阅读说明：本技术 一种用于SAR干涉图处理的Goldstein金字塔构建方法 (Goldstein pyramid construction method for SAR interferogram processing ) 是由 庄会富 范洪冬 邓喀中 谭志祥 迟博文 张克非 郑美楠 于 2020-11-27 设计创作，主要内容包括：本发明公开了一种用于SAR干涉图处理的Goldstein金字塔构建方法,适用于图像处理领域。把SAR影像复数干涉图I0设为Goldstein金字塔的第0层图像G_0；设置对复数干涉图G_0进行Goldstein金字塔变换的最大层数L；设置频率域Goldstein低通滤波器的核大小N；利用大小为N的频率域Goldstein滤波器对金字塔第i层复数干涉图G_i做滑移滤波处理；对滤波后的干涉图进行下采样得到第i+1层Goldstein金字塔G_(i+1)；重复上述步骤直到得到第L层Goldstein金字塔G_L,把第0-L层金字塔图像进行组合得到Goldstein金字塔。其可以生成SAR影像复数干涉图的Goldstein金字塔图像,实现对复数干涉图的多尺度描述,为进一步提升复数干涉图滤波、相位解缠等处理的效果提供了一种有效的多尺度变换方法,对SAR影像复数干涉图构建金字塔的效果好。(The invention discloses a Goldstein pyramid construction method for SAR interferogram processing, which is suitable for the field of image processing. Setting SAR image complex interference pattern I0 as layer 0 image G of Goldstein pyramid 0 (ii) a Setting a pair of complex interferograms G 0 Performing the maximum layer number L of Goldstein pyramid transformation; setting the kernel size N of a frequency domain Goldstein low-pass filter; pyramid ith layer complex interference pattern G using a frequency domain Goldstein filter of size N i Performing slipping filtering treatment; for filtered interferograms Downsampling to obtain the i +1 layer Goldstein pyramid G i+1 (ii) a Repeating the steps until obtaining the L-th layer Goldstein pyramid G L And combining the 0-L layer pyramid images to obtain a Goldstein pyramid. Which can generate SAR image complex stemThe Goldstein pyramid image related to the image realizes multi-scale description of the complex interferogram, provides an effective multi-scale transformation method for further improving the processing effects of filtering, phase unwrapping and the like of the complex interferogram, and has a good effect of building a pyramid on the SAR image complex interferogram.)

1. A Goldstein pyramid construction method for SAR interferogram processing is characterized by comprising the following steps:

a, acquiring a complex interference pattern I0 with w columns and h rows, wherein A is the real part of the complex interference pattern I0, B is the imaginary part of the complex interference pattern I0, and j represents an imaginary unit;

b, establishing a Goldstein pyramid, and setting the maximum layer number L of Goldstein pyramid transformation according to the column number w and the row number h of the complex interference pattern I0;

c, setting the original SAR image complex interference pattern I0 as a layer 0 image G of a Goldstein pyramid₀；

d, setting the kernel size N of a frequency domain Goldstein low-pass filter, utilizing the Goldstein low-pass filter to suppress the high-frequency part of the complex interferogram, reserving the low-frequency part of the complex interferogram, and transforming the complex interferogram from a space domain to a frequency domain by using common Fourier transform or fast Fourier transform;

e using frequency domain Goldstein filter with kernel size N to pyramid i-th layer complex interference pattern G_iPerforming slip filtering processing to obtain filtered interferogram

f pairs of filtered interferogramsDownsampling to obtain the i +1 layer Goldstein pyramid G_i+1；

G repeating the steps (e) and (f) until the L-th layer Goldstein pyramid G is obtained_LAnd combining the 0-L layer pyramid images from bottom to top to obtain the Goldstein pyramid with the layer number of L + 1.

2. The Goldstein pyramid construction method for SAR interferogram processing according to claim 1, characterized by: the maximum layer number L in the step c meets the condition that:andwherein log is logarithmic sign.

3. The Goldstein pyramid construction method for SAR interferogram processing according to claim 1, characterized by: the kernel size N of the frequency domain Goldstein filter is different according to the selected Fourier transform method, and either the condition d1 or d2 is satisfied:

d1 when the Goldstein filtering algorithm is implemented using ordinary Fourier transform, N is odd, N is greater than or equal to 3 and less than or equal to min (w, h) and N belongs to N^*In which N is^*Denotes a positive integer, min (·) denotes taking the minimum value, in particular, N is proposed to be set to N ═ 5;

d2 when the Goldstein filter algorithm is implemented using fast fourier transform, N is an even number, 2 ≦ N ≦ min (w, h), in particular, N is suggested to be set to N ≦ 4 or N ≦ 6.

