KR102501303B1 - Method of similarity measurement and recognition between targets of SAR images - Google Patents

Abstract

A method for calculating similarity of targets and a method for identifying a target according to the present invention are provided. The method for calculating similarity of targets comprises the steps of: receiving a target image; extracting target scattering points (g) from the target image and generating a set of target scattering points (G) based on the positions of the target scattering points (g); generating a target world view vector (WVV) descriptor matrix (G') from the set of target scattering points (G) using a WVV; obtaining a template WVV descriptor matrix (B') from template information; circularly moving the target WVV descriptor matrix (G') in the row direction by a plurality of shift values (k) within a preset range to generate a plurality of moving WVV descriptor matrices (G'_((k))); and calculating a maximum shift value (k') at which a maximum similarity value (Smax(k)) between each of the plurality of moving WVV descriptor matrices (G'_((k))) and the template WVV descriptor matrix (B') is maximum, and determining a final similarity (S') based on the maximum similarity value (Smax(k')) at this time.

Description

Method of similarity measurement and recognition between targets of SAR images

The present invention relates to a similarity calculation method between a target of a SAR image and a pre-stored template, and relates to a similarity calculation method using a World View Vector (WVV) descriptor that is robust to rotational transformation of target scattering points and a target identification method using the same. .

Automatic target recognition (ATR) is an important research topic in many applications such as surveillance, homeland security and military affairs. A lot of research has been done on SAR-ATR technology for automatically detecting and identifying ground targets from Synthetic Aperture Radar (SAR) images.

Since the characteristics of SAR images change depending on the observation angle or the environment around the target, it is very difficult to automatically identify targets in various environments. Therefore, it is necessary to develop SAR-ATR technology that extracts the features of the target of interest from the SAR image and is robust to feature changes according to the observation angle or the surrounding environment of the target.

Recently, studies have been conducted to identify targets in SAR images using various observation vectors extracted from scattering points. Scattering points in SAR images are very useful as feature vectors for identification because they show unique electromagnetic characteristics depending on the material of the target.

In general, rotational transformation of the scattering point may occur due to an error in the step of estimating the azimuth of the target. The rotation transformation of the scattering points lowers the similarity value between the target scattering points and the template scattering points.

An object of the present invention is to provide a method for calculating a similarity between a target of a SAR image and a pre-stored template.

An object to be solved by the present invention is to provide a method for identifying a target of a SAR image based on similarity with a pre-stored template.

A method for calculating the similarity of a target according to an embodiment of the present invention includes receiving a target image, extracting target scattering points (g) from the target image, and targeting the target scattering points (g) based on the locations of the target scattering points (g). Generating a scattering point set (G), generating a target WVV descriptor matrix (G′) from the target scattering point set (G) using WVV (World View Vector), template WVV descriptor matrix (from template information) Obtaining a plurality of shifted WVV descriptor matrices (G' (k) by circularly moving the target WVV descriptor matrix (G') in a row direction by a plurality of shift values ( _k ) within a preset range, respectively. ), and a maximum shift value at which a maximum similarity value (Smax _(k) ) between each of the plurality of shifted WVV descriptor matrices (G′ (k)) and the template WVV descriptor matrix (B′) is maximized ( k') is calculated, and the final similarity (S') is determined based on the maximum similarity value (Smax(k')) at this time.

According to an example, the number of target scattering points (g) in the target scattering point set (G) may be N, and the target WVV descriptor matrix (G') may be an Nx360 matrix. The values of the ith row of the target WVV descriptor matrix (G′) are the position of the ith target scattering point ( _gi ) as the origin and the remaining target scattering points (g ₁ to g _i-1 , g _i+1 to g _N ) can be determined based on the distance and angle.

According to another example, the obtaining of the template WVV descriptor matrix (B′) may include receiving a template image from a database, extracting template scattering points (b) from the template image, and obtaining the template scattering points (b). ), and generating a template WVV descriptor matrix (B′) from the template scattering point set (B) using the WVV. .

