CN111353371A - Shoreline extraction method based on spaceborne SAR images - Google Patents

Abstract

A coastline extraction method based on a spaceborne SAR image, comprising the following steps: S1: firstly select a SAR image to read the SAR image; S2: preprocess the image, enlarge the original SAR image, and intercept the image with the coastline S3: Use Kmeans clustering and segmentation method to process the intercepted image with Kmeans algorithm; S4: Combine morphological processing, perform morphological closing operation, select structural elements, first dilate and then erode; S5: Use Canny boundary The extraction operator is used to extract the coastline; S6: analyze the extraction result, and superimpose the extracted coastline with the original map to verify the correctness of the coastline extraction result. Compared with other edge extraction algorithms of the same type, the coastline extracted by this method has better continuity and smoothness, and has a high degree of agreement with the real coastline.

Description

Shoreline extraction method based on spaceborne SAR images

technical field

The invention relates to a remote sensing image processing method, in particular to a coastline extraction method based on spaceborne SAR images.

Background technique

Satellite airborne synthetic aperture radar is now the basic earth observation tool for satellite remote sensing. Compared with traditional optical remote sensing and hyperspectral remote sensing, satellite airborne synthetic aperture radar has all-weather and high-resolution imaging capabilities, especially in cloudy areas. while taking images.

With the advancement of technology, the future spaceborne synthetic aperture radar will definitely achieve more functions, realize all high-resolution broadband multi-mode microwave imaging, realize the miniaturization of artificial synthetic aperture radar and reduce the cost.

In order to obtain as much information as possible at the minimum cost, although the technical difficulties of satellite airborne synthetic aperture radar breakthrough still exist, with the continuous development of technology, satellite airborne synthetic aperture radar inevitably enters a new development period.

The penetrability of synthetic aperture radar enables it to obtain images that reflect the microwave scattering characteristics of the target. The advantage of this characteristic makes synthetic aperture radar an important method to obtain ground object information. In addition, since SAR is coherent imaging, the characteristics of coherent imaging enable SAR images to be synthesized by aperture. The SAR images obtained by this method have high resolution and can provide detailed map information.

Coastline detection is very important in coastal zones and can be used for a variety of activities such as geographic mapping, automated navigation, coastal erosion and monitoring. Shoreline extraction from Synthetic Aperture Radar (SAR) imagery is becoming increasingly popular due to its wide coverage and all-weather capability. However, high-accuracy coastline extraction remains a challenging problem due to speckle and insufficient contrast due to strong ocean currents, shadows or specific coast types, i.e. sandy coasts, etc. In recent decades, many methods have been proposed by many people at home and abroad for coastline detection of SAR images. It is roughly divided into the following categories: boundary tracking method, active contour method, Level Set algorithm, Markovian segmentation method, wavelet transform method, etc. Among these algorithms, the horizontal cut-set algorithm is higher than other algorithms in the comprehensive evaluation of detection speed and detection effect. Although some people have improved it to improve the detection speed, its implementation principle is quite complicated, and the detection speed is still relatively high. It is slow and cannot meet the needs of practical engineering. There are also methods for extracting coastlines from SAR images. For example, the domestic patent with the application number of CN201610621676.1, "A Method for Extracting Shorelines in SAR Coastal Images", determines the ray by confirming the geometric center of the ocean area and taking this point as the starting point to make rays. The boundary points on the coast are connected in turn to obtain the coastline. This method is suitable for large-scale images, and is not suitable for the case where the image is segmented, and further research is needed.

SUMMARY OF THE INVENTION

In view of this, the present invention provides a coastline extraction method based on spaceborne SAR images to solve the problem of coastline extraction, and the technical solution is as follows:

A method for coastline extraction based on spaceborne SAR images, comprising the following steps:

S1: First select a SAR image to read the SAR image;

S2: Preprocess the image, enlarge the original SAR image, and capture a part of the image with the coastline;

S3: Use the Kmeans clustering and segmentation method to process the captured image with the Kmeans algorithm;

S4: Combine morphological processing, perform morphological closing operation, select structural elements, first expand and then corrode;

S5: Use Canny boundary extraction operator for coastline extraction;

S6: Analyze the extracted results, overlay the extracted coastline with the original map, and verify the correctness of the coastline extraction results.

Further, in step S3, in the Kmeans clustering algorithm, the selected K value is 10.

Further, in step S3, the algorithm processing further includes performing segmentation and binarization according to the clustering mode of the 10th cluster of the K-means cluster map, and performing filling and denoising processing on the binarized image.