4. G for SAR interferogram processing according to claim 1The method for constructing the oldstein pyramid is characterized by comprising the following steps: the pyramid layer number i used in the step e meets the condition: i is more than or equal to 0 and less than or equal to L-1, and the interference pattern G is processed_iThe slip step used in the slip filtering process is S1.

5. The Goldstein pyramid construction method for SAR interferogram processing according to claim 1, characterized by: f, aligning the filtered interferogramsThe down-sampling is performed by removing the image directlyEven rows and even columns, usingThe odd rows and the odd columns of G_i+1。

6. The Goldstein pyramid construction method for SAR interferogram processing according to claim 1, characterized by: the combination mode of the Goldstein pyramids in the step G is that the Goldstein pyramids are arranged from bottom layer to top layer according to the layer number from small to large, and the 0 th layer pyramid image G₀At the bottom, L-th pyramid image G_LAt the topmost layer.

Technical Field

The invention relates to a Goldstein pyramid construction method, in particular to a Goldstein pyramid construction method for SAR interferogram processing, which is suitable for the technical field of image processing.

Background

Pyramid construction belongs to the field of image processing, is an image multi-scale expression technology, can show information of images under different resolutions, and is used in the fields of image filtering, image fusion, image super-resolution reconstruction, motion recognition, face recognition and the like. The pyramid of an image is a set of images arranged in a pyramid shape with gradually decreasing resolution from the bottom to the top of the pyramid, and is derived from the same original image. The higher the level of the image in the pyramid, the smaller the image and the lower the resolution.

The SAR interferometry technology utilizes interference phase information to accurately measure a digital elevation model of an earth surface target and the micro deformation of a radar visual line. Because the data acquired by the SAR sensor is not influenced by weather, the SAR sensor can realize continuous observation on the ground all day long, so that the SAR interferometric technology has wide application in fields such as landslide, settlement monitoring, earthquake deformation and the like. However, due to the influence of spatial-temporal decorrelation, thermal noise decorrelation, or the like, a large amount of phase noise exists in an interferogram obtained from an actual image, and it is necessary to suppress noise in the interferogram by filtering. In addition, phase unwrapping is required before surface deformation can be extracted from the interferogram. The pyramid is constructed in the process of processing interferograms such as filtering, phase unwrapping and the like, and is an effective multi-scale and multi-resolution analysis method.

The Gaussian pyramid is the most classical map image pyramid construction method, and the original image is filtered by using a two-dimensional Gaussian filter, and then the first layer of pyramid image is obtained by adopting an image down-sampling method. A two-dimensional gaussian filter is a spatial domain filter that has a good effect on common real domain images containing gaussian noise. However, the SAR interferogram is a complex image, and the gaussian pyramid extending from the real domain to the complex domain does not use the frequency characteristics of the complex image

Disclosure of Invention

Aiming at the defects of the technology, the Goldstein pyramid construction method for SAR interferogram processing is rigorous in theory and good in construction effect.

In order to achieve the above purpose, the Goldstein pyramid construction method for SAR interferogram processing of the present invention comprises the following steps:

a, acquiring a complex interference pattern I0 with w columns and h rows, wherein A is the real part of the complex interference pattern I0, B is the imaginary part of the complex interference pattern I0, and j represents an imaginary unit;

b, establishing a Goldstein pyramid, and setting the maximum layer number L of Goldstein pyramid transformation according to the column number w and the row number h of the complex interference pattern I0;

c, setting the original SAR image complex interference pattern I0 as a layer 0 image G of a Goldstein pyramid₀；

d, setting the kernel size N of a frequency domain Goldstein low-pass filter, utilizing the Goldstein low-pass filter to suppress the high-frequency part of the complex interferogram, reserving the low-frequency part of the complex interferogram, and transforming the complex interferogram from a space domain to a frequency domain by using common Fourier transform or fast Fourier transform;

e using frequency domain Goldstein filter with kernel size N to pyramid i-th layer complex interference pattern G_iPerforming slip filtering processing to obtain filtered interferogram

f pairs of filtered interferogramsDownsampling to obtain the i +1 layer Goldstein pyramid G_i+1；

G repeating the steps (e) and (f) until the L-th layer Goldstein pyramid G is obtained_LAnd combining the 0-L layer pyramid images from bottom to top to obtain the Goldstein pyramid with the layer number of L + 1.

The maximum layer number L in the step c meets the condition that:wherein log is logarithmic sign.