According to another example, the method for calculating the similarity of the target may include a moving WVV descriptor matrix (G' _(k1) ), which is one of the plurality of moving WVV descriptor matrices (G' _(k) ), and the template WVV descriptor matrix (B'). ) may further include calculating a similarity matrix S(G' _(k1) , B') between. The moving WVV descriptor matrix (G' _(k1) ) is an N × 360 matrix consisting of g' _(k1)it (i = 1, ..., N; t = 1, 2, ..., 360), The template WVV descriptor matrix (B') is an M × 360 matrix consisting of b' _jt (j = 1, ..., M; t = 1, 2, ..., 360), and the similarity matrix (S( G′ _(k1) , B′)) may be an N×M matrix composed of s _(k1)ij (i=1, ..., N; j=1, ..., M). The s _(k1)ij can be calculated as s _(k1)ij = (360-Σ _t │g' _(k1)it -b' _jt │ ² )/360.

According to another example, the method of calculating the similarity of the target may include the moving WVV descriptor matrix G' (k1) and the template WVV descriptor matrix by using the similarity matrix S(G' _(k1) , B' _). A step of calculating a maximum similarity value (Smax(k1)) between (B') may be further included. When N≥M, the maximum similarity value Smak(k1) may be calculated by Smax(k1)=Σ _j max({s _(k1)ij } _i ). When N<M, the maximum similarity value Smak(k1) may be calculated by Smax(k1)=Σ _i max({s _(k1)ij } _j ).

According to another example, the maximum shift value k' may be calculated by k' = arg max Smax(k). The final similarity (S') may be calculated by S'=Smax(k').

A computer program according to an embodiment of the present invention is stored in a medium in order to execute the above-described similarity calculating method of a target using a computing device.

A target identification method according to an embodiment of the present invention includes receiving a target image, extracting target scattering points (g) from the target image, and target scattering points (g) based on the positions of the target scattering points (g). Generating a set (G), generating a target WVV descriptor matrix (G') from the target scattering point set (G) using a World View Vector (WVV), and generating the target WVV descriptor matrix (G') Generating a plurality of shifted WVV descriptor matrices (G′ _(k) ) by circularly moving each of a plurality of shift values (k) in a row direction within a preset range, template WVV descriptors from each of a plurality of template information stored in the database Obtaining a matrix (B'), wherein a maximum similarity value (Smax(k)) between each of the plurality of moving WVV descriptor matrices (G' _(k) ) and the template WVV descriptor matrix (B') is maximized. Calculating the shift value (k') and determining the final similarity (S') based on the maximum similarity value (Smax(k')) at this time, and a plurality of templates WVV each obtained from the plurality of template information and identifying a target of the target image based on a template WVV descriptor matrix having the highest final similarity (S') to the target image among descriptor matrices.

According to an example, each of the plurality of template information may include a type of template target and a template image of the template target.

According to another example, the target identification method is a degree of similarity between a moving WVV descriptor matrix (G' _(k1) ), which is one of the plurality of moving WVV descriptor matrices (G' _(k) ), and the template WVV descriptor matrix (B'). A step of calculating the matrix S(G' _(k1) , B') may be further included. The moving WVV descriptor matrix (G' _(k1) ) is an N × 360 matrix consisting of g' _(k1)it (i = 1, ..., N; t = 1, 2, ..., 360), The template WVV descriptor matrix (B') is an M × 360 matrix consisting of b' _jt (j = 1, ..., M; t = 1, 2, ..., 360), and the similarity matrix (S( G′ _(k1) , B′)) may be an N×M matrix composed of s _(k1)ij (i=1, ..., N; j=1, ..., M). The s _(k1)ij can be calculated as s _(k1)ij = (360-Σ _t │g' _(k1)it -b' _jt │ ² )/360.

According to another example, the target identification method uses the similarity matrix S(G' _(k1) , B') to determine the moving WVV descriptor matrix G' _(k1 ) and the template WVV descriptor matrix (B'). ') may further include calculating a maximum similarity value Smax(k1). When N≥M, the maximum similarity value Smak(k1) is calculated by Smax(k1)=Σ _j max({s _(k1)ij } _i ), and when N<M, the maximum similarity value (Smak(k1)) can be calculated by Smax(k1)=Σ _i max({s _(k1)ij } _j ).

According to another example, when N≥M, the final similarity (S') is calculated by S'=Smax(k')/M, and when N<M, the final similarity (S') is S It can be calculated by '= Smax(k')/N.

A computer program according to another embodiment of the present invention is stored in a medium to execute the above-described target identification method using a computing device.

