Airborne wide-angle staring SAR slow moving target detection method based on improved ViBe

Abstract

The invention discloses an airborne wide-angle staring SAR (Synthetic Aperture Radar) low-speed moving target detection method based on improved ViBe, which utilizes the structural information of a shadow and the prior information that a detection target is the shadow, extracts an area of interest in combination with a multi-frame detection result, and obtains better low-speed moving target detection performance. According to the method, the shadow of the slow moving target in the airborne wide-angle staring SAR image can be effectively detected, so that high-precision slow moving target detection is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of radar, and more specifically, to a method for detecting slow moving targets in airborne wide-angle staring SAR based on improved ViBe. Background Technique

[0002] Synthetic Aperture Radar (SAR) technology is a radar detection technology that can work all day and all weather. In recent years, in order to improve the dynamic reconnaissance and surveillance capabilities of SAR for key areas, researchers have proposed wide-angle staring SAR imaging technology represented by Circular SAR (CSAR). In this SAR imaging mode, the radar carrying platform flies along special trajectories such as arcs, circles, and curves, and at the same time, the radar antenna beam always points to the area to be imaged and observed, so as to achieve multi-(full)-azimuth angle imaging and long-time staring imaging detection. Compared with the traditional straight-trajectory SAR imaging, the wide-angle staring SAR imaging has many advantages such as multi-(full)-angle imaging, video imaging, long-time staring imaging, and three-dimensional imaging, thus effectively making up for the deficiencies of the traditional straight-trajectory SAR imaging. Similarly, combining wide-angle staring SAR with Ground Moving Target Indication (GMTI) forms wide-angle staring SAR-GMTI technology. Compared with the traditional straight-trajectory SAR-GMTI technology, the wide-angle staring SAR-GMTI technology takes into account the advantages of both wide-angle staring SAR imaging and GMTI. Therefore, it can perform multi-angle continuous imaging and tracking surveillance on moving targets in the observation area, so as to obtain high-precision moving target detection, parameter estimation, and positioning results, and further provide more accurate dynamic reconnaissance and surveillance information.

[0003] In traditional linear trajectory SAR-GMTI technology, moving target detection is mainly carried out in the radar echo signal domain. The commonly used moving target detection methods can be mainly divided into single-channel methods and multi-channel methods. The multi-channel moving target detection methods mainly include Space Time Adaptive Processing (STAP), Displaced Phase Center Antenna (DPCA), and Along-track Interferometry (ATI). The advantage of multi-channel moving target detection methods is that they can detect slow moving targets submerged by stationary clutter. However, the cost of multi-channel radar systems is relatively high, and there are problems such as channel mismatch. The single-channel moving target detection method is mainly the Doppler filtering method. By performing Doppler filtering processing, clutter can be suppressed to achieve moving target detection. However, the single-channel moving target detection method is helpless in detecting slow moving targets submerged in stationary clutter.

[0004] In recent years, indirect GMTI technology based on ground moving target shadow detection has gradually attracted the attention of researchers. Due to the shielding effect of moving targets on electromagnetic waves, etc., moving targets will leave shadows in SAR images. The shadows of moving targets can reflect the true positions of moving targets. Therefore, by detecting the shadows of moving targets, the detection of moving targets can be indirectly achieved. In the detection of moving target shadows in wide-angle staring SAR, the most commonly used method is the background difference method. This method first estimates the background image using multiple frames of images, and then calculates the difference between the background image and each frame of image to obtain the detection result of moving target shadows. However, this method has some deficiencies. First, when there are many moving target shadows in the scene, the estimated background image is not accurate. Based on the inaccurate background image, using the background difference method to detect moving target shadows will result in high missed alarms and false alarms, leading to poor detection performance of moving target shadows. In addition, the background difference method is a detection method based on individual pixels, ignoring the structural information of moving target shadows. Moving target shadow detection based on deep learning is also a popular research direction, but such methods require a large amount of wide-angle staring SAR sample data to train the deep convolutional neural network, and there are problems with poor generalization. Summary of the Invention

[0005] The present invention provides an airborne wide-angle staring SAR slow moving target detection method based on improved ViBe, which solves the technical problem that it is difficult for airborne wide-angle staring SAR to detect slow moving targets with high precision in the prior art.

