High-precision video SAR (Synthetic Aperture Radar) moving target detection method based on information geometry theory

Abstract

The invention discloses a high-precision video SAR (Synthetic Aperture Radar) moving target detection method based on an information geometry theory in the technical field of SAR moving target shadow detection. The high-precision video SAR moving target detection method based on the information geometry theory comprises the following steps of performing shadow enhancement and background suppression on a video SAR image sequence; rOI region extraction based on an information geometry theory; and through error detection suppression, a false ROI region is effectively removed. According to the high-precision video SAR moving target detection method based on the information geometry theory, on the premise of ensuring the detection precision, the detection probability of video SAR moving target shadow is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of SAR moving target shadow detection, and in particular to a high-precision video SAR moving target detection method based on information geometry theory. Background Art

[0002] Video SAR has a wide range of applications in military decision-making, smart cities, traffic control, and other fields due to its ability to continuously monitor changes in the target area. In video SAR systems, due to the high carrier frequency of the radar, the Doppler modulation of the echo of the moving target is very sensitive to the movement of the target. Even a small movement will cause a large offset and defocusing of the moving target imaging in the SAR image, but the shadow left by the moving target can reflect the actual position and status information of the moving target in the SAR image sequence. Therefore, compared with traditional SAR systems, in video SAR image sequences, the shadow of the moving target is easier to observe, and using the shadow of the moving target for detection is a more direct and effective means.

[0003] At present, there are three main types of methods in the traditional video SAR moving target shadow detection technology: moving target detection method based on difference, constant false alarm rate (CFAR) detector and threshold segmentation.

[0004] The difference-based detection method first estimates the background image through multiple frames of continuous video SAR images, then performs a difference operation on the background image and the current image to be detected to obtain a binary image of the foreground, and then uses the inter-frame difference algorithm to extract the motion information of the previous and next frames and perform an AND operation on the binary image obtained by the background difference, and finally performs morphological processing on the result of the previous step to extract the shadow area of ​​the moving target. Although the difference-based detection method is effective, there are some limitations: first, it has high requirements for the background modeling algorithm, and the background is estimated by using several previous frames, so the detection frame will lag behind; second, because the inter-frame difference algorithm only compares the difference between the two frames, it ignores the information of the overlapping part, so it is easy to have holes and double-sided phenomena in the shadow area of ​​the moving target, resulting in missed detection or false alarm.

[0005] The detection method based on CFAR can effectively avoid the above limitations. CFAR can select a suitable area in the background noise of the video SAR image and determine an appropriate detection threshold based on the statistical characteristics of the noise in this area. When the target area exceeds this threshold, the target is considered to be detected. However, in a multi-target environment, the reference area may contain multiple target signals, which will reduce the detection performance of the method; at the same time, in a non-uniform noise environment, the performance of CFAR will also be affected.

[0006] The detection method based on threshold segmentation can binarize the video SAR image using the image threshold segmentation method (such as OTSU method, minimum error method, etc.) to obtain the shadow area of ​​the moving target in the image. However, the threshold segmentation method is based on the histogram information of the image for segmentation, ignoring the spatial information and edge information of the video SAR image, so it is easy to have problems of missed detection and too high false alarm. Summary of the invention

[0007] In order to solve the above problems, the purpose of the present invention is to provide a high-precision video SAR moving target detection method based on information geometry theory, to solve the technical problems of the shortcomings of the traditional video SAR moving target shadow detection method, and to improve the detection probability of the video SAR moving target shadow while ensuring the detection accuracy.

[0008] In order to achieve the above object, the technical solution of the present invention is as follows:

[0009] The present invention provides a high-precision video SAR moving target detection method based on information geometry theory, comprising the following steps:

[0010] S1. Perform shadow enhancement and background suppression on video SAR image sequences;

[0011] S2. ROI region extraction based on information geometry theory;

[0012] S3. Effectively remove false ROI areas through false detection suppression.

[0013] Furthermore, the S1 comprises the following steps:

[0014] S11. Calculate the initial subspace {U, V} of the video SAR image sequence using the principal component analysis algorithm;

[0015] S12. Set the number of mixed Gaussians, and model the data matrix X composed of the video SAR image sequence based on a mixed Gaussian low-rank representation model, that is:

[0016]

[0017] in, represents the set of all Gaussian component weights, represents the set of variances of all Gaussian components;

[0018] S13. Update the low-rank matrix U*V using online subspace learning technology T , and subtract it from the original video SAR image to obtain a foreground image containing the shadow of the video SAR moving target;

[0019] S14. Using the alternating direction multiplier algorithm to remove the strong scattered object imaging contained in the foreground image, a video SAR image sequence with shadow enhancement and background suppression, namely, an enhanced image, is obtained.

[0020] Further, the S2 comprises the following steps:

[0021] S21. All pixels in the neighborhood of the coordinate pixel of the enhanced image form a neighborhood system, and construct a statistical distribution describing the point;

[0022] S22. Calculate the JS divergence to describe the geometric distance in the Gaussian statistical manifold to analyze the pattern and capture the local information features in the image;

[0023] S23. Traverse each pixel of the image, obtain the JS divergence matrix corresponding to the enhanced image, and calculate the joint probability distribution JPDF;

[0024] S24. Use the minimum cross entropy algorithm to select the threshold and extract the ROI area of ​​the enhanced image.

[0025] Further, the S21 includes the following steps:

[0026] S211. All pixels in the N*N neighborhood of the enhanced image at the (x, y) pixel form a neighborhood system, calculate the mean μ and variance σ of all pixels in the neighborhood system, and use Gaussian distribution to describe the statistical distribution of the neighborhood system:

[0027]

[0028] S212. Taking the gray intensity value at the current (x, y) pixel as the mean μ′ and a very small variance σ′, construct a statistical distribution describing the point:

[0029]

[0030] Further, the S22 comprises the following steps:

[0031] S221. By parameter θ=(μ,σ) T The probability distribution family formed constitutes a statistical model S = {P(x|θ)|θ∈Θ}, and S forms a differentiable Gaussian statistical manifold under a certain topological structure;

[0032] S222. Calculate JS divergence, use JS divergence to describe the geometric distance between different probability distributions in the Gaussian statistical manifold, and perform pattern analysis;

[0033] S223. Normalize the numerical value of JS divergence. On the Gaussian statistical manifold, JS divergence explains the geometric relationship between the pixel grayscale value at the current position and the distribution of its neighborhood system, helping to capture the local information features in the image.

[0034] Furthermore, the JS divergence calculation formula in S222 is as follows:

[0035]

[0036] in, is the average distribution of the two distributions, and KL(pq) is defined as:

[0037]

[0038] Further, the S23 comprises the following steps:

[0039] S231. Traverse each pixel of the image to obtain the JS divergence matrix corresponding to the enhanced image;

[0040] S232. Let n ij is the number of joint occurrences between grayscale value and JS divergence, and n ij Each element in is divided by the total number of pixels in the enhanced image to obtain the normalized joint probability distribution JPDF.

[0041] Further, the S3 comprises the following steps:

[0042] S31. Calculate the gradient image of the current frame, and use the gradient image to remove the erroneous ROI area;

[0043] S32. Use data association algorithm to suppress false alarms of multiple frames.

[0044] Further, the S31 includes the following steps:

[0045] S311. Use the Sobel operator to detect the edge of the image, calculate the relative change of the gray level of the original video SAR image, and obtain the gradient image of the current original SAR image;

[0046] S312. Import the gradient image, calculate the overlap area between the gradient image and the ROI region, and remove the ROI region whose overlap area is greater than a threshold.

[0047] Further, the S32 comprises the following steps:

[0048] S321. Extract the centroid position of each ROI region and the corresponding time frame number as a feature vector;

[0049] S322. Set a spatial distance threshold T Iand time threshold T T ,Through these two thresholds, the detection points within the time threshold frame are associated to identify possible real targets;

[0050] S323. Detection points with strong consistency in multiple frames are considered to be valid targets, and only detection points that meet a certain number of associations and spatial changes in the center of mass can be retained.

[0051] By adopting the above technical solution, the present invention has the following advantages:

[0052] 1. The present invention provides a high-precision video SAR moving target detection method based on information geometry theory. By performing shadow enhancement and background suppression operations on video SAR image sequences, the representation of moving target shadows is enhanced, while strong scattered objects and low scattering areas in the background are effectively suppressed.

[0053] 2. The present invention provides a high-precision video SAR moving target detection method based on information geometry theory. The geometric distance between the current pixel position and its neighborhood system distribution is calculated based on information geometry theory, revealing the changes in local structure, and forming a joint probability distribution with the gray value distribution. The ROI area is extracted using the minimum cross entropy, greatly improving the detection performance of the moving target shadow.