According to the present invention, the similarity between the target scattering points and pre-stored template scattering points can be accurately calculated even if the target scattering points are rotated.

1 shows a conceptual diagram of a target identification system according to an embodiment of the present invention.
2A shows an exemplary target image.
FIG. 2B exemplarily shows a target image in which rotation transformation has occurred.
3 shows exemplary template information.
4 schematically shows the internal configuration of a target identification device according to an embodiment of the present invention.
5 is a flowchart illustrating a method for calculating similarity between targets according to an embodiment of the present invention.
FIG. 6 shows one example among a plurality of target WVV descriptors reconstructed from a plurality of target scattering points extracted from the target image of FIG. 2a.
7 is a flowchart illustrating a method for calculating similarity between targets according to an embodiment of the present invention.
8 shows the degree of error of the rotation conversion angle estimated by the method for calculating the similarity of targets according to the degree of random movement of scattering point coordinates according to the random movement amount.
9 shows the error degree of the rotation conversion angle estimated by the target similarity calculation method according to the rotation conversion angle between [-10°, 10°].
10 shows a result of comparing the final similarity estimated by the method of calculating the similarity of a target according to the rotation conversion angle between [-10°, 10°] with the similarity estimated by the conventional method.
FIG. 11 shows a result of comparing the final similarity estimated by the method of calculating the similarity of targets according to the random movement amount of scattering point coordinates with the similarity estimated by the conventional method.

Hereinafter, various embodiments will be described in detail so that those skilled in the art can easily implement the present disclosure with reference to the accompanying drawings. However, since the technical idea of the present disclosure may be implemented in various forms, it is not limited to the embodiments presented herein. In describing the embodiments presented in this specification, if it is determined that a detailed description of related known technologies may obscure the gist of the technical idea of the present disclosure, detailed descriptions of the known technologies will be omitted. The same or similar components are assigned the same reference numerals, and duplicate descriptions thereof will be omitted.

In this specification, when an element is described as being “connected” to another element, this includes not only the case of being “directly connected” but also the case of being “indirectly connected” with another element intervening therebetween. When an element "includes" another element, this means that it may further include another element without excluding another element in addition to the other element unless otherwise stated.

Some embodiments may be described as functional block structures and various processing steps. Some or all of these functional blocks may be implemented with any number of hardware and/or software components that perform a particular function. For example, functional blocks of the present disclosure may be implemented by one or more microprocessors or circuit configurations for a predetermined function. The functional blocks of this disclosure may be implemented in a variety of programming or scripting languages. The functional blocks of this disclosure may be implemented as an algorithm running on one or more processors. The functions performed by the function blocks of the present disclosure may be performed by a plurality of function blocks, or the functions performed by the plurality of function blocks in the present disclosure may be performed by one function block. In addition, the present disclosure may employ prior art for electronic environment setting, signal processing, and/or data processing.

1 shows a conceptual diagram of a target identification system according to an embodiment of the present invention.

Referring to FIG. 1 , the target identification system 100 receives a target image 110, compares a target of the target image 110 with a plurality of template information 132 previously stored in a database 130, and compares the target image ( 110) includes a target identification device 120 for identifying the target.

The target image 110 may be a SAR image generated using a synthetic aperture radar (SAR). The target image 110 may be an image of a part corresponding to a target of interest among SAR images. 2A shows an exemplary target image 110a.

The characteristics of the target image 110 may vary according to an angle at which the target is observed by SAR or a part of the target image 110 that is obscured by an environment around the target, such as a tree or surrounding terrain. Also, rotation conversion may occur in the step of estimating the azimuth of the target. FIG. 2B exemplarily shows a target image 110b on which rotation transformation has occurred. As shown in FIG. 2B , when a rotation transformation occurs in the target image 110b, similarity with the template information 132 stored in the database 130 may decrease.

The template information 132 is target information that is previously identified and is compared with the target of the target image 110 . In this specification, a target to be identified is referred to as a target, and a target having related information stored in the database 130 is referred to as a template or template target. 3 shows exemplary template information 132 . As shown in FIG. 3 , the template information 132 includes a template target type 132a and a template image 132b.