[0006] To solve the above technical problems, the technical solution of the present invention is as follows:

[0007] The present invention provides a method for detecting slow-moving targets in airborne wide-angle staring SAR based on improved ViBe, comprising the following steps:

[0008] Perform binarization, morphological processing, connected component analysis, and shadow information detection based on superpixel segmentation on each frame of wide-angle staring SAR image in sequence to obtain a first region of interest image;

[0009] Remove the shadows of stationary targets in the first region of interest image by calculating the local contrast;

[0010] Fill the shadow regions of moving targets in the first region of interest image after removing the shadows of stationary targets to obtain a reconstructed background image;

[0011] Perform multiple non-repetitive samplings on the pixels within a preset-size neighborhood in the reconstructed background image to obtain an initial background model;

[0012] According to the initial background model, calculate the Euclidean distance between each pixel to be classified and the pixels in the background model sample set in the color space, and determine whether the pixel to be classified belongs to the foreground or the background, wherein the background model sample set is the neighborhood pixel value sampling set of the pixel to be classified;

[0013] After screening the pixels determined to be the foreground according to the gray value of the pixels, perform morphological filtering to obtain the detection result of each frame of wide-angle staring SAR image;

[0014] Sum up the detection results of each frame of wide-angle staring SAR image, use morphological processing and connected component analysis to extract a second region of interest, and ignore the detection results outside the second region of interest to obtain the slow-moving target detection result.

[0015] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0016] The present invention uses superpixel segmentation technology to extract the shadow information and local contrast information of the wide-angle staring SAR image to realize background reconstruction. Secondly, modify the definition of distance in image pixel classification, add image pixel screening, and improve the ViBe algorithm using morphological filtering to make it more suitable for the shadow detection of moving targets in wide-angle staring SAR images. Finally, jointly extract the region of interest from the detection results of multiple frames, which can suppress the vast majority of false alarms and realize the high-precision detection of slow-moving targets in airborne wide-angle staring SAR. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic flow chart of a method for detecting slow-moving targets in airborne wide-angle staring SAR based on improved ViBe provided by an embodiment of the present invention;

[0018] Figure 2 Flow chart of background reconstruction provided by an embodiment of the present invention;

[0019] Figure 3 Schematic diagram of the result after binarization and morphological filtering provided by an embodiment of the present invention;

[0020] Figure 4 Schematic diagram of the shadow detection window provided by an embodiment of the present invention;

[0021] Figure 5 Flow chart of screening the first region of interest using shadow information provided by an embodiment of the present invention;

[0022] Figure 6 Schematic diagram of the shadows and bodies of stationary and moving targets provided by an embodiment of the present invention;

[0023] Figure 7 Schematic diagram of the traditional local contrast detection window;

[0024] Figure 8 Schematic diagram of the local contrast detection window based on superpixel segmentation provided by an embodiment of the present invention;

[0025] Figure 9 Flow chart of screening the region of interest using the obtained local contrast information provided by an embodiment of the present invention;

[0026] Figure 10 Schematic diagram of shadow filling provided by an embodiment of the present invention;

[0027] Figure 11 Flow chart of improving the ViBe algorithm provided by an embodiment of the present invention;

[0028] Figure 12 Schematic diagram of the background model sample set provided by an embodiment of the present invention;

[0029] Figure 13 Schematic diagram of the image pixel classification process provided by an embodiment of the present invention;

[0030] Figure 14 Schematic diagram of suppressing false alarm targets outside the second region of interest provided by an embodiment of the present invention;

[0031] Figure 15 Schematic diagram of the shadow detection result of the observation scene provided by an embodiment of the present invention;

[0032] Figure 16 Schematic diagram of the result after removing the shadow of the stationary target provided by an embodiment of the present invention;

[0033] Figure 17Schematic diagram of the background reconstruction result of the observation scenario provided by the embodiment of the present invention.

[0034] Figure 18 Schematic diagram of the detection result of the slow-moving target shadow. Detailed implementation manners

[0035] The drawings are only for illustrative purposes and should not be construed as a limitation of this patent;

[0036] To better illustrate this embodiment, some components in the drawings are omitted, enlarged or reduced, and do not represent the size of the actual product;

[0037] For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0038] The technical solutions of the present invention will be further described below with reference to the drawings and embodiments.

[0039] Embodiment 1

[0040] This embodiment provides a method for detecting slow-moving targets in airborne wide-angle staring SAR based on improved ViBe. As Figure 1 shown, it includes the following steps:

[0041] Perform binarization, morphological processing, connected component analysis, and shadow information detection based on superpixel segmentation on each frame of the wide-angle staring SAR image in sequence to obtain the first image of interest;

[0042] Remove the shadows of stationary targets in the first image of interest by calculating the local contrast;

[0043] Fill the shadow areas of moving targets in the first image of interest after removing the shadows of stationary targets to obtain a reconstructed background image;

[0044] Perform multiple non-repetitive samplings on the pixels within a neighborhood of a preset size in the reconstructed background image to obtain an initial background model;

[0045] According to the initial background model, calculate the Euclidean distance between each pixel to be classified and the pixels in the background model sample set in the color space, and determine whether the pixel to be classified belongs to the foreground or the background, where the background model sample set is the neighborhood pixel value sampling set of the pixel to be classified;

[0046] After screening the pixels determined to be the foreground according to the gray value of the pixels, perform morphological filtering to obtain the detection result of each frame of the wide-angle staring SAR image;

[0047] Sum up the detection results of each frame of wide-angle staring SAR image, and use morphological processing and connected component analysis to extract the second region of interest, and ignore the detection results outside the second region of interest to obtain the slow-moving target detection result.

[0048] The embodiment of the present invention proposes a method for detecting slow-moving targets in airborne wide-angle staring SAR based on improved ViBe. This method utilizes the structural information of shadows and the prior information that the detection target is a shadow, and jointly extracts the region of interest from the detection results of multiple frames, obtaining better performance in detecting slow-moving targets. The present invention can effectively detect the shadows of slow-moving targets in airborne wide-angle staring SAR images, thereby achieving high-precision slow-moving target detection.

[0049] Embodiment 2

[0050] On the basis of Embodiment 1, the embodiment of the present invention further describes the specific process of obtaining the reconstructed background image.

[0051] Obtaining the reconstructed background image includes the following steps:

[0052] Perform binarization, morphological processing, connected component analysis, and shadow information detection based on superpixel segmentation on each frame of wide-angle staring SAR image in turn to obtain the first region of interest image;

[0053] Remove the shadows of stationary targets in the first region of interest image by calculating the local contrast;

[0054] Fill the shadow regions of moving targets in the first region of interest image after removing the shadows of stationary targets to obtain the reconstructed background image.

[0055] The ViBe algorithm has the ability to initialize the model using a single-frame image, good performance in detecting moving targets, and a fast background model update strategy. However, if there are moving targets in the initial frame image during the background modeling of the ViBe algorithm, the "ghost" phenomenon will appear in the detection results of subsequent dozens of frames or even hundreds of frames. To avoid the "ghost" phenomenon, it is necessary to reconstruct the background of the initial frame image and remove the shadows of moving targets in the initial frame image. The flowchart of background reconstruction is as Figure 2 shown, which can be mainly divided into three steps: shadow detection, removal of stationary target shadows, and filling of shadow pixels.

[0056] In a further embodiment, performing binarization, morphological processing, connected component analysis, and shadow information extraction based on superpixel segmentation on the initial frame wide-angle staring SAR image in turn to obtain the first region of interest image includes:

[0057] After binarizing the initial frame wide-angle staring SAR image using the OTSU algorithm, obtain a binary image;

[0058] Process the binary image using morphological processing and connected component analysis to obtain a preliminary first image of interest, where the preliminary first image of interest includes a number of first regions of interest;

[0059] Process the preliminary first image of interest using shadow information detection based on superpixel segmentation to obtain the first image of interest.

[0060] Image binarization is the most commonly used and convenient method to obtain the region of interest (ROI). To obtain the shadow information in the wide-angle staring SAR image, it is first necessary to use the binarization algorithm to extract the darker regions in the image. In the binarization algorithm, the OTSU algorithm is a binarization algorithm with higher efficiency and better automation, so it is widely used. Its basic principle is to divide the pixels in the image into foreground and background using a gray threshold, and the selected threshold needs to ensure the maximum variance between the two types of pixels. After obtaining the binary image using the OTSU algorithm, there will be a large number of false alarms in the image. From features such as area, aspect ratio, and rectangularity, it can be found that some false alarm regions are obviously not the shadows of moving targets. Therefore, using morphological processing and connected component analysis can effectively remove these regions. Binarization and morphological filtering are as Figure 3 shown.

[0061] After binarization and morphological filtering, the darker regions of interest can be obtained. However, the darker regions are not necessarily shadows, and there may also be road surfaces with a relatively small radar scattering coefficient. The real shadow region will be darker than the neighborhood, while the false shadow regions (such as road surfaces with a relatively small radar scattering coefficient) have little difference in brightness from the neighborhood. Taking advantage of this, a shadow detection method based on superpixel segmentation can be used to distinguish between the two.

[0062] Superpixel segmentation is an image preprocessing technique that aggregates adjacent pixels with similar colors and textures into pixel blocks.

[0063] The Simple Linear Iterative Clustering (SLIC) algorithm is a commonly used superpixel segmentation algorithm. This algorithm can not only obtain good superpixel segmentation results but also has a low computational load.

[0064] In a further embodiment, the using shadow information detection based on superpixel segmentation includes:

[0065] Define the three-dimensional coordinates of any pixel point i as [x i , y i , I i , then the distance D between any two pixel points i and jij :

[0066]

[0067] In the formula, d c represents the intensity spatial distance, d s represents the position spatial distance, S represents the expected superpixel size, and W represents the maximum pixel intensity distance; when W is small, it indicates that the intensity spatial distance is more important, and the superpixel blocks better retain the edge structure of the image. When W is large, it indicates that the position spatial distance is more important, and the generated superpixels are more regular. Therefore, superpixel segmentation is performed on the original image marked with the region of interest.