[0054] 3. The present invention provides a high-precision video SAR moving target detection method based on information geometry theory, which utilizes edge information and dynamic information between multiple frames to effectively remove false ROI areas and improve the detection accuracy of moving target shadows. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a flow chart of the high-precision video SAR moving target detection method based on information geometry theory of the present invention;

[0056] Figure 2 It is a flow chart of ROI region extraction of the present invention;

[0057] FIG3( a ) is an original video SAR image of the present invention;

[0058] FIG3( b ) is an enhanced image of the present invention;

[0059] Figure 4 The extracted ROI region of the present invention;

[0060] Figure 5 It is the detection result of the moving target of each frame of the present invention. DETAILED DESCRIPTION

[0061] The technical solution of the present invention is described in detail below in conjunction with the drawings in the specification. It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0062] Figure 1 The flowchart of the high-precision video SAR moving target detection method based on information geometry theory is shown; the high-precision video SAR moving target detection method based on information geometry theory can be divided into three parts as a whole: including the following steps: Figure 1 As shown:

[0063] S1. Perform shadow enhancement and background suppression on video SAR image sequences;

[0064] Among them, S1 includes the following specific steps:

[0065] S11. Calculate the initial subspace {U, V} of the video SAR image sequence using the principal component analysis algorithm;

[0066] S12. Set the number of mixed Gaussians, and model the data matrix X composed of the video SAR image sequence based on a mixed Gaussian low-rank representation model, that is:

[0067]

[0068] in, represents the set of all Gaussian component weights, represents the set of variances of all Gaussian components;

[0069] S13. Update the low-rank matrix U*V using online subspace learning technology T , and subtract it from the original video SAR image to obtain a foreground image containing the shadow of the video SAR moving target;

[0070] S14. Using the alternating direction multiplier algorithm to remove the strong scattered object imaging contained in the foreground image, a video SAR image sequence with shadow enhancement and background suppression is obtained, which is referred to as enhanced image.

[0071] S2. ROI region extraction based on information geometry theory;

[0072] Among them, S2 includes the following steps: Figure 2As shown:

[0073] S21. All pixels in the neighborhood of the coordinate pixel of the enhanced image form a neighborhood system, and construct a statistical distribution describing the point;

[0074] Wherein, S21 includes the following specific steps:

[0075] S211. All pixels in the N*N neighborhood of the enhanced image at the (x, y) pixel form a neighborhood system, calculate the mean μ and variance σ of all pixels in the neighborhood system, and use Gaussian distribution to describe the statistical distribution of the neighborhood system:

[0076]

[0077] S212. Taking the gray intensity value at the current (x, y) pixel as the mean μ′ and a very small variance σ′, construct a statistical distribution describing the point:

[0078]

[0079] S22. Calculate the JS divergence to describe the geometric distance in the Gaussian statistical manifold to analyze the pattern and capture the local information features in the image;

[0080] Wherein, S22 includes the following specific steps:

[0081] S221. By parameter θ=(μ,σ) T The probability distribution family formed can form a statistical model S = {P(x|θ)|θ∈Θ}. S forms a differentiable Gaussian statistical manifold under a certain topological structure. Compared with Euclidean space and Euclidean distance metric, the statistical manifold and the metric based on geodesic distance on it can better reflect the geometric position and similarity of each point.

[0082] S222. Jensen–Shannon divergence (JS divergence) is used to describe the geometric distance between different probability distributions in the Gaussian statistical manifold for pattern analysis.

[0083]

[0084] in, is the average distribution of the two distributions, and KL(p||q) is the Kullback-Leibler divergence, which is defined as:

[0085]

[0086] S223. The JS divergence value is normalized to 0-255. On the Gaussian statistical manifold, the JS divergence provides a non-Euclidean "distance" measure. Its performance on the manifold can explain the geometric relationship between the pixel grayscale value at the current position and the distribution of its neighborhood system, reveal changes in local structure, and help capture local information features in the image, which are often "target areas" different from the background.

[0087] S23. Traverse each pixel of the image, obtain the JS divergence matrix corresponding to the enhanced image, and calculate the joint probability distribution JPDF;

[0088] S23 includes the following specific steps:

[0089] S231. Traverse each pixel of the image to obtain the JS divergence matrix corresponding to the enhanced image;

[0090] S232. Let n ij is the number of joint occurrences between grayscale value and JS divergence, and n ij Each element in is divided by the total number of pixels in the enhanced image to obtain the normalized joint probability distribution JPDF.

[0091] S24. Based on the obtained joint probability distribution JPDF, the minimum cross entropy algorithm is used to calculate JPDF = {jpdf ij ; i, j = 0, 1, ..., L-1} the best gray value threshold T G With JS divergence threshold T JS , extract the ROI area of ​​the enhanced image.

[0092] S3. Effectively remove false ROI areas through false detection suppression (assisted by road edge information and inter-frame correlation).

