An automatic ghost detection algorithm for SAR images

Abstract

The invention proposes an automatic ghost detection algorithm for SAR images, which comprises the following steps: filtering a SAR image; obtaining statistical features of the filtered SAR image, and calculating two thresholds T1 and T2 of a main object and a ghost according to the statistical features; performing main object binarization processing and ghost binarization processing on the SAR image according to the thresholds T1 and T2, respectively, to obtain binary images of the main object and the ghost; performing expansion processing on the two-scene binary images; performing edge extraction on the expanded images, respectively, to obtain the edge of the main object and the edge of the ghost; respectively screening the extracted main object edge and ghost edge, and eliminating short and invalid edges; performing correlation calculation on the screened main object edge and ghost edge, setting a threshold T, judging whether the SAR image has a ghost by the threshold, and if so, marking the ghost area and the main object position corresponding to it on the image.

Description

Technical Field

[0001] The invention belongs to the technical field of SAR image recognition and matching, and in particular relates to an automatic detection algorithm for SAR image ghosting. Background Art

[0002] Synthetic aperture radar (SAR) is a high-resolution imaging radar that measures distance by the time difference between transmitting electromagnetic pulses and receiving target echoes, and uses the target echo energy as the grayscale value of the image to form a SAR image. Even under extremely low visibility weather conditions, it can still obtain high-resolution images similar to optical photography.

[0003] When SAR is imaging, the echo signal from outside the observation area will interfere with the echo signal in the observation area. When the distance difference between the echo signal target in the observation area and the echo signal target outside the observation area to the radar is relatively small, ghosting will be formed on the image, especially for some strong targets, which are prone to ghosting. The ghosting problem will seriously affect the image quality.

[0004] The application efficiency of SAR images mainly depends on the quality of the acquired images. Therefore, the quality assessment of SAR images is a bridge connecting the development of SAR systems and the application of SAR images. Ghosting detection of SAR images is an important step in the quality assessment of SAR images. Ghosting detection of SAR images can provide a basis for the design of SAR system to remove ghosting and provide feedback information for improving the design, thereby effectively removing ghosting. It can also help users use SAR data, so that they can clearly know the ghosting area of ​​the image and avoid being affected by it when using it.

[0005] Currently, there is relatively little theoretical research on ghost detection in SAR images. Ghost identification in SAR images is mostly based on visual judgment, which is subjective and lacks scientific basis. Summary of the invention

[0006] In view of this, the purpose of the present invention is to propose an automatic detection algorithm for SAR image ghosting, which can be used to determine whether there is ghosting in the SAR image, and if so, give the ghost position.

[0007] In order to achieve the above technical purpose, the technical solution adopted by the present invention is an image recognition and matching ghost detection algorithm that detects ghosts by extracting the edges of the main object and the ghost and calculating the correlation of the edges. The overall steps are as follows:

[0008] Step 1: Filter the SAR image;

[0009] Step 2: Obtain the statistical features of the filtered SAR image, and calculate two thresholds T1 and T2 of the main object and the ghost according to the statistical features;

[0010] Step 3: According to T1 and T2, the SAR image is subjected to binarization processing of the main object and the ghost respectively, so as to form a binary image of the main object and the ghost;

[0011] Step 4: dilate the above two binary images;

[0012] Step 5: Extract the edges of the image after dilation processing to obtain the edge of the main object and the edge of the ghost;

[0013] Step 6: Filter the extracted main object edges and ghost edges respectively to remove short and invalid edges;

[0014] Step 7: Calculate the correlation between the edge of the main object and the ghost edge after screening, set the threshold T, and use the threshold to determine whether there is a ghost in the SAR image. If so, mark the ghost area and its corresponding main object position on the image.