The type of template target 132a may include, for example, a target category such as a tank, an armored vehicle, and a truck, and vehicle model names such as a K-1 tank and a K200 armored vehicle. Template image 132b may be a SAR image of a template target. As shown in FIG. 3 , the template image 132b varies according to the type 132a of the template target, but it is difficult to easily distinguish it.

The template information 132 may include information generated from the template image 132b. For example, the template information 132 may include template WVV descriptor matrices respectively generated from the template images 132b instead of the template images 132b.

The target identification device 120 compares the target image 110 with each template image 132b to calculate a degree of similarity, and calculates the target image 110 based on the template image 132b having the highest similarity with the target image 110. target can be identified. The target identification device 120 may calculate similarity using World View Vector (WVV) used by Murtagh (1992) for feature vector-based pattern recognition.

The mark identification device 120 may be implemented with at least one computing device including at least one processor and memory. The processor typically controls the overall operation of the mark identification device 120 . A processor may perform basic arithmetic, logic, and input/output operations, and execute program codes stored in memory, for example. The processor may execute a method for calculating similarity between targets according to an embodiment of the present invention. The processor may execute the target identification method according to an embodiment of the present invention.

The memory is a recording medium readable by a processor and may include a permanent mass storage device such as RAM, ROM, and a disk drive. An operating system and at least one program or application code may be stored in the memory. The memory may store program codes for the mark identification device 120 to execute the similarity calculation method of targets according to the present invention. The memory may store program codes for the mark identification device 120 to execute the target identification method of the target according to the present invention.

4 schematically shows the internal configuration of the target identification device 120 according to an embodiment of the present invention.

Referring to FIG. 4 , the target identification device 120 includes a similarity calculation unit 122 and a target identification unit 124 .

The similarity calculation unit 122 may extract target scattering points from the target image 110 and generate a target scattering point set based on the locations of the target scattering points. The similarity calculating unit 122 may generate a target WVV descriptor matrix from the target scattering point set using WVV. The similarity calculating unit 122 may generate a plurality of shifted WVV descriptor matrices by circularly moving the target WVV descriptor matrix in a row direction by a plurality of shift values k within a preset range. The similarity calculation unit 122 may obtain a template WVV descriptor matrix from each of the plurality of template information 132 stored in the database 130 . The similarity calculation unit 122 may calculate a maximum shift value at which the maximum similarity value between each of the plurality of moving WVV descriptor matrices and the template WVV descriptor matrix is maximized, and determine the final similarity based on the maximum similarity value at this time.

The target identification unit 124 determines the target image 110 based on the template WVV descriptor matrix having the highest final similarity with the target image 120 among the plurality of template WVV descriptor matrices obtained from the plurality of template information 132 respectively. can identify.

5 is a flowchart illustrating a method for calculating similarity between targets according to an embodiment of the present invention.

Referring to FIG. 5 , the target identification device 120 receives a target image (eg, 110a of FIG. 2a) (S110). The target image 110a may be a part corresponding to a target of interest among SAR images.

The similarity calculating unit 122 may extract target scattering points g from the target image 110a and generate a target scattering point set G based on the positions of the target scattering points g (S120). ). N target scattering points g may be extracted in the target image 110a.

The target scattering point set G extracted from the target image 110a may consist of the positions (x, y) of N target scattering points g as in [Equation 1].

[Equation 1]

G = {(x _i , y _i )|i=1, 2, ..., N}

The similarity calculating unit 122 may generate a target WVV descriptor matrix G′ from the target scattering point set G using WVV (S130).

FIG. 6 shows a target WVV descriptor generated based on one scattering point among target scattering points extracted from the target image of FIG. 2a.

Referring to FIG. 6, WVV converts scattering point location (x, y) information into polar coordinates (r, θ), and then converts the ith target scattering point (g _i ), one of N target scattering points, to the origin Defined as N-1 vectors (r _ik , θ _ik ) composed of the sizes and angles of the remaining target scattering points (g ₁ to g _i-1 , g _i+1 to g _N ), and as shown in [Equation 2] is expressed

[Equation 2]

WVV _i ={(r _ik , θ _ik )|k=1, 2, ..., N; k≠i; θ _ik ≤ θ _ik+1 }

The similarity calculation unit 122 linearly interpolates N-1 vectors (r _ik , θ _ik ) at an equal angle (eg, 1°), normalizes the vector size in order not to be affected by the scale, and calculates [Equation 3] and Similarly, a target WVV descriptor (G' _i ) consisting of 360 observation vectors (g _it ) is created.