The kernel size N of the frequency domain Goldstein filter is different according to the selected Fourier transform method, and the conditions of d1 or d2 are satisfied:

d1) when the Goldstein filtering algorithm is implemented by using the ordinary Fourier transform, N is an odd number, N is more than or equal to 3 and less than or equal to min (w, h), and N belongs to N^*In which N is^*Denotes a positive integer, min (·) denotes taking the minimum value, in particular, N is proposed to be set to N ═ 5;

d2) when the Goldstein filtering algorithm is implemented using fast Fourier transform, N is an even number, 2 ≦ N ≦ min (w, h), in particular, N is proposed to be set to N ≦ 4 or N ≦ 6.

The pyramid layer number i used in the step e meets the condition: i is more than or equal to 0 and less than or equal to L-1, and the interference pattern G is processed_iThe slip step length used in the slip filtering process is S-1

F, aligning the filtered interferogramsThe down-sampling is performed by removing the image directlyEven rows and even columns, usingThe odd rows and the odd columns of G_i+1。

The combination mode of the Goldstein pyramid in the step g is that the combination modes are arranged from bottom layer to top layer and from small layer to large layer, and the 0 th layerPyramid image G₀At the bottom, L-th pyramid image G_LAt the topmost layer.

Has the advantages that: the Goldstein pyramid image processing method utilizes Goldstein filtering theory, transforms a complex interferogram from a space domain to a frequency domain through Fourier transform, further completes low-pass filtering required by pyramid construction in the frequency domain, combines an image downsampling theory to obtain the Goldstein pyramid image, fully utilizes the frequency characteristics of the complex interferogram compared with a Gaussian pyramid using a space domain two-dimensional Gaussian filter, is more suitable for pyramid image construction in SAR complex interferogram processing, has perfect theoretical support, avoids the problems of more noise residue and more phase information loss in the Gaussian pyramid image construction of the SAR complex interferogram, and improves the reliability of data analysis and processing of the SAR complex interferogram under different scales and resolutions. An effective multi-scale graph transformation method is provided for further improving the processing effects of complex interference graph filtering, phase unwrapping and the like.

Drawings

FIG. 1 is a flow chart of a Goldstein pyramid construction method for SAR interferogram processing according to the present invention;

FIG. 2 is a Goldstein pyramid complex interferogram I0 of the present invention for SAR interferogram processing, also layer 0G of the Goldstein pyramid image₀A schematic diagram of (a);

FIG. 3 is a layer 1G of a Goldstein pyramid image of the Goldstein pyramid of the present invention for SAR interferogram processing₁A schematic diagram of (a);

FIG. 4 is a layer 2G of a Goldstein pyramid image of the Goldstein pyramid of the present invention for SAR interferogram processing₂A schematic diagram of (a);

FIG. 5 is a layer 3G of a Goldstein pyramid image of the Goldstein pyramid of the present invention for SAR interferogram processing₃Schematic representation of (a).

Detailed Description

The invention is described in further detail below with reference to the following figures and examples:

as shown in fig. 1, the Goldstein pyramid construction method for SAR interferogram processing of the present invention comprises the following steps:

a, acquiring a complex interference pattern I0 with w columns and h rows, wherein A is the real part of the complex interference pattern I0, B is the imaginary part of the complex interference pattern I0, and j represents an imaginary unit;

b, establishing a Goldstein pyramid, and setting the maximum layer number L of Goldstein pyramid transformation according to the column number w and the row number h of the complex interference pattern I0;

c, setting the original SAR image complex interference pattern I0 as a layer 0 image G of a Goldstein pyramid₀(ii) a The maximum layer number L meets the condition that:andwherein log is logarithmic sign;

d, setting the kernel size N of a frequency domain Goldstein low-pass filter, utilizing the Goldstein low-pass filter to suppress the high-frequency part of the complex interferogram, reserving the low-frequency part of the complex interferogram, and transforming the complex interferogram from a space domain to a frequency domain by using common Fourier transform or fast Fourier transform;

e using frequency domain Goldstein filter with kernel size N to pyramid i-th layer complex interference pattern G_iPerforming slip filtering processing to obtain filtered interferogramThe setting condition of the kernel size N of the frequency domain Goldstein filter is that the pyramid layer number i meets the condition: i is more than or equal to 0 and less than or equal to L-1, and the interference pattern G is processed_iThe slippage step length used in the slippage filtering processing is S ═ 1;