[0068] Aggregate the pixel points in the preliminary first image of interest where D ij is less than the threshold into superpixels to obtain a first superpixel image;

[0069] Construct a shadow detection window for each superpixel in the first superpixel image. The shadow detection window includes a protection window and a reference window. As Figure 4 shown, which shows the case where both the protection window and the reference window are two-level neighborhoods, there are:

[0070]

[0071] In the formula, H0 is the hypothesis that the target superpixel s is a background region, H1 is the hypothesis that the target superpixel s is a shadow region, χ represents a preset threshold, and m0 and m1 respectively represent the average brightness of the target superpixel and the average brightness of the reference window;

[0072] The role of the protection window is to prevent the pixels in the reference window from including shadow regions. The reference window is used to statistically calculate the average brightness of background clutter and compare the average brightness of the reference window with the average brightness of the target superpixel s. When the ratio of the average brightness of the target superpixel s to the average brightness of the reference window is less than the preset threshold, the target superpixel s is determined to be a shadow region. Otherwise, the target superpixel s is determined to be a background region.

[0073] If the superpixel is a background region, all the pixels that make up the superpixel are background pixels; if the superpixel is a shadow region, all the pixels that make up the superpixel are shadow pixels;

[0074] Using the shadow detection method based on superpixel segmentation, the shadow information of the entire wide-angle staring SAR image can be obtained. Then, the obtained shadow information is used to screen the region of interest. If a region does not contain shadow pixels, it is considered that the region is probably not a shadow region. Based on this premise, the non-shadow regions in the region of interest are removed. The specific screening method is:

[0075] Determine whether each first region of interest in the preliminary first image of interest includes shadow pixels. If it includes shadow pixels, remove the first region of interest; if it does not include shadow pixels, retain the first region of interest; thereby obtaining the first image of interest.

[0076] Therefore, screening using shadow information can effectively eliminate non-shadow dark regions. The process of screening regions of interest using shadow information is as Figure 5 shown.

[0077] Among the regions of interest obtained through the above binarization, morphological filtering, and screening using shadow information, there are not only shadows of moving targets but also shadows of stationary targets. When subsequently filling the shadows to reconstruct the background, only the shadows of moving targets need to be filled, and the shadows of stationary targets do not need to be filled. Therefore, it is necessary to remove the shadows of stationary targets and only retain the shadow regions of moving targets. Local contrast can distinguish the effective features of stationary target shadows and moving target shadows. As Figure 6 shown, stationary targets do not deviate from their true positions in wide-angle staring SAR images. Therefore, there is an image of the stationary target itself in the neighborhood of the shadow of the stationary target, with a relatively high local contrast. Due to its own movement, a moving target will deviate from its original position in wide-angle staring SAR images. Therefore, the shadow of the moving target is still located near the true position of the moving target. Therefore, there is no image of the moving target itself in the neighborhood of the shadow of the moving target, and the local contrast is low. By utilizing the differences in local contrast features between stationary target shadows and moving target shadows, the shadow regions of stationary targets and moving targets can be effectively distinguished.

[0078] In a further embodiment, the calculation of the local contrast includes:

[0079] Aggregate the pixel points in the first image of interest with D ij less than the threshold into superpixels to obtain a second superpixel image;

[0080] Regarding the superpixels in the second superpixel image that include the first region of interest as the central region of the detection window, and then regarding the region directly connected to the central region as the neighborhood to obtain a local contrast detection window;

[0081] As Figure 7As shown in the figure, the traditional local contrast detection window takes the minimum bounding rectangle of the region of interest as the central region of the detection window, and then forms a complete detection window by constructing an eight-neighborhood of the same size as the central region. This detection window ignores the edge structures of shadows and targets. The central region contains not only shadow pixels but also background clutter pixels. Similarly, for the neighborhood containing the target, in addition to containing target pixels, it also contains background clutter pixels. In addition, there will also be cases where the target spans two or even more neighborhoods. All of the above problems will affect the calculation of local contrast. Therefore, the embodiment of the present invention redesigned the local contrast detection window based on superpixel segmentation. Superpixel segmentation can well preserve the edge structure information of shadows and targets in the image, and perform superpixel segmentation on the original image marked with the region of interest. Then, the superpixel containing the region of interest is used as the central region of the detection window, and the region directly adjacent to the central region is used as the neighborhood to complete the construction of the local contrast detection window. The local contrast detection window based on superpixel segmentation is as Figure 8 shown. By making use of the characteristic that superpixel segmentation can well preserve the edge structure information of the image, the central region and the neighborhood containing the target contain as few background clutter pixels as possible, and a more accurate local contrast result can be calculated.

[0082] The calculation of the local contrast C is as follows:

[0083]

[0084] In the formula, G c is the average gray value of the pixels in the central region, G ni is the average gray value of the pixels in the i-th neighborhood, and k is the total number of neighborhoods. Since the target may appear in all directions of the shadow, when calculating the local contrast, the neighborhood with the highest average gray value of the pixels should be selected for calculation.

