CN115184934B - SAR image imaging projection plane extraction method - Google Patents

Abstract

The invention relates to an SAR image imaging projection plane extraction method, in particular to SAR image processing. The invention aims to solve the problem of image defocusing caused by multiple ship postures in SAR images. The process is as follows: 1. obtaining SAR images, which are marked as I (m, n); 2. setting G _I When I (m, n) is greater than or equal to G _I When the scattering point is extracted from I (m, n), the j-th scattering point is recorded as s _j Forming a scattering point set, extracting all J scattering points from I (m, n) and forming a new SAR image by the corresponding position and amplitude of each scattering point3. Constructing a parameter matrix; 4. obtaining a new parameter matrix; 5. calculating root mean square error; 6. comparing the root mean square error with the precision; if the root mean square error is less than or equal to the precision, an optimal parameter matrix and an optimal classification result are obtained, and seven is executed; otherwise, let i=i+1, repeatedly execute four to six; 7. and estimating the image amplitude to obtain an image sequence. The SAR image processing method and device are used in the SAR image processing field.

Description

SAR image imaging projection plane extraction method

Technical Field

The present invention relates to SAR image processing.

Background

The synthetic aperture radar (synthetic aperture radar, SAR) has important practical value for ship target imaging, and is widely applied to the fields of sea area supervision, ship detection and the like. However, when the ship target makes complex motions, a plurality of imaging projection planes (imaging projection planes, IPPs) are generated during imaging, so that a plurality of ship postures appear in the SAR image, the imaging quality is reduced, and the target feature extraction effect is affected. In order to solve the problem, expert scholars at home and abroad propose a plurality of optimal imaging data segment selection methods to determine a single IPP, obtain radar images containing single ship postures and improve imaging quality. But these methods typically require the generation of image sequences, increasing computational complexity and reducing image resolution.

Disclosure of Invention

The invention aims to solve the problem of image defocusing caused by a plurality of ship postures in SAR images, and provides an SAR image imaging projection plane extraction method.

The SAR image imaging projection plane extraction method specifically comprises the following steps:

firstly, obtaining SAR images by using an SAR imaging method, and marking the SAR images as I (m, n);

wherein m=1, 2, …, N _a For azimuth position, n=1, 2, …, N _r Is distance position N _a As azimuth unit number, N _r Is the number of distance units;

step two, setting amplitude threshold G _I When I (m, n) is greater than or equal to G _I Extracting scattering points from SAR image I (m, n), and recording the j-th scattering point position as s _j ＝[m _j ,n _j ] ^T Form scattering point set s= { S ₁ ,s ₂ ,…,s _J Extracting all J scattering points and the corresponding position s of each scattering point from the SAR image I (m, n) _j Amplitude I (m) _j ,n _j ) Construction of new SAR image

Wherein m is _j And n _j The azimuth position and the distance position of the jth scattering point are respectively, j=1, 2, …, J is the number of the scattering points, and T represents transposition;

in the method, in the process of the invention,and have->Representing the jth scattering point in the kth SAR image +.>The amplitude of (a); k=1, 2, …, K is a cluster number;

step three, the scattering point set S can be divided into K clusters, the initialization iteration number i=0, and the parameter matrix Θ is initialized ⁽⁰⁾ Setting iteration precision delta of an EM algorithm;

C _k representing the kth category (cluster), each cluster modeled as a two-dimensional gaussian distribution, denoted μ _k For the center of the kth cluster, Σ _k For the covariance matrix of the kth cluster, P (C _k ) Characterizing the prior probability of selecting the kth cluster;

μ _k 、Σ _k and P (C) _k ) Forming a parameter matrix

Step four, updating the parameter matrix by using an EM algorithm to obtain a new parameter matrix Θ ⁽ⁱ⁺¹⁾ ；

Step five, calculating the root mean square error RMSE (Θ) ⁽ⁱ⁾ ,Θ ⁽ⁱ⁺¹⁾ )；

Step six, comparing the root mean square error RMSE (Θ) ⁽ⁱ⁾ ,Θ ⁽ⁱ⁺¹⁾ ) And the size of the precision delta;

if RMSE (Θ) ⁽ⁱ⁾ ,Θ ⁽ⁱ⁺¹⁾ ) Delta less than or equal to the value, the optimal parameter matrix theta is obtained ^opt ＝Θ ⁽ⁱ⁺¹⁾ And an optimal classification result P ^opt (s _j ∈C _k |s _j )＝P ⁽ⁱ⁺¹⁾ (s _j ∈C _k |s _j ；Θ ⁽ⁱ⁾ ) Executing the seventh step; otherwise, let i=i+1, repeatedly executing step four to step six;

