SAR image adaptive threshold segmentation method and device

Abstract

The invention discloses an SAR image adaptive threshold segmentation method and device, and relates to the technical field of remote sensing image processing.The method comprises the steps that firstly, a large target area in an SAR image is extracted through rough threshold segmentation, a background point is automatically selected according to a target in the image, and a section line from the center of the target to the background point is extracted; a threshold value corresponding to the target is calculated according to the section line, adaptive blocking is carried out on the image according to the target, a local threshold value is automatically selected, and adaptive threshold value segmentation of the SAR image is realized; compared with an existing self-adaptive threshold segmentation method and a semi-automatic global threshold segmentation method, the image is subjected to self-adaptive partitioning by taking a large target as a center, a section line crossing the target and the background is automatically drawn for the partitioned image, and a local threshold is automatically extracted; according to the method, the problems of regular grid division, automatic threshold extraction, manual line drawing and the like in the existing method are solved, the SAR image threshold segmentation effect is greatly improved, and the method can be applied to water extraction, shadow detection, bright target detection and the like.

Description

Technical Field

[0001] The invention relates to the technical field of remote sensing image processing, and in particular to a SAR image adaptive threshold segmentation method and device. Background Art

[0002] In reality, compared with optical remote sensing images, SAR images have the advantages of penetrating clouds and fog and collecting information all day and all weather; quickly and accurately obtaining ground information from SAR images is of great significance to surface monitoring, ecological assessment, disaster monitoring and early warning, etc.; threshold segmentation of SAR images is currently the most basic and most commonly used technical method for extracting ground object information; this method uses different thresholds to distinguish different ground objects in the image and extract the target ground object from the background; for example: water body information extraction, coastline extraction, etc.; however, there is a large amount of coherent speckle noise in SAR images and this noise is multiplicative noise. Compared with the additive noise handled in the optical field, the removal of multiplicative noise is more complicated. Coupled with the unique statistical distribution characteristics of SAR, threshold segmentation of SAR images is more difficult than threshold segmentation of optical images.

[0003] In practical applications, methods for obtaining thresholds can be divided into two categories: manual extraction and automatic extraction. The manual extraction method requires the operator to check the intensity value between the target object and the background many times, and roughly give a threshold based on past experience. The image also needs to be tested and compared repeatedly. This method relies purely on manual visual interpretation. Although it can sometimes achieve good accuracy, the workload is large and the final threshold is random. When a large number of SAR images are used for threshold selection, this method is less efficient. The automatic threshold extraction method is more stable and objective than manual extraction, and data can be processed in batches. It is common to improve the OTSU method and apply it to SAR images. However, due to the unique statistical characteristics of SAR images, the SAR intensity obeys a unimodal negative exponential distribution and there is no valley value, so the segmentation effect is not ideal.

[0004] In response to the above problems, some researchers have proposed a semi-automatic threshold segmentation method. This method requires manual drawing of one or more straight lines across the water and land areas, and then uses a greedy algorithm to automatically determine the boundary between bright and dark targets, and uses the pixel value at the boundary as the threshold. However, this method requires manual line drawing, and the position and number of the lines are subjective. At the same time, the global threshold is used, which makes it difficult to segment when the boundaries between bright and dark targets in different areas are very different. Therefore, the current threshold segmentation method has poor results and is difficult to apply to water body extraction, shadow detection, bright target detection and other directions. Summary of the invention

[0005] The embodiment of the present invention provides a SAR image adaptive threshold segmentation method and device, which can solve the problem that the threshold segmentation method currently adopted in the prior art has a poor effect.

