CN113192078B - Rapid sea surface SAR image superpixel segmentation method based on density - Google Patents

Rapid sea surface SAR image superpixel segmentation method based on density Download PDF

Info

Publication number
CN113192078B
CN113192078B CN202110420778.8A CN202110420778A CN113192078B CN 113192078 B CN113192078 B CN 113192078B CN 202110420778 A CN202110420778 A CN 202110420778A CN 113192078 B CN113192078 B CN 113192078B
Authority
CN
China
Prior art keywords
image
sub
pixel
sea surface
density
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110420778.8A
Other languages
Chinese (zh)
Other versions
CN113192078A (en
Inventor
李刚
王学谦
刘瑜
何友
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tsinghua University
Naval Aeronautical University
Original Assignee
Tsinghua University
Naval Aeronautical University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Tsinghua University, Naval Aeronautical University filed Critical Tsinghua University
Priority to CN202110420778.8A priority Critical patent/CN113192078B/en
Publication of CN113192078A publication Critical patent/CN113192078A/en
Application granted granted Critical
Publication of CN113192078B publication Critical patent/CN113192078B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10032—Satellite or aerial image; Remote sensing
    • G06T2207/10044—Radar image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Radar Systems Or Details Thereof (AREA)
  • Image Analysis (AREA)

Abstract

The invention provides a density-based fast sea surface SAR image superpixel segmentation method, and belongs to the field of synthetic aperture radar image processing. In the training stage, pure clutter sub-images and pure sea surface target-containing sub-images with the same size are obtained, the density vector of each sub-image is calculated, a weighted sparse optimization problem is established and solved, and an optimal weight vector and an optimal deviation are obtained; in the testing stage, acquiring a sea surface SAR image and dividing the sea surface SAR image into a plurality of sub-images, calculating a corresponding label of each sub-image by using the optimal weight vector and the optimal deviation, and judging whether the sub-image possibly contains a target or not; and performing superpixel segmentation on all sub-images possibly containing the target to obtain a final superpixel segmentation result of the sea surface SAR image. According to the method, a large number of sea clutter areas are deleted quickly before the SAR image superpixel segmentation, and then the remaining areas are finely segmented, so that the calculation speed and the storage efficiency of the conventional superpixel segmentation method are improved remarkably.

Description

Rapid sea surface SAR image superpixel segmentation method based on density
Technical Field
The invention belongs to the field of Synthetic Aperture Radar (SAR) image processing, and particularly relates to a density-based fast sea surface SAR image superpixel segmentation method which can be used for superpixel generation of a sea surface SAR image, namely preprocessing before detection at present and accelerating the target segmentation and detection process.
Background
A Synthetic Aperture Radar (SAR) belongs to an active imaging sensor and can provide high-resolution images of targets such as sea surface ships and the like. Compared with passive sensors such as optical sensors, infrared sensors and the like, the SAR image is not affected by illumination and weather, and has all-weather and all-time monitoring capability. The super-pixel segmentation of the SAR image is beneficial to improving the speed of subsequent target detection and reducing the influence of speckle noise, and has important application in military civil aspects such as military sea defense, sustainable fishery and the like.
In recent years, many experts and scholars have proposed various methods for super-pixel segmentation of SAR images. However, the existing superpixel segmentation method ignores the characteristic that the density of the target pixel in the sea-level SAR image is extremely low, and results in a lower segmentation speed and a larger data storage burden.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a density-based fast sea surface SAR image superpixel segmentation method. According to the method, a large number of sea clutter areas are deleted quickly before the SAR image superpixel segmentation, and then the remaining areas are finely segmented, so that the calculation speed and the storage efficiency of the conventional superpixel segmentation method are improved remarkably.
The invention provides a density-based fast sea surface SAR image superpixel segmentation method which is characterized by comprising a training stage and a testing stage; in the training stage, firstly, a plurality of pure clutter sub-images with the same size and a plurality of pure sub-images containing sea surface targets are respectively obtained, and the density vector of each sub-image is calculated; then, establishing a weighted sparse optimization problem based on the density vectors of all the sub-images and solving the weighted sparse optimization problem to obtain an optimal weight vector and an optimal deviation; in the testing stage, acquiring a sea surface SAR image and dividing the sea surface SAR image into a plurality of subimages, calculating a label corresponding to each subimage of the sea surface SAR image by using the optimal weight vector and the optimal deviation, and judging whether the subimages possibly contain a sea surface target or not through the label; and finally, performing superpixel segmentation on all sub-images possibly containing the sea surface target to obtain a final superpixel segmentation result of the sea surface SAR image. The method comprises the following steps:
1) A training stage;
1-1) determining sparsity K and determining a scale factor alpha epsilon (0,1);
respectively obtain Q 1 Zhang Chun clutter sub-image and Q 2 Two pure subimages containing sea surface targets are identical in size, and the size of each subimage is M 1 ×M 2 Wherein M is 1 Representing the length of the sub-picture, M 2 Represents the width of the sub-image;
setting image block size P 1 ×P 2 In which P is 1 Length, P, of representative image block 2 Width, P, of representative image block 1 Quilt M 1 Integer removal of P 2 Quilt M 2 Trimming;
1-2) dividing each sub-image according to the set image block size for the pure clutter sub-image and the pure sub-image containing the sea surface target obtained in the step 1-1), and calculating the density value of each image block in each sub-image;
for any sub-image, the density value calculation expression of each image block in the sub-image is as follows:
Figure BDA0003027760500000021
where ρ is i Indicating the density value of the ith image block in the subimage; i, j denote the indices of the different image blocks in the sub-image, N (patc) Representing the number of image blocks in the sub-picture,
Figure BDA0003027760500000022
representing the absolute value of the difference between the average gray scale of the ith image block and the average gray scale of the jth image block, d (c) Is a constant for the sub-image
Figure BDA0003027760500000023
Indi (-) represents an indicator function whose output is 1 when the function input is less than 0, and 0 otherwise;
1-3) sorting the density values of all image blocks in each sub-image from small to large by using the result of the step 1-2) to obtain a density vector of the sub-image;
for any sub-image, the density vector expression is as follows:
Figure BDA0003027760500000024
wherein
Figure BDA0003027760500000025
1-4) utilizing the results of the step 1-3) to establish the following weighted sparse optimization problem:
Figure BDA0003027760500000026
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA0003027760500000027
a set of density vectors representing all sub-images,
Figure BDA0003027760500000028
density vector representing the qth sub-image, Q =1, …, Q being the total number of sub-images, Q = Q 1 +Q 2 ；
Figure BDA0003027760500000029
A set of labels representing all of the sub-images,
Figure BDA00030277605000000210
a label representing the qth sub-image, wherein
Figure BDA00030277605000000211
Indicating that the qth sub-image contains a sea surface object,
Figure BDA00030277605000000212
representing that only sea clutter exists in the qth sub-image; Λ is a diagonal matrix of Q × Q, the qth diagonal element in Λ is
Figure BDA00030277605000000213
Figure BDA00030277605000000214
Representing the number of sea surface targets in the q sub-image;
Figure BDA00030277605000000215
representing a weight vector, b = b × 1,b representing a bias,
Figure BDA00030277605000000216
respectively representing an optimal weight vector and an optimal deviation;
1-5) solving the weighted sparse optimization problem in the step 1-4) by utilizing an orthogonal matching pursuit algorithm to obtain an optimal weight vector
Figure BDA0003027760500000031
And an optimum deviation;
2) A testing stage;
2-1) obtaining a sea surface SAR image; setting a super-pixel size S;
2-2) uniformly segmenting the sea surface SAR image in the step 2-1) into M 1 ×M 2 A plurality of sub-images of a size;
2-3) for each sub-image obtained in the step 2-2), according to the set image block size P 1 ×P 2 Dividing each sub-image, and calculating the density value of each image block in each sub-image;
for any sub-image, the density value calculation expression of each image block in the sub-image is as follows:
Figure BDA0003027760500000032
2-4) sorting the density values of all image blocks in each sub-image from small to large by using the result of the step 2-3) to obtain a density vector rho of the sub-image;
2-5) calculating the label y corresponding to each sub-image by using the result of the step 2-4),
Figure BDA0003027760500000033
then, y is determined: if y is larger than 0, keeping the sub-image corresponding to the label y; if y is less than or equal to 0, deleting the sub-image corresponding to the label y;
2-6) carrying out super-pixel segmentation on each sub-image reserved in the step 2-5) to finally obtain a super-pixel segmentation result of the sub-image; the method comprises the following specific steps:
2-6-1) uniformly setting the center of the super pixel at intervals S in the horizontal and vertical directions of each sub-image for the sub-images retained through the step 2-5); then moving the center of each super pixel to the position with the minimum gradient value in the 8 neighborhoods of the super pixel; setting kappa n = + ∞, representing the initial distance of each pixel from its corresponding superpixel center, where n represents the index value of the pixel, n =1,2, …, M 1 ×M 2 ；
2-6-2) for each superpixel center, sequentially taking each pixel in a 2S multiplied by 2S neighborhood of the superpixel center as a current pixel, and calculating the distance between the current pixel and the superpixel center, wherein the expression is as follows:
Figure BDA0003027760500000034
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA0003027760500000035
represents the spatial distance, (x) 1 ,y 1 ) And (x) 2 ,y 2 ) Respectively representing the spatial positions of the center of the superpixel and the current pixel, d (inte) ＝|δ 1 -δ 2 | represents the gray scale distance, δ 1 ,δ 2 Respectively representing the gray values of the center of the super pixel and the current pixel,
Figure BDA0003027760500000036
the distance of the density is indicated by,
Figure BDA0003027760500000037
respectively representing the density values of the center of the super pixel and the current pixel; wherein each pixel is 1 × 1 in sizeCalculating the density values of the image blocks corresponding to the super-pixel center and the current pixel by using the formula (3) to obtain
Figure BDA0003027760500000041
If D is<κ n Then the current pixel is determined to belong to the center of the superpixel and k is updated n = D; otherwise, judging that the current pixel does not belong to the center of the super pixel; after all the current pixels are judged, updating a pixel set belonging to the center of each super pixel;
2-6-3) respectively averaging the spatial positions, the gray scales and the densities of all pixels in the pixel set belonging to the center of each super pixel by using the result of the step 2-6-2) and then taking the average as the updated spatial position, the gray scale and the density of the center of the super pixel;
2-6-4) repeating the steps 2-6-2) to 2-6-3) until the upper limit T of the repetition times is reached and the pixel set belonging to the center of each super pixel is updated;
2-6-5) processing the result of the step 2-6-4) by adopting a nearest neighbor post-processing method to obtain a final pixel set corresponding to each superpixel center in all the sub-images reserved in the step 2-5), and completing the superpixel segmentation of the sea surface SAR image.
The invention has the characteristics and beneficial effects that:
the existing SAR image superpixel segmentation method has the defects that each region in the SAR image needs to be finely segmented, and the target in the sea surface SAR image only occupies a small number of pixels, so that the calculation running efficiency of the superpixel detection method is low. The invention provides a fast sea surface SAR image superpixel segmentation method based on density characteristics based on low density characteristics of ship targets in a sea surface SAR image, a large number of clutter areas in the image are rapidly filtered before fine segmentation, and then fine segmentation is carried out on the areas which are not filtered, so that the storage efficiency of the segmentation speed is remarkably accelerated, and the fast response capability of China to the ship targets on the sea surface is expected to be improved.
Drawings
FIG. 1 is an overall flow chart of the method of the present invention.
FIG. 2 is a diagram illustrating effects of an embodiment of the present invention.
Detailed Description
The invention provides a density-based fast sea surface SAR image superpixel segmentation method, the overall flow is shown in figure 1, the method is divided into a training stage and a testing stage, and the method comprises the following steps:
1) A training stage;
1-1) determining sparsity K for determining the number of nonzero elements of the weighted sparse classifier, wherein K =4 in the embodiment;
determining a scale factor α ∈ (0,1) for calculating density characteristics, α =0.2 in this example;
separately obtain Q 1 Zhang Chun clutter sub-image and acquisition Q 2 Two pure subimages containing sea surface targets are identical in size, and the size of each subimage is M 1 ×M 2 Wherein M is 1 Representing the length of the sub-picture, M 2 Represents the width of the sub-image; the number of the two sub-images is approximately the same;
sub-image size M in the present embodiment 1 ×M 2 Select 300X 300, Q 1 ＝Q 2 =60, image source is an existing public dataset.
Setting image block size P 1 ×P 2 In which P is 1 Length, P, of the representative image block 2 Width, P, of representative image block 1 Needs to be M 1 Removal of P 2 Needs to be M 2 Trimming; the image block size in this embodiment is 10 × 10.
1-2) dividing each sub-image according to the set image block size for the pure clutter sub-image and the pure sub-image containing the sea surface target obtained in the step 1-1), and calculating the density value of each image block in each sub-image;
for any sub-image, the density value calculation expression of each image block in the sub-image is as follows:
Figure BDA0003027760500000051
where ρ is i Indicating the density value of the ith image block in the subimage; i, j denote the indices of the different image blocks in the sub-image, N (patc) Representing the number of image blocks in the sub-picture,
Figure BDA0003027760500000052
representing the absolute value of the difference between the average gray scale of the ith image block and the average gray scale of the jth image block, d (c) Is a constant for the sub-image
Figure BDA0003027760500000053
Indi (-) represents an indicator function whose output is 1 when the function input is less than 0 and 0 otherwise.
1-3) sorting the density values of all image blocks in each subimage from small to large by using the result of the step 1-2) to obtain a density vector of the subimage;
for any sub-image, the density vector expression is as follows:
Figure BDA0003027760500000054
wherein
Figure BDA0003027760500000055
1-4) utilizing the results of the step 1-3) to establish the following weighted sparse optimization problem:
Figure BDA0003027760500000056
wherein the content of the first and second substances,
Figure BDA0003027760500000057
a set of density vectors representing all sub-images,
Figure BDA0003027760500000058
representing the density vector of the qth sub-image, Q =1, …, Q being the total number of sub-images,Q＝Q 1 +Q 2 ；
Figure BDA0003027760500000059
a set of labels representing all of the sub-images,
Figure BDA00030277605000000510
a label representing the qth sub-image, wherein
Figure BDA00030277605000000511
Indicating that the qth sub-image contains a sea surface object,
Figure BDA00030277605000000512
representing that only sea clutter exists in the q-th sub-image; Λ is a Q × Q diagonal matrix, the qth diagonal element in Λ is
Figure BDA00030277605000000513
Figure BDA00030277605000000514
Representing the number of sea surface targets in the current q-th sub-image;
Figure BDA0003027760500000061
representing a weight vector, b = b × 1,b representing a bias,
Figure BDA0003027760500000062
respectively, an optimal weight vector and an optimal bias.
1-5) solving the weighted sparse optimization problem of the step 1-4);
the optimization problem can be solved by using an Orthogonal Matching Pursuit (OMP) algorithm in documents A.Tropp and A.C.Gilbert, "Signal Recovery From Random Measurements Via Orthogonal Matching Pursuit," IEEE Transactions on Information Theory, vol.53, no.12, pp.4655-4666, and Dec.2007 "to obtain an optimal weight vector
Figure BDA0003027760500000063
And the optimum deviation
Figure BDA0003027760500000064
2) A testing stage;
2-1) obtaining a sea surface SAR image, wherein the number of pixels of the image is N;
setting a super pixel size S (S can be set to 25% of the number of pixels occupied by the ship, such as S = 20);
2-2) uniformly segmenting the sea surface SAR image in the step 2-1) into M 1 ×M 2 A plurality of sub-images of a size;
2-3) for each sub-image obtained in the step 2-2), according to the set image block size P 1 ×P 2 Dividing each sub-image, and calculating the density value of each image block in each sub-image;
for any sub-image, the density value calculation expression of each image block in the sub-image is as follows:
Figure BDA0003027760500000065
2-4) sorting the density values of all image blocks in each subimage from small to large by using the result of the step 2-3) to obtain a density vector rho of the subimage;
for any sub-image, the density vector expression is as follows:
Figure BDA0003027760500000066
wherein
Figure BDA0003027760500000067
2-5) using the results of step 2-4), for each sub-image, its label y is calculated,
Figure BDA0003027760500000068
then, y is determined: if y is more than 0, keeping the sub-image corresponding to the label y (the sub-image may contain the target and needs to be subjected to next super-pixel segmentation); and if y is less than or equal to 0, deleting the sub-image corresponding to the label y.
2-6) for the sub-images (y is more than 0) reserved in the step 2-5), performing superpixel segmentation on each sub-image, and performing the following local k-means clustering operation based on the spatial distance, the gray level distance and the density distance to finally obtain a superpixel segmentation result of the sub-image; the method comprises the following specific steps:
2-6-1) for the sub-images retained through the step 2-5), the super pixel centers are uniformly arranged at intervals S along the horizontal and vertical directions of each sub-image. Each superpixel center is then moved to the position within its 8 neighbors with the smallest gradient value. Setting kappa n = + ∞, representing the initial distance of each pixel from its corresponding superpixel center, where n represents the index value of the pixel, n =1,2, …, M 1 ×M 2 ；
2-6-2) for each superpixel center, sequentially taking each pixel in a 2S multiplied by 2S neighborhood as a current pixel, and calculating the distance between the current pixel and the superpixel center, wherein the distance is defined as follows:
Figure BDA0003027760500000071
wherein the content of the first and second substances,
Figure BDA0003027760500000072
represents the spatial distance (x) 1 ,y 1 ) And (x) 2 ,y 2 ) Respectively representing the spatial positions of the center of the superpixel and the current pixel, d (inte) ＝|δ 1 -δ 2 I denotes the gray scale distance, delta 1 ,δ 2 Respectively representing the gray values of the center of the super pixel and the current pixel,
Figure BDA0003027760500000073
the distance of the density is represented by,
Figure BDA0003027760500000074
respectively representing the center of the superpixel and the density of the current pixelAnd (4) measuring values. When each pixel is taken as an image block with the size of 1 × 1, the formula for calculating the density value of the pixel can use the formula (3), and the only difference is that P is adopted at this time 1 ＝P 2 ＝1。
If D is<κ n Then the current pixel is determined to belong to the center of the superpixel and k is updated n = D; otherwise, judging that the current pixel does not belong to the center of the super pixel; and after all the current pixels are judged, updating the pixel set belonging to the center of each super pixel.
2-6-3) respectively averaging the spatial positions, the gray scales and the densities of all pixels in the pixel set belonging to the center of the super pixel by using the result of the step 2-6-2) and then taking the average as the updated spatial position, the gray scale and the density of the center of the super pixel.
2-6-4) repeating the steps 2-6-2) to 2-6-3) until the upper limit T of the repetition times is reached, wherein T is more than or equal to 10 times, and the pixel set belonging to the center of each super pixel is updated completely.
2-6-5) using the nearest neighbor post-processing steps in the documents r.achanta, a.shaji, k.smith, a.lucchi, p.fua and s.s.sutrunk, "SLIC superior pixels coordinated to State-of-the-Art superior Methods," IEEE Transactions on Pattern Analysis and Machine Analysis, vol.34, no.11, pp.2274-2282, nov.2012), to enhance the connectivity of the Superpixel internal pixels for the results of steps 2-6-4) to obtain the final pixel set corresponding to each Superpixel center in the sub-image containing the target retained in step 2-5), and then the sea surface SAR image is completely Superpixel segmented.
Fig. 2 is a sea surface SAR image (3000 × 3000 pixels). For the image, the Superpixel segmentation time of the existing method (r.achanta, a. Shaji, k.smith, a.lucchi, p.fua and s.ssussturn, "SLIC Superpixels matched to State-of-the-Art Superpixel Methods," IEEE Transactions on Pattern Analysis and Machine Analysis, vol.34, no.11, pp.2274-2282, nov.2012.) is 2643 seconds, and the Superpixel segmentation time of the proposed method is 78 seconds, so the invention can greatly shorten the Superpixel segmentation time.

Claims (2)

1. A fast sea surface SAR image superpixel segmentation method based on density is characterized in that the method comprises a training phase and a testing phase; in the training stage, firstly, a plurality of pure clutter sub-images with the same size and a plurality of pure sub-images containing sea surface targets are respectively obtained, and the density vector of each sub-image is calculated; then, establishing a weighted sparse optimization problem based on the density vectors of all the sub-images and solving the weighted sparse optimization problem to obtain an optimal weight vector and an optimal deviation; in the testing stage, acquiring a sea surface SAR image and dividing the sea surface SAR image into a plurality of subimages, calculating a label corresponding to each subimage of the sea surface SAR image by using the optimal weight vector and the optimal deviation, and judging whether the subimages possibly contain a sea surface target or not through the label; and finally, performing superpixel segmentation on all sub-images possibly containing sea surface targets to obtain a final superpixel segmentation result of the sea surface SAR image.
2. A method as claimed in claim 1, characterized in that the method comprises the following steps:
1) A training stage;
1-1) determining sparsity K and determining a scale factor alpha epsilon (0,1);
separately obtain Q 1 Zhang Chun clutter sub-image and Q 2 Two pure subimages containing sea surface targets are identical in size, and the size of each subimage is M 1 ×M 2 Wherein M is 1 Representing the length of the sub-picture, M 2 Represents the width of the sub-image;
setting image block size P 1 ×P 2 In which P is 1 Length, P, of representative image block 2 Width, P, of representative image block 1 Quilt M 1 Integer removal of P 2 Quilt M 2 Trimming;
1-2) dividing each sub-image according to the set image block size for the pure clutter sub-image and the pure sub-image containing the sea surface target obtained in the step 1-1), and calculating the density value of each image block in each sub-image;
for any sub-image, the density value calculation expression of each image block in the sub-image is as follows:
Figure FDA0003027760490000011
where ρ is i Indicating the density value of the ith image block in the sub-image; i, j denote the indices of the different image blocks in the sub-image, N (patc) Representing the number of image blocks in the sub-picture,
Figure FDA0003027760490000012
representing the absolute value of the difference between the average gray scale of the ith image block and the average gray scale of the jth image block, d (c) Is a constant for the sub-image
Figure FDA0003027760490000013
Indi (-) represents an indicator function whose output is 1 when the function input is less than 0, and 0 otherwise;
1-3) sorting the density values of all image blocks in each sub-image from small to large by using the result of the step 1-2) to obtain a density vector of the sub-image;
for any sub-image, the density vector expression is as follows:
Figure FDA0003027760490000014
wherein
Figure FDA0003027760490000015
1-4) utilizing the results of the step 1-3) to establish the following weighted sparse optimization problem:
Figure FDA0003027760490000021
wherein the content of the first and second substances,
Figure FDA0003027760490000022
representing all subgraphsA set of density vectors of the image,
Figure FDA0003027760490000023
density vector representing the Q-th sub-image, Q =1, …, Q being the total number of sub-images, Q = Q 1 +Q 2 ；
Figure FDA0003027760490000024
A set of labels representing all of the sub-images,
Figure FDA0003027760490000025
a label representing the qth sub-image, wherein
Figure FDA0003027760490000026
Indicating that the qth sub-image contains a sea surface object,
Figure FDA0003027760490000027
representing that only sea clutter exists in the qth sub-image; Λ is a diagonal matrix of Q × Q, the qth diagonal element in Λ is
Figure FDA0003027760490000028
Figure FDA0003027760490000029
Representing the number of sea surface targets in the q sub-image;
Figure FDA00030277604900000210
representing a weight vector, b = b × 1,b representing a bias,
Figure FDA00030277604900000211
respectively representing an optimal weight vector and an optimal deviation;
1-5) solving the weighted sparse optimization problem in the step 1-4) by utilizing an orthogonal matching pursuit algorithm to obtain an optimal weight vector
Figure FDA00030277604900000212
And an optimum deviation;
2) A testing stage;
2-1) obtaining a sea surface SAR image; setting a super-pixel size S;
2-2) uniformly segmenting the sea surface SAR image in the step 2-1) into M 1 ×M 2 A plurality of sub-images of a size;
2-3) for each sub-image obtained in the step 2-2), according to the set image block size P 1 ×P 2 Dividing each sub-image, and calculating the density value of each image block in each sub-image;
for any sub-image, the density value calculation expression of each image block in the sub-image is as follows:
Figure FDA00030277604900000213
2-4) sorting the density values of all image blocks in each sub-image from small to large by using the result of the step 2-3) to obtain a density vector rho of the sub-image;
2-5) calculating the label y corresponding to each sub-image by using the result of the step 2-4),
Figure FDA00030277604900000214
then, y is determined: if y is greater than 0, keeping the subimage corresponding to the label y; if y is less than or equal to 0, deleting the sub-image corresponding to the label y;
2-6) carrying out superpixel segmentation on each sub-image reserved in the step 2-5) to finally obtain a superpixel segmentation result of the sub-image; the method comprises the following specific steps:
2-6-1) uniformly setting the center of the super pixel at intervals S in the horizontal and vertical directions of each sub-image for the sub-images retained through the step 2-5); then moving the center of each super pixel to the position with the minimum gradient value in the 8 neighborhoods of the super pixel; setting kappa n = + ∞, representing each pixel to which it correspondsInitial distance beyond pixel center, where n represents index value of pixel, n =1,2, …, M 1 ×M 2 ；
2-6-2) for each superpixel center, sequentially taking each pixel in a 2S multiplied by 2S neighborhood of the superpixel center as a current pixel, and calculating the distance between the current pixel and the superpixel center, wherein the expression is as follows:
Figure FDA0003027760490000031
wherein the content of the first and second substances,
Figure FDA0003027760490000032
represents the spatial distance, (x) 1 ,y 1 ) And (x) 2 ,y 2 ) Respectively representing the spatial positions of the center of the superpixel and the current pixel, d (inte) ＝|δ 1 -δ 2 | represents the gray scale distance, δ 1 ,δ 2 Respectively representing the gray values of the center of the super pixel and the current pixel,
Figure FDA0003027760490000033
the distance of the density is indicated by,
Figure FDA0003027760490000034
respectively representing the density values of the center of the super pixel and the current pixel; wherein, each pixel is respectively used as an image block with the size of 1 multiplied by 1, and the density values of the image blocks corresponding to the center of the super pixel and the current pixel are calculated by using the formula (3) to obtain
Figure FDA0003027760490000035
If D is<κ n Then the current pixel is determined to belong to the center of the superpixel and k is updated n = D; otherwise, judging that the current pixel does not belong to the center of the super pixel; after all the current pixels are judged, updating a pixel set belonging to the center of each super pixel;
2-6-3) respectively averaging the spatial positions, the gray scales and the densities of all pixels in the pixel set belonging to the center of each super pixel by using the result of the step 2-6-2) and then taking the average as the updated spatial position, the gray scale and the density of the center of the super pixel;
2-6-4) repeating the steps 2-6-2) to 2-6-3) until the upper limit T of the repetition times is reached and the pixel set belonging to the center of each super pixel is updated;
2-6-5) processing the result of the step 2-6-4) by adopting a nearest neighbor post-processing method to obtain a final pixel set corresponding to each superpixel center in all the sub-images reserved in the step 2-5), and completing the superpixel segmentation of the sea surface SAR image.
CN202110420778.8A 2021-04-19 2021-04-19 Rapid sea surface SAR image superpixel segmentation method based on density Active CN113192078B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110420778.8A CN113192078B (en) 2021-04-19 2021-04-19 Rapid sea surface SAR image superpixel segmentation method based on density

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110420778.8A CN113192078B (en) 2021-04-19 2021-04-19 Rapid sea surface SAR image superpixel segmentation method based on density

Publications (2)

Publication Number Publication Date
CN113192078A CN113192078A (en) 2021-07-30
CN113192078B true CN113192078B (en) 2022-12-02

Family

ID=76977787

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110420778.8A Active CN113192078B (en) 2021-04-19 2021-04-19 Rapid sea surface SAR image superpixel segmentation method based on density

Country Status (1)

Country Link
CN (1) CN113192078B (en)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105654442A (en) * 2015-12-31 2016-06-08 大连理工大学 Noise elimination method for impact noise image
CN106204482A (en) * 2016-07-08 2016-12-07 桂林电子科技大学 Based on the mixed noise minimizing technology that weighting is sparse

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105654442A (en) * 2015-12-31 2016-06-08 大连理工大学 Noise elimination method for impact noise image
CN106204482A (en) * 2016-07-08 2016-12-07 桂林电子科技大学 Based on the mixed noise minimizing technology that weighting is sparse

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
Density-Aware Feature Embedding for Face Clustering;Senhui Guo等;《2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)》;20200805;全文 *
Spatial Density Peak Clustering for Hyperspectral Image Classification With Noisy Labels;Bing Tu等;《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》;20190731;第57卷(第7期);全文 *
基于SLIC超像素分割的SAR图像海陆分割算法;李智等;《雷达科学与技术》;20170815(第04期);全文 *

Also Published As

Publication number Publication date
CN113192078A (en) 2021-07-30

Similar Documents

Publication Publication Date Title
CN112380952B (en) Power equipment infrared image real-time detection and identification method based on artificial intelligence
CN112700429B (en) Airport pavement underground structure disease automatic detection method based on deep learning
CN107016677B (en) Cloud picture segmentation method based on FCN and CNN
CN106780485B (en) SAR image change detection method based on super-pixel segmentation and feature learning
CN110009010B (en) Wide-width optical remote sensing target detection method based on interest area redetection
CN108596055B (en) Airport target detection method of high-resolution remote sensing image under complex background
CN110263705A (en) Towards two phase of remote sensing technology field high-resolution remote sensing image change detecting method
CN108898065B (en) Deep network ship target detection method with candidate area rapid screening and scale self-adaption
CN108961229A (en) Cardiovascular OCT image based on deep learning easily loses plaque detection method and system
CN110097101B (en) Remote sensing image fusion and coastal zone classification method based on improved reliability factor
Wang et al. Combing Single Shot Multibox Detector with transfer learning for ship detection using Chinese Gaofen-3 images
CN108960190B (en) SAR video target detection method based on FCN image sequence model
CN110414509B (en) Port docking ship detection method based on sea-land segmentation and characteristic pyramid network
CN113468968B (en) Remote sensing image rotating target detection method based on non-anchor frame
CN110728706B (en) SAR image fine registration method based on deep learning
CN108734103A (en) The detection of moving target and tracking in satellite video
CN111310756A (en) Damaged corn particle detection and classification method based on deep learning
KR101941043B1 (en) Method for Object Detection Using High-resolusion Aerial Image
CN109829426B (en) Railway construction temporary building monitoring method and system based on high-resolution remote sensing image
CN109035196A (en) Image local fuzzy detection method based on conspicuousness
CN107194917A (en) In-orbit SAR image change detection based on DAP and ARELM
Zhang et al. Contextual squeeze-and-excitation mask r-cnn for sar ship instance segmentation
CN114119621A (en) SAR remote sensing image water area segmentation method based on depth coding and decoding fusion network
CN113192078B (en) Rapid sea surface SAR image superpixel segmentation method based on density
CN112435249A (en) Dynamic small target detection method based on periodic scanning infrared search system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant