CN115205678A

CN115205678A - SAR image building rapid extraction method

Info

Publication number: CN115205678A
Application number: CN202210794683.7A
Authority: CN
Inventors: 谢聪; 庄龙; 雷志勇; 朱睿希
Original assignee: CETC 14 Research Institute
Current assignee: CETC 14 Research Institute
Priority date: 2022-07-07
Filing date: 2022-07-07
Publication date: 2022-10-18

Abstract

The invention relates to a method for quickly extracting an SAR image building, which comprises the following steps: s1, preprocessing, namely filtering an image; s2, coarse extraction, namely segmenting the filtered image, and classifying the image into a target class and a background class; and S3, fine extraction is carried out, the area, the perimeter, the length-width ratio and the neighborhood characteristics of the target object are extracted, the false alarm target is removed by utilizing the characteristics, and the building target in the image is reserved. According to the SAR image building rapid extraction method, the SAR image is more uniform and smooth through pretreatment, and a building region with clear and complete outline can be obtained, so that a basis is provided for subsequent target fine extraction; through the rough extraction and the fine extraction, false alarm targets of similar buildings in the SAR image can be effectively reduced.

Description

SAR image building rapid extraction method

Technical Field

The invention relates to the field of image segmentation, in particular to a method for quickly extracting an SAR image building.

Background

Synthetic Aperture Radar (SAR) as an active imaging system can penetrate through cloud layers to generate ground information and is not influenced by atmospheric conditions to acquire data in a large range all day long, so that the SAR is widely applied to the aspects of earth surface resource detection, land monitoring, city monitoring, disaster monitoring and the like. The building is used as an important artificial target on the ground, and the position, the outline and the change of the surface building are monitored all day long by utilizing the SAR image, so that important basic information can be provided for military target detection and urban planning.

Building extraction of SAR images remains a challenging task because the acquired images contain a large amount of speckle noise, which adversely affects building extraction. In recent decades, researchers have proposed many SAR image segmentation methods, including an image clustering method, a threshold method, a method based on a Markov Random Field (MRF), and a level set method. A high-resolution SAR image building segmentation algorithm based on an improved Markov random field model is provided ("electronics newspaper"). According to the characteristics of SAR image spot noise and building complex texture characteristics, the image is initially segmented by using a maximum likelihood criterion method of a multi-scale Markov random field model, a new neighborhood potential function model based on Gabor texture similarity is provided, and an iterative condition model algorithm is adopted for building segmentation. Through carrying out a building extraction experiment on a plurality of real SAR images, the result shows that the method based on the improved Markov random field model can extract clear building outlines.

The building extraction method based on the Markov random field model has huge calculated amount, is influenced by speckle noise of images, and is difficult to popularize in practical engineering application. Compared with the segmentation method of the Markov random field model, the level set segmentation algorithm can better adapt to the topological relation change in the image, so that a more stable result can be obtained in building segmentation. The Chan-Vese (CV) model achieves good results in image segmentation by using smooth areas of the image to represent the object and the background.

When the image segmentation of a complex scene is faced, the segmentation precision of the level set method is usually reduced obviously. The method combining multi-source image fusion can provide more information for image segmentation. An energy minimization-based space-borne SAR image building segmentation method (the automated science and technology) combines a high-resolution optical remote sensing image to provide an energy minimization-based space-borne SAR image building segmentation model. Firstly, expressing SAR image data by using Gamma distribution so as to construct a conditional probability energy item to push an evolution curve to evolve to a target edge; secondly, a priori shape energy term is constructed based on a priori knowledge, and an evolution curve is continuously deformed under the promotion of the energy term, so that the image segmentation precision under a complex background can be improved.

Therefore, the existing image segmentation method needs higher calculation cost or optical auxiliary data, and is difficult to automatically and effectively extract buildings from images in practical application, so that a rapid SAR image building extraction method is urgently needed.

Disclosure of Invention

In order to solve the prior technical problem, the invention provides a method for quickly extracting an SAR image building.

The invention specifically comprises the following contents: a method for quickly extracting an SAR image building comprises the following steps:

s1, image preprocessing: the SAR coherent imaging mechanism enables the obtained image to contain a large amount of speckle noise, and the speckle noise brings adverse effects to subsequent processing such as image target detection and identification. Therefore, the image is filtered by adopting the non-local sparse model, so that the speckle noise of the image is reduced, and meanwhile, the detailed information such as the edge, the contour and the like of the image can be kept.

The non-local sparse model noise reduction method comprises the following steps:

a) Firstly, taking an N multiplied by N neighborhood of each pixel in an image as an image block, calculating the similarity of each image block and other image blocks, and constructing a similar set, wherein N is an integer;

b) Secondly, analyzing the structure and the size of the similar set, and constructing a dictionary of sparse representation by combining the similar set;

c) Then, according to the designed sparse representation dictionary, performing sparse decomposition and sparse reconstruction on the similar set of the images by adopting an SOMP (Simultaneous Orthogonal Matching Pursuit) algorithm;

d) And finally, calculating the average value of all the noise reduction results of each pixel in the image, and taking the average value as the final noise reduction result of the pixel, thereby obtaining the noise reduction result of the whole image.

S2, coarse target extraction: after the filtering processing of the non-local sparse model, the background in the image is more uniform and smooth, the building outline is clearer and more complete, and the distinguishability from the background is higher. And (3) segmenting the image by adopting an automatic threshold segmentation method (OTSU algorithm), and classifying the image into two categories, namely a target category and a background category.

The OTSU segmentation method comprises the following steps:

a) Firstly, assuming an initial segmentation threshold, dividing each pixel of an image into a foreground (namely a target) and a background, and respectively calculating the proportion of the total number of pixels of the image occupied by the foreground pixel and the background pixel and the average gray level;

b) Secondly, solving the inter-class variance value of the foreground pixel and the background pixel;

c) And finally, finding out a threshold value which enables the inter-class variance to be maximum by adopting a traversal method, namely the finally selected segmentation threshold value.

S3, target fine extraction: the result of the coarse extraction of the target in S2 includes the building and the false alarm target, in order to reduce the false alarm target, the target is expanded and corroded by using morphological processing, and the area, the perimeter, the aspect ratio, the neighborhood (such as shadow) features and the like of the target object are respectively extracted by using an object-oriented idea, the false alarm target is removed by using the features, and the building target in the image is retained.

According to the SAR image building rapid extraction method, the SAR image is more uniform and smooth through pretreatment, and a building region with clear and complete outline can be obtained, so that a basis is provided for subsequent target fine extraction; through the rough extraction and the fine extraction, false alarm targets of similar buildings in the SAR image can be effectively reduced.

Drawings

The following further explains embodiments of the present invention with reference to the drawings.

FIG. 1 is a flow chart of the SAR image building fast extraction method of the present invention;

fig. 2 is a diagram of the result of processing an image according to the present invention, in which (a) is an original SAR image, and (b) is a schematic diagram of a non-local sparse model filtering process; (c) is a schematic image segmentation diagram; and (d) a schematic diagram of the extraction result of the building.

Detailed Description

As shown in fig. 1, the method for quickly extracting a building from an SAR image of the present invention is disclosed. The present invention is illustrated by taking a certain SAR image as an example. Experimental data a single-polarized SAR image (size 1000 × 1000) with a typical feature scene was selected, and as shown in fig. 2 (a), the study area was a suburban area including features such as buildings, roads, farmlands, trees, shadows, and the like. The SAR image is used for carrying out a building extraction experiment, and the concrete steps are as follows:

(1) Image pre-processing

The image is filtered by adopting a non-local sparse model, so that the speckle noise of the image is reduced, and meanwhile, the detailed information such as the edge, the outline and the like of the image can be kept. Sparse representation is implemented by designing an effective dictionary and reconstructing an image signal with a small amount of information, so as to implement image denoising, as shown in fig. 2 (b).

a) Firstly, taking an N multiplied by N neighborhood of each pixel in an image as an image block, taking an M multiplied by M range with a pixel point as a center as a search frame, calculating the Euclidean distance between each image block and an original image block in the search frame, and taking a similar image block when the distance is smaller than a certain threshold value, so as to solve a similar set of the pixel point in the image, wherein M and N are positive integers, and values are taken according to actual needs;

b) Secondly, the structure and the size of the similar set are analyzed, and a dictionary of sparse representation is constructed by combining the similar set. When the similarity set is smaller than a certain threshold value, selecting a dictionary used by sparse representation in the K-SVD method as a dictionary of the similarity set; and when the similarity set is larger than a certain threshold value, solving the dictionary of the similarity set by combining an SOMP algorithm iterative computation method.

c) Then, according to the designed sparse representation dictionary, carrying out sparse decomposition and reconstruction on the similar set of the image by adopting an SOMP algorithm, and solving a sparse coefficient matrix of the image;

d) And finally, each pixel point in the original image comprises a plurality of denoising results, all denoising results of each pixel are summed and averaged to be used as the final denoising result of the pixel, and thus the denoising result of the whole image is obtained.

(2) Coarse extraction of target

After the filtering processing of the non-local sparse model, the background in the image is more uniform and smooth, the building outline is clearer and more complete, and the distinguishability from the background is higher. And (3) segmenting the image by adopting an automatic threshold segmentation method (OTSU algorithm), and classifying the image into two categories, namely a target category and a background category, as shown in figure 2 (c).

The image OTSU threshold segmentation method comprises the following steps:

a) Firstly, assuming an initial segmentation threshold value, dividing each pixel of an image into a foreground (namely a target) and a background, and respectively calculating the proportion of the total pixel number of the image occupied by the foreground pixel and the background pixel and the average gray level;

c) And finally, finding out a threshold value which enables the inter-class variance to be maximum by adopting a traversal method, namely the finally selected segmentation threshold value. And (4) performing binary segmentation on the image by using the final threshold, wherein if the pixel value in the image is greater than the threshold, the image is taken as a target (1 is taken), and if the pixel value is less than the threshold, the image is taken as a background (0 is taken).

(3) Target refinement extraction

And (3) the result of the coarse target extraction in the step (2) comprises buildings and false alarm targets, and in order to reduce the false alarm targets, the targets are subjected to expansion and corrosion by using morphological processing. Selecting a window with the size of J multiplied by J from the image, and performing expansion and corrosion treatment on each pixel in the image to connect the split building outlines and reduce isolated false alarm targets. J is a positive integer and takes a value according to actual needs.

By adopting the idea of object-oriented, each object in the image is defined as an object block, the characteristics of the area, the perimeter, the aspect ratio, the shape index, the shadow characteristic and the like of the object are respectively extracted for each object block, the false alarm object is removed by utilizing the characteristics, the building object in the image is reserved, and finally the extraction result is shown in fig. 2 (d).

The SAR image building rapid extraction method has the following effects:

(1) The building target contour can be quickly extracted from the SAR image

Buildings in SAR images typically appear to be composed of several strongly scattering points or lines, exhibiting a fragmented, discrete state. The image is denoised and enhanced by using the non-local sparse model, so that the SAR image is more uniform and smooth, a building region with clear and complete outline can be obtained, and a foundation is provided for the subsequent accurate extraction of the target.

(2) False alarm targets of similar buildings in SAR images can be reduced

Building-like, strongly scattering objects often exist in SAR images, and these objects can form false alarm phenomena in the building extraction process. The method comprises the steps of obtaining a target rough extraction result by performing OTSU threshold segmentation on a preprocessed image, then processing the rough extraction result by adopting a morphological method, and effectively removing a false alarm target in the image by combining the characteristics of the area, the perimeter, the length-width ratio, the shadow characteristic and the like based on an object-oriented idea.

The SAR image is filtered by adopting a non-local sparse model, so that split and discrete buildings in the image are aggregated into a complete strong scattering target, and a multi-feature discrimination method is combined, so that the special shape features and shadow features of the buildings are fully considered, the false alarm phenomenon is reduced, the purpose of effectively and quickly extracting the buildings from the image is achieved, and the SAR image extracting method has a good practical effect.

In the above description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. The foregoing description is only a preferred embodiment of the invention, which can be embodied in many different forms than described herein, and therefore the invention is not limited to the specific embodiments disclosed above. And that those skilled in the art may, using the methods and techniques disclosed above, make numerous possible variations and modifications to the disclosed embodiments, or modify equivalents thereof, without departing from the scope of the claimed embodiments. Any simple modification, equivalent change and modification of the above embodiments according to the technical essence of the present invention are within the scope of the technical solution of the present invention.

Claims

1. A method for quickly extracting an SAR image building is characterized by comprising the following steps: the method comprises the following steps:

s1, preprocessing, namely filtering an image;

s2, coarse extraction, namely segmenting the filtered image, and classifying the image into a target type and a background type;

and S3, performing fine extraction, extracting the area, the perimeter, the length-width ratio and the neighborhood characteristics of the target object, removing the false alarm target by using the characteristics, and keeping the building target in the image.

2. The SAR image building fast extraction method according to claim 1, characterized in that: in the S1, a non-local sparse model is adopted to filter the image, so that speckle noise of the image is reduced, and detailed information such as image edges and contours is kept.

3. The SAR image building fast extraction method according to claim 2, characterized in that: the method for filtering the image by adopting the non-local sparse model comprises the following steps:

firstly, taking an N multiplied by N neighborhood of each pixel in an image as an image block, calculating the similarity of each image block and other image blocks, and constructing a similar set, wherein N is an integer;

secondly, analyzing the structure and the size of the similar set, and constructing a dictionary of sparse representation by combining the similar set;

then, according to the designed sparse representation dictionary, carrying out sparse decomposition and sparse reconstruction on the similar set of the image by adopting an SOMP algorithm;

and finally, calculating the average value of all the noise reduction results of each pixel in the image, and taking the average value as the final noise reduction result of the pixel, thereby obtaining the noise reduction result of the whole image.

4. The SAR image building fast extraction method according to claim 1, characterized in that: and S2, segmenting the image by adopting an automatic threshold segmentation method.

5. The SAR image building fast extraction method according to claim 4, characterized in that: the image segmentation method adopting an automatic threshold segmentation method comprises the following steps:

firstly, assuming an initial segmentation threshold value, dividing each pixel of an image into a foreground pixel and a background pixel, wherein the foreground is a target, and calculating the proportion of the foreground pixel to the background pixel to the total pixel number of the image and the average gray level respectively;

secondly, solving the inter-class variance value of the foreground pixel and the background pixel;

and finally, finding out a threshold value which enables the inter-class variance to be maximum by adopting a traversal method, namely the finally selected segmentation threshold value.

6. The SAR image building fast extraction method according to claim 1, characterized in that: and S3, expanding and corroding the target by using morphological processing on the crude target extraction result processed in the S2, extracting the area, the perimeter, the length-width ratio, the neighborhood characteristics and the like of the target object respectively by using an object-oriented idea, removing the false alarm target by using the characteristics, and keeping the building target in the image.