[0093] S3 includes the following specific steps:

[0094] S31. Calculate the gradient image of the current frame, and use the gradient image to remove the erroneous ROI area;

[0095] Wherein, S31 includes the following specific steps:

[0096] S311. The ROI region extracted from the ROI region does not contain the erroneous ROI region of the moving target shadow, and most of these erroneous ROI regions are distributed in the edge region of the image. The Sobel operator is used to detect the edge of the image, and the relative change of the gray level of the original video SAR image is calculated to obtain the gradient image of the current original SAR image.

[0097] S312. Introduce the gradient image, calculate the overlap area between the gradient image and the ROI area, andA The ROI area is removed.

[0098] S32. Most of the erroneous ROI areas at the edges have been eliminated, but a small part still exists. Therefore, the data association algorithm is used to suppress false alarms in multiple frames.

[0099] Wherein, S32 includes the following specific steps:

[0100] S321. Extract the centroid position of each ROI region and the corresponding time frame number as a feature vector;

[0101] S322. Set a spatial distance threshold T I and time threshold T T ,Through these two thresholds, the detection points within the time threshold frame are associated to identify possible real targets;

[0102] S323. Detection points with strong consistency in multiple frames are considered to be valid targets, and only detection points that meet a certain number of associations and spatial changes in the center of mass can be retained.

[0103] Finally, through the above three parts, high-precision moving target detection of video SAR can be achieved.

[0104] In a specific embodiment, the implementation steps are as follows:

[0105] S1. A low-rank representation model based on mixed Gaussian is used to perform shadow enhancement and background suppression on the original video SAR image to obtain an enhanced image sequence. In this example, the 56th frame video SAR image and the enhanced image are selected as examples as shown in Figure 3(a) and Figure 3(b). From the comparison between Figure 3(a) and Figure 3(b), it can be seen that the shadow of the moving target in the enhanced image is more obvious and has a higher contrast with its surroundings. At the same time, the background of the original video SAR image is suppressed, so that the subsequent ROI area extraction operation will not be interfered by its background.

[0106] S2. Set the neighborhood size to 20*20, use information geometry theory to extract the JS divergence feature matrix of the image, combine it with the image grayscale value, calculate the joint probability distribution JPDF of the image, and use the minimum cross entropy algorithm to calculate the optimal grayscale value threshold T G With JS divergence threshold T JS , and the pixel positions that are smaller than both thresholds are the ROI areas. Count the connected domains in the ROI area, and delete the connected domains with more than 500 pixels and less than 100 pixels. Record the centroid position information and frame number information of the connected domain and proceed to the next step. The overall process of this step is as follows Figure 2 As shown, the extracted ROI area example diagram is as follows Figure 4 As shown, from Figure 4 It can be seen that almost all shadow areas of moving targets are detected.

[0107] S3. Use the Sobel operator to calculate the gradient amplitude of the current original video SAR image and generate a gradient image with a gradient amplitude of more than 50. Calculate the overlapping area between the gradient image and the ROI area. If the overlapping area is greater than 50% of the area of ​​the ROI area itself, it is an incorrect ROI area and is removed. Then extract the centroid position and time frame number of the ROI area of ​​each frame, and calculate the Euclidean distance between the detection result of the current frame and the detection result of the next frame frame by frame. If it is found that the distance between a certain target and the target in the next frame is less than the spatial distance threshold (T I =30), the two targets are considered to belong to the same track and are marked as the same group. If not, they are searched in subsequent frames until the time threshold (T T =5). Finally, the number of associations for each target is counted, and the change between the first and last centroid positions is detected. If the moving distance is large and the number of associations is greater than or equal to 8, the target is considered to be a valid target. The detection results of the moving targets in each frame are as follows: Figure 5 As shown, the red box is the detection box of the present invention, and the green box is the real label box.

[0108] Finally, it should be pointed out that although the present invention has been described with reference to the current specific embodiments, ordinary technicians in this technical field should realize that the above embodiments are only used to illustrate the present invention and are not used as a limitation of the present invention. Various equivalent changes or substitutions can be made without departing from the concept of the present invention. Therefore, as long as the changes and modifications to the above embodiments are within the essential spirit of the present invention, they will fall within the scope of the claims of the present invention.

Claims

1. A high-precision video SAR moving target detection method based on information geometry theory, characterized in that: The following steps are involved: S1. Perform shadow enhancement and background suppression on video SAR image sequences; S2. ROI region extraction based on information geometry theory; S3. Effectively remove false ROI areas through false detection suppression.

2. The high-precision video SAR moving target detection method based on information geometry theory according to claim 1 is characterized in that: The S1 comprises the following steps: S11. Calculate the initial subspace {U, V} of the video SAR image sequence using the principal component analysis algorithm; S12. Set the number of mixed Gaussians, and model the data matrix X composed of the video SAR image sequence based on a mixed Gaussian low-rank representation model, that is: in, represents the set of all Gaussian component weights, represents the set of variances of all Gaussian components; S13. Update the low-rank matrix U*V using online subspace learning technology T , and subtract it from the original video SAR image to obtain a foreground image containing the shadow of the video SAR moving target; S14. Using the alternating direction multiplier algorithm to remove the strong scattered object imaging contained in the foreground image, a video SAR image sequence with shadow enhancement and background suppression, namely, an enhanced image, is obtained.

3. The high-precision video SAR moving target detection method based on information geometry theory according to claim 1 is characterized in that: The S2 comprises the following steps: S21. All pixels in the neighborhood of the coordinate pixel of the enhanced image form a neighborhood system, and construct a statistical distribution describing the point; S22. Calculate the JS divergence to describe the geometric distance in the Gaussian statistical manifold to analyze the pattern and capture the local information features in the image; S23. Traverse each pixel of the image, obtain the JS divergence matrix corresponding to the enhanced image, and calculate the joint probability distribution JPDF; S24. Use the minimum cross entropy algorithm to select the threshold and extract the ROI area of ​​the enhanced image.

4. The high-precision video SAR moving target detection method based on information geometry theory according to claim 3 is characterized in that: The S21 comprises the following steps: S211. All pixels in the N*N neighborhood of the enhanced image at the (x, y) pixel form a neighborhood system, calculate the mean μ and variance σ of all pixels in the neighborhood system, and use Gaussian distribution to describe the statistical distribution of the neighborhood system: S212. Taking the gray intensity value at the current (x, y) pixel as the mean μ′ and a very small variance σ′, construct a statistical distribution describing the point:

5. The high-precision video SAR moving target detection method based on information geometry theory according to claim 3 is characterized in that: The S22 comprises the following steps: S221. By parameter θ=(μ,σ) T The probability distribution family formed constitutes a statistical model S = {P(x|θ)|θ∈Θ}, and S forms a differentiable Gaussian statistical manifold under a certain topological structure; S222. Calculate JS divergence, use JS divergence to describe the geometric distance between different probability distributions in the Gaussian statistical manifold, and perform pattern analysis; S223. Normalize the numerical value of JS divergence. On the Gaussian statistical manifold, JS divergence explains the geometric relationship between the pixel grayscale value at the current position and the distribution of its neighborhood system, helping to capture the local information features in the image.

6. The high-precision video SAR moving target detection method based on information geometry theory according to claim 5 is characterized in that: The JS divergence calculation formula in S222 is as follows: in, is the average distribution of the two distributions, and KL(p||q) is defined as:

7. The high-precision video SAR moving target detection method based on information geometry theory according to claim 3 is characterized in that: The S23 comprises the following steps: S231. Traverse each pixel of the image to obtain the JS divergence matrix corresponding to the enhanced image; S232. Let n ij is the number of joint occurrences between grayscale value and JS divergence, and n ij Each element in is divided by the total number of pixels in the enhanced image to obtain the normalized joint probability distribution JPDF.

8. The high-precision video SAR moving target detection method based on information geometry theory according to claim 3 is characterized in that: The S3 comprises the following steps: S31. Calculate the gradient image of the current frame, and use the gradient image to remove the erroneous ROI area; S32. Use data association algorithm to suppress false alarms of multiple frames.

9. The high-precision video SAR moving target detection method based on information geometry theory according to claim 8 is characterized in that: The S31 comprises the following steps: S311. Use the Sobel operator to detect the edge of the image, calculate the relative change of the gray level of the original video SAR image, and obtain the gradient image of the current original SAR image; S312. Import the gradient image, calculate the overlap area between the gradient image and the ROI region, and remove the ROI region whose overlap area is greater than a threshold.

10. The high-precision video SAR moving target detection method based on information geometry theory according to claim 8, characterized in that: The S32 comprises the following steps: S321. Extract the centroid position of each ROI region and the corresponding time frame number as a feature vector; S322. Set a spatial distance threshold T I and time threshold T T ,Through these two thresholds, the detection points within the time threshold frame are associated to identify possible real targets; S323. Detection points with strong consistency in multiple frames are considered to be valid targets, and only detection points that meet a certain number of associations and spatial changes in the center of mass can be retained.

Images