d1) when the Goldstein filtering algorithm is implemented by using the ordinary Fourier transform, N is an odd number, N is more than or equal to 3 and less than or equal to min (w, h), and N belongs to N^*In which N is^*Denotes a positive integer, min (·) denotes taking the minimum value, in particular, N is proposed to be set to N ═ 5;

d2) when the Goldstein filtering algorithm is implemented using fast Fourier transform, N is an even number, 2 ≦ N ≦ min (w, h), in particular, N is proposed to be set to N ≦ 4 or N ≦ 6;

f pairs of filtered interferogramsDownsampling to obtain the i +1 layer Goldstein pyramid G_i+1(ii) a Filtered interferogramThe down-sampling is performed by removing the image directlyEven rows and even columns, usingThe odd rows and the odd columns of G_i+1(ii) a The pyramid layer number i satisfies the condition: i is more than or equal to 0 and less than or equal to L-1, and the interference pattern G is processed_iThe slippage step length used in the slippage filtering processing is S ═ 1;

g repeating the steps (e) and (f) until the L-th layer Goldstein pyramid G is obtained_LCombining 0-L layer pyramid images from bottom to top to obtain a Goldstein pyramid multi-scale image with the layer number of L +1, wherein the Goldstein pyramid is arranged from bottom to top according to the layer number from small to large in a combining mode, and the 0-L layer pyramid image G₀At the bottom, L-th pyramid image G_LAt the topmost layer.

The first embodiment,

Step a: acquiring a complex interferogram I0 (a + B × j) with the column number w being 512 and the row number h being 512 as shown in fig. 2, wherein a is a real part of the complex interferogram I0, B is an imaginary part of the complex interferogram I0, and j represents an imaginary unit;

step b: the SAR image complex interference pattern I0 shown in FIG. 2 is set as the layer 0 image G of Goldstein pyramid₀I.e. the complex interferogram I0 and the layer 0 image G of the Goldstein pyramid₀Are completely the same;

step c: according to the condition that the maximum layer number L of Goldstein pyramid transformation needs to meetAndsetting a pair of complex interferograms G₀The maximum layer number L of Goldstein pyramid transformation is 3, wherein log is a logarithmic sign;

step d: the mathematical expression of the complex interferogram for the Goldstein low-pass filter H (p, q) at position (p, q) is H (p, q) ═ S { | F (p, q) | }^αWhere F (p, q) is frequency domain data after fourier transform, | F (p, q) | is a power spectrum of F (p, q), α is a filtering parameter, S {. is a smoothing factor, H (p, q) can suppress a high frequency part of a complex interferogram, a low frequency part of the complex interferogram is retained, a kernel size N of a frequency domain Goldstein filter is different according to a selected fourier transform method, and a condition d1 or d2 is satisfied, in this embodiment, the complex interferogram is transformed from a spatial domain to a frequency domain using ordinary fourier transform, and the condition d1 is satisfied by setting N to 5:

d1) when the Goldstein filtering algorithm is implemented by using the ordinary Fourier transform, N is an odd number, N is more than or equal to 3 and less than or equal to min (w, h), and N belongs to N^*In which N is^*Denotes a positive integer, min (·) denotes taking the minimum value, in particular, N is proposed to be set to N ═ 5;

d2) when the Goldstein filtering algorithm is implemented using fast Fourier transform, N is an even number, 2 ≦ N ≦ min (w, h), in particular, N is proposed to be set to N ≦ 4 or N ≦ 6;

step e: if the pyramid level i satisfies the condition: i is more than or equal to 0 and less than or equal to L-1, and a pair interference pattern G is arranged_iWhen the sliding filtering is carried out, the sliding step length S is equal to 1, and a frequency domain Goldstein filter with the kernel size N being equal to 5 is utilized to carry out the complex interference pattern G on the ith layer of the pyramid_iPerforming slip filtering processing to obtain filtered interferogram

Step f: for filtered interferogramsUsing direct removal of imagesDown-sampling in even rows and even columns usingThe odd rows and the odd columns form an i +1 th layer Goldstein pyramid G_i+1

Step g: repeating steps (e) and (f) until a lth 3-level Goldstein pyramid G is obtained as shown in fig. 5_LArranging the 0-L layer Goldstein pyramid images in a combination mode from bottom layer to top layer from small layer to large layer to obtain a Goldstein pyramid multi-scale image with the layer number of L +1, such as a 0-layer pyramid image G shown in FIG. 2₀At the bottom, L-th pyramid image G_LAt the top level, the resulting layer 1 Goldstein pyramid image is shown in FIG. 3, and the layer 2 Goldstein pyramid image is shown in FIG. 4.