[0015] The above step 1 further includes the following sub-steps:

[0016] Step S11: Modeling the SAR image as F=(T+N)·S; wherein F is the received backscattering intensity, T is the signal intensity, N is the additive noise, and S is the multiplicative speckle noise;

[0017] Step S12: converting the multiplicative speckle noise into additive noise by logarithmic transformation: log(F)=log(T+N)+log(S);

[0018] Step S13: filtering out the speckle noise, and obtaining a SAR image with the speckle removed by exponential transformation.

[0019] The above step 2 obtains the statistical features of the SAR image: the statistical features such as the mean, median, 75th percentile, and maximum value of the image grayscale value can be counted, and the value range of a certain type of objects in the image can also be counted. According to the statistical features, a threshold T1 for distinguishing the main object from the ghost and a threshold T2 for distinguishing the ghost from the background are established.

[0020] The step 3 can be divided into two parts:

[0021] First, the SAR image is binarized according to the threshold value:

[0022] For each pixel

[0023] if (current pixel gray value>T1)

[0024] Current pixel grayscale value = 65535;

[0025] else

[0026] The gray value of the current pixel = 0

[0027] Obtain a binary image of the SAR image with respect to the main object;

[0028] Then, the following binarization algorithm is performed on the SAR image according to the threshold:

[0029] For each pixel

[0030] if (current pixel gray value>T1)

[0031] Current pixel gray value = 0;

[0032] else if (current pixel gray value>=T2)

[0033] Current pixel grayscale value = 65535;

[0034] else

[0035] Current pixel gray value = 0

[0036] The binary image of the SAR image ghost is obtained.

[0037] The specific algorithm of the expansion process in step 4 above is:

[0038] For each pixel:

[0039] if (current pixel grayscale value = 65535)

[0040] Grayscale values ​​of the 8 pixels around the current pixel = 65535

[0041] The edge extraction algorithm in step 5 above uses an edge extraction operator to perform convolution calculation on the expanded image to extract the edge of the main object and the edge of the ghost.

[0042] The above step 6 further includes the following sub-steps:

[0043] Step S61: storing the edges of the main object and the ghost image and calculating their lengths;

[0044] Step S62: Set a length threshold K, retain edges with a length greater than K, and remove edges with a length less than or equal to K.

[0045] The above step 7 further includes the following sub-steps:

[0046] Step S71: dividing the edges retained in step 6 into two categories: main object edges and ghost edges;

[0047] Step S72: sorting the main object edge and the ghost edge from long to short according to their lengths;

[0048] Step S73: Calculate the correlation between each main object edge and each ghost edge according to the sequence;

[0049] Step S74: setting a correlation threshold T, and judging whether there is ghosting in the SAR image by the threshold. If there is, marking the ghosting area and the main object corresponding to it on the image.

[0050] The advantages and beneficial effects of the present invention are as follows:

[0051] (1) Currently, ghost detection of SAR images still relies on visual judgment. The present invention proposes a SAR image ghost detection algorithm that can automatically perform ghost detection.

[0052] (2) Ghost detection of SAR images is an important step in SAR image quality assessment. The present invention can scientifically detect ghosts in SAR images and provide a basis and feedback information for improving the SAR system.

[0053] (3) The present invention extracts the statistical features of the SAR image to select the threshold value, and then detects ghosting. It is objective and scientific, can analyze specific problems specifically, and the algorithm has strong applicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A schematic diagram of a technical flow of a SAR image ghosting detection algorithm according to an embodiment of the present invention;

[0055] Figure 2 An edge extraction operator used for edge extraction in an embodiment of the present invention: Roberts operator. DETAILED DESCRIPTION

[0056] The specific implementation of the present invention will be further described below in conjunction with the accompanying drawings.

[0057] During SAR imaging, echo signals from outside the observation area interfere with echo signals in the observation area, which results in ghosting of strong targets. When the contrast between the target and the background is obvious, the ghosting will be very obvious. An obvious feature of SAR ghosting images is that the main object is brighter, while its ghost is slightly darker. Therefore, ghosting detection is performed based on this feature.