[0006] The embodiment of the present invention provides a SAR image adaptive threshold segmentation method and device, comprising the following steps:

[0007] Obtain signal strength or amplitude maps of SAR images;

[0008] Extracting dark targets from the binary image of the signal intensity map or amplitude map of the SAR image, determining the minimum bounding rectangle corresponding to the dark target, and performing a binary operation on the dark target according to the minimum bounding rectangle corresponding to the dark target to obtain the divided dark target;

[0009] For each dark target after segmentation, a background point is selected, and a profile line from the center of the dark target after segmentation to the background point is extracted, and a local threshold of the dark target of the corresponding segmentation is determined according to the profile line;

[0010] Divide the signal intensity map or amplitude map of the SAR image into blocks to obtain a block image;

[0011] Detect the block dark target in each block image. If there is only one block dark target, the local threshold corresponding to the block dark target is used as the segmentation threshold of the block image; if there are multiple block dark targets, the local threshold corresponding to the largest block dark target is used as the segmentation threshold of the block image; if there is no block dark target, the local threshold corresponding to the block dark target closest to the geometric center of the block image is used as the segmentation threshold of the block image;

[0012] Each block image is binarized according to the corresponding local threshold to obtain the threshold segmentation image of the SAR image.

[0013] Preferably, the determining of the minimum bounding rectangle corresponding to the dark target includes:

[0014] Determine the boundary points of the dark target object, and form a point set of the dark target with all the boundary points;

[0015] The minAreaRect function in OpenCV software is used to calculate the point set of the dark target to obtain the minimum enclosing rectangle of the dark target.

[0016] Preferably, the binary operation of the dark object according to the minimum circumscribed rectangle corresponding to the dark object comprises the following steps:

[0017] Calculate the minimum bounding rectangle of each target in the SAR image;

[0018] If the aspect ratio of the smallest circumscribed rectangle is ≥ 2, divide the rectangle into two equal parts along the direction perpendicular to the long side;

[0019] Generate a scan line in the length and width directions of the minimum circumscribed rectangle, count the number of pixel value changes along the scan line, and if the number of changes is greater than twice, divide the rectangle into two parts along the scan line;

[0020] The new target obtained after the binary division is judged again. If it meets the binary division conditions, the binary division is continued until all targets do not meet the binary division conditions.

[0021] Preferably, the determining of the local threshold corresponding to the dark target block comprises the following steps:

[0022] Get the background points corresponding to each dark target in the segmented SAR image. If the number of background points is 0, skip the dark target and continue to calculate the threshold between the next dark target and the background.

[0023] Connect the geometric center of the minimum circumscribed rectangle corresponding to the dark target with the background point, and estimate the image profile line through which the connecting line passes;

[0024] According to the gradient function, the gradients of adjacent pixels are calculated starting from both ends of the profile line;

[0025] According to the gradient, pixels are merged from both ends of the profile line to the center and the gradient is updated until the pixel pairs meet;

[0026] The mean of the pixel pairs at the meeting point is taken as the threshold, and the average of the thresholds corresponding to all background points is taken as the local threshold of the block image target.

[0027] Preferably, the step of obtaining the background point corresponding to each dark target in the segmented SAR image comprises the following steps:

[0028] For each dark target extracted, calculate its minimum circumscribed rectangle, and further calculate the image coordinates of the long and short side lengths a, b and the geometric center o;

[0029] Take two points along the short side whose distance from the geometric center o is equal to the short side length b. They are defined as candidate background points.

[0030] Starting from the geometric center o, the number of pixel changes is determined along the line connecting the candidate background point and the geometric center. If there is no change, the background point is removed; if it changes only once, the point is determined as a background point; if the number of changes reaches 2 times or more, the midpoint from the first change to the second change position is taken as the background point;

[0031] If the number of background points is 0, then take the point whose distance from the geometric center o to the long side length a along the long side direction and define it as the candidate background point.

[0032] Preferably, the step of estimating the image profile line through which the connecting line passes comprises the following steps:

[0033] A buffer zone is automatically generated on both sides of the line. The buffer zone takes the drawn line as the central axis, and the distance between the two ends and the central axis is L.

[0034] Extract the pixel values ​​covered by the buffer zone and calculate the mean of all pixels on the vertical line of the central axis to form a mean line, which is the estimated profile line.

[0035] Preferably, the calculating the gradient of adjacent pixels comprises the following steps:

[0036] Let the section line be X, x=(x1, x2, x3, ...x n ) is a set of pixels on the profile line X, the total number of pixels is n, and the gradient of adjacent pixels is calculated using the gradient function; where the gradient function used for the head end is:

[0037] f(M k ,x k+1 )=|x k+1 -M k | k=1,2,...,n-1

[0038] Where M1=x1,M k is the weighted mean, and the formula is:

[0039]

[0040] The gradient function used at the tail end is:

[0041] f(x h-1 ,M h )=|x h-1 -M h |h=n,n-1,...,2

[0042] Among them, M n =x n , M h is the weighted mean, and the formula is:

[0043]

[0044] Preferably, the step of combining pixels from both ends of the profile line toward the center and updating the gradient according to the gradient until the pixel pairs meet includes:

[0045] Using the calculated gradient f as the comparison criterion, the end with the smaller gradient on the left and right ends is merged forward by one pixel, and the gradient of the merged pixel set and its adjacent pixels is recalculated. The larger end is not merged, and comparisons are made successively until the left and right ends meet.

[0046] Preferably, the obtaining of the block images comprises:

[0047] The basic block size B is set. For each dark target, the judgment is made based on the minimum bounding rectangle of the dark target. When any side length of the bounding rectangle of the dark target is greater than B, the SAR image range corresponding to the bounding rectangle is set as a block image. When two or more bounding rectangles overlap, the overlapping bounding rectangles are merged into one rectangle. For the remaining part of the SAR image, the blocks are divided according to the regular grid.

[0048] The embodiment of the present invention further provides a SAR image adaptive threshold segmentation device, comprising:

[0049] SAR imaging module, used to obtain the signal strength map or amplitude map of SAR images;

[0050] An image segmentation module is used to extract dark targets from the binary image of the signal intensity map or amplitude map of the SAR image, determine the minimum bounding rectangle corresponding to the dark target, and perform a binary operation on the dark target according to the minimum bounding rectangle corresponding to the dark target to obtain the dark target after segmentation;

[0051] For each dark target after segmentation, a background point is selected, and a profile line from the center of the dark target after segmentation to the background point is extracted, and a local threshold of the dark target of the corresponding segmentation is determined according to the profile line;

[0052] Divide the signal intensity map or amplitude map of the SAR image into blocks to obtain a block image;

[0053] The threshold acquisition module is used to detect the block dark target in each block image. If there is only one block dark target, the local threshold corresponding to the block dark target is used as the segmentation threshold of the block image; if there are multiple block dark targets, the local threshold corresponding to the largest block dark target is used as the segmentation threshold of the block image; if there is no block dark target, the local threshold corresponding to the block dark target closest to the geometric center of the block image is used as the segmentation threshold of the block image;

[0054] The image segmentation module is used to binarize each block image according to the corresponding local threshold value to obtain the threshold segmentation image of the SAR image.

[0055] The embodiment of the present invention provides a SAR image adaptive threshold segmentation method and device, which has the following beneficial effects compared with the prior art:

[0056] The present invention first extracts a larger target area in the SAR image through rough threshold segmentation and opening and closing operations, then automatically selects background points and extracts the profile line from the target center to the background point according to the target in the image, and calculates the threshold corresponding to the target based on the profile line, and finally adaptively divides the image into blocks according to the target, and automatically selects the local threshold to achieve adaptive threshold segmentation of the SAR image. Compared with manually drawing lines on the entire image and taking a global threshold, the present invention divides the image into blocks and then automatically draws lines on the block images using background points and profile lines, thereby overcoming the subjectivity and arbitrariness of manual work, and the fewer dark targets in the image after the block can make the boundary difference between bright and dark targets smaller, making it very easy to segment, and finally the segmentation threshold is used to binarize the block images and combine them to obtain the segmentation result of the SAR image, which greatly improves the effect of SAR image threshold segmentation and can be applied to water body extraction, shadow detection, bright target detection and other directions.

[0057] Moreover, compared with global single threshold segmentation, the present invention adopts automatic image segmentation and performs multi-threshold extraction and segmentation, which can effectively improve the accuracy of light and dark boundary extraction detection, reduce the false detection rate, have better detection effect, and have stronger stability.

[0058] Moreover, compared with the adaptive threshold segmentation of regular grid blocking, the present invention performs size adaptive blocking for larger targets in the image, and through the two steps of coarse threshold segmentation and precise threshold segmentation, it can effectively avoid the problem of large segmentation deviation in the automatic segmentation method, which may occur in the regular grid blocking, that is, the blocks are full of bright areas or dark areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 The present invention provides a schematic diagram of the overall process of a SAR image adaptive threshold segmentation method and device. DETAILED DESCRIPTION

[0060] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific embodiments disclosed below.

[0061] See also Figure 1 The embodiment of the present invention provides a SAR image adaptive threshold segmentation method, comprising the following steps:

[0062] Step 1: Preliminary binarization:

[0063] Calculate the intensity map or amplitude map of the original SAR image and perform preliminary binarization based on the Otsu method.

[0064] Step 2: Dark target extraction and segmentation:

[0065] Erosion and dilation processing is performed on the binary image to remove noise and small objects while retaining larger objects in the image; dark objects are extracted and their minimum bounding rectangles are calculated. For dark objects with complex shapes or long and narrow shapes, binary operations are performed based on their minimum bounding rectangles.

[0066] Specifically, the method for performing binary operation on dark targets is:

[0067] Generate scan lines from the length and width of the minimum circumscribed rectangle, and count the number of pixel value changes along the scan lines. If the number of changes is greater than twice, or the aspect ratio of the minimum circumscribed rectangle is ≥2, divide the rectangle into two equal parts along the direction perpendicular to the long side. At this time, the dark target inside the rectangle is also divided into two parts; judge the new dark target obtained after the binary division again. If it meets the binary division conditions, continue to binary division until all dark targets do not meet the binary division conditions.

[0068] Step 3: Calculation of local threshold and its position:

[0069] The threshold between the dark target and the background after each block is calculated, and the geometric center of the smallest circumscribed rectangle of the dark target is used as the position of the threshold.

[0070] Specifically, the method for calculating the local threshold is:

[0071] ① Background point selection: For each dark target extracted, two background points are calculated and selected.

[0072] ② Automatic connection: Automatically connect the geometric center of the minimum circumscribed rectangle and the background point.

[0073] ③ Profile estimation: Estimate the values ​​of the image profile pixels through which the connecting line passes.

[0074] ④ Gradient calculation: Starting from the beginning and end of the profile line, calculate the gradient f of adjacent pixels according to the gradient function.

[0075] ⑤Pixel merging: Merge pixels from both ends of the profile line toward the center according to the gradient, and update the gradient until they meet.

[0076] ⑥Threshold calculation: The threshold is calculated based on the value of the pixel pair at the meeting point and the gradient. Since there are two background points, there are two corresponding thresholds. Finally, the average value of these two thresholds is taken as the local threshold corresponding to the dark target.

[0077] Specifically, the above-mentioned method for selecting background points is:

[0078] For each dark target extracted, calculate the short side length l of its minimum circumscribed rectangle and the coordinates of the geometric center o, and then select points along the short side direction whose distance from the geometric center is the short side length. There are two such points, which are defined as candidate background points.

[0079] Starting from the geometric center o, the number of pixel changes is determined along the line connecting the candidate background point and the geometric center. If it changes only once, the point is determined as a background point; if the number of changes reaches 2 times or more, the midpoint from the first change to the second change position is taken as the background point.

[0080] Specifically, the above-mentioned profile estimation method is:

[0081] A buffer zone is automatically generated on both sides of the line. The buffer zone takes the drawn line as the central axis, and the distance between the two ends and the central axis is L.

[0082] Extract the pixel values ​​covered by the buffer zone, calculate the mean of all pixels on the vertical line of the central axis, and form a mean line, which is the estimated profile line.

[0083] Specifically, the above-mentioned gradient calculation method is:

[0084] Let the section line be X, x=(x1, x2, x3, ...x n ) is a set of pixels on the profile line X, the total number of pixels is n, and the gradient of adjacent pixels is calculated using the gradient function; where the gradient function used for the head end is:

[0085] f(M k ,x k+1 )=|x k+1 -M k | k=1,2,...,n-1

[0086] Where: M1 = x1; M k is the weighted mean, and the formula is:

[0087]

[0088] The gradient function used at the tail end is:

[0089] f(x h-1 ,M h )=|x h-1 -M h |h=n,n-1,...,2

[0090] Where: M n =x n ;M h is the weighted mean, and the formula is:

[0091]

[0092] Specifically, the above pixel merging method is:

[0093] Using the calculated gradient f as the comparison criterion, the end with the smaller gradient on the left and right ends merges one pixel forward, and recalculates the gradient of the merged pixel set and its adjacent pixels. The larger end is not merged, and this process is repeated until the left and right ends meet.

[0094] Specifically, the above threshold calculation method is:

[0095] Perform multiple operations on the values ​​and gradients of the pixel pairs at the intersection to obtain several thresholds. Users can select the best threshold according to different application scenarios. There are four calculation methods:

[0096] ① When the encounter pixel pair (x i , x i+1 ), the arithmetic mean of the two pixels is calculated as the final threshold T; the formula is as follows:

[0097]

[0098] ② When the encounter pixel pair (x i , x i+1 ), the smaller value of the two pixels is selected as the final threshold T.

[0099] ③When the encounter pixel pair (x i , x i+1 ), the weighted average of the two pixels is calculated as the final threshold T; its formula is as follows:

[0100]

[0101] ④ Sort the gradients calculated on the left and right sides of the profile line from small to large, select j gradients from the larger side, and calculate the arithmetic mean of the 2j pixels corresponding to the j gradients as the final threshold T.

[0102] Step 4: Image segmentation:

[0103] Divide the SAR intensity map or amplitude map into blocks.

[0104] Specifically, the image segmentation method is:

[0105] Set the basic block size B. For the minimum bounding rectangle of the dark target detected in the previous step, if any side length is greater than B, extract the original image range corresponding to the rectangle as a block image. If two or more minimum bounding rectangles overlap, merge the overlapping bounding rectangles into one block.

[0106] The remaining image parts in the original image are divided according to the regular grid. The specific division method is as follows:

[0107] Assuming that the number of rows of the entire image is H and the number of columns is W, the number of blocks in the column direction is l=round(H / B), where round means rounding off, and the number of blocks in the row direction is r=round(W / B).

[0108] If l=0, then let l=1, and if r=0, then let r=1.

[0109] Then the number of rows of the actual block image is h=round(H / l), and the number of columns is w=round(W / r).

[0110] Finally, the remaining part of the entire image is divided into blocks of size h*w. Due to rounding, the actual block size of the last row and the last column will change slightly. The number of rows of the last row of block images is: Hh*l, and the number of columns of the last column of block images is: Ww*r.

[0111] Step 5: Image binarization:

[0112] For each SAR image after segmentation, a suitable local threshold is selected as the threshold of the block image, and the threshold is used to binarize the SAR image after segmentation. After each block image is binarized, the entire image is binarized.

[0113] Specifically, the threshold selection method for the block image is:

[0114] Detect the dark target in each block image. If there is only one dark target, the threshold corresponding to the dark target is used as the segmentation threshold of the block image; if there are multiple dark targets, the threshold corresponding to the largest dark target is used as the segmentation threshold of the block image; if there is no dark target, the threshold corresponding to the dark target closest to the geometric center of the block image is used as the segmentation threshold of the block image.

[0115] The purpose of the present invention is to quickly and accurately obtain multiple local thresholds for distinguishing target objects from backgrounds through technical means such as image segmentation, selection of bright and dark target points, and automatic threshold extraction, and automatically perform adaptive image segmentation, thereby effectively solving the problem that the threshold is difficult to automatically determine in the current SAR image threshold segmentation.

[0116] Compared with manual threshold selection, especially when processing large quantities of data, the method of the present invention effectively overcomes the disadvantage of excessive subjectivity of manually obtained results and greatly improves the efficiency and accuracy of threshold extraction.

[0117] Compared with semi-automatic threshold selection, the method of the present invention automates the marking and blocking process, further reducing the random influence that may be caused by human participation, obtaining higher-precision extraction results, and can be applied to images of different sensors, different regions, and different resolutions.

[0118] Compared with global single threshold segmentation, the method of the present invention uses automatic image segmentation and performs multi-threshold extraction and segmentation, which can effectively improve the accuracy of light and dark boundary extraction detection, reduce the false detection rate, have better detection effect, and have better stability.

[0119] The embodiment of the present invention further provides a SAR image adaptive threshold segmentation device, comprising:

[0120] The SAR image module is used to obtain the signal strength map or amplitude map of the SAR image.

[0121] The image segmentation module is used to extract dark targets in the binary image of the signal intensity map or amplitude map of the SAR image, determine the minimum bounding rectangle corresponding to the dark target, and perform a binary operation on the dark target according to the minimum bounding rectangle corresponding to the dark target to obtain the dark target after segmentation.

[0122] For each dark target after segmentation, background points are selected, and the profile line from the center of the dark target after segmentation to the background point is extracted, and the local threshold of the dark target after segmentation is determined according to the profile line.

[0123] The signal intensity map or amplitude map of the SAR image is divided into blocks to obtain a block image.

[0124] The threshold acquisition module is used to detect the block dark targets in each block image. If there is only one block dark target, the local threshold corresponding to the block dark target is used as the segmentation threshold of the block image; if there are multiple block dark targets, the local threshold corresponding to the largest block dark target is used as the segmentation threshold of the block image; if there is no block dark target, the local threshold corresponding to the block dark target closest to the geometric center of the block image is used as the segmentation threshold of the block image.

[0125] The image segmentation module is used to binarize each block image according to the corresponding local threshold value to obtain the threshold segmentation image of the SAR image.

[0126] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A SAR image adaptive threshold segmentation method, characterized in that: The following steps are involved: Obtain signal strength or amplitude maps of SAR images; Extracting dark targets from the binary image of the signal intensity map or amplitude map of the SAR image, determining the minimum bounding rectangle corresponding to the dark target, and performing a binary operation on the dark target according to the minimum bounding rectangle corresponding to the dark target to obtain the divided dark target; For each dark target after segmentation, a background point is selected, and a profile line from the center of the dark target after segmentation to the background point is extracted, and a local threshold of the dark target of the corresponding segmentation is determined according to the profile line; Divide the signal intensity map or amplitude map of the SAR image into blocks to obtain a block image; Detect the dark target in each block image. If there is only one dark target, use the local threshold corresponding to the dark target as the segmentation threshold of the block image. If there are multiple dark objects in blocks, the local threshold corresponding to the largest dark object in the block is used as the segmentation threshold of the block image; If there is no dark target in the block, the local threshold corresponding to the dark target in the block closest to the geometric center of the block image is used as the segmentation threshold of the block image; Each block image is binarized according to the corresponding local threshold to obtain the threshold segmentation image of the SAR image.

2. The SAR image adaptive threshold segmentation method according to claim 1, characterized in that: The step of determining the minimum bounding rectangle corresponding to the dark target includes: Determine the boundary points of the dark target object, and form a point set of the dark target with all the boundary points; The minAreaRect function in OpenCV software is used to calculate the point set of the dark target to obtain the minimum enclosing rectangle of the dark target.

3. The SAR image adaptive threshold segmentation method according to claim 1, characterized in that: The binary operation of the dark object according to the minimum circumscribed rectangle corresponding to the dark object comprises the following steps: Calculate the minimum bounding rectangle of each target in the SAR image; If the aspect ratio of the smallest circumscribed rectangle is ≥ 2, divide the rectangle into two equal parts along the direction perpendicular to the long side; Generate a scan line in the length and width directions of the minimum circumscribed rectangle, count the number of pixel value changes along the scan line, and if the number of changes is greater than twice, divide the rectangle into two parts along the scan line; The new target obtained after the binary division is judged again. If it meets the binary division conditions, the binary division is continued until all targets do not meet the binary division conditions.

4. The SAR image adaptive threshold segmentation method according to claim 1, characterized in that: The method of determining the local threshold of the dark target of the corresponding block includes the following steps: Get the background points corresponding to each dark target in the segmented SAR image. If the number of background points is 0, skip the dark target and continue to calculate the threshold between the next dark target and the background. Connect the geometric center of the minimum circumscribed rectangle corresponding to the dark target with the background point, and estimate the image profile line through which the connecting line passes; According to the gradient function, the gradients of adjacent pixels are calculated starting from both ends of the profile line; According to the gradient, pixels are merged from both ends of the profile line to the center and the gradient is updated until the pixel pairs meet; The mean of the pixel pairs at the meeting point is taken as the threshold, and the average of the thresholds corresponding to all background points is taken as the local threshold of the block image target.

5. The SAR image adaptive threshold segmentation method according to claim 4, characterized in that: The method of obtaining the background point corresponding to each dark target in the segmented SAR image comprises the following steps: For each dark target extracted, calculate its minimum circumscribed rectangle, and further calculate the image coordinates of the long and short side lengths a, b and the geometric center o; Take two points along the short side whose distance from the geometric center o is equal to the short side length b. They are defined as candidate background points. Starting from the geometric center o, the number of pixel changes is determined along the line connecting the candidate background point and the geometric center. If there is no change, the background point is removed; if it changes only once, the point is determined as a background point; if the number of changes reaches 2 times or more, the midpoint from the first change to the second change position is taken as the background point; If the number of background points is 0, then take the point whose distance from the geometric center o to the long side length a along the long side direction and define it as the candidate background point.

6. The SAR image adaptive threshold segmentation method according to claim 4, characterized in that: The step of estimating the image profile line through which the connecting line passes comprises the following steps: A buffer zone is automatically generated on both sides of the line. The buffer zone takes the drawn line as the central axis, and the distance between the two ends and the central axis is L. Extract the pixel values ​​covered by the buffer zone and calculate the mean of all pixels on the vertical line of the central axis to form a mean line, which is the estimated profile line.

7. The SAR image adaptive threshold segmentation method according to claim 4, characterized in that: The step of calculating the gradient of adjacent pixels comprises the following steps: Let the section line be X, x=(x1, x2, x3, ...x n ) is a set of pixels on the profile line X, the total number of pixels is n, and the gradient of adjacent pixels is calculated using the gradient function; where the gradient function used for the head end is: f(M k ,x k+1 )=|x k+1 -M k | k=1,2,...,n-1 Where M1=x1,M k is the weighted mean, and the formula is: The gradient function used at the tail end is: f(x h-1 ,M h )=|x h-1 -M h |h=n,n-1,...,2 Among them, M n =x n , M h is the weighted mean, and the formula is:

8. The SAR image adaptive threshold segmentation method according to claim 4, characterized in that: The step of combining pixels from both ends of the profile line toward the center and updating the gradient according to the gradient until the pixel pairs meet includes: Using the calculated gradient f as the comparison criterion, the end with the smaller gradient on the left and right ends is merged forward by one pixel, and the gradient of the merged pixel set and its adjacent pixels is recalculated. The larger end is not merged, and comparisons are made successively until the left and right ends meet.

9. The SAR image adaptive threshold segmentation method according to claim 1, characterized in that: The step of obtaining the block images comprises: The basic block size B is set. For each dark target, the judgment is made based on the smallest bounding rectangle of the dark target. When any side length of the bounding rectangle of the dark target is greater than B, the SAR image range corresponding to the bounding rectangle is set as a block image. When two or more bounding rectangles overlap, the overlapping bounding rectangles are merged into one rectangle. For the remaining part of the SAR image, the blocks are divided according to the regular grid.

10. A SAR image adaptive threshold segmentation device, characterized in that: include: SAR imaging module, used to obtain the signal strength map or amplitude map of SAR images; An image segmentation module is used to extract dark targets from the binary image of the signal intensity map or amplitude map of the SAR image, determine the minimum bounding rectangle corresponding to the dark target, and perform a binary operation on the dark target according to the minimum bounding rectangle corresponding to the dark target to obtain the dark target after segmentation; For each dark target after segmentation, a background point is selected, and a profile line from the center of the dark target after segmentation to the background point is extracted, and a local threshold of the dark target of the corresponding segmentation is determined according to the profile line; Divide the signal intensity map or amplitude map of the SAR image into blocks to obtain a block image; A threshold acquisition module is used to detect the dark target in each block image. If there is only one dark target, the local threshold corresponding to the dark target is used as the segmentation threshold of the block image. If there are multiple dark objects in blocks, the local threshold corresponding to the largest dark object in the block is used as the segmentation threshold of the block image; If there is no dark target in the block, the local threshold corresponding to the dark target in the block closest to the geometric center of the block image is used as the segmentation threshold of the block image; The image segmentation module is used to binarize each block image according to the corresponding local threshold value to obtain the threshold segmentation image of the SAR image.

Images