[0085] In a further embodiment, removing the shadow of the stationary target in the first image of interest by calculating the local contrast includes:

[0086] Calculating the local contrast of all superpixels in the second superpixel image and performing binarization according to the preset threshold of the local contrast to divide the pixels into high local contrast pixels and low local contrast pixels;

[0087] Calculating the local contrast of all superpixels in the image and performing binarization according to the preset threshold, and the local contrast information of the image can be obtained. Using the obtained local contrast information to screen the region of interest, if a region does not contain high local contrast pixels, it is considered that this region is probably the shadow of the moving target. Based on this premise, the shadow of the moving target in the region of interest is screened out, and the shadow of the stationary target is removed. The specific screening method is as follows:

[0088] Determine whether each first region of interest in the first image of interest includes high local contrast pixels. If not, retain the first region of interest; if so, remove the first region of interest.

[0089] Use local contrast information to filter out the shadows of stationary targets effectively. The process of filtering the regions of interest using local contrast information is as Figure 9 shown.

[0090] In a further embodiment, after shadow detection and removal of the shadows of stationary targets, the shadow regions of moving targets can be obtained. Next, it is necessary to fill the shadow regions of the moving targets to complete background reconstruction. According to the Markov random field theory, the background of the shadow regions can be reconstructed using the neighboring pixels of the shadows. The specific shadow filling method is:

[0091] Randomly sample the neighborhood of the shadow region, and then fill the sampled neighborhood pixel values into the shadow region to obtain the reconstructed background image.

[0092] The schematic diagram of shadow filling is as Figure 10 shown.

[0093] Embodiment 3

[0094] Based on Embodiment 1 and Embodiment 2, this embodiment further describes the specific process of improving the ViBe algorithm.

[0095] The improved ViBe algorithm includes:

[0096] Perform multiple non-repetitive samplings on the pixels within a preset size neighborhood in the reconstructed background image to obtain an initial background model;

[0097] According to the initial background model, calculate the Euclidean distance between each pixel to be classified and the pixels in the background model sample set in the color space, and determine whether the pixel to be classified belongs to the foreground or the background, where the background model sample set is the sampling set of the neighborhood pixel values of the pixel to be classified;

[0098] After screening the pixels determined to be the foreground according to the gray value of the pixels, perform morphological filtering to obtain the detection result of each frame of wide-angle staring SAR image;

[0099] Sum up the detection results of each frame of wide-angle staring SAR image, and use morphological processing and connected component analysis to extract the second region of interest, and ignore the detection results outside the second region of interest to obtain the detection result of slow moving targets.

[0100] The ViBe algorithm is a non-parametric background modeling method with fast response, strong robustness, and good detection effect. The ViBe algorithm directly uses the frequency of background sampling points to fit the background probability distribution at that point, without being limited to a certain distribution or parameter. After completing the background reconstruction, there are no moving target shadows in the scene. Therefore, a background model can be constructed based on the reconstructed background image, and the ViBe algorithm can be used to continue the shadow detection of moving targets in wide-angle staring SAR images. The original ViBe algorithm is for moving target detection in optical videos and does not utilize the prior information that the detected targets in wide-angle staring SAR images are shadows. Therefore, the present invention improves the ViBe algorithm to make it more suitable for the shadow detection of moving targets in wide-angle staring SAR images. The improved ViBe algorithm includes six steps: background model initialization, image pixel classification, pixel screening, morphological filtering, background update, and false alarm suppression, as Figure 11 shown.

[0101] In a further embodiment, according to the initialized background model, the Euclidean distance between each pixel point to be classified and the pixel points in the background model sample set in the color space is calculated to determine whether the pixel point to be classified belongs to the foreground or the background, including:

[0102] The original ViBe algorithm uses a random strategy to repeat 20 times within a 3×3 neighborhood, and each time a pixel point is randomly selected. A total of 20 pixel points are selected as the background sample set of the central pixel, thus completing the initialization of the background model. Therefore, the ViBe algorithm has strong robustness and can effectively cope with noise interference. However, when randomly selecting 20 pixel points within each 8-neighborhood, there will inevitably be duplicate selected pixels. To address this problem, the present invention chooses to expand the neighborhood and perform 20 non-repetitive samplings within a 5×5 neighborhood to complete the initialization of the background model, as Figure 12 shown.

