G4U-SLIC, and 0.2 for Y4R-SLIC to 0.082.

[0102] The novel superpixel segmentation algorithm of this invention has significant effects and obvious performance advantages. It is particularly suitable for accurately and efficiently distinguishing small-sized ship targets and sea clutter in high sea states and complex sea clutter scenarios.

[0103] Figure 7 The superpixel segmentation result of the superpixel segmentation algorithm of the present invention is generated by combining RS scattering power decomposition, the color encoding of the present invention and the improved SLIC algorithm. Figure 8 This is a superpixel generated by combining RS scattering power decomposition, industry-standard color coding, and the original SLIC algorithm. Figure 9 This is a superpixel generated by combining Y4R scattering power decomposition, industry-standard color coding, and the original SLIC algorithm. Figure 10 This is a superpixel generated by combining G4U scattering power decomposition, industry-standard color coding, and the original SLIC algorithm. Figure 11 This is the superpixel segmentation result obtained by the original simple linear iterative clustering (SLIC) algorithm.

[0104] from Figures 7 to 11The experimental results clearly show that the superpixel segmentation effect of combining polarization decomposition with the SLIC algorithm is significantly better than that of the original SLIC algorithm. Furthermore, compared with SLIC algorithms based on Y4R scattering power decomposition and G4U scattering power decomposition, the SLIC algorithm based on RS scattering power decomposition can perform superpixel segmentation while preserving more details. Compared with other methods, the superpixel edges obtained by combining RS scattering power decomposition, the color coding of this invention, and the improved SLIC algorithm are smoother and more closely aligned. The superpixel segmentation algorithm of this invention can not only accurately distinguish small-sized ship targets and sea clutter but also better conforms to the actual scattering mechanism.

[0105] This invention also provides a PolSAR image superpixel segmentation system based on polarization decomposition, applicable to the above-mentioned methods, comprising:

[0106] The decomposition module is used to perform RS scattering power decomposition on the polarization coherence matrix corresponding to the input polarization synthetic aperture radar image to obtain surface scattering power, secondary scattering power, volume scattering power and directed dipole power.

[0107] The encoding module is used to perform RGB color encoding on the scattering power to generate a color-coded image, where the red channel represents the sum of the secondary scattering power and the directed dipole power, the green channel represents the volume scattering power, and the blue channel represents the surface scattering power.

[0108] The segmentation module is used to perform superpixel segmentation on color-coded images based on a linear iterative clustering algorithm, generating segmentation results with edge-fitting.

[0109] The system also includes a distance metric module, which uses a weighted distance function to replace the Euclidean distance function in the linear iterative clustering algorithm during the linear iterative clustering process.

[0110] The present invention also provides a computer program including instructions for performing the method described above. When executed by a computing device, the computer program is used to implement a PolSAR image superpixel segmentation method based on polarization decomposition.

[0111] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described; only preferred embodiments of the present invention are illustrated. The descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. As long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0112] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this invention should be determined by the appended claims.

Claims

1. A radar image superpixel segmentation method based on polarization decomposition, characterized in that: Includes the following steps: Step S1: Perform RS scattering power decomposition on the polarization coherence matrix corresponding to the input polarization synthetic aperture radar image to obtain surface scattering power, secondary scattering power, volume scattering power and directed dipole power; Step S2: Encode the four powers using RGB color to obtain a color-coded image. The encoding method is as follows: the red channel represents the sum of the secondary scattering power and the directed dipole power, the green channel represents the volume scattering power, and the blue channel represents the surface scattering power. Step S3: Apply the linear iterative clustering (SLIC) algorithm to the color-coded image to perform superpixel segmentation according to the set number of superpixels, and obtain superpixel segmentation results with regular shapes and matching edges.

2. The radar image superpixel segmentation method based on polarization decomposition according to claim 1, characterized in that: In step S3, peak detection is performed on the color component histogram of the color-coded image to obtain the number of peaks, and the number of superpixels is set to be the product of the number of peaks and the set shrinkage ratio.

3. The radar image superpixel segmentation method based on polarization decomposition according to claim 1, characterized in that: The Euclidean distance function in the linear iterative clustering SLIC algorithm described in step S3 is replaced with a weighted distance function.

4. The radar image superpixel segmentation method based on polarization decomposition according to claim 3, characterized in that: The weighted distance function The expression is: ； ； ； ； In the formula, This represents the closest distance between the center points of two adjacent superpixels; Indicates the compactness coefficient; Indicates the first The coordinates of each pixel; Indicates the first The coordinates of the cluster centers; Represents pixels The coherence matrix; Indicates the first The coherence matrix of the cluster centers; Indicates the search for the trace; Indicates the number of pixels; This indicates the set number of superpixels.

5. The radar image superpixel segmentation method based on polarization decomposition according to claim 1, characterized in that: Edge recall rate and undersegmentation error are used as evaluation metrics for superpixel edge attachment capability.

6. The radar image superpixel segmentation method based on polarization decomposition according to claim 5, characterized in that: The edge recall rate The expression is: ； In the formula, This indicates the number of superpixel edge pixels that overlap with the real edge; This represents the number of pixels at the actual edge.

7. The radar image superpixel segmentation method based on polarization decomposition according to claim 5, characterized in that: The undersegmentation error The expression is: ； In the formula, Indicates the number of pixels; Represents a set of superpixels; Represents the true set of partitions.

8. A radar image superpixel segmentation system based on polarization decomposition, applicable to the method described in any one of claims 1 to 7, characterized in that, include: The decomposition module is used to perform RS scattering power decomposition on the polarization coherence matrix corresponding to the input polarization synthetic aperture radar image to obtain surface scattering power, secondary scattering power, volume scattering power and directed dipole power. The encoding module is used to perform RGB color encoding on the scattering power to generate a color-coded image, where the red channel represents the sum of the secondary scattering power and the directed dipole power, the green channel represents the volume scattering power, and the blue channel represents the surface scattering power. The segmentation module is used to perform superpixel segmentation on the color-coded image based on a linear iterative clustering algorithm to generate a segmentation result with edge-fitting.

9. A radar image superpixel segmentation system based on polarization decomposition according to claim 8, characterized in that: The system also includes a distance metric module, which uses a weighted distance function to replace the Euclidean distance function in the linear iterative clustering algorithm during the linear iterative clustering process.

10. A computer program, characterized in that: The program includes instructions for performing the method as described in any one of claims 1 to 7, wherein the computer program, when executed by a computing device, is used to implement a polarization decomposition-based radar image superpixel segmentation method.

Images