wherein P is ⁽ⁱ⁺¹⁾ (s _j ∈C _k |s _j ；Θ ⁽ⁱ⁾ ) Is latent variable, expressed in parameter matrix theta ⁽ⁱ⁾ The jth scattering point belongs to class C _k Probability of P ^opt (s _j ∈C _k |s _j ) For the optimal classification result, the j scattering point belongs to the category C _k Probability of (2);

step seven, estimating the image amplitude to obtain an image sequence

The beneficial effects of the invention are as follows:

the invention provides a novel SAR image imaging projection plane extraction method for reducing the calculation complexity and guaranteeing the image resolution, which can finish IPP extraction in an SAR image domain and obtain an SAR image containing a single ship gesture by utilizing a cluster analysis technology. The method can keep ship target information to the maximum extent, improves SAR image quality, ensures azimuth resolution, is beneficial to subsequent target feature extraction, classification and identification, and has practical application value.

Drawings

FIG. 1 is a flow chart of a new SAR image imaging projection plane extraction method;

FIG. 2 shows a ship target scattering point model used in the present invention in the first embodiment, wherein the X axis is the azimuth direction, the Y axis is the distance direction, and the Z axis is the vertical sea level direction;

FIG. 3a is a SAR image obtained by the method of the present invention in the first embodiment;

FIG. 3b is a set of scattering points extracted by the method of the present invention in accordance with the first embodiment;

FIG. 4 shows the classification result of scattering points obtained by the method of the present invention in the first embodiment;

FIG. 5 is a SAR image 1 comprising a single ship target attitude obtained by the method of the present invention in accordance with the first embodiment;

fig. 6 is a SAR image 2 comprising a single ship target pose obtained by the method of the present invention in embodiment one.

Detailed Description

The first embodiment is as follows: referring to fig. 1, the specific procedure of the method for extracting the imaging projection plane of the SAR image according to the present embodiment is as follows:

firstly, obtaining SAR images by using an SAR imaging method, and marking the SAR images as I (m, n);

wherein m=1, 2, …, N _a For azimuth position, n=1, 2, …, N _r Is distance position N _a As azimuth unit number, N _r Is the number of distance units;

step two, setting amplitude threshold G _I When I (m, n) is greater than or equal to G _I Extracting scattering points from SAR image I (m, n), and recording the j-th scattering point position as s _j ＝[m _j ,n _j ] ^T Form scattering point set s= { S ₁ ,s ₂ ,…,s _J Extracting all J scattering points and the corresponding position s of each scattering point from the SAR image I (m, n) _j Amplitude I (m) _j ,n _j ) Construction of new SAR image

Wherein m is _j And n _j The azimuth position and the distance position of the jth scattering point are respectively, j=1, 2, …, J is the number of the scattering points, and T represents transposition;

in the method, in the process of the invention,and have->Representing the jth scattering point in the kth SAR image +.>The amplitude of (a); k=1, 2, …, K is a cluster number;

step three, the scattering point set S can be divided into K clusters, the initialization iteration number i=0, and the parameter matrix Θ is initialized ⁽⁰⁾ Setting iteration precision delta of an EM algorithm;

C _k representing the kth category (cluster), each cluster modeled as a two-dimensional gaussian distribution, denoted μ _k For the center of the kth cluster, Σ _k For the covariance matrix of the kth cluster, P (C _k ) Characterizing the prior probability of selecting the kth cluster;

μ _k 、Σ _k and P (C) _k ) Forming a parameter matrix

Step four, updating the parameter matrix by using an EM algorithm to obtain a new parameter matrix Θ ⁽ⁱ⁺¹⁾ ；

Step five, calculating the root mean square error RMSE (Θ) ⁽ⁱ⁾ ,Θ ⁽ⁱ⁺¹⁾ )；

Step six, comparing the root mean square error RMSE (Θ) ⁽ⁱ⁾ ,Θ ⁽ⁱ⁺¹⁾ ) And the size of the precision delta;

if RMSE (Θ) ⁽ⁱ⁾ ,Θ ⁽ⁱ⁺¹⁾ ) Delta less than or equal to the value, the optimal parameter matrix theta is obtained ^opt ＝Θ ⁽ⁱ⁺¹⁾ And an optimal classification result P ^opt (s _j ∈C _k |s _j )＝P ⁽ⁱ⁺¹⁾ (s _j ∈C _k |s _j ；Θ ⁽ⁱ⁾ ) Executing the seventh step; otherwise, let i=i+1, repeatedly executing step four to step six;

wherein P is ⁽ⁱ⁺¹⁾ (s _j ∈C _k |s _j ；Θ ⁽ⁱ⁾ ) Is latent variable, expressed in parameter matrix theta ⁽ⁱ⁾ The jth scattering point belongs to class C _k Probability of P ^opt (s _j ∈C _k |s _j ) For the optimal classification result, the j scattering point belongs to the category C _k Probability of (2);

step seven, estimating the image amplitude to obtain an image sequence

The second embodiment is as follows: the present embodiment differs from the specific embodiment in that the step two has an amplitude threshold G _I Can be obtained by the OTSU method.