[0103] Define the background model sample set M(x):

[0104] M(x) = {v1, …, v i , …, v N}, 1 ≤ i ≤ N

[0105] where v i is the pixel value obtained by sampling in the neighborhood of pixel point x, and N is the number of non-repetitive samplings;

[0106] After completing the initialization of the background model, the next step is to classify the image pixels, that is, to determine whether the image pixels belong to the foreground or the background. The process of image pixel classification is as Figure 13 shown;

[0107] Calculate the Euclidean distance d between each pixel point to be classified and the pixel points in the background model sample set in the color spacei (x):

[0108] d i (x) = v i -v(x), 1 ≤ i ≤ N

[0109] Wherein, v(x) is the gray value of the pixel point x;

[0110] Determine whether the pixel point to be classified belongs to the foreground or the background:

[0111]

[0112] If the number of samples with a distance less than R between the pixel points in the background model sample set M(x) and the current pixel point x is less than the threshold Min, the pixel point x is judged as the foreground; otherwise, the pixel point is judged as the background. Wherein, Min is the sample threshold, and R is the distance threshold, which is determined by the dispersion degree of the gray values of the background model. The calculation formula is:

[0113]

[0114] Wherein, m represents the median of |v i -v i+1 |.

[0115] In a further embodiment, combining the characteristic that the average pixels in the shadow area of the moving target in the wide-angle staring SAR image are lower than the background pixel value, a background threshold T can be preset to screen the pixels classified as the foreground, and the high-gray-value pixels that obviously do not belong to the shadow are reclassified as the background. According to the gray value of the pixel point, the pixels judged as the foreground are screened, and the screening is carried out by the following formula:

[0116]

[0117] Wherein, if the gray value of the pixel point is higher than the background threshold T, the pixel point is reclassified as the background. If the gray value of the pixel point is lower than the background threshold T, the pixel point is correctly classified, and T is the background threshold.

[0118] After reclassification by the background threshold T, the obtained result still contains areas such as speckle noise and road edges that are not the shadow of the moving target. Therefore, the influence brought by these factors can be eliminated by using morphological processing and connected component analysis.

[0119] In a further embodiment, during the process of detecting moving targets using wide-angle staring SAR images, the observation scene may change due to various factors, such as jitter of the observation angle, falling of leaves, etc. Therefore, when using the improved ViBe algorithm for moving target detection, the background model needs to be updated in real time to make the detection results more accurate. The background model update method adopted by the improved ViBe algorithm is the conservative update method. Only when a pixel is classified as the background, will this pixel be included in the background model set. Therefore, after obtaining the detection results of the initial frame wide-angle staring SAR image, the following steps are further included:

[0120] Update the initialized background model using the conservative update method, and at the same time, use the conservative update method for real-time update during subsequent detections to obtain the updated background model.

[0121] In a further embodiment, false alarm suppression is performed by combining the detection results of multiple frames. The motion trajectories of real moving targets are generally regular, while false moving targets will appear disorderly in the detection results of multiple frames. According to the above characteristics, first sum up all the initially obtained frame detection results. If it is a real moving target, a long strip-shaped connected region will be obtained in the summation result; then use morphological processing and connected component analysis to extract the region of interest;

[0122] Finally, ignore the detection results outside the region of interest, thereby suppressing false alarm targets outside the region of interest, as Figure 14 shown.

[0123] In a further embodiment, for each frame of wide-angle staring SAR image, shadow information is extracted using a shadow detection method based on superpixel segmentation, and then the shadow information is used to suppress false alarms within the second region of interest.

[0124] Embodiment 4

[0125] In this embodiment, the effectiveness and practicality of the methods described in Embodiments 1 to 3 are verified through experimental verification using measured data.

[0126] To prove the effectiveness of the airborne wide-angle staring SAR slow moving target detection method based on the improved ViBe proposed by the present invention, 100 frames of SAR images (frames 153 to 252) are selected for experiments using the airborne wide-angle staring SAR data published by the laboratory. The method proposed in the embodiments of the present invention is verified through the experimental data processing results.

[0127] Use the background reconstruction algorithm to perform background reconstruction on a single frame of wide-angle staring SAR image. This algorithm mainly includes 3 steps: shadow detection, removal of stationary target shadows, and filling of shadow pixels. The shadow detection results are as Figure 15 shown. From Figure 15It can be found that the OTSU algorithm can effectively extract the dark regions in the image. After morphological filtering, regions that are clearly not the shadows of moving targets are removed according to features such as area, aspect ratio, and rectangularity. Finally, shadow information is extracted using a shadow detection method based on superpixel segmentation, and non-shadow regions can be effectively removed.

[0128] The result after removing the shadows of stationary targets is as Figure 16 shown. After performing superpixel segmentation on the image, the local contrast information of the image is extracted. Then, the shadows of stationary targets can be effectively removed using the extracted local contrast information, leaving only the shadows of moving targets. Finally, the shadow regions of moving targets are filled using neighboring pixels, thus completing the background reconstruction, as Figure 17 shown. After the background reconstruction is completed, there are no longer shadows of moving targets in the background.