[0058] Ghost detection of SAR images is actually a process of pattern recognition and matching. First, the image is filtered, and then the original image is divided into two branches. The left branch detects the slightly brighter main object, and the right branch detects the slightly darker ghost (if a ghost exists). Then, edge detection is performed on each of them. Finally, by finding the correlation between the edge of the main object and the edge of the ghost, it is determined whether there is a ghost.

[0059] Figure 1 The technical flow chart of a SAR image ghost detection algorithm according to an embodiment of the present invention is shown, which mainly includes seven steps: filtering, statistical feature extraction, binarization, dilation, edge extraction, edge screening and correlation processing. The implementation details of these seven steps in the embodiment will be described below in conjunction with the following figures.

[0060] Step 1: Filter the SAR image.

[0061] The SAR image is modeled as F = (T + N) S, where F is the received backscatter intensity, T is the signal intensity, N is the additive noise, and S is the multiplicative speckle noise. The multiplicative speckle noise is converted into additive noise by logarithmic transformation: log(F) = log(T + N) + log(S). The speckle noise is filtered out by filtering, and the speckle-filtered SAR image is obtained by exponential transformation.

[0062] Here, Lee filtering is used to filter out the additive noise of SAR images. Of course, other filtering methods can also be selected in practical applications. The optional filtering methods include: enhanced Lee filtering, median filtering, mean filtering, Gaussian filtering, etc. The selection of filtering window size and basic parameters should be determined according to the specific situation of the image. A large window and a high intensity will increase the smoothness of the image, and the corresponding image edge and detail information will not be well preserved. If a small window and low intensity filter is used, the noise can be filtered out while retaining the detail information.

[0063] Step 2: Obtain statistical characteristics of SAR images.

[0064] The purpose of obtaining the statistical features of SAR images is to obtain the grayscale value distribution of the image through the features. Because ghosting mainly occurs in areas with larger grayscale values, the thresholds T1 and T2 can be set scientifically through the grayscale value distribution of the image. The T1 threshold is mainly used to identify areas with larger grayscale values ​​in the image, so that ghosting can be detected in this area; because the ghost is slightly darker than the main object, the T2 threshold is set, and the approximate area of ​​the ghost (if ghosting exists) is identified in conjunction with the T1 threshold. Therefore, the statistical features that need to be obtained include the mean, median, 75th percentile and maximum value of the grayscale. Through these statistical features, we can know the approximate value range of the grayscale value of the strong target, so as to set the T1 and T2 thresholds. This threshold set for ghost detection of the entire image sacrifices part of its detection accuracy because it takes the entire image into consideration; of course, if you want to perform ghost detection on a certain thing in the image (such as a building area), you can also use some prior knowledge of the building area as a feature (such as the grayscale value range of the building area, etc.) to set a threshold for a specific thing to detect whether there is a ghost in the specific thing. This method of setting the threshold will be more accurate and have a better effect on detecting its ghost, but it can only detect ghosts on specific things, but not on the entire image.

[0065] Step 3: Binarize the SAR image.

[0066] In order to realize ghost detection, it is necessary to perform double detection on the main object and the ghost. Figure 1 In the figure, there are two branches, the left branch detects the slightly brighter main object, and the right branch detects the slightly darker ghost (if the ghost exists), so it is necessary to perform binarization on the main object and the ghost respectively.

[0067] The binary processing algorithm corresponding to the main object is described as follows: for each pixel

[0068] if (current pixel gray value>T1)

[0069] Current pixel grayscale value = 65535;

[0070] else

[0071] Current pixel gray value = 0

[0072] The binary processing algorithm corresponding to the ghost part is as follows: for each pixel

[0073] if (current pixel gray value>T1)

[0074] Current pixel gray value = 0;

[0075] else if (current pixel gray value>=T2)

[0076] Current pixel grayscale value = 65535;

[0077] else

[0078] Current pixel gray value = 0

[0079] Step 4: Dilate the binarized image.

[0080] After the binarization process, due to the selection of the threshold, the grayscale value of the extracted area will be discontinuous, and there will be obvious particles, which will directly affect the following edge extraction. Here, the morphological expansion algorithm is used to eliminate the influence of the particles as much as possible, and the internal connectivity of the area is achieved, making the edge of the area more prominent, which is convenient for subsequent extraction.

[0081] The algorithm is as follows: For each pixel:

[0082] if (current pixel grayscale value = 65535)

[0083] Grayscale of 8 pixels around the current pixel = 65535

[0084] Step 5: Edge extraction

[0085] Edge extraction is a very critical part of the algorithm of the present invention. How to effectively extract the required edges is crucial and has a great impact on the following steps. The edge extraction algorithm uses an edge extraction operator to perform convolution calculation on the expanded image to extract the edges of the main object and the edges of the ghost. Figure 2 The Roberts operator can be used for edge extraction to obtain relatively thin edges, which is beneficial for subsequent processing. In addition, other edge extraction operators can also be used, such as Sobel operator, Prewitt operator, Kirsch operator, Robinson operator, etc.

[0086] Step 6: Edge screening

[0087] After edge extraction, although the contour edge is extracted to a certain extent, too many small edge fragments are generated, which affect the following operations. Therefore, we need to filter the extracted edges and remove the short and useless edges.

[0088] The embodiment adopts the method of establishing a linked list for edges: the extracted edge pixels are linked one by one, and a relatively long chain can be established for a longer contour edge, while only a very short chain can be established for a small fragmented edge, so that a longer valid contour edge can be retained by setting a threshold. Algorithm for establishing a linked list for edges:

[0089] (1) For each pixel, if the grayscale value of the pixel is invalid, continue searching until a valid pixel that has not been marked is encountered. If the grayscale of the pixel is a valid value and has not been marked, add the pixel to the linked list and mark the position of the pixel to indicate that it has been linked. The next time the element is accessed, it will be skipped. This starts a linked list.

[0090] (2) Search the pixels in the 8 adjacent positions of the pixel. If a valid pixel is found that is not marked, add the pixel to the current linked list and mark the pixel to indicate that it has been linked. Return to this step and continue until a pixel is found to be valid and not marked, but there are no valid and unmarked pixels in its 8 adjacent pixels. This chaining is terminated, a linked list is completed, and go to step (1) to continue searching until all pixels have been searched.

[0091] Step 7: Correlation Processing

[0092] Perform correlation processing on these chained edges, and calculate the correlation between the edges of the main object and the ghost edges respectively. If the correlation is greater than a certain threshold, the two edges are said to be similar, and thus there is a ghost in the image. The positions of the two edges are the positions of the ghost and the corresponding main object. If the correlation is less than a certain threshold, there is no ghost in the image.

[0093] The algorithm for correlation processing proposed here is as follows:

[0094] (1) First, sort the linked lists containing the main object edges and the ghost edges separately, with the longest edge at the front, and so on.

[0095] (2) First, determine the edge list A of the main object

[0096] (3) Then take the ghost edge list B in turn, and then translate this list to the list taken in (2).

[0097] (4) After the translation, find the distance between the corresponding points of the two linked lists, sum the distances of all corresponding points, and then divide the sum by the length of the linked lists to obtain a value Diff (the smaller the Diff, the greater the similarity between the two edges). If Diff is less than a certain threshold T, the ghost image is found and the algorithm ends.

[0098] Otherwise, go to (3) and continue. If all the ghost linked lists are not similar to linked list A, then change linked list A to the next main object linked list and continue to go to (3).

[0099] The above embodiments of the present invention are for illustration only and do not represent the quality of the embodiments. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0100] The present invention is described in detail through the above-mentioned implementation modes, and those skilled in the art can clearly understand the embodiment methods. The above implementation modes are not used to limit the present invention, and the present invention is not limited to the above-mentioned embodiments. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An automatic ghost detection algorithm for SAR images, characterized in that: The algorithm includes: Step 1: Filter the SAR image; Step 2: Obtain the statistical features of the filtered SAR image, and calculate two thresholds T1 and T2 of the main object and the ghost according to the statistical features; Step 3: According to T1 and T2, the SAR image is subjected to binarization processing of the main object and the ghost respectively, so as to form a binary image of the main object and the ghost; Step 4: dilate the above two binary images; Step 5: Extract the edges of the image after dilation processing to obtain the edge of the main object and the edge of the ghost; Step 6: Filter the extracted main object edges and ghost edges respectively to remove short and invalid edges; Step 7: Calculate the correlation between the edge of the main object and the ghost edge after screening, set the threshold T, and use the threshold to determine whether there is a ghost in the SAR image. If so, mark the ghost area and its corresponding main object position on the image.

2. The SAR image ghosting automatic detection algorithm according to claim 1, characterized in that: The step 1 further comprises the following sub-steps: Step S11: Modeling the SAR image as F=(T+N)·S; wherein F is the received backscattering intensity, T is the signal intensity, N is the additive noise, and S is the multiplicative speckle noise; Step S12: converting the multiplicative speckle noise into additive noise by logarithmic transformation: log(F)=log(T+N)+log(S); Step S13: filtering out the speckle noise, and obtaining a SAR image with the speckle removed by exponential transformation.

3. The SAR image ghosting automatic detection algorithm according to claim 1, characterized in that: The statistical features of the SAR image obtained in step 2 include: the mean, median, 75th percentile, maximum value of the image grayscale value, or the value range of a certain type of objects in the image, and a threshold T1 for distinguishing the main object from the ghost, and a threshold T2 for distinguishing the ghost from the background are established based on the statistical features.

4. The SAR image ghosting automatic detection algorithm according to claim 1, characterized in that: The step 3 can be divided into two parts: (1) The following binarization algorithm is performed on the SAR image according to the threshold T1: For each pixel if (current pixel gray value>T1) Current pixel grayscale value = 65535; else Current pixel gray value = 0 Obtain a binary image of the SAR image with respect to the main object; (2) According to the thresholds T1 and T2, the SAR image is subjected to the following binarization algorithm: For each pixel if (current pixel gray value>T1) Current pixel gray value = 0; else if (current pixel gray value>=T2) Current pixel grayscale value = 65535; else Current pixel gray value = 0 The binary image of the SAR image ghost is obtained.

5. The SAR image ghosting automatic detection algorithm according to claim 1, characterized in that: The specific algorithm of the expansion process in step 4 is: For each pixel: if (current pixel grayscale value = 65535) The grayscale value of the 8 pixels around the current pixel = 65535.

6. The SAR image ghosting automatic detection algorithm according to claim 1, characterized in that: The edge extraction algorithm in step 5 uses an edge extraction operator to perform convolution calculation on the expanded image to extract the edge of the main object and the edge of the ghost.

7. The SAR image ghosting automatic detection algorithm according to claim 1, characterized in that: The step 6 further comprises the following sub-steps: Step S61: storing the edges of the main object and the ghost image and calculating their lengths; Step S62: Set a length threshold K, retain edges with a length greater than K, and remove edges with a length less than or equal to K.

8. The SAR image ghosting automatic detection algorithm according to claim 1, characterized in that: The step 7 further comprises the following sub-steps: Step S71: dividing the edges retained in step 6 into two categories: main object edges and ghost edges; Step S72: sorting the main object edge and the ghost edge from long to short according to their lengths; Step S73: Calculate the correlation between each main object edge and each ghost edge according to the sequence; Step S74: setting a correlation threshold T, and judging whether there is ghosting in the SAR image by the threshold. If there is, marking the ghosting area and the corresponding main object position on the image.

Images