The OTSU method divides the image into a background part and a target part, and evaluates the advantages and disadvantages of the amplitude threshold by taking the inter-class variance as a standard, so that the minimum error classification probability is ensured. The OTSU method has the advantages of automatic selection of an amplitude threshold, stable performance and the like, and is widely used in the field of image processing.

Other steps and parameters are the same as in the first embodiment.

And a third specific embodiment: the difference between this embodiment and one of the first to second embodiments is that the EM algorithm in the fourth step is implemented as follows:

step four, first: implementing step E, updating latent variable P ⁽ⁱ⁺¹⁾ (s _j ∈C _k |s _j ；Θ ⁽ⁱ⁾ )，P ⁽ⁱ⁺¹⁾ (s _j ∈C _k |s _j ；Θ ⁽ⁱ⁾ ) The calculation mode of (a) is as follows:

wherein P is ⁽ⁱ⁺¹⁾ (s _j ∈C _k |s _j ；Θ ⁽ⁱ⁾ ) Is latent variable, expressed in parameter matrix theta ⁽ⁱ⁾ The jth scattering point belongs to class C _k Probability of P ⁽ⁱ⁾ (s _j |s _j ∈C _k ；Θ ⁽ⁱ⁾ ) Represented in a parameter matrix Θ ⁽ⁱ⁾ The function value of the lower conditional probability;

step four, two: implementing M step, updating parameter matrix theta ⁽ⁱ⁺¹⁾ ，Θ ⁽ⁱ⁺¹⁾ The calculation mode of (a) is as follows:

and P ⁽ⁱ⁺¹⁾ (C _k ) Composing a parameter matrix->

Other steps and parameters are the same as in one of the first to second embodiments.

The specific embodiment IV is as follows: this embodiment differs from one to three embodiments in that the root mean square error RMSE (Θ ⁽ⁱ⁾ ,Θ ⁽ⁱ⁺¹⁾ ) The calculation mode of (a) is as follows:

in the method, in the process of the invention,covariance matrix for kth cluster in ith iteration result +.>R=1, 2, s=1, 2.

Other steps and parameters are the same as in one to three embodiments.

Fifth embodiment: the difference between the present embodiment and one to four embodiments is that the image amplitude in the seventh step may be estimated as:

other steps and parameters are the same as in one of the first to fifth embodiments.

The following examples are used to verify the benefits of the present invention:

embodiment one:

the following examples are used to verify the benefits of the present invention:

in the embodiment, SAR imaging is completed by combining the method with a Chirp Scaling (CS) algorithm, parameters of an airborne SAR system are shown in table 1, and a selected scattering point model is shown in fig. 2.

FIG. 3a is a SAR image generated by CS algorithm, it can be seen that two IPPs appear in the SAR image, resulting in azimuth defocusing; fig. 3b is a set of scattering points extracted from a SAR image.

Classifying the scattering points in fig. 3b gives the classification result shown in fig. 4, and it can be seen that the scattering points belonging to both IPPs are correctly classified.

According to the scattering point classification result in fig. 4, the image amplitude is estimated to obtain the SAR images shown in fig. 5 and 6, and the SAR image only comprises the single gesture of the ship target, so that the method provided by the invention can extract the IPP, improve the SAR image quality and verify the effectiveness of the proposed method.