[0129] Figure 18The detection results of slow-moving targets obtained by using six different methods are shown. From left to right are the 153rd, 178th, 203rd, 228th, and 252nd frames of the airborne wide-angle staring SAR images. From top to bottom are the original images, the detection results of the background difference method, the detection results of the original ViBe algorithm, the detection results of the improved ViBe algorithm (without false alarm suppression), the detection results of the background difference method (with false alarm suppression), the detection results of the original ViBe algorithm (with false alarm suppression), and the detection results of the improved ViBe algorithm. The true positions of the moving targets are marked with red boxes in the original images and the detection results of all methods. It includes the background difference method, the original ViBe algorithm, the improved ViBe algorithm (without false alarm suppression), the background difference method (with false alarm suppression), the original ViBe algorithm (with false alarm suppression), and the improved ViBe algorithm. Compare the improved ViBe algorithm, the original ViBe algorithm (with false alarm suppression), and the background difference method (with false alarm suppression). There is only one missed detection in the detection results of the improved ViBe algorithm, and no false alarms occur; there are two missed detections and one false alarm in the detection results of the original ViBe algorithm (with false alarm suppression); there is one missed detection and two false alarms in the detection results of the background difference method (with false alarm suppression). Among these three moving target shadow detection methods, the improved ViBe algorithm has the best detection performance for slow-moving targets. Comparing the background difference method, the original ViBe algorithm, and the improved ViBe algorithm (without false alarm suppression), it can also be found that the improved ViBe algorithm has the fewest missed detections and false alarms, and the detection performance is the best. It verifies that the proposed airborne wide-angle staring SAR slow-moving target detection method based on the improved ViBe can effectively detect the shadows of slow-moving targets. Comparing the improved ViBe algorithm and the improved ViBe algorithm (without false alarm suppression), the original ViBe algorithm (with false alarm suppression) and the original ViBe algorithm, the background difference method (with false alarm suppression) and the background difference method, it can be found that after adding false alarm suppression to the three methods, the false alarms outside the region of interest and the false alarms in the non-shadow regions within the region of interest can be removed, making the number of false alarms significantly reduced, thus verifying the correctness and effectiveness of the proposed false alarm suppression method of the present invention. Table 1 summarizes the detection results of moving targets in 100 frames of airborne wide-angle staring SAR images. It can be found from Table 1 that the number of missed detection targets and false alarm targets of the improved ViBe algorithm is the least, and the detection rate of slow-moving targets reaches 97.66%, with only 8 false alarms occurring. After adding false alarm suppression to the three methods, the number of false alarm targets is significantly reduced. The above experimental results further verify that the proposed airborne wide-angle staring SAR slow-moving target detection method based on the improved ViBe has good detection performance for slow-moving targets.

[0130] Table 1 Statistics of Detection Results of Slow-Moving Targets

[0131]

[0132] Like or similar reference numerals correspond to like or similar components;

[0133] The terms used to describe the positional relationship in the drawings are for illustrative purposes only and should not be construed as a limitation of this patent;

[0134] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, rather than limitations on the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to enumerate all the implementation manners here. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the claims of the present invention. This patent application is supported by the Shenzhen Science and Technology Plan (Project No.: JCYJ20240813151238049), the Shenzhen Science and Technology Plan (Project Nos.: 202206193000001, 20220815171723002), the Guangdong Basic and Applied Basic Research Foundation (Project No.: 2023A1515011588), and the project "Research on Airborne Staring SAR Moving Target Detection and Tracking Algorithm" of Beijing Institute of Radio Measurement (Contract No.: 20242467).

Claims

1. An airborne wide-angle staring SAR slow moving target detection method based on improved ViBe, characterized in that, It includes the following steps: Successively perform binarization, morphological processing, connected component analysis, and shadow information detection based on superpixel segmentation on each frame of wide-angle staring SAR image to obtain a first region of interest image; Remove the shadows of stationary targets in the first region of interest image by calculating the local contrast; Fill the shadow regions of moving targets in the first region of interest image after removing the shadows of stationary targets to obtain a reconstructed background image; Perform multiple non-repetitive samplings on the pixels within a preset-size neighborhood in the reconstructed background image to obtain an initial background model; According to the initial background model, calculate the Euclidean distance between each pixel to be classified and the pixels in the background model sample set in the color space, and determine whether the pixel to be classified belongs to the foreground or the background, where the background model sample set is the neighborhood pixel value sampling set of the pixel to be classified; After screening the pixels determined to be the foreground according to the gray value of the pixels, perform morphological filtering to obtain the detection result of each frame of wide-angle staring SAR image; Sum up the detection results of each frame of wide-angle staring SAR image, use morphological processing and connected component analysis to extract a second region of interest, and ignore the detection results outside the second region of interest to obtain the slow-moving target detection result.

2. The airborne wide-angle staring SAR slow moving target detection method based on improved ViBe according to claim 1, characterized in that, Successively perform binarization, morphological processing, connected component analysis, and shadow information extraction based on superpixel segmentation on the initial frame of wide-angle staring SAR image to obtain a first region of interest image, including: After binarizing the initial frame of wide-angle staring SAR image using the OTSU algorithm, obtain a binary image; Process the binary image using morphological processing and connected component analysis to obtain a preliminary first region of interest image, where the preliminary first region of interest image includes several first regions of interest; Process the preliminary first region of interest image using shadow information detection based on superpixel segmentation to obtain the first region of interest image.

3. The airborne wide-angle staring SAR slow moving target detection method based on improved ViBe according to claim 2, characterized in that, The use of shadow information detection based on superpixel segmentation includes: Define the three-dimensional coordinates of any pixel point i as [x i , y i , I i . Then the distance D ij between any two pixel points i and j is: where d c represents the intensity spatial distance, d s represents the position spatial distance, S represents the expected superpixel size, and W represents the maximum pixel intensity distance; Aggregate the pixel points in the preliminary first image of interest that are D ij less than the threshold into superpixels to obtain a first superpixel image; Construct a shadow detection window for each superpixel in the first superpixel image, and the shadow detection window includes a protection window and a reference window, where: In the formula, H0 is the hypothesis that the target superpixel s is a background region, H1 is the hypothesis that the target superpixel s is a shadow region, χ represents a preset threshold, and m0 and m1 respectively represent the average brightness of the target superpixel and the average brightness of the reference window; If the superpixel is a background region, all pixels constituting the superpixel are background pixels; if the superpixel is a shadow region, all pixels constituting the superpixel are shadow pixels; Judge whether each first region of interest in the preliminary first region of interest image includes shadow pixels. If it includes shadow pixels, remove the first region of interest; if it does not include shadow pixels, retain the first region of interest; to obtain the first region of interest image.

4. The airborne wide-angle staring SAR slow moving target detection method based on improved ViBe according to claim 3, characterized in that, The calculation of the local contrast includes: Aggregate the pixel points in the first image of interest that are D ij less than the threshold into superpixels to obtain a second superpixel image; Use the superpixels in the second superpixel image that include the first region of interest as the central region of the detection window, and then use the region directly connected to the central region as the neighborhood to obtain a local contrast detection window; The calculation of the local contrast C is: Where G c is the average gray value of pixels in the central region, G ni is the average gray value of pixels in the i-th neighborhood, and k is the total number of neighborhoods.

5. The airborne wide-angle staring SAR slow moving target detection method based on improved ViBe according to claim 4, characterized in that Removing the shadows of stationary targets in the first image of interest by calculating the local contrast, including: Calculating the local contrast of all superpixels in the second superpixel image, and performing binarization according to a preset threshold of the local contrast to divide the pixels into high local contrast pixels and low local contrast pixels; Judging whether there are high local contrast pixels in each first region of interest in the first image of interest. If not, keep the first region of interest; if so, remove the first region of interest.

6. The airborne wide-angle staring SAR slow moving target detection method based on improved ViBe according to claim 1, characterized in that Filling the shadow regions of moving targets in the first image of interest after removing the shadows of stationary targets to obtain a reconstructed background image, including: Randomly sampling the neighborhood of the shadow region, and then filling the pixel values of the sampled neighborhood into the shadow region to obtain a reconstructed background image.

7. The airborne wide-angle staring SAR slow moving target detection method based on improved ViBe according to claim 1, characterized in that, According to the initialized background model, calculating the Euclidean distance between each pixel point to be classified and the pixel points in the background model sample set in the color space, and judging whether the pixel point to be classified belongs to the foreground or the background, including: Defining the background model sample set M(x): M(x) = {v1, …, v i , …, v N}, 1 ≤ i ≤ N where v i is the pixel value obtained by sampling in the neighborhood of pixel point x, and N is the number of non-repeated samplings; Calculate the Euclidean distance d in the color space between each pixel point to be classified and the pixel points in the background model sample set i (x): d i f(x) = v i -v(x), 1 ≤ i ≤ N In the formula, v(x) is the gray value of the pixel point x; Judging whether the pixel point to be classified belongs to the foreground or the background: In the formula, Min is the sample threshold, and R is the distance threshold, which is given by the following formula: where m represents the median of |v i - v i+1 |.

8. The airborne wide-angle staring SAR slow moving target detection method based on improved ViBe according to claim 1, wherein Screening the pixel points judged as the foreground according to the gray value of the pixel points, and screening by the following formula: In the formula, T is the background threshold.

9. The airborne wide-angle staring SAR slow moving target detection method based on improved ViBe according to claim 1, characterized in that, After obtaining the detection result of the initial frame wide-angle staring SAR image, the following steps are further included: Updating the initialized background model using the conservative update method, and at the same time, using the conservative update method to update in real time during subsequent detections to obtain an updated background model.

10. The airborne wide-angle staring SAR slow moving target detection method based on improved ViBe according to claim 1, characterized in that, Extracting shadow information from each frame of wide-angle staring SAR image using the shadow detection method based on superpixel segmentation, and then suppressing false alarms in the second region of interest using the shadow information.

Images