In step S4, the selected structural element is 4.

In step S4, dilation and erosion are to calculate an area in the image, involving features of lines and points. In the image, dilation is an area that expands to the surrounding area, while erosion is to reduce the surrounding area from the same time.

In step S5, the Canny boundary extraction operator performs coastline extraction, including the following steps:

①Denoise: The original image data must first be convolved with a two-dimensional Gaussian filter template;

(2) Gradient calculation: The use of the derivative operator obtains the respective derivatives G _X and G _Y of the grayscale image in two directions, and the magnitude of the gradient |G| and the direction θ can be calculated through the obtained derivatives:

③ Gradient direction determination: Calculate the direction of the edge, and divide the gradient direction of the edge with multiple angles to find the adjacent pixels in the pixel direction;

④ Traverse the entire image: when the gray value of the two pixels is not the maximum value of the gray value of the pixels before and after the gradient direction, the gray value of the pixel is 0, that is, it is not an edge;

⑤ Two thresholds are obtained by accumulating the histogram. Above the threshold must be an edge, and below the threshold it should not be an edge; if the detection is in the middle of the two thresholds, then whether the edge pixels in the adjacent pixels of the pixel are determined by the edge There is no higher threshold for a pixel to judge, if it is, it is an edge; otherwise it is not an edge.

The invention can facilitate the extraction of coastlines from SAR images with complex scenes and large geographic coverage, and has strong automation and adaptability.

Description of drawings

Fig. 1 is the technical roadmap of this method;

Figure 2 is the original image of Sentinel-1A;

Figure 3 is a cut out part of the coastline;

Figure 4 is the processing result of kmeans algorithm;

Fig. 5 is a binary image that is segmented;

Figure 6 is a random matrix;

Figure 7 is the first step of clustering;

Fig. 8 is the clustering iteration step;

Figure 9 is the clustering demonstration result;

Figure 10 is the denoising result;

Figure 11 is the result of morphological processing;

Figure 12 is the image before noise reduction;

Figure 13 is a partial enlargement of the image before noise reduction;

Figure 14 is the extraction result of the canny operator;

Figure 15 is the final result of coastline extraction.

Detailed ways

The coastline extraction method based on the spaceborne SAR image provided by the present invention, as shown in FIG. 1 , includes the following steps: 1) first select a SAR image, and read the SAR image; 2) preprocess the image, The SAR image is enlarged, and a part of the image with the coastline is intercepted; 3) Kmeans algorithm processing is performed on the intercepted image using the Kmeans clustering segmentation method; the processing also includes the clustering mode of the 10th cluster according to the K-means cluster map Perform segmentation and binarization, and fill and denoise the binarized image; 4) Combine morphological processing, perform morphological closing operation, select four structural elements, first dilate and then erode; 5) Use Canny boundary The extraction operator is used to extract the coastline; 6) Analyze the extraction result, that is, superimpose the extracted coastline with the original map to verify the correctness of the coastline extraction result. The following is a detailed description of each step:

one

As shown in Figure 2, it is a raw SAR image to be processed. The experimental data of this scene is the image of Sentinel-1A in the Yellow Sea area. The imaging time is 2017.12.05, the polarization mode is VV, and the pixel size is 15×15. The information of SAR images is mainly affected by backscattering. The brightness of the image represents the intensity of backscattering. The rougher the inner surface of the pixel, the stronger the backscattering. The smooth surface produces specular reflection, and the backscattering is weak, resulting in low brightness.

two

On the basis of Figure 2, a part of the coastline is intercepted, that is, Figure 3. As one of the key technologies of digital image processing, image segmentation undoubtedly plays an important role. Image segmentation provides a solid foundation for subsequent image processing, recognition, analysis, and understanding by extracting meaningful feature locations in images, such as image edges and image regions. Currently, many different methods for edge extraction or image segmentation have been developed for various images. However, none of these methods can be applied to all situations, which is the main reason why the research of image segmentation is still one of the hot spots in image processing.

three

As shown in Figure 4, the intercepted image in Figure 3 is processed by the Kmeans algorithm. After the Kmeans algorithm is processed, the distinction between the intensity of the ocean and the land increases. Then, according to the clustering pattern of the 10th cluster in the K-means clustering map Perform segmentation binarization to obtain Figure 5.

Among the simplest clustering algorithms, there is no doubt that Kmeans cluster segmentation has a place. This is a very typical unsupervised learning algorithm. It is mainly used to automatically group similar samples into a category. The biggest difference between a clustering algorithm and a classification algorithm is that the clustering algorithm is an unsupervised learning algorithm that can be processed automatically. Classification algorithms are supervised learning algorithms that require human intervention.

The difficulty of the Kmeans algorithm is that it needs to set different k values to obtain different clustering results. However, the uncertainty of the value of k is a shortcoming of this algorithm. Often in order to achieve good experimental results, multiple attempts are required to select the optimal k value. In this experiment, the selected K value is 10. When K is larger or smaller, the effect of clustering is not as good as 10, and it is more scattered.

The basic idea of Kmeans is to gather k points into space and classify the objects closest to them. Through the iterative method, the value of each cluster center is continuously updated, and the whole process of update iteration will not stop until the best clustering result appears. If it can be represented by a formula, assuming that the cluster is divided into (C ₁ , C ₂ , .....C _i ), it can be expressed as:

where Ci is the number of clusters, x is the sample point in _{Ci , ui} is the centroid of _{Ci (} the mean of all samples in Ci), and SSE is the clustering error of all samples, representing the clustering effect good or bad. In order to simply understand this process, this paper will conduct a small experiment to demonstrate the operation process.

First, use the randn function to generate three normally distributed random matrices, as shown in Figure 6, and call them the original images.

Next, k random cluster center points are generated. For simplicity, set k to be the same as the number of matrices, that is, k=3. The next step is to calculate the distance of each data point to these centers individually. Each data point regards the center point with the shortest distance as its own category. At this point, each data point has its own corresponding center point. However, as can be seen from Figure 7, the clustering at this time is not accurate.

At this time, the position of the center point is recalculated. Calculate the center of all blue points and move the blue center point to the calculated position. At this time, the green point in Figure 8 will be classified to the new blue center point, thus being dyed blue. The green and red center points are also moved to new positions.

After the above steps are repeated continuously, until the center point iteration becomes stable, the iteration is stopped, and the algorithm ends. The clustering results are obtained, as shown in Figure 9.

The samples of the same color are clustered into a cluster, and finally three clusters are formed, so that the samples within the same cluster have high similarity and the difference between different clusters is high.

Immediately after clustering the images, binarize the processed images. The simple point of image binarization is to represent all points of the entire image as 0 or 255, that is, the effect of either black or white, turning the entire image into a distinct black and white effect. In other words, the original gray value image is converted into a black and white binary image, and global and local features can be seen by choosing an appropriate threshold. Points whose gray level is greater than or equal to the threshold are judged as one class, and the gray level is expressed as 255. On the contrary, the points whose gray level is lower than the threshold value will be judged as another class, and the gray level is expressed as 0 (that is, background or something irrelevant).

By dynamically adjusting the threshold value, the binarization of the image can also be made dynamic, so that the effect on the segmented image can be observed and dynamic analysis can be realized. By using some closed or connected boundaries to define non-overlapping parts, an ideal binary image can be obtained. This is a crucial step for digital image processing. Binary image plays an important role in digital image processing. Especially in some aspects of image processing, many systems need to be processed by binary image.

To process and analyze binary images, binarizing grayscale images is an obvious first step. The binary image is obtained by conversion, so that the image properties are only related to black or white points in the next image processing, and the multi-level nature of pixels does not need to be considered, which can simplify the processing process.

It can be seen that there is a lot of obvious noise on the binarized image. The point map and the small holes are all over the whole image. In order to proceed to the next steps, it must first be filled and denoised. After the denoising process, Figure 10 is obtained.

Four

Obviously, there is still a problem with the coastline. Next, the morphological closing operation is used, and the selection of structural elements is repeatedly tried for the experimental effect. Finally, it is found that when the structural element is 4, the denoising results are better, and the denoising of the remaining structural elements. no significant effect. Dilate and then etch, resulting in Figure 11.

As we all know, it is impossible to remove all noises in general. In order to convert an image from a grayscale image to a binary image, some noise is inevitable. This situation can cause difficulties in image extraction.

There are many types of noise in binary images. The most representative are dots and small holes, as shown in Figure 12. Figure 13 is a cutout of the image before noise reduction in this experiment (ie, Figure 12). Dot diagrams and pinholes refer to pixel-connected components and relatively small area zero-pixel-connected components. Generally, etching and dilation treatments are effective in removing these connections.

Set theory is the mathematical foundation and language used in mathematical morphology. There are several basic operations in image morphological processing called dilation, erosion, opening and closing, which have their own characteristics in binary images and grayscale images. These basic operations can also be derived and incorporated into various algorithms of mathematical morphology.

为膨胀运算符号，D_f和D_g分别为f和g的定义域，则f被g腐蚀和膨胀可表示为：Let f(x,y) be the input image, g(i,j) is the structuring element, Θ is the erosion symbol,

For the dilation operation symbol, D _f and D _g are the domains of f and g respectively, then f is corroded and dilated by g and can be expressed as:

f(x,y)Θg(i,j)=min{f(x+i,y+j)-g(i,j)|(x+i,y+j)∈D _f ,(i,j )∈D _g } (2)

表示开运算，·表示闭运算，其定义分别为：Assume

Represents an open operation, and · represents a closed operation, and the definitions are:

The opening operation is usually used to remove bright details smaller than the structuring elements, while keeping the large overall bright features unchanged; the closing operation is usually used to remove the dark details smaller than the structuring elements while keeping the highlight feature elements unchanged.

Dilation and erosion are the computation of a region in an image (features of lines and points). In the image, dilation is the area that expands around, while erosion is the reduction of the surrounding area from the same time. It is worth noting that, in general, dilation and erosion are not mutual, i.e. they can be used in a cascade. After dilation and erosion, or after erosion enlarges, it is often impossible to restore the original image, but it produces new forms of transformation. This is also known as a morphological opening and closing operation.

Obviously, image noise filtering is an integral part of image preprocessing. Morphological noise filters are constructed by combining opening and closing operations. By using structuring elements to open the set, the noise blocks around the target in the binary image mentioned above can be eliminated; on the contrary, the closed operation can be used to remove the noise pinholes in the target.

In the above method, the selection of structural elements is the top priority, which must be larger than all the noises (including noise blocks and noise holes) in the binary image. In conducting this study, multiple structural elements were repeatedly tried, and a parameter size of 4 was finally determined. And when the structural element takes this value, the experiment has a better effect.

When actually performing image processing, the open operation is usually used to eliminate sizes smaller than structuring elements. Try to keep the overall grayscale values and bright areas of the image larger than structuring elements while maintaining detail. Conversely, the closure operation is used to remove black details smaller than structuring elements. Of course, it is also necessary to keep the gray values of the image and dark areas larger than the structuring elements.

Through the above description, a simple conclusion can be drawn. At the same time, the operation of filtering out various noises can be achieved by combining the two operations of opening operation and closing operation. Moreover, if multiple structural elements can be combined with opening and closing operations, it is possible to filter out various noises. Get great results in preserving image detail.

five

The coastline in Figure 11 is extracted using the Canny operator, as shown in Figure 14. Finally, in order to verify the correctness of the coastline extraction results, the extracted coastlines are superimposed with the original map, as shown in Figure 15.

Edge detection technology is very important for processing digital images, because the edge is the boundary line between the object and the background to be extracted, and the edge can be extracted to distinguish the object and the background. In an image, a boundary represents the end of a feature region and the beginning of another feature region. The internal characteristics or properties of a bounded boundary are the same, but the internal characteristics or properties of different regions are different. Certain image features, including grayscale, color, or texture features, aid in edge detection. Edge detection is actually a detection of changes in image features.

The classic approach to edge extraction is to examine the grayscale variation of each pixel of the image in a specific area. Use the first or second directional derivative near the edge to detect edges using a simple method. This method is called edge detection.

The basic idea of edge detection is to determine whether a pixel is located at the edge of an object by detecting the state of each pixel and its neighbors. If each pixel lies on the boundary of the object, the gray value of adjacent cells will vary more.

The boundary extracted by the Canny operator is relatively continuous and smooth. Therefore, this paper chooses the Canny operator as the edge operator for coastline extraction.

The Canny edge operator is optimal as an edge detection operator. This is widely used in many image processing fields. Canny's edge check has the following requirements for the detection operator: low error rate: the real edge points are not lost as much as possible, so as to avoid the edge being judged as a non-edge point; high position accuracy: the detected edge should be as close to the real edge as possible ; each edge point has a unique response, resulting in a single-pixel-wide edge. The specific implementation steps of the Canny edge operator are as follows:

①Denoising: The edge detection algorithm cannot process the unprocessed original image, so the original data must be convolved with the Gaussian smoothing template first. The resulting image is a bit blurry compared to the original, a necessary cost to remove noise.

②Gradient calculation: The use of the derivative operator can easily obtain the respective derivatives G _X and G _Y of the grayscale image in two directions, and the magnitude and direction of the gradient can be calculated through the obtained derivatives:

G _X and G _Y represent the respective derivatives of the grayscale image in the x and y directions.

③ Gradient direction determination: Calculate the direction of the edge and divide the gradient direction of the edge at multiple angles (eg, 0°, 45°, 90°, and 135°) to find out the adjacent pixels in the pixel direction.

④ Traverse the entire image: When the gray value of two pixels is not the maximum value of the gray value of the pixels before and after the gradient direction, the gray value of the pixel is 0, that is, it is not an edge.

⑤ Two thresholds are obtained by accumulating the histogram. Obviously, above the threshold must be an edge, below the threshold it should not be an edge. If the detection is in the middle of the two thresholds, then whether an edge pixel in the pixel's neighbors is judged by the edge pixel not having a higher threshold, if it is, it is an edge; otherwise it is not an edge.

six,

Looking at Figure 15, it can be seen that the extracted coastline has good continuity and excellent smoothness. Although the accuracy is not high enough in some places, it can be seen that it basically coincides with the coastline.

Compared with the commonly used boundary tracking methods, the Kmeans clustering and segmentation method is slightly worse than the boundary tracking method in accuracy, but in terms of applicability, it is much stronger than the former. Since the boundary tracking method can only detect the coastline given the SAR image, assuming that the complete coastline is segmented by the image, the use of the boundary tracking method will cause the embarrassing situation that only half of the coastline can be traced and the other half of the coastline cannot be displayed. The Kmeans method does not have this limitation, and is more adaptable to the segmentation of coastlines by images, and is simpler and more convenient than the boundary tracking method.

The method of Kmeans clustering and segmentation combined with morphology to extract coastlines proposed in this paper is suitable for automatically extracting coastlines in SAR images. The experimental results show that the influence of speckle noise can be effectively alleviated by combining Kmeans and morphological processing methods. Furthermore, the method proposed in this paper has advantages in maintaining the continuity of coastlines in complex scenes.

Claims (6)

1. A coastline extraction method based on a satellite-borne SAR image comprises the following steps:

s1: firstly, selecting an SAR image map, and reading the SAR image;

s2: preprocessing an image, amplifying an original SAR image, and intercepting a part of image with a coastline;

s3: performing Kmeans algorithm processing on the intercepted image by using a Kmeans clustering segmentation method;

s4: combining morphological processing, performing morphological closing operation, selecting structural elements, expanding and corroding;

s5: carrying out coastline extraction by using a Canny boundary extraction operator;

s6: and analyzing the extracted result, overlapping the extracted coastline with the original map, and verifying the correctness of the coastline extraction result.

2. The coastline extraction method based on spaceborne SAR images as claimed in claim 1, wherein: in step S3, the K value selected in the Kmeans clustering algorithm is 10.

3. The coastline extraction method based on spaceborne SAR images as claimed in claim 2, wherein: in step S3, the algorithm processing further includes segmenting and binarizing according to the clustering pattern of the 10 th cluster of the K-means cluster map, and performing filling and denoising processing on the binarized image.

4. The coastline extraction method based on spaceborne SAR images as claimed in claim 1, wherein: in step S4, the structural element is selected to be 4.

5. The coastline extraction method based on spaceborne SAR images as claimed in claim 1, wherein: in step S4, dilation and erosion are calculated for a region in the image, relating to the features of lines and points, in which dilation is a region that expands around and erosion is a region that reduces around from the same time.

6. The coastline extraction method based on spaceborne SAR images as claimed in claim 1, wherein: in step S5, the Canny boundary extraction operator performs coastline extraction, and includes the following steps:

① denoising, wherein the original image data must be convolved with a two-dimensional Gaussian filter template;

② gradient calculation of derivative operatorUsing derivation of respective derivatives G in two directions of the gray-scale image_XAnd G_YFrom the derivative obtained, the magnitude | G | and direction θ of the gradient can be calculated:

③ gradient direction determination, calculating the direction of the edge and dividing the gradual change direction of the edge by a plurality of angles to find out the adjacent pixels in the pixel direction;

④ traversing the whole image, when the gray values of two pixels are not the maximum value of the gray values of the pixels before and after the gradient direction, the gray value of the pixel is 0, i.e. not the edge;

⑤ derives two thresholds by accumulating histograms, above which it must be an edge and below which it should not be, if the detection is in the middle of the two thresholds, then whether the edge pixels in the pixels' neighbours are judged by the edge pixels not having a higher threshold, if so, an edge, otherwise, no edge.

Images