Table 1 on-board SAR system parameters

Claims (5)

1. The SAR image imaging projection plane extraction method is characterized by comprising the following steps of: the method comprises the following specific processes:

   firstly, obtaining SAR images by using an SAR imaging method, and marking the SAR images as I (m, n);

   wherein m=1, 2, …, N _a For azimuth position, n=1, 2, …, N _r Is distance position N _a As azimuth unit number, N _r Is the number of distance units;

   step two, setting amplitude threshold G _I When I (m, n) is greater than or equal to G _I Extracting scattering points from SAR image I (m, n), and recording the j-th scattering point position as s _j ＝[m _j ,n _j ] ^T Form scattering point set s= { S ₁ ,s ₂ ,…,s _J Extracting all J scattering points and the corresponding position s of each scattering point from the SAR image I (m, n) _j Amplitude I (m) _j ,n _j ) Construction of new SAR image

   Wherein m is _j And n _j The azimuth position and the distance position of the jth scattering point are respectively, j=1, 2, …, J is the number of the scattering points, and T represents transposition;

   in the method, in the process of the invention,and have-> Representing the jth scattering point in the kth SAR image +.>The amplitude of (a); k=1, 2, …, K is a cluster number;

   step three, the scattering point set S can be divided into K clusters, the initialization iteration number i=0, and the parameter matrix Θ is initialized ⁽⁰⁾ Setting iteration precision delta of an EM algorithm;

   C _k representing the kth category, each cluster is modeled as a two-dimensional gaussian distribution, denoted μ _k For the center of the kth cluster, Σ _k For the covariance matrix of the kth cluster, P (C _k ) Characterizing the prior probability of selecting the kth cluster;

   μ _k 、Σ _k and P (C) _k ) Forming a parameter matrix

   Step four, updating the parameter matrix by using an EM algorithm to obtain a new parameter matrix Θ ⁽ⁱ⁺¹⁾ ；

   Step five, calculating the root mean square error RMSE (Θ) ⁽ⁱ⁾ ,Θ ⁽ⁱ⁺¹⁾ )；

   Step six, comparing the root mean square error RMSE (Θ) ⁽ⁱ⁾ ,Θ ⁽ⁱ⁺¹⁾ ) And the size of the precision delta;

   if RMSE (Θ) ⁽ⁱ⁾ ,Θ ⁽ⁱ⁺¹⁾ ) Delta less than or equal to the value, the optimal parameter matrix theta is obtained ^opt ＝Θ ⁽ⁱ⁺¹⁾ And an optimal classification result P ^opt (s _j ∈C _k |s _j )＝P ⁽ⁱ⁺¹⁾ (s _j ∈C _k |s _j ；Θ ⁽ⁱ⁾ ) Executing the seventh step; otherwise, let i=i+1, repeatedly executing step four to step six;

   wherein P is ⁽ⁱ⁺¹⁾ (s _j ∈C _k |s _j ；Θ ⁽ⁱ⁾ ) Is a potential changeThe quantity is expressed in a parameter matrix theta ⁽ⁱ⁾ The jth scattering point belongs to class C _k Probability of P ^opt (s _j ∈C _k |s _j ) For the optimal classification result, the j scattering point belongs to the category C _k Probability of (2);

   step seven, estimating the image amplitude to obtain an image sequence
2. 2. The SAR image imaging projection plane extraction method according to claim 1, wherein: the amplitude threshold G in the second step _I Can be obtained by the OTSU method.
3. 3. The SAR image imaging projection plane extraction method according to claim 2, wherein: the EM algorithm in the fourth step is specifically implemented as follows:

   step four, first: implementing step E, updating latent variable P ⁽ⁱ⁺¹⁾ (s _j ∈C _k |s _j ；Θ ⁽ⁱ⁾ )，P ⁽ⁱ⁺¹⁾ (s _j ∈C _k |s _j ；Θ ⁽ⁱ⁾ ) The calculation mode of (a) is as follows:

   wherein P is ⁽ⁱ⁺¹⁾ (s _j ∈C _k |s _j ；Θ ⁽ⁱ⁾ ) Is latent variable, expressed in parameter matrix theta ⁽ⁱ⁾ The jth scattering point belongs to class C _k Probability of P ⁽ⁱ⁾ (s _j |s _j ∈C _k ；Θ ⁽ⁱ⁾ ) Represented in a parameter matrix Θ ⁽ⁱ⁾ The function value of the lower conditional probability;

   step (a)And IV, two: implementing M step, updating parameter matrix theta ⁽ⁱ⁺¹⁾ ，Θ ⁽ⁱ⁺¹⁾ The calculation mode of (a) is as follows:

   and P ⁽ⁱ⁺¹⁾ (C _k ) Composing a parameter matrix->
4. 4. The SAR image imaging projection plane extraction method according to claim 3, wherein: the root mean square error RMSE (Θ) ⁽ⁱ⁾ ,Θ ⁽ⁱ⁺¹⁾ ) The calculation mode of (a) is as follows:

   in the method, in the process of the invention,covariance matrix for kth cluster in ith iteration result +.>R=1, 2, s=1, 2.
5. 5. The SAR image imaging projection plane extraction method according to claim 4, wherein: the image amplitude in the seventh step can be estimated as: