A land-sea segmentation method, system, electronic device and medium

Abstract

The application provides a sea-land segmentation method, system, electronic equipment and medium, and belongs to the field of SAR image semantic segmentation. The sea-land segmentation method comprises the following steps: acquiring a synthetic aperture radar image; clustering pixel points in the synthetic aperture radar image based on a simulated plant growth algorithm and a simple linear iterative clustering algorithm to obtain a plurality of superpixels; and using a Dice coefficient to enhance the connectivity of the plurality of superpixels to obtain a sea-land segmentation result. The simulated plant growth algorithm and the simple linear iterative clustering algorithm are combined, and the Dice coefficient is used to enhance the connectivity of the superpixels, thereby improving the accuracy of sea-land segmentation of the SAR image.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of SAR image semantic segmentation, in particular to a sea-land segmentation method and system based on a simulated plant growth algorithm and a simple linear iterative clustering algorithm, an electronic device and a medium. BACKGROUND

[0002] Compared with optical imaging, synthetic aperture radar has the advantage of all-weather imaging, so that SAR (Synthetic Aperture Radar) images play an important role in strategic target identification and detection, disaster control, land resource monitoring, sea area management, map surveying and ship target identification. In recent years, SAR images have gradually attracted attention in the field of sea area management. One of the main problems is the sea-land segmentation and coastline monitoring problem. Due to long-term river sediment accumulation, reclamation and other reasons, the coastline is constantly changing. Effective monitoring of the changes in the coastline has certain practical significance for dynamic monitoring of sea area use.

[0003] The currently more typical method is a threshold-based segmentation algorithm. This algorithm has the characteristics of simplicity, efficiency and easy implementation, and is favored by scholars in many segmentation technologies. The main idea of the threshold segmentation technology is to select a suitable threshold from the gray value, and to divide the image into several different classes of pixel points by comparing the threshold with other gray values. The key to this method is the selection of the threshold. An optimal threshold or a combination of optimal threshold groups is found. However, this method only considers the gray value of the pixel itself and generally does not consider the spatial features. Therefore, it is sensitive to noise and has low precision in sea-land segmentation. It is difficult to use alone in practical applications and is usually combined with other methods.

[0004] In order to better realize the practical role of SAR images in detecting coastline changes, it is crucial to accurately and quickly segment the sea and land in SAR images. SUMMARY

[0005] The purpose of the present application is to provide a sea-land segmentation method, system, electronic device and medium, which can improve the precision of sea-land segmentation of SAR images.

[0006] To achieve the above purpose, the present application provides the following solutions:

[0007] A sea-land segmentation method, comprising:

[0008] obtaining a synthetic aperture radar image; the synthetic aperture radar image includes a sea portion and a land portion;

[0009] based on a simulated plant growth algorithm and a simple linear iterative clustering algorithm, clustering the pixel points in the synthetic aperture radar image to obtain a plurality of superpixels;

[0010] The connectivity of the plurality of superpixels is enhanced by using a Dice coefficient to obtain a land-sea segmentation result.

[0011] Optionally, pixel points in the synthetic aperture radar image are clustered based on a simulated plant growth algorithm and a simple linear iterative clustering algorithm to obtain a plurality of superpixels, specifically including:

[0012] An initial number of superpixels is determined;

[0013] A plurality of initial superpixel center points are determined according to a total number of pixel points in the synthetic aperture radar image and the initial number of superpixels;

[0014] Gradient values of each pixel point in the synthetic aperture radar image are calculated;

[0015] A plurality of optimal superpixel center points are determined by using a simulated plant growth algorithm according to the gradient values of each pixel point and the plurality of initial superpixel center points;

[0016] For any pixel point, a color domain distance between the pixel point and each optimal superpixel center point is calculated according to a color pixel value of the pixel point in a gray space and color pixel values of each optimal superpixel center point in the gray space;

[0017] A spatial domain distance between the pixel point and each optimal superpixel center point is calculated according to a coordinate point of the pixel point on a plane space and coordinate points of each optimal superpixel center point on the plane space;

[0018] A total distance between the pixel point and each optimal superpixel center point is determined according to the color domain distance and the spatial domain distance between the pixel point and each optimal superpixel center point;

[0019] Each pixel point is clustered according to the total distance between each pixel point and each optimal superpixel center point to obtain a plurality of superpixels.

[0020] Optionally, the plurality of initial superpixel center points are determined according to a total number of pixel points in the synthetic aperture radar image and the initial number of superpixels, specifically including:

[0021] The synthetic aperture radar image is divided into a plurality of regions according to the total number of pixel points in the synthetic aperture radar image and the initial number of superpixels;

[0022] For any region, a center pixel point in the region is taken as an initial superpixel center point of the region to obtain a plurality of initial superpixel center points.

[0023] Optionally, the plurality of optimal superpixel center points are determined by using a simulated plant growth algorithm according to the gradient values of each pixel point and the plurality of initial superpixel center points, specifically including:

[0024] For any initial superpixel center point, gradient values of each pixel point in the initial superpixel center point setting range are determined according to the initial superpixel center point, and a morphon concentration value of each pixel point in the initial superpixel center point setting range is calculated;

[0025] The pixel point with the maximum morphon concentration value in the initial superpixel center point setting range is taken as an optimal superpixel center point.

[0026] Optionally, the setting range is 3*3; and the following formula is used to calculate the morphon concentration value of the pixel point with the coordinate (x, y):

[0027]

[0028] wherein, T xy is the morphon concentration value of the pixel point with the coordinate (x, y), grad xy is the gradient value of the pixel point with the coordinate (x, y).

[0029] Optionally, the Dice coefficient is used to strengthen the connectivity of the multiple superpixels to obtain the sea-land segmentation result, and specifically includes:

[0030] The Dice coefficient is used to calculate the similarity between any two superpixels;

[0031] Two superpixels with a similarity less than a set threshold value are merged into one superpixel to obtain the sea-land segmentation result.

[0032] To achieve the above object, the present application further provides the following schemes:

[0033] A sea-land segmentation system, comprising:

[0034] An image acquisition unit, configured to acquire a synthetic aperture radar image; the synthetic aperture radar image includes a sea portion and a land portion;

[0035] A clustering unit, connected with the image acquisition unit, configured to cluster pixel points in the synthetic aperture radar image based on a simulated plant growth algorithm and a simple linear iterative clustering algorithm to obtain multiple superpixels;

[0036] A strengthening unit, connected with the clustering unit, configured to use the Dice coefficient to strengthen the connectivity of the multiple superpixels to obtain a sea-land segmentation result.

[0037] To achieve the above object, the present application further provides the following schemes:

[0038] An electronic device, comprising a memory and a processor, wherein the memory is configured to store a computer program, and the processor is configured to run the computer program to enable the electronic device to perform the above sea-land segmentation method.

[0039] To achieve the above object, the application further provides the following scheme.

[0040] A computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the sea-land segmentation method.

[0041] According to the embodiments of the application, the following technical effects are achieved.

[0042] The application combines the simulated plant growth algorithm and the simple linear iterative clustering algorithm, clusters the pixel points in the synthetic aperture radar image, obtains a plurality of superpixels, and in the process of strengthening connectivity, the Dice coefficient is used to merge the superpixels that are too small with the adjacent superpixels, so that the sea-land segmentation precision of the SAR image is improved. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings in the following description are only some embodiments of the application, and for those skilled in the art, other drawings can be obtained from these drawings without creative labor.

[0044] Figure 1 The flow chart of the sea-land segmentation method of the application;

[0045] Figure 2 The overall flow chart of the sea-land segmentation based on the simulated plant growth algorithm, the simple linear iterative clustering algorithm and the Dice coefficient;

[0046] Figure 3 The module schematic diagram of the sea-land segmentation system of the application.

[0047] Symbol explanation:

[0048] Image acquisition unit-1, clustering unit-2, strengthening unit-3. DETAILED DESCRIPTION

[0049] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only some of the embodiments of the application, but not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0050] The application aims to provide a sea-land segmentation method, system, electronic device and medium, which combines a simulated plant growth algorithm with a SLIC (Simple Linear Iterative Clustering) algorithm to cluster pixel points and improve the accuracy of sea-land segmentation of a SAR image.

[0051] In order to make the above-mentioned purposes, features and advantages of the application more obvious and easy to understand, the application will be further described in detail below with reference to the drawings and specific embodiments.

[0052] Embodiment one

[0053] As shown in the figure, the embodiment provides a sea-land segmentation method, which comprises: Figure 1

[0054] S1: Obtain a synthetic aperture radar image. In this embodiment, the synthetic aperture radar image is taken from two data sets of MSAR and SSDD. The synthetic aperture radar image includes a sea portion and a land portion.

[0055] S2: Cluster pixel points in the synthetic aperture radar image based on a simulated plant growth algorithm and a simple linear iterative clustering algorithm to obtain a plurality of superpixels.

[0056] S3: Use a Dice coefficient to strengthen the connectivity of the plurality of superpixels to obtain a sea-land segmentation result.

[0057] Further, S2 specifically comprises:

[0058] S21: Initialize the number of superpixels.

[0059] S22: Determine a plurality of initial superpixel center points according to the total number of pixel points in the synthetic aperture radar image and the number of superpixels.

[0060] Specifically, the synthetic aperture radar image is divided into a plurality of regions according to the total number of pixel points in the synthetic aperture radar image and the number of superpixels. For any region, the center pixel point in the region is taken as the initial superpixel center point of the region to obtain a plurality of initial superpixel center points.

[0061] In this embodiment, the clustering center (initial superpixel center point) is first initialized, the number of superpixels is preset according to the total number of pixel points in the synthetic aperture radar image, it is assumed that the synthetic aperture radar image has N pixel points, the number of pre-segmented superpixels is K, and the size of the segmented superpixels is N / K.

[0062] S23: Calculate the gradient value of each pixel point in the synthetic aperture radar image.

[0063] ​S24: According to the gradient value of each pixel point and the plurality of initial superpixel center points, the plurality of optimal superpixel center points are determined by using the simulated plant growth algorithm.

[0064] Specifically, the superpixel center points are reset in the c*c range of the initial superpixel center points. Considering that if the calculation range is too large, the calculation amount will be increased, thereby affecting the segmentation efficiency, therefore the value of c is 3. For any initial superpixel center point, according to the gradient value of each pixel point in the initial superpixel center point setting range (3*3), the morphogen concentration value of each pixel point in the initial superpixel center point setting range is calculated.

[0065] In this embodiment, the initial superpixel center points are optimized by using the simulated plant growth algorithm, each pixel point is regarded as a cell, and the absolute value of the gradient value of each pixel point is regarded as the morphogen of the cell. The morphogen concentration value of the pixel point with coordinate (x, y) in the 3*3 region is:

[0066]

[0067] Wherein, T xy is the morphogen concentration value of the pixel point with coordinate (x, y), grad xy is the gradient value of the pixel point with coordinate (x, y).

[0068] The pixel point with the maximum morphogen concentration value in the setting range of each initial superpixel center point is regarded as the optimal superpixel center point.

[0069] Obviously, T 11 +T 12 +T 13 +T 21 +...+T 33 =1, a probability space is generated, and after the morphogen concentration value is established, the mechanism of the plant's optical rotation is established. The high and low of the morphogen concentration value represents the size of the optimal growth opportunity of the cell in the probability space, that is, the cell with higher morphogen concentration value will have higher optimal growth opportunity, therefore the pixel point with the highest morphogen concentration will be the new superpixel center point.

[0070] After the superpixel center points are reset, a fixed size neighborhood is generated, that is, all pixel points in the 2S*2S region are assigned a class label (that is, each pixel point is classified into a superpixel). Then, the distance of each pixel point in the region to the superpixel center point is calculated. S is the average area of the superpixel.

[0071] S25: For any pixel point, according to the color pixel value of the pixel point in the gray space and the color pixel value of each optimal superpixel center point in the gray space, the color domain distance between the pixel point and each optimal superpixel center point is calculated.

[0072] Specifically, the color domain distance between the pixel point j and the optimal superpixel center point i is calculated by using the following formula:

[0073] d c,ij = (l j -l i ) 2 + (a j -a i ) 2 + (b j -b i ) 2 ;

[0074] wherein d c,ij is the color domain distance between the pixel point j and the optimal superpixel center point i, l j is the brightness of the pixel point j in the gray space, a j is the color domain between red and green of the pixel point j in the gray space, b j is the color domain between yellow and blue of the pixel point j in the gray space, l i is the brightness of the optimal superpixel center point i in the gray space, a i is the color domain between red and green of the optimal superpixel center point i in the gray space, and b i is the color domain between yellow and blue of the optimal superpixel center point i in the gray space.

[0075] S26: According to the coordinate points of the pixel points on the plane space and the coordinate points of the optimal superpixel center points on the plane space, the spatial domain distance between the pixel points and the optimal superpixel center points is calculated.

[0076] Specifically, the spatial domain distance between the pixel point j and the optimal superpixel center point i is calculated by using the following formula:

[0077] d s,ij = (x j -x i ) 2 + (y j -y i ) 2 ;

[0078] wherein d s,ij is the spatial domain distance between the pixel point j and the optimal superpixel center point i, x j is the horizontal coordinate of the pixel point j on the plane space, y j is the vertical coordinate of the pixel point j on the plane space, x i is the horizontal coordinate of the optimal superpixel center point i on the plane space, and y i is the vertical coordinate of the optimal superpixel center point i on the plane space.

[0079] S27: determining a total distance between the pixel point and each optimal superpixel center point according to the color domain distance and the spatial domain distance between the pixel point and each optimal superpixel center point.

[0080] Specifically, the total distance between the pixel point j and the optimal superpixel center point i is calculated by using the following formula:

[0081]

[0082] wherein D ij is the total distance between the pixel point j and the optimal superpixel center point i, Nc is the maximum color domain distance, N s is the maximum spatial domain distance in the superpixel, and m is a constant. N c is the maximum color domain distance, each superpixel cluster follows the same standard, the maximum color domain distance changes with the image and changes with the cluster, so a fixed constant m = 10 is taken instead of N c . The final total distance is as follows:

[0083]

[0084] In order to simplify the calculation complexity, the above formula is optimized as:

[0085] D ij = (S 2 *d c,ij ) + (m 2 *d s,ij ).

[0086] Since each pixel point is searched by multiple optimal superpixel center points, each pixel point can obtain the total distance with all the optimal superpixel center points around it, and the pixel point belongs to the superpixel centered on the optimal superpixel center point when the total distance is the smallest.

[0087] S28: clustering each pixel point according to the total distance between each pixel point and each optimal superpixel center point to obtain multiple superpixels.

[0088] As Figure 2 shown, in the process of iterative clustering of each pixel point, if the class label (belonging superpixel) of any pixel point changes, the number of superpixels is determined again.

[0089] After iterative optimization, problems such as too small superpixels may occur, and the superpixels smaller than a set threshold are set as X min , and the superpixels adjacent to X i (i = 1, 2,..., n), n is the number of superpixels adjacent to X minThe number of adjacent superpixels, two pixels with a color domain distance less than a certain threshold are classified into the same superpixel.

[0090] In the embodiment, the superpixels after the preliminary clustering are connected by using the Dice coefficient to strengthen the connectivity, and the too small superpixels are merged with adjacent superpixels, and finally the sea-land segmentation of the SAR image is realized.

[0091] Further, the S3 specifically comprises:

[0092] S31: the similarity between any two superpixels is calculated by using the Dice coefficient.

[0093] Specifically, the similarity D(X min ,X i ) between the superpixel X min and the superpixel X i is calculated by using the following formula:

[0094]

[0095] It should be noted that X min and X i represent the pixel point set of the corresponding superpixel.

[0096] S32: two superpixels with a similarity less than a set threshold are merged into one superpixel to obtain the sea-land segmentation result, that is, the segmented SAR image.

[0097] In order to verify the segmentation effect of the present application, the boundary tracking algorithm and the threshold-based segmentation algorithm are used to segment the SAR image respectively, and the segmentation results are compared with the segmentation result of the present application, it can be obtained that the boundary tracking algorithm cannot accurately identify the clear coastline, and the threshold-based segmentation algorithm is difficult to identify the coastline without complete contour. The segmentation method used in the present application can better preserve the quality of the image, and the whole image is segmented, and the coastline can be better identified.

[0098] The present application uses the simulated plant growth algorithm, optimizes the clustering part of the SLIC algorithm by using the phototropism growth mechanism of plants, optimizes the selection of the clustering center according to the distribution of auxin, so as to obtain better superpixel center points, iteratively clusters the pixel points according to the set distance function, and merges the too small superpixels with their adjacent superpixels by using the Dice coefficient when strengthening the connectivity, and finally realizes the fast and accurate sea-land segmentation of the SAR image.

[0099] Embodiment two

[0100] In order to perform the method corresponding to the above-mentioned embodiment one to realize the corresponding functions and technical effects, a sea-land segmentation system is provided below.

[0101] As Figure 3 shown, the sea-land segmentation system provided by the embodiment comprises an image acquisition unit 1, a clustering unit 2 and a strengthening unit 3.

[0102] The image acquisition unit 1 is configured to acquire a synthetic aperture radar image.

[0103] The clustering unit 2 is connected with the image acquisition unit 1, and the clustering unit 2 is configured to cluster pixel points in the synthetic aperture radar image based on a simulated plant growth algorithm and a simple linear iterative clustering algorithm, to obtain a plurality of superpixels.

[0104] The strengthening unit 3 is connected with the clustering unit 2, and the strengthening unit 3 is configured to strengthen connectivity of the plurality of superpixels by using a Dice coefficient, to obtain a sea-land segmentation result.

[0105] Compared with the prior art, the sea-land segmentation system provided by the embodiment has the same beneficial effects as the sea-land segmentation method provided by the first embodiment, which will not be repeated here.

[0106] Embodiment Three

[0107] The embodiment provides an electronic device, comprising a memory and a processor, the memory is configured to store a computer program, and the processor is configured to run the computer program to enable the electronic device to execute the sea-land segmentation method of the first embodiment.

[0108] Optionally, the electronic device can be a server.

[0109] In addition, the embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the sea-land segmentation method of the first embodiment.

[0110] In the specification, each embodiment is described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts of each embodiment can be referred to each other.

[0111] The principles and implementation manners of the present application are described by using specific examples in this paper, and the above description of the embodiments is only used to help understand the method and core idea of the present application; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation manner and application range will be changed. In conclusion, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A land-sea segmentation method, characterized by, The sea-land segmentation method comprises: acquiring a synthetic aperture radar image; the synthetic aperture radar image comprises a sea portion and a land portion; based on a simulated plant growth algorithm and a simple linear iterative clustering algorithm, clustering the pixel points in the synthetic aperture radar image to obtain a plurality of superpixels, specifically comprising: initializing the number of superpixels; determining a plurality of initial superpixel center points according to the total number of pixel points in the synthetic aperture radar image and the number of superpixels; calculating the gradient value of each pixel point in the synthetic aperture radar image; For any initial superpixel center point, the morpheme concentration value of each pixel within a set range of the initial superpixel center point is calculated based on the gradient values ​​of each pixel within that set range; the set range is 3. 3; The coordinates are calculated using the following formula: x , y The morpheme concentration value of the pixel: ;in, T xy The coordinates are ( x , y The morpheme concentration value of the pixel. grad xy The coordinates are ( x , y The gradient value of the pixel; setting the pixel point with the maximum morphological concentration value in the range of each initial superpixel center point as the optimal superpixel center point; for any pixel point, calculating the color domain distance between the pixel point and each optimal superpixel center point according to the color pixel value of the pixel point in the gray space and the color pixel value of each optimal superpixel center point in the gray space; calculating the spatial domain distance between the pixel point and each optimal superpixel center point according to the coordinate point of the pixel point on the plane space and the coordinate point of each optimal superpixel center point on the plane space; determining the total distance between the pixel point and each optimal superpixel center point according to the color domain distance and the spatial domain distance between the pixel point and each optimal superpixel center point; clustering each pixel point according to the total distance between each pixel point and each optimal superpixel center point to obtain a plurality of superpixels; using the Dice coefficient to enhance the connectivity of the plurality of superpixels to obtain a sea-land segmentation result.

2. The land-sea segmentation method of claim 1, wherein, According to the total number of pixel points in the synthetic aperture radar image and the number of superpixels, a plurality of initial superpixel center points are determined, specifically comprising: According to the total number of pixel points in the synthetic aperture radar image and the number of superpixels, the synthetic aperture radar image is divided into a plurality of regions; for any region, the center pixel point in the region is taken as the initial superpixel center point of the region to obtain a plurality of initial superpixel center points.

3. The land-sea segmentation method of claim 1, wherein, Using the Dice coefficient to enhance the connectivity of the plurality of superpixels to obtain a sea-land segmentation result, specifically comprising: using the Dice coefficient to calculate the similarity between any two superpixels; two superpixels with a similarity less than a set threshold are merged into one superpixel to obtain a sea-land segmentation result.

4. A sea-land division system for use in the sea-land division method according to any one of claims 1 to 3, characterized by The sea-land segmentation system comprises: an image acquisition unit for acquiring a synthetic aperture radar image; the synthetic aperture radar image comprises a sea portion and a land portion; a clustering unit connected with the image acquisition unit, configured to cluster the pixel points in the synthetic aperture radar image based on a simulated plant growth algorithm and a simple linear iterative clustering algorithm to obtain a plurality of superpixels; a strengthening unit connected with the clustering unit, configured to use the Dice coefficient to enhance the connectivity of the plurality of superpixels to obtain a sea-land segmentation result.

5. An electronic device, comprising: comprising a memory and a processor, the memory being used to store a computer program, and the processor running the computer program to make the electronic device execute the sea-land segmentation method of any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, It stores a computer program, which is executed by a processor to implement the sea-land segmentation method of any one of claims 1 to 3.

Images