An island shoreline detection method based on phase consistency random walk

Abstract

The present application relates to the technical field of coast water edge line detection, and is an island water edge line detection method based on phase consistency random walk, comprising: converting a single polarization SAR image into a phase consistency image by using a two-dimensional logarithmic Gabor filter; obtaining a sea-land membership function by using an FCM method, and creating a global sea-land priori; constructing a cross-sea bridge marker field based on LSD, and creating a marker field priori of a non-water-permeable structure connected to an island; introducing a super-pixel layer to construct a water edge line detection model; inputting the single polarization SAR image into the water edge line detection model to obtain an output label of each pixel, and realizing island water edge line detection. The present application systematically solves the problems of coherent noise interference in the single polarization SAR image, difficulty in seed point initialization of the traditional random walk method, insufficient water edge line extraction precision caused by complex island contour, and invalidity of gray scale similarity in edge detection caused by large differences in ground object backscattering, and can effectively extract the island coast line of the single polarization SAR image.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of water edge line detection, and particularly relates to a sea island water edge line detection method based on phase consistency random walk. BACKGROUND

[0002] At present, the sea island coastline extraction of single-polarized SAR images can be divided into seven methods: edge detection (ED) based method, threshold based method, region merging (RM) based method, partial differential equation (PDE) based method, deep learning (DL) based method, Markov random field (MRF) based method and superpixel (SP) based method.

[0003] The edge detection based method has the advantages of simple principle, easy understanding and strong operability in water edge line extraction, and shows good performance in processing SAR images. However, the threshold of the edge detection based method is difficult to control, and false edges are prone to occur.

[0004] The threshold based method has the advantages of simple calculation and fast operation speed, but is not suitable for detecting complex water edge lines with uneven contrast.

[0005] The region merging based method has the advantages of simple principle and easy implementation, but it is difficult to automatically select seed points, and cannot solve the problem of low-contrast region extraction.

[0006] The partial differential equation based method has the advantages of relatively complete theory, certain automatic evolution and good edge fitting, but has defects such as sensitivity to initial contour, slow image calculation speed and the need for manual adjustment of parameters.

[0007] The deep learning based method has the advantages of strong feature learning ability and end-to-end learning process, but has deficiencies in sample quantity, automaticity and generalization ability.

[0008] The Markov random field based method can make full use of context and structure information, but has insufficient detection ability for low-contrast and uneven regions, and slow calculation speed.

[0009] The superpixel based method takes superpixels as primitives, which can greatly reduce the computational complexity in subsequent processing, but the precision of complex coastline is unstable.

[0010] In addition, in the VV monopolar SAR image, the backscattering coefficients of the land objects such as beaches and forests on the islands present spatial non-uniform characteristics, so that the contrast of the waterline area of the islands is extremely low, in addition, the coherent speckle of the complex sea surface exists, which causes the false detection problem of the existing algorithm, in addition, the island-land and island-island bridges present non-uniform characteristics and even part of the bridges appear to be broken, which brings difficulties to the research on the detection of the waterline of the islands based on the SAR image. Therefore, it is necessary to provide a method to solve the interference of the coherent noise in the monopolar SAR image, the difficulty of seed point initialization, the insufficient extraction accuracy of the waterline, and the similarity failure problem. SUMMARY

[0011] According to the technical problem proposed above, a waterline detection method of islands based on phase consistency random walk is provided. The present application starts from the phase consistency domain, in the prior construction, for the complex multi-island environment, the fuzzy C clustering mean (FCM) is used to calculate the membership degree of land and sea to provide global prior, at the same time, the line segment detection (LSD) based on phase consistency is used to provide a method for detecting the water-penetrating structures such as cross-sea bridges. The similarity measurement is established based on phase consistency, the superpixel layer is increased to better utilize the image information in the non-local neighborhood, and two prior construction state transition matrices are introduced to complete the construction and solution of the RW model based on the SAR image.

[0012] The technical means adopted by the present application are as follows:

[0013] A waterline detection method of islands based on phase consistency random walk, comprising: converting a monopolar SAR image into a phase consistency image by using a two-dimensional logarithmic Gabor filter; obtaining a sea-land membership function by using an FCM method to create a global prior of sea and land; constructing a cross-sea bridge label field based on LSD to create a label field prior of non-water-penetrating structures connected with the islands; introducing a superpixel layer to construct a waterline detection model; inputting the monopolar SAR image into the waterline detection model to obtain an output label of each pixel, and realizing the detection of the waterline of the islands.

[0014] Further, the logarithmic Gabor filter is defined as:

[0015]

[0016] wherein, and respectively represent the angular frequency and direction corresponding to the image pixel at the position in the polar coordinates, and respectively represent the scale and direction of the Gabor filter, and respectively denote the center frequency and the center direction, for different center frequencies, is the standard deviation of the Gaussian function in the angular direction;

[0017] The log-Gabor filter in the spatial domain is represented by the inverse Fourier transform:

[0018]

[0019] where, and denote the real and imaginary parts of the current log-Gabor wavelet at scale and direction and respectively; the response of the image to the spatial filter is denoted by:

[0020]

[0021] where, denotes the convolution; denotes the pixel value of the image at position , denotes the real part response at scale and direction and respectively, denotes the imaginary part response at scale and direction and respectively.

[0022] Further, the obtaining the phase coherence map specifically comprises: based on the response of the image to the spatial filter, the phase coherence of each direction is defined as:

[0023]

[0024] where, , denote the direction and scale of the log-Gabor filter respectively; denotes the interval parameter, denotes the number of scales; is the estimated noise level, is a constant, denotes that the contained quantity is equal to itself when its value is positive, otherwise it is zero; the amplitude response is denoted by:

[0025]

[0026] denotes the phase deviation function of a log-Gabor filter at scale ,direction ,is defined as:

[0027]

[0028]

[0029] wherein and are denoted as:

[0030]

[0031]

[0032] the weight function is defined as:

[0033]

[0034] wherein is a gain factor to control the sharpness of the filter, is a cut-off value, is a frequency response extension, denoted as:

[0035]

[0036] wherein N is the total number of scales considered, is the amplitude of the n-th scale at x, is the amplitude with the maximum response at x.

[0037] Further, the phase consistency map is transformed from image pixels to a phase consistency measure :

[0038]

[0039] wherein denotes a mathematical function of phase consistency.

[0040] Further, the sea-land global prior is created, specifically comprising: obtaining the membership of sea-land by the FCM classifier using the measure of phase consistency as input, the membership is between , defining as the membership of the n-th node of phase consistency, in order to embody the sea-land two kinds of prior, defining , then introducing it as a prior into the waterline detection model.

[0041] Furthermore, the creation of a priori marker field for the impermeable structures connected to the island specifically includes: assuming the existence of a line segment detection function. ,get line segment , To locate the bridge, the following conditions must be met: any two line segments must be in opposite directions and have similar absolute values; the distance between the two line segments must be similar; and each line segment must be long enough to mark the bridge. Defined as:

[0042]

[0043] in, Represents line segment direction, This represents the calculation of the perpendicular distance between two line segments. This represents the calculation of the length of a line segment. , and These represent the thresholds for angle, distance, and length, respectively.

[0044] Furthermore, the waterline detection model specifically includes: an assumed phase consistency map. have There are 10 nodes, each representing 10 nodes. One element, while based on the phase consistency graph Generated by superpixels Each node is represented by the mean of each superpixel, and the data is integrated. Element nodes and superpixel nodes, totaling A weighted graph is generated using nodes, represented as follows: ,in It is a set of nodes. It is a set of edges. The weight matrix has edge connections; in the node set any node Indicates the first indivual Element nodes or superpixel nodes, in the edge set any edge in Indicates at node In the four-neighbor or eight-neighbor domains of the node The connection, It consists of labeled and unlabeled nodes. This represents a set of labeled nodes. , This indicates that there are K labels, and two category labels are needed for land-sea segmentation. Represents a set of uncategorized labels;

[0045] Assumption The empty domain of a domain element is ,Depend on The set of spatial domains composed of superpixels of a domain is :

[0046]

[0047] in, Indicates the first The spatial set of the n superpixels, the nth Each superpixel The spatial range is represented as .

[0048] Furthermore, based on the aforementioned phase consistency diagram Connections between nodes in the domain, connections between nodes in the superpixel domain, and The connection relationships between nodes in the domain and corresponding nodes in the superpixel domain are defined, and the weight matrix is ​​defined. for:

[0049]

[0050] Among them, the weight , , , They are defined as follows:

[0051]

[0052]

[0053]

[0054]

[0055] in, Phase coherence diagram Nodes in the domain i The value, Phase coherence diagram Nodes in the domain j The value, Indicates the first i The mean value of each superpixel. Indicates the first j The mean of each superpixel;

[0056] based on Domain elements and superpixels have the same spatial location, and nodes degree Defined as:

[0057]

[0058] Introducing cluster membership , the node Prior weights are defined as:

[0059]

[0060] where, is a regular parameter, the transition probability is defined as:

[0061]

[0062] where, and are the weights of two nodes;

[0063] Based on the transition probability , the probability of going from node to node m with label is:

[0064]

[0065] where, when , , otherwise ;

[0066] By setting the vector , the vector expression of the probability is:

[0067]

[0068] where, , the transition probability matrix is defined as:

[0069]

[0070] is a vector, where:

[0071]

[0072] For any node , the average reaching probability is calculated, and its vector representation is:

[0073]

[0074] The label is obtained by calculation:

[0075] .

[0076] Compared with the prior art, the present application has the following advantages:

[0077] The island waterline detection method based on phase consistency random walk provided by the present application converts a single-polarized SAR image into a phase consistency image by using a two-dimensional logarithmic Gabor filter; obtains a sea-land membership function by using an FCM method, and creates a global sea-land priori; constructs a cross-sea bridge marker field based on LSD, and creates a marker field priori of a non-water-permeable structure connected with an island; and introduces a super-pixel layer to construct a waterline detection model. The present application can effectively extract the island coastline of a single-polarized SAR image, and systematically solves the problems of coherent noise interference in a single-polarized SAR image, difficulty in seed point initialization of a traditional random walk method, insufficient waterline extraction precision caused by a complex island contour, and invalidity of gray scale similarity in edge detection caused by large differences in ground object backscattering.

[0078] The island waterline detection method based on phase consistency random walk provided by the present application fully utilizes the constructed phase consistency image, establishes similarity measurement through phase consistency, introduces two priors to construct a state transition matrix, and adopts a model architecture of 'transform domain-prior-walk' in a phase consistency domain, thereby avoiding extracting the boundaries of two sides of a cross-sea bridge as waterlines, and improving the accuracy and practicability of waterline detection.

[0079] The island waterline detection method based on phase consistency random walk provided by the present application is not only suitable for island waterline detection, but also can be theoretically extended to multi-target extraction of single-polarized SAR images in other complex scenes, and has certain universality.

[0080] Based on the above reasons, the present application can be widely popularized in the field of waterline detection. BRIEF DESCRIPTION OF DRAWINGS

[0081] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without any creative labor.

[0082] Figure 1 The figure is a method architecture diagram of the island waterline detection method based on phase consistency random walk of the present application.

[0083] Figure 2 The figure is a comparison diagram of island waterline extraction in the embodiment of the present application. DETAILED DESCRIPTION

[0084] It should be noted that the embodiments and features of the embodiments in the present application can be combined with each other in the case of no conflict. The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings and in combination with the embodiments.

[0085] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings of the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, but not all the embodiments. The description of the at least one exemplary embodiment below is actually only illustrative, but not as any limitation on the present application and its application or use. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0086] It should be noted that the terms used herein are only intended to describe specific embodiments, and are not intended to limit the exemplary embodiments according to the present application. As used herein, the singular form is intended to include the plural form unless the context clearly indicates otherwise, and it should also be understood that when the terms "comprise" and / or "include" are used in the specification, there is a presence of a feature, step, operation, device, component and / or combination thereof.

[0087] Unless specifically stated otherwise, the relative arrangement of the components and steps illustrated in these embodiments and the numerical expressions and values set forth herein are not limiting of the scope of the present application. It should be understood that the various parts of the drawings are not necessarily drawn to scale, and that, for the purpose of convenience and clarity, not all components and steps can be shown in a given figure. Techniques, methods, and apparatus known to those of ordinary skill in the relevant art can not be discussed in detail, but are intended to be a part of the specification when discussing the various embodiments. In all examples shown and discussed herein, any specific values are intended to be exemplary only and are not limiting the scope of the application. Other examples of the exemplary embodiments can have different values. It should be noted that like numbers and letters refer to like elements throughout the several views of the drawings and that the exemplified embodiments can not reflect the specific numbers of the drawings. Thus, specific embodiments are not limited to the exemplified examples, but include other examples that can result in a different result.

[0088] As Figure 1As shown, the present application provides an island waterline detection method based on phase coherence random walk, in the VV single polarization SAR image, the backscattering coefficient of island beach and forest and other objects presents spatial non-uniform characteristics, so that the contrast of island waterline area is extremely low, in addition, the coherent speckle of complex sea surface exists, which causes the false detection problem of existing algorithm, in addition, the island-land and island-island bridge presents non-uniform characteristics and even part of the fracture, which brings difficulties to the research of island waterline detection based on SAR image, and an effective solution is urgently needed. Through research, it is found that the SAR image obtained by phase coherence can greatly reduce the interference caused by these problems.

[0089] The single polarization SAR image is converted into a phase coherence map by using a two-dimensional log Gabor filter; in specific implementation, as a preferred embodiment of the present application, the log Gabor filter is defined as:

[0090]

[0091] wherein, and respectively represent the corresponding angular frequency and direction of the image pixel at position in polar coordinates, and respectively represent the scale and direction of the Gabor filter, and respectively represent the center frequency and center direction, for different center frequencies, is the standard deviation of the Gaussian function in the angle direction;

[0092] As a frequency domain filter, the log Gabor filter is represented by inverse Fourier transform in space domain:

[0093]

[0094] wherein, and respectively represent the real part and the imaginary part of the current log Gabor wavelet at scale and direction and respectively; the response of the image to the space domain filter is represented as:

[0095]

[0096] wherein, represents convolution; represents the pixel value of the image at position , represents in scale and direction and real part response, denotes imaginary part response in scale and direction and .

[0097] In practice, as a preferred embodiment of the present application, the phase coherence map is obtained specifically comprising:

[0098] Based on the response of the image through a spatial filter, the phase coherence in each direction is defined as:

[0099]

[0100] wherein, , denote the direction and scale of the log-Gabor filter respectively; denotes the interval parameter, denotes the number of scales; is set to 6, i.e. the interval is , is set to 4. is usually set to 3, is a very small constant to avoid division by zero. denotes the included quantity is equal to itself when its value is positive, otherwise zero; the amplitude response denotes:

[0101]

[0102] denotes the phase deviation function of the log-Gabor filter in scale and direction , is defined as:

[0103]

[0104]

[0105] wherein, and denote:

[0106]

[0107]

[0108] The weight function is defined as:

[0109]

[0110] in, The gain factor, which controls the sharpness of the filter, is set to 10. The cutoff value is set to 0.4. The frequency response extension is expressed as:

[0111]

[0112] Where N is the total number of scales considered. It is the amplitude at the nth scale at point x. It is the amplitude at which the maximum response is located at x.

[0113] Based on the above analysis, the image Phase consistency can be obtained in multiple directions. This means that in position... Existing vector Current literature generally uses multi-directional averaging to describe the phase consistency at a given location, which often introduces blurring at weak edges; the idea is similar to Gaussian smoothing. To highlight phase consistency information in a specific direction, in a preferred embodiment of this invention, the phase consistency map is transformed from image pixels into a phase consistency metric. :

[0114]

[0115] in, A mathematical function representing phase consistency.

[0116] FCM (Fixed Component Clustering) is a soft clustering method. Its core idea is to determine the membership degree between sample points and cluster centers based on iterative optimization of the objective function. The membership degree is used to represent the relationship between data points, thereby determining the cluster to which each sample point belongs. In this invention, FCM is used to obtain the sea-land membership function, creating a global sea-land prior. Specifically, in a preferred embodiment, a phase consistency metric is used as input to obtain the sea-land membership degree through the FCM classifier. This membership degree... Between, definition For phase consistency The membership degree of a node is defined to reflect the two priors of land and sea. This was then introduced as a priori into the waterline detection model.

[0117] In multi-island scenarios, permeable structures are mainly cross-sea bridges, which can usually be represented by line segments. Therefore, determining their marker field directly involves detecting the corresponding line segments. This section proposes a method for constructing a cross-sea bridge marker field based on LSD (Low Segment Detection). The method involves constructing an LSD-based cross-sea bridge marker field and creating prior marker fields for non-permeable structures connected to islands. Specifically, as a preferred embodiment of this invention, it is assumed that a line segment detection function exists. ,get line segment , To locate the bridge, the following conditions must be met: any two line segments must be in opposite directions and have similar absolute values; the distance between the two line segments must be similar; and each line segment must be long enough to mark the bridge. Defined as:

[0118]

[0119] in, Represents line segment direction, This represents the calculation of the perpendicular distance between two line segments. This represents the calculation of the length of a line segment. , and These represent the thresholds for angle, distance, and length, respectively.

[0120] In traditional RW (Rapid Wave Wiring), the walking path often connects local pixels. Pixel-based approaches can only compare adjacent pixels pairwise, and in SAR images with speckle noise, they often stop walking when encountering local noise points. Introducing a superpixel layer addresses this problem. Compared to local and global prior connections, superpixel connections better utilize image information within non-local neighborhoods. A waterline detection model is constructed using a superpixel layer; in a preferred embodiment of this invention, a phase consistency map is assumed... have There are 10 nodes, each representing 10 nodes. One element, while based on the phase consistency graph Generated by superpixels Each node is represented by the mean of each superpixel, and the data is integrated. Element nodes and superpixel nodes, totaling A weighted graph is generated using nodes, represented as follows: ,in It is a set of nodes. It is a set of edges. The weight matrix has edge connections; in the node set any node Indicates the first indivualElement node or superpixel node, any edge in edge set between nodes in the four-neighborhood or eight-neighborhood of node , the connection between node and node is divided into three cases, the connection between element nodes, the connection between superpixel nodes and the connection between superpixel and element node, the connection between element nodes is the same as the connection between pixels in general RW theory, the connection between superpixel nodes is only the connection between the center of each superpixel and the center of another superpixel, and the connection between superpixel and element node is special, the center of superpixel only connects with the elements in this superpixel region.

[0121] composed of labeled and unlabeled nodes, denotes the set of labeled nodes, , denotes that there are K labels, two class labels are needed for sea-land segmentation, denotes the set of class labels;

[0122] Suppose the empty space of domain element is , the empty space set composed of superpixels in domain is :

[0123]

[0124] wherein, denotes the spatial set of the th superpixel, the spatial range of the th superpixel in is denoted as .

[0125] According to the phase consistency graph the connection of nodes in the domain, the connection of nodes in the superpixel domain and the connection relationship between nodes in the domain and corresponding nodes in the superpixel domain, the weight matrix is defined as:

[0126]

[0127] wherein, the weights , , , are respectively defined as:

[0128]

[0129] ​​

[0130]

[0131]

[0132] wherein, denotes the phase consistency map denotes the value of the node i in the domain, denotes the phase consistency map denotes the value of the node j in the domain, denotes the mean value of the i -th superpixel, denotes the mean value of the j -th superpixel; the four weights describing the relevance of the nodes and describe the spatial relationship and the similarity measure of the domain elements to the superpixels.

[0133] Based on the fact that the domain elements and the superpixels have the same spatial position, the degree of the node is defined as:

[0134]

[0135] However, only relying on to complete the RW will have the general problem of RW, that is, when the walking node is too far from the seed node, the walking probability is too small, thus causing the problem of segmentation error, therefore, the clustering membership is introduced, and the prior weight of the node is defined as:

[0136]

[0137] wherein, is a regular parameter, which is a constant greater than 0 and less than 1, and the transition probability is defined as:

[0138]

[0139] wherein, and are the weights of the two nodes;

[0140] Based on the transition probability , the probability of walking from the node to the node m with the label is:

[0141]

[0142] ​​wherein, when , , otherwise ;

[0143] By setting the vector , the vector expression of the probability is:

[0144]

[0145] wherein, , the transition probability matrix is defined as:

[0146]

[0147] is a vector, wherein:

[0148]

[0149] For any node , the average reaching probability is calculated, and the vector representation is:

[0150]

[0151] The label is obtained by calculation:

[0152] .

[0153] The monopolar SAR image is input into the water edge line detection model to obtain the output label of each pixel, and the water edge line detection of the island is realized:

[0154] The VV polarization image , the label , , , , , ;

[0155] (1) According to the measurement of phase consistency, generate a PCI image, and generate a superpixel image;

[0156] (2) Generate a label vector , , generate a BF according to the bridge marking field ;

[0157] (3) Generate a weight matrix , and obtain according to the FCM algorithm;

[0158] (4) Calculate the transition matrix and ;

[0159] (5) Solve linear equations ;

[0160] (6) Normalize arrival probability ;

[0161] (7) Obtain the output label of each pixel. Output: output the island waterline label for each pixel .

[0162] Embodiments

[0163] As shown in Figure 2 , in this embodiment, the waterline detection results of the Ground Truth algorithm, the KM-ACM algorithm, the HED-Unet algorithm, the SPEC algorithm and the detection result of the algorithm of the present application are compared, Figure 2 , (a) is the original image, (b) is the Ground Truth (true value map), (c) is the running result of the KM-ACM algorithm, (d) is the running result of the HED-Unet algorithm, (e) is the running result of the SPEC algorithm, and (f) is the running result of the algorithm of the present application. From Figure 2 , it can be seen that the waterline detected by the HED-Unet extends to the land in many places, the extraction result of the SPEC has some errors compared with the real waterline, the extraction result of the KM-ACM is better in overall adhesion, but the recognition effect is slightly poor in some small sharp places, and the extraction result of the PCRW is basically consistent with the real waterline, and the extraction effect is the best.

[0164] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. An island shoreline detection method based on phase consistency random walk, characterized in that, The method comprises the following steps: The monopolar SAR image is converted into a phase consistency map by using a two-dimensional logarithmic Gabor filter; the logarithmic Gabor filter is defined as: wherein, and respectively denote the image pixel at position corresponding angular frequency and direction in polar coordinates, and respectively denote the scale and orientation of the Gabor filter, and respectively denote the center frequency and center orientation, is kept constant for different center frequencies, is the standard deviation of the Gaussian function in the angular direction; The logarithmic Gabor filter is expressed by inverse Fourier transform in the spatial domain: in, and These represent the current logarithmic Gabor wavelet at scale and direction, respectively. and The real and imaginary parts; the image The response after the spatial filter is expressed as: wherein, denotes a convolution; denotes the pixel value of the image at position , denotes the real part response at scale and direction and , denotes the imaginary part response at scale and direction and ; The FCM method is used to obtain a sea-land membership function, and a global sea-land priori is created; A cross-sea bridge marker field based on LSD is constructed, and a marker field priori of a non-water-permeable structure connected with the island is created; A superpixel layer is introduced to construct a waterline detection model; The monopolar SAR image is input into the waterline detection model, and an output label of each pixel is obtained, so that the waterline detection of the island is realized.

2. The phase coherence based random walk based island shoreline detection method of claim 1, wherein, The phase consistency map is obtained, specifically comprising: Based on image The phase coherence in each direction is defined as the response of a spatial filter that passes only the phase information of the direction of interest. wherein , respectively denote the orientation and the scale of the log-Gabor filter; denotes the interval parameter, denotes the number of scales; is the estimated noise level, is a constant, denotes the included quantity which is equal to itself when its value is positive, otherwise zero; the amplitude response denotes: denotes the phase deviation function of a log-Gabor filter at scale , orientation , is defined as: wherein and are represented by: The weight function is defined as: wherein is a gain factor for controlling the filter sharpness, is a cutoff value, is a frequency response extension, expressed as: where N is the total number of scales considered, is the amplitude of the n-th scale at x, is the amplitude with the maximum response at x.

3. The phase coherence-based random walk based island shoreline detection method of claim 2, wherein, transforming the phase coherence map from image pixels to a phase coherence metric : wherein denotes a mathematical function of phase coherence.

4. The phase coherence-based random walk based island shoreline detection method of claim 3, wherein, The global sea-land priori is created, specifically comprising: Using a phase consistency metric as input, the sea-land membership degree is obtained through an FCM classifier. This membership degree... Between, definition For phase consistency The membership degree of a node is defined to reflect the two priors of land and sea. This was then introduced as a priori into the waterline detection model.

5. The phase coherence based random walk based island shoreline detection method of claim 1, wherein, The marker field priori of the non-water-permeable structure connected with the island is created, specifically comprising: Assume that there is a line segment detection function , get line segment , , positioning the bridge needs to meet: any two line segments opposite direction and absolute value close, two line segment distance close, each line segment length is long enough, will mark the field defined as: wherein, represents the direction of a line segment , represents the perpendicular distance calculation of two line segments, represents the length calculation of a line segment, , and represent the threshold values of angle, distance and length, respectively.

6. The phase coherence based random walk based island shoreline detection method of claim 1, wherein, The waterline detection model, specifically comprising: phase consistency graph has nodes, each node represents one element, meanwhile based on the phase consistency graph generated by superpixels nodes, each node is represented by the mean value of each superpixel, integrating element nodes and superpixel nodes, a total of nodes, a weighted graph is generated by using the nodes, denoted as where is the set of nodes, is the set of edges, is the weight matrix with edge connection; in the node set any one node represents the th element node or superpixel node, in the edge set any one edge represents the connection between the node and the node in the four-neighborhood or eight-neighborhood of the node composed of labeled and unlabeled nodes, represents the set of labeled nodes, , represents K labels, two class labels are needed for sea-land segmentation, represents the set of no class labels; Assume The empty space of a domain element is The empty space of a domain element is The empty space of a domain element is : wherein, represents a spatial set of the th superpixel, the th superpixel is represented in the spatial range of .​ 7. The phase coherence-based random walk based island shoreline detection method of claim 6, wherein, According to the phase consistency map connections of nodes in the domain, connections of nodes in the superpixel domain, and connections of nodes in the domain and corresponding nodes in the superpixel domain, defining a weight matrix is defined as: wherein the weights , , , are defined as follows: wherein, representing a phase consistency map nodes in the domain i value of, representing a phase consistency map nodes in the domain j value of, representing a mean value of the i super-pixel, representing a mean value of the j super-pixel; Based on Domain elements have the same spatial location with superpixels, the node Degree is defined as: Introducing cluster membership The nodes Prior weights are defined as: wherein is a rule parameter defining the transition probability as: wherein and are the weights of the two nodes; Based on transition probabilities From node The probability of walking to a node m with label is: wherein when , , otherwise ; By setting the vector , the probability vector expression is: wherein , transition probability matrix is defined as: , is a vector, wherein: For any node , the average reach probability is calculated, whose vector representation is: The label is obtained by calculation. 。

Images