[go: up one dir, main page]

CN107256399B - Gamma distribution superpixel-based method and superpixel TMF-based SAR image coastline detection method - Google Patents

Gamma distribution superpixel-based method and superpixel TMF-based SAR image coastline detection method Download PDF

Info

Publication number
CN107256399B
CN107256399B CN201710449129.4A CN201710449129A CN107256399B CN 107256399 B CN107256399 B CN 107256399B CN 201710449129 A CN201710449129 A CN 201710449129A CN 107256399 B CN107256399 B CN 107256399B
Authority
CN
China
Prior art keywords
superpixel
point
value
seed
neighborhood
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.)
Expired - Fee Related
Application number
CN201710449129.4A
Other languages
Chinese (zh)
Other versions
CN107256399A (en
Inventor
史晓非
王智罡
刘玲
丁星
马海洋
冯建德
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.)
Dalian Maritime University
Original Assignee
Dalian Maritime 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 Dalian Maritime University filed Critical Dalian Maritime University
Priority to CN201710449129.4A priority Critical patent/CN107256399B/en
Publication of CN107256399A publication Critical patent/CN107256399A/en
Application granted granted Critical
Publication of CN107256399B publication Critical patent/CN107256399B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/13—Satellite images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Astronomy & Astrophysics (AREA)
  • Remote Sensing (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

本发明公开了一种基于Gamma分布超像素算法和基于超像素TMF的SAR图像海岸线检测算法，通过读取图像I，输入种子点数k，并根据输入的种子点数k和图像的大小确定种子点的位置以及每个种子点周围搜索区域的大小，直至超像素的边界不再发生变化，输入所述超像素图像，更新种子点并计算势能，最终输出海岸线结果。本发明在SAR图像海岸线检测算法中提出了一种全新的TMF算法，能够良好的解决中心像素和邻域像素之间的相对位置关系的问题，而解决前一问题需要引入超像素方法。

Figure 201710449129

The present invention discloses a SAR image coastline detection algorithm based on a Gamma distribution superpixel algorithm and a superpixel TMF, by reading an image I, inputting the number of seed points k, and determining the position of the seed points and the size of the search area around each seed point according to the input number of seed points k and the size of the image, until the boundary of the superpixel no longer changes, inputting the superpixel image, updating the seed points and calculating the potential energy, and finally outputting the coastline result. The present invention proposes a new TMF algorithm in the SAR image coastline detection algorithm, which can well solve the problem of the relative position relationship between the central pixel and the neighborhood pixel, while solving the former problem requires the introduction of a superpixel method.

Figure 201710449129

Description

Gamma distribution superpixel-based method and superpixel TMF-based SAR image coastline detection method
Technical Field
The invention relates to a method for detecting a coastline of an SAR image of a probability factor TMF (TMF) of a super-pixel, belonging to the field of coastline detection.
Background
Synthetic Aperture Radar (SAR) in microwave remote sensing is an active microwave detector, and imaging is performed by synthesizing a larger equivalent antenna Aperture through a smaller-sized real antenna Aperture by using a synthetic Aperture principle, a signal processing method and a pulse compression technology. Compared with optical imaging, the synthetic aperture radar has the advantages of all-weather imaging and the like, so that the SAR image plays an important role in the fields of strategic target identification and detection, disaster control, state and soil resource monitoring, sea area use management, map mapping, ship target identification and the like. In recent years, SAR images gradually attract attention in the field of sea area management, one of the main problems is the problem of coastline detection, and the coastline is continuously changed due to long-time accumulation of river sediment, sea reclamation and other reasons, so that the change of the coastline is effectively monitored, and the method has certain practical significance in dynamic monitoring of sea area use.
So far, the coastline detection method has been greatly developed. Common methods for coastline detection include a super-pixel segmentation method based on graph theory, a super-pixel formation method based on gradient rise, a triple Markov random field, and a method most suitable for processing SAR image classification and segmentation in the existing Markov random field, and the advantages of the method are mainly embodied in that the introduction of an auxiliary field can convert a global non-stationary SAR image into some local stationary images, and then the local stationary images are processed. However, the method is mainly considered based on the pixel angle, so that the method is easily influenced by speckle noise, time complexity is high, less texture information can be utilized, initialization of the auxiliary field is realized through a threshold value, large errors can be caused to a segmentation result if the auxiliary field is not accurate enough, meanwhile, the potential energy function only considers the relation between the label value of the central pixel and the neighborhood pixels and the auxiliary field value, and ignores the similarity between the neighborhood pixels and the central pixel and the relative position relation between the central pixel and the neighborhood pixels.
Disclosure of Invention
The invention aims at the problem, provides a Gamma distribution superpixel-based method and a superpixel TMF-based SAR image coastline detection method, and is characterized by comprising the following steps:
S1: reading an image I, and inputting a seed point number k;
S2: according to the input seed pointDetermining the position of the seed point and the size of a search area around each seed point by the number k and the size of the image;
S3: selecting a neighborhood window of 5 pixels multiplied by 5 pixels around each seed point, calculating the mean value in each point sub local window in the neighborhood window as the characteristic of the current point, determining a point set C which is similar to the seed point texture characteristic in the local window through a clustering method for the characteristic, and calculating the similar point set C:
Figure GDA0002763444200000021
Figure GDA0002763444200000022
Figure GDA0002763444200000023
wherein, (x, y) represents the coordinates of the pixel points in the local window, I (x, y) represents the pixel values of the pixel points in the local window, W represents the local window, N represents the pixel in C, and mus(x, y) represents the mean value within a local window of pixels at (x, y) locations,
Figure GDA0002763444200000028
represents μ of each point in CsMean, σ denotes the variance of the mean within the local window;
S4: traversing the image according to step S3Calculating the feature of each neighborhood point, and calculating d according to the formulai,jFinally by comparison di,jForming a super pixel;
d isi,j
Figure GDA0002763444200000024
Wherein, muiRepresents the mean value, μ, of the seed point ijMeans, σ, representing neighborhood point jiRepresenting the variance, σ, of the mean of i within a local windowjVariance representing mean of j within local window
Figure GDA0002763444200000025
Wherein xiLine coordinate, y, representing seed point iiColumn coordinate, x, representing seed point ijLine coordinate, y, representing neighborhood point jjColumn coordinates representing neighborhood point j
Figure GDA0002763444200000026
Wherein m represents dci,jAnd dsi,jA weight coefficient therebetween;
the distance S between the adjacent seed points
Figure GDA0002763444200000027
Wherein N represents the total number of image pixels, and k represents the number of seed points;
S5: updating the positions of the seed points to be the mean value of all the point positions contained in each type of super pixels;
S6: repeating step S3-S5Stopping repeating until the boundary of the super pixel is not changed any more, and outputting a super pixel image;
S7: inputting the superpixel image, initially clustering the superpixels into 2 classes by using a kmeans method, and taking the clustering result as an initial marking field XspAnd calculating potential energy Wsp(Xsp,Usp)；
S8: updating the marker field XspAnd auxiliary field UspAnd updating the parameter set theta by an ICM method and an SG method:
S9: repeating step S7-S8Until the mark field is no longer changed, outputtingAnd (5) going out of the seashore to obtain a result.
Further, step S3The clustering method of (1): selecting any two points in the neighborhood as seed points A and B, calculating the absolute value of the difference value of the characteristics from each point to the two points A and B in a local window, and when the absolute value of the difference value between the point A and the point B is smaller than the absolute value of the difference value between the point A and the point B, enabling the current point to belong to the same class as the point A; when the absolute value of the difference between the point A and the point B is larger than that of the difference between the point A and the point B, two types with larger contrast are formed.
Further, the potential energy Wsp(Xsp,Usp)：
Figure GDA0002763444200000031
Wherein x issLabels, x, representing the central superpixel stLabel, C, representing a neighborhood superpixel tspRepresenting a set of superpixel potential masses, fcosCosine value, W, representing the angle between two vectorsedge(xs,xt) And representing the weight value of the neighborhood superpixel t, wherein the expression is as follows:
Figure GDA0002763444200000032
wherein L iss,tRepresenting the length, L, of the common boundary between the central superpixel s and the neighborhood superpixel tsRepresenting the perimeter of the central superpixel s.
Further, the
Figure GDA0002763444200000033
And
Figure GDA0002763444200000034
Figure GDA0002763444200000035
Figure GDA0002763444200000036
wherein, XspMarker field, U, representing a superpixelspAuxiliary fields, W, representing super-pixelssp(Xxp,Usp) Representing potential energy function, SP representing set of superpixels, L representing view of SAR image, IspRepresenting the value of a superpixel, i.e. the mean, mu, of all pixel values in a superpixeliRepresenting the mean of the i-th class superpixels.
Further, the SG method parameter updating formula is:
Figure GDA0002763444200000041
where q denotes the number of iterations, A is the number of constant control iterations, x0And u0Values, x, representing the initial tag value and the auxiliary fieldq+1And uq+1Represents the label value and auxiliary field value obtained from the (q + 1) th iteration, p represents the p-th parameter in the parameter set, WfWhat is shown is a function of the potential energy,
Figure GDA0002763444200000042
the estimated value of the q-th order parameter is shown.
The invention has the advantages that: the invention provides a brand-new TMF method in the SAR image coastline detection method, which can well solve the problem of relative position relation between a central pixel and a neighborhood pixel, and needs to introduce a superpixel method for solving the former problem. The super-pixel method provided by the invention is used for solving the problem, and finally the super-pixel-based TMF method is provided by combining the advantages of the two methods.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a schematic overall flow chart of the present invention.
FIG. 2 is a shoreline detection chart of the present invention
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the following describes the technical solutions of the embodiments of the present invention clearly and completely with reference to the accompanying drawings in the embodiments of the present invention:
as shown in fig. 1, a method for detecting a coastline of an SAR image based on a Gamma distribution superpixel method and a superpixel TMF includes the following steps: s1: reading an image I, and inputting a seed point number k;
S2: determining the position of the seed point and the size of a search area around each seed point according to the input number k of the seed points and the size of the image;
S3: selecting a neighborhood window of 5 pixels multiplied by 5 pixels around each seed point, calculating the mean value in each point sub local window in the neighborhood window as the characteristic of the current point, determining a point set C which is similar to the seed point texture characteristic in the local window through a clustering method for the characteristic, and calculating the similar point set C:
Figure GDA0002763444200000051
Figure GDA0002763444200000052
Figure GDA0002763444200000053
wherein (A), (B), (C), (D), (C), (x, y) represents the coordinates of the pixel points in the local window, I (x, y) represents the pixel values of the pixel points in the local window, W represents the local window, N represents the pixel in C, mus(x, y) represents the mean value within a local window of pixels at (x, y) locations,
Figure GDA0002763444200000054
represents μ of each point in CsMean, σ denotes the variance of the mean within the local window;
S4: traversing the image according to step S3Calculating the feature of each neighborhood point, and calculating d according to the formulai,jComparison of di,jForming a super pixel; in the present embodiment, it is preferred that,
Figure GDA0002763444200000055
wherein, muiRepresents the mean value, μ, of the seed point ijMeans, σ, representing neighborhood point jiRepresenting the variance, σ, of the mean of i within a local windowjVariance representing mean of j within local window
Figure GDA0002763444200000056
Wherein xiLine coordinate, y, representing seed point iiColumn coordinate, x, representing seed point ijLine coordinate, y, representing neighborhood point jjColumn coordinates representing neighborhood point j
Figure GDA0002763444200000057
Wherein m represents dci,jAnd dsi,jThe weight coefficient between the weight of the first and second groups,
in the present embodiment, S
Figure GDA0002763444200000058
S5: updating the positions of the seed points to be the mean value of all the point positions contained in each type of super pixels;
S6: repeating step S3-S5Stopping repeating until the boundary of the super pixel is not changed any more, and outputting a super pixel image;
S7: inputting a superpixel image, initially clustering the superpixels into 2 classes by using a kmeans method, and taking the clustering result as an initial marking field XspAnd calculating potential energy Wsp(Xsp,Usp)；
S8: updating the marker field XspAnd auxiliary field UspAnd updating the parameter set theta by an ICM method and an SG method:
S9: repeating step S7-S8And outputting the result of the coastline until the marking field is not changed any more.
As a preferred embodiment, step S3The medium clustering method comprises the following steps: selecting any two points in the neighborhood as seed points A and B, calculating the absolute value of the difference value of the characteristics from each point to the two points A and B in a local window, and when the absolute value of the difference value between the point A and the point B is smaller than the absolute value of the difference value between the point A and the point B, enabling the current point to belong to the same class as the point A; when the absolute value of the difference between the point A and the point B is larger than that of the difference between the point A and the point B, two types with larger contrast are formed. It is understood that in other embodiments, step S3The method of calculating the local window may be determined according to actual requirements and actual accuracy requirements.
In the present embodiment, potential energy Wsp(Xsp,Usp)：
Figure GDA0002763444200000061
Wherein x issLabels, x, representing the central superpixel stLabel, C, representing a neighborhood superpixel tspRepresenting a set of superpixel potential masses, fcosCosine value, W, representing the angle between two vectorsedge(xs,xt) Represents a neighborhoodThe weight of the domain superpixel t is expressed as follows:
Figure GDA0002763444200000062
wherein L iss,tRepresenting the length, L, of the common boundary between the central superpixel s and the neighborhood superpixel tsRepresenting the perimeter of the central superpixel s.
In the present embodiment, it is preferred that,
Figure GDA0002763444200000063
and
Figure GDA0002763444200000064
Figure GDA0002763444200000065
Figure GDA0002763444200000066
wherein, XspMarker field, U, representing a superpixelspAuxiliary fields, W, representing super-pixelssp(Xxp,Usp) Representing potential energy function, SP representing set of superpixels, L representing view of SAR image, IspRepresenting the value of a superpixel, i.e. the mean, mu, of all pixel values in a superpixeliRepresenting the mean of the i-th class superpixels.
In this embodiment, the SG method parameter update formula is:
Figure GDA0002763444200000067
where q denotes the number of iterations, A is the number of constant control iterations, x0And u0Values, x, representing the initial tag value and the auxiliary fieldq+1And uq+1Representing the tag value and the value of the auxiliary field obtained from the (q + 1) th iteration,p denotes the p-th parameter in the parameter set, WfWhat is shown is a function of the potential energy,
Figure GDA0002763444200000071
the estimated value of the q-th order parameter is shown.
Parameter setting in the examples:
seed numbers of Envasat images are respectively set to be 250, 250, 300 and 250, seed numbers k of Radarsat images are respectively set to be 250, 250, 300 and 300, and seed numbers k of Terra images are respectively set to be 100. The weight coefficient between and is 0.5, and the local window size is 5 maximum iterations. The initial value of (1) is 0.6, the value of (0.6), the iteration number of the outer iteration is 10, the iteration number of the inner iteration is 10, and the threshold value for judging no change is 0.0001.
The parameters for the comparative test GMRF were set as follows: the number of iterations of the conditional iteration mode is 5 and the size of the filter window is. Comparative experiments the parameters for the superpixel-based TMF were set as follows: the seed numbers of the Envisat images are 250, 250, 300 and 250, the seed numbers k of the Radarsat images are 250, 250, 300 and 300, and the seed number of the Terra image is 100. dcAnd dsThe weight coefficient m between is 0.5, the local window size is 3 pixels × 3 pixels, and the maximum number of iterations is 5. Alpha is alphaxIs 1, alphauaHas a value of 0.6, alphaubThe value of (a) is 0.6, the number of iterations of the outer iteration is 10, the number of iterations of the inner iteration is 10, and the threshold value for judging that no change occurs is 0.0001.
Example (b):
the comparison of the performance of the method mainly adopts root mean square error RMSE and QA (average accuracy) as precision analysis indexes, firstly RMSE comparison is carried out, and the calculation formula is as follows:
Figure GDA0002763444200000072
where RMSE represents the average error between hand-drawn coastline and various methods of coastline extraction, x1kRepresenting coasts drawn by artificial handAnd extracting the pixel value of the k-th position pixel in the binary image of the line extraction result. x is the number of2kThe pixel value of the k-th position pixel in the binary image representing the coastline extraction result obtained by the theoretical model, and N represents the number of image pixels. Smaller values of RMSE indicate closer to the true coastline, higher accuracy.
Performance comparisons were performed for Envisat, Terra, and Radarsat images, respectively. The experimental images are Envisat, Terra and Radarsat images, the contrast performance is on the premise that all methods can effectively detect the coastline, and when the coastline cannot be effectively detected by the methods, the accuracy is expressed in a positive and infinite manner and does not participate in contrast analysis. The RMSE pair for the method is shown in table 1 for the Envisat image.
Size of GMRF Super pixel TMF The patented method
323×371 +∞ +∞ 0.0758
374×365 +∞ 0.0697 0.0658
324×238 0.0988 0.0982 0.0967
206×235 0.0994 0.1006 0.0580
From the data in the table, it can be seen that, for an image GMRF method and a conventional superpixel TMF method which have high noise intensity and poor uniformity, the seashore line cannot be correctly identified, the GMRF method cannot solve the image of a region having similar sea texture characteristics in land, and the conventional superpixel TMF method has a defect in edge fitting degree. For the three methods of detecting the coastline well by the image with more uniform sea surface, the RMSE value of the method is smaller than that of the other methods, and the method is slightly higher than that of the comparison method. The RMSE pair for the method is shown in table 2 for the Terra images.
Size of GMRF Super pixel TMF The patented method
115×71 0.1207 0.1044 0.0946
51×80 +∞ 0.1291 0.1328
126×126 +∞ 0.1185 0.1094
134×212 0.1226 0.0929 0.0906
From the above experimental results, it can be seen that the three methods can be used for processing the image with relatively simple texture and relatively uniform sea surface, but the method is still superior to the contrast method for the image with low contrast. The RMSE pair for the method is shown in table 3 for the Radarsat image.
Figure GDA0002763444200000081
Figure GDA0002763444200000091
From the above table, it can be seen that the calculation accuracy of the method is better than that of the two comparison methods, and the detection result is better than that of the pixel-based method due to the fact that the super-pixel contains more information than the pixel, and meanwhile, the method has the problem of low edge fitting degree in the traditional super-pixel method, so that the performance of the method is better than that of the traditional super-pixel TMF method. Table 4 shows the Envisat image contrast.
Figure GDA0002763444200000092
The performance index of QA for the Terra image is shown in table 5.
Figure GDA0002763444200000093
The performance index for QA for the Radarsat image is shown in table 6.
Figure GDA0002763444200000094
Figure GDA0002763444200000101
Where + ∞indicatesthat the image cannot be efficiently identified using the model. Wherein a smaller value of RMSE indicates a smaller difference between the detected coastline and the actual coastline, and a closer proximity between the identified coastline and the actual coastline. QA represents the percentage of correctly recognized pixels in the entire image, and a larger value indicates that the more correctly recognized pixels, the more accurate the coastline obtained by segmentation. According to the detection results of SAR images formed by three satellites of Envisat, Terra and Radarsat, the detection accuracy of most images is superior to that of a comparison method.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered to be within the technical scope of the present invention, and the technical solutions and the inventive concepts thereof according to the present invention should be equivalent or changed within the scope of the present invention.

Claims (5)

1.一种基于Gamma分布超像素方法及超像素TMF的SAR图像海岸线检测方法，其特征在于包括如下步骤：1. a SAR image coastline detection method based on Gamma distribution superpixel method and superpixel TMF, is characterized in that comprising the steps: S1：读取图像I，并输入种子点数k；S 1 : read the image I, and input the number of seed points k; S2：根据输入的种子点数k和图像的大小确定种子点的位置以及每个种子点周围搜索区域的大小；S 2 : Determine the position of the seed points and the size of the search area around each seed point according to the input seed point number k and the size of the image; S3：在每一个种子点周围选择一个5像素×5像素的邻域窗，计算邻域窗内每一个种子点局部窗内的均值作为当前点的特征，对该特征通过聚类方法确定局部窗内和种子点纹理特征相似的点集C并计算相似点集C：S 3 : Select a 5-pixel × 5-pixel neighborhood window around each seed point, calculate the mean value in the local window of each seed point in the neighborhood window as the feature of the current point, and determine the local The point set C with similar texture features of the seed point in the window and the similar point set C are calculated:
Figure FDA0002763444190000011
Figure FDA0002763444190000011
Figure FDA0002763444190000012
Figure FDA0002763444190000012
Figure FDA0002763444190000013
Figure FDA0002763444190000013
 其中，(x,y)表示局部窗内像素点坐标，I(x,y)表示局部窗内像素点的像素值，W表示局部窗，N1表示C中像素个数，μs(x,y)表示(x,y)位置像素局部窗内的均值，
Figure FDA0002763444190000014
表示C中每一个点的μs均值，σ表示的是局部窗内均值的方差；Among them, (x, y) represents the coordinates of the pixel point in the local window, I(x, y) represents the pixel value of the pixel point in the local window, W represents the local window, N 1 represents the number of pixels in C, μ s (x, y) represents the mean value in the local window of the (x, y) position pixel,
Figure FDA0002763444190000014
represents the μ s mean of each point in C, and σ represents the variance of the mean in the local window; S4：遍历所述图像，计算出每一个邻域点的特征，计算种子点和邻域点位置的欧氏距离，与统计量的距离的加权距离di,j，比较di,j形成超像素；S 4 : Traverse the image, calculate the feature of each neighborhood point, calculate the Euclidean distance between the seed point and the neighborhood point, and the weighted distance d i,j of the distance from the statistic, compare d i,j to form superpixel; 所述邻域点和种子点之间的距离dci,j：The distance dc i,j between the neighborhood point and the seed point:
Figure FDA0002763444190000015
Figure FDA0002763444190000015
 其中，μi表示种子点i的均值，μj表示邻域点j的均值，σi表示局部窗内i的均值的方差，σj表示局部窗内j的均值的方差where μ i represents the mean value of seed point i, μ j represents the mean value of neighborhood point j, σ i represents the variance of the mean value of i in the local window, σ j represents the variance of the mean value of j in the local window
Figure FDA0002763444190000016
Figure FDA0002763444190000016
 其中xi表示种子点i的行坐标，yi表示种子点i的列坐标，xj表示邻域点j的行坐标，yj表示邻域点j的列坐标where x i represents the row coordinate of the seed point i, yi represents the column coordinate of the seed point i, x j represents the row coordinate of the neighborhood point j, y j represents the column coordinate of the neighborhood point j
Figure FDA0002763444190000017
Figure FDA0002763444190000017
 其中，m表示dci,j与dsi,j之间的权重系数；Among them, m represents the weight coefficient between dc i,j and ds i,j ; 相邻种子点之间的距离SThe distance S between adjacent seed points
Figure FDA0002763444190000021
Figure FDA0002763444190000021
 其中，N表示图像像素总数，k表示种子点数；Among them, N represents the total number of image pixels, and k represents the number of seed points; S5：更新种子点的位置为每一类超像素中所包含的所有点位置的均值；S 5 : The position of the updated seed point is the mean value of the positions of all points contained in each type of superpixel; S6：重复步骤S3-S5直到所述超像素的边界不再发生变化，停止重复，输出超像素图像； S6 : Repeat steps S3 - S5 until the boundary of the superpixel no longer changes, stop repeating, and output a superpixel image; S7：输入所述超像素图像，对所述超像素使用kmeans方法初始聚成2类，并将该聚类结果当作初始的标记场Xsp并计算势能Wsp(Xsp,Usp)；S7: Input the superpixel image, use the kmeans method to initially cluster the superpixels into two categories, take the clustering result as the initial marking field Xsp and calculate the potential energy Wsp ( Xsp , Usp ) ; S8：更新标记场Xsp和辅助场Usp，并通过ICM方法和SG方法对参数集θ进行更新：S 8 : Update the marker field X sp and the auxiliary field U sp , and update the parameter set θ by the ICM method and the SG method: S9：重复步骤S7-S8直到标记场不再发生变化为止，输出海岸线结果。 S9 : Repeat steps S7 - S8 until the marker field no longer changes, and output the result of the coastline. 2.根据权利要求1所述的一种基于Gamma分布超像素方法及超像素TMF的SAR图像海岸线检测方法，其特征还在于：步骤S3中所述聚类方法：2. a kind of SAR image coastline detection method based on Gamma distribution superpixel method and superpixel TMF according to claim 1, is characterized in that: clustering method described in step S 3 : 选取邻域中任意两个点作为种子点A和B，计算局部窗内，每一个点到A和B两个点的特征的差值的绝对值，当点与点A的差值的绝对值小于与B的差值的绝对值，则当前点与A点属于同一类；当点与点A的差值的绝对值大于与B的差值的绝对值，则当前点与B点属于同一类,则形成对比度较大的两类。Select any two points in the neighborhood as seed points A and B, and calculate the absolute value of the difference between the features of each point and the two points A and B in the local window. When the absolute value of the difference between the point and point A If it is less than the absolute value of the difference with B, then the current point and point A belong to the same category; when the absolute value of the difference between point and point A is greater than the absolute value of the difference with B, then the current point and point B belong to the same category , the two types with larger contrast are formed. 3.根据权利要求1所述的一种基于Gamma分布超像素方法及超像素TMF的SAR图像海岸线检测方法，其特征还在于：3. a kind of SAR image coastline detection method based on Gamma distribution superpixel method and superpixel TMF according to claim 1, is characterized in that: 所述势能Wsp(Xsp,Usp)：The potential energy W sp (X sp ,U sp ):
Figure FDA0002763444190000022
Figure FDA0002763444190000022
 其中,xs表示中心超像素s的标签，xt表示邻域超像素t的标签，Csp表示超像素势团集合，fcos表示两向量之间夹角的余弦值，Wedge(xs,xt)表示邻域超像素t的权值，其表达式如下：Among them, x s represents the label of the central superpixel s, x t represents the label of the neighborhood superpixel t, C sp represents the superpixel potential set, f cos represents the cosine value of the angle between the two vectors, W edge (x s ,x t ) represents the weight of the neighborhood superpixel t, and its expression is as follows:
Figure FDA0002763444190000023
Figure FDA0002763444190000023
 其中，Ls,t表示中心超像素s和邻域超像素t公共边界的长度，Ls表示中心超像素s的周长。Among them, L s, t represents the length of the common boundary between the central superpixel s and the neighborhood superpixel t, and L s represents the perimeter of the central superpixel s. 4.根据权利要求1所述的一种基于Gamma分布超像素方法及超像素TMF的SAR图像海岸线检测方法，其特征还在于：所述超像素更新后的标记场
Figure FDA0002763444190000031
和超像素更新后的辅助场
Figure FDA0002763444190000032
4. a kind of SAR image coastline detection method based on Gamma distribution superpixel method and superpixel TMF according to claim 1, it is characterized in that: the mark field after described superpixel update
Figure FDA0002763444190000031
and superpixel updated auxiliary field
Figure FDA0002763444190000032
Figure FDA0002763444190000033
Figure FDA0002763444190000033
Figure FDA0002763444190000034
Figure FDA0002763444190000034
 其中，Xsp表示超像素的标记场，Usp表示超像素的辅助场，Wsp(Xxp,Usp)表示势能函数，SP表示超像素的集合，L表示SAR图像的视数，Isp表示超像素的值即超像素中所有像素值的均值，μi表示第i类超像素的均值。where X sp represents the marker field of the superpixel, U sp represents the auxiliary field of the superpixel, W sp (X xp , U sp ) represents the potential energy function, SP represents the set of superpixels, L represents the viewing number of the SAR image, and I sp The value representing the superpixel is the mean value of all pixel values in the superpixel, and μi represents the mean value of the i -th type of superpixel. 5.根据权利要求1所述的一种基于Gamma分布超像素方法及超像素TMF的SAR图像海岸线检测方法，其特征还在于：所述SG方法参数更新公式为：5. a kind of SAR image coastline detection method based on Gamma distribution superpixel method and superpixel TMF according to claim 1, is characterized in that: described SG method parameter update formula is:
Figure FDA0002763444190000035
Figure FDA0002763444190000035
 其中，q表示迭代的次数，A是一个常量控制迭代的次数，x0和u0表示初始的标签值和辅助场的值，xq+1和uq+1表示第q+1次迭代得到的标签值和辅助场的值，p表示的是参数集中的第p个参数，Wf表示的是势能函数，
Figure FDA0002763444190000036
表示的是第q次参数的估计值。Among them, q represents the number of iterations, A is a constant controlling the number of iterations, x 0 and u 0 represent the initial label value and the value of the auxiliary field, and x q+1 and u q+1 represent the q+1th iteration to obtain The label value of and the value of the auxiliary field, p represents the p-th parameter in the parameter set, W f represents the potential energy function,
Figure FDA0002763444190000036
represents the estimated value of the qth parameter.
CN201710449129.4A 2017-06-14 2017-06-14 Gamma distribution superpixel-based method and superpixel TMF-based SAR image coastline detection method Expired - Fee Related CN107256399B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710449129.4A CN107256399B (en) 2017-06-14 2017-06-14 Gamma distribution superpixel-based method and superpixel TMF-based SAR image coastline detection method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710449129.4A CN107256399B (en) 2017-06-14 2017-06-14 Gamma distribution superpixel-based method and superpixel TMF-based SAR image coastline detection method

Publications (2)

Publication Number Publication Date
CN107256399A CN107256399A (en) 2017-10-17
CN107256399B true CN107256399B (en) 2021-01-15

Family

ID=60024091

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710449129.4A Expired - Fee Related CN107256399B (en) 2017-06-14 2017-06-14 Gamma distribution superpixel-based method and superpixel TMF-based SAR image coastline detection method

Country Status (1)

Country Link
CN (1) CN107256399B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110096961B (en) * 2019-04-04 2021-03-02 北京工业大学 A Superpixel-Level Semantic Annotation Method for Indoor Scenes
CN110097101B (en) * 2019-04-19 2022-09-13 大连海事大学 Remote sensing image fusion and coastal zone classification method based on improved reliability factor

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105389826A (en) * 2015-12-23 2016-03-09 国家海洋局第一海洋研究所 High-resolution SAR remote sensing extraction method for coastline of coral island
CN106023179A (en) * 2016-05-13 2016-10-12 江苏科技大学 SAR image coastline extracting method based on geometric active contour model
CN106156758A (en) * 2016-07-29 2016-11-23 辽宁工程技术大学 A kind of tidal saltmarsh method in SAR coast chart picture

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105389826A (en) * 2015-12-23 2016-03-09 国家海洋局第一海洋研究所 High-resolution SAR remote sensing extraction method for coastline of coral island
CN106023179A (en) * 2016-05-13 2016-10-12 江苏科技大学 SAR image coastline extracting method based on geometric active contour model
CN106156758A (en) * 2016-07-29 2016-11-23 辽宁工程技术大学 A kind of tidal saltmarsh method in SAR coast chart picture

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
SAR图像海岸线检测与地物分类;魏明月;《中国优秀硕士学位论文全文数据库 信息科技辑》;20170315;第2017年卷(第3期);第I136-2135页 *
Superpixel-based coastline extraction in SAR images with speckle noise removal;Xiaofang Liu et al.;《2016 IEEE International Geoscience and Remote Sensing Symposium (IGARSS)》;20161103;全文 *

Also Published As

Publication number Publication date
CN107256399A (en) 2017-10-17

Similar Documents

Publication Publication Date Title
CN107301661B (en) High-resolution remote sensing image registration method based on edge point features
CN110659591B (en) SAR image change detection method based on twin network
Tang et al. Compressed-domain ship detection on spaceborne optical image using deep neural network and extreme learning machine
CN110334762B (en) Feature matching method based on quad tree combined with ORB and SIFT
CN110097101B (en) Remote sensing image fusion and coastal zone classification method based on improved reliability factor
CN103034863B (en) The remote sensing image road acquisition methods of a kind of syncaryon Fisher and multiple dimensioned extraction
CN107292922B (en) A method of it is registrated for optics with diameter radar image
CN111027446B (en) Coastline automatic extraction method of high-resolution image
CN111008585B (en) Ship target detection method based on self-adaptive layered high-resolution SAR image
CN107301649B (en) An Algorithm for Shoreline Detection in Region Merged SAR Images Based on Superpixels
CN110889843B (en) SAR image ship target detection method based on maximum stable extremal region
CN111145228A (en) Heterogeneous image registration method based on local contour point and shape feature fusion
CN110598613B (en) A kind of highway fog monitoring method
CN111046772A (en) Multi-temporal satellite remote sensing island shore line and development and utilization information extraction method
CN107067405A (en) Based on the preferred Remote Sensing Image Segmentation of yardstick
CN107909002B (en) Sea and land segmentation method of infrared remote sensing image based on coastline matching
CN109635789B (en) High-resolution SAR image classification method based on intensity ratio and spatial structure feature extraction
Wang et al. A novel multi-scale segmentation algorithm for high resolution remote sensing images based on wavelet transform and improved JSEG algorithm
CN101634706A (en) Method for automatically detecting bridge target in high-resolution SAR images
CN111353371A (en) Shoreline extraction method based on spaceborne SAR images
CN108647658A (en) A kind of infrared imaging detection method of high-altitude cirrus
Chen et al. Contour refinement and EG-GHT-based inshore ship detection in optical remote sensing image
CN107256399B (en) Gamma distribution superpixel-based method and superpixel TMF-based SAR image coastline detection method
CN110310263B (en) A method for detecting residential areas in SAR images based on saliency analysis and background priors
CN108765440B (en) Line-guided superpixel coastline extraction method of single-polarized SAR image

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
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20210115

Termination date: 20210614

CF01 Termination of patent right due to non-payment of annual fee