[Formula 3]

G' _i ={g' _it |t=1, ..., 360}

If there are multiple target scattering points corresponding to the same angle, the size of the observation vector may be designated as the maximum value among them. By repeating this process as many times as the number of target scattering points (N), N target scattering points (g ₁ to g _N ) are formed as a target WVV descriptor matrix (G') consisting of 360 observation vector sets, respectively, as shown in [Equation 4]. can be reconstructed into

[Formula 4]

G'={g' _it i=1, ..., N; t=1, ..., 360}

As shown in Equation 4, the target WVV descriptor matrix (G') is an Nx360 matrix composed of g' _it . The values of the ith row of the target WVV descriptor matrix (G′) are based on the position of the ith target scattering point ( _gi ) as the origin and the rest of the target scattering points (g ₁ to g _i-1 , g _i+1 to g _N ) can be determined based on the distance and angle to

FIG. 6 shows 56 first to 56th target WVV descriptors (G' 1 to 56) reconstructed _from 56 first to 56th target scattering points (g ₁ to g ₅₆ ) extracted from the target image 110a of FIG. 2a. An example of the 23rd target WVV descriptor (G' ₂₃ ) among G' ₅₆ ) is shown. The 23rd target WVV descriptor (G' ₂₃ ) is composed of 360 observation vectors (g' _23t ) generated by linear interpolation at 1˚ intervals based on the 23rd target scattering point (g ₂₃ ). Since the target WVV descriptor (G' _i ) represents the distance to each target scattering point at every 1° interval with the i th target scattering point (gi ₎ as the origin in the spatial domain, the WVV descriptor between the two scattering point groups Using the matrix, information on the position of scattering points as well as structural similarity can be determined.

Referring back to FIG. 5 , the target identification device 120 receives template information 132 from the database 130, and the similarity calculator 122 calculates the template WVV descriptor matrix B' from the template information 132. It can be obtained (S140).

The template information 132 may be one of a plurality of template information 132 previously stored in the database 130 . The template information 132 may be the template image 132b shown in FIG. 3 . The similarity calculation unit 122 may extract template scattering points (b) from the template image 132b and generate a template scattering point set (B) based on the locations of the template scattering points (b). It is assumed that there are M template scattering points (b) in the template scattering point set (B).

The similarity calculation unit 122 uses WVV to generate the target WVV descriptor matrix (G′) from the target scattering point set (G), and the template WVV descriptor matrix from the template scattering point set (B) using WVV in the same manner as the process of generating the target WVV descriptor matrix (G′). (B') can be produced.

A template scattering point set (B) consisting of M template scattering points (b ₁ to b _M ) can be reconstructed into a template WVV descriptor matrix (B′) each consisting of 360 observation vector sets as shown in [Equation 5]. .

[Formula 5]

B'={b' _jt |j=1, ..., M; t=1, ..., 360}

The template WVV descriptor matrix (B') is an Mx360 matrix consisting of b' _jt . The values of the jth row of the template WVV descriptor matrix (B′) are the location of the jth template scattering point (b _j ) as the origin and the remaining target scattering points (b ₁ to b _j-1 , b _j+1 to b _M ) can be determined based on the distance and angle to

The similarity calculating unit 122 circularly shifts the target WVV descriptor matrix G' in the row direction by a plurality of shift values k within a preset range, respectively, so that the plurality of shifted WVV descriptor matrices G' _{( k)} ) can be generated (S150). The shifted WVV descriptor matrix (G' _(k) ) corresponding to the shift value (k) is g' _(k)it (i = 1, ..., N; t = 1, 2, ..., 360) It may be an N × 360 matrix consisting of

The shift value k may be a preset search interval, eg, an angle between a first angle (eg, -L ₁ ) and a second angle (eg, L ₂ ). Each of L1, L2 and k may be an integer greater than or equal to 0. For example, the shift value (k) may be a value of 1° interval between -10° and 10°. For example, when the shift value k is 1, the (1st) shifted WVV descriptor matrix G' ₍₁ ) circulates the target WVV descriptor matrix G' by 1 position in the row direction (horizontal direction). may have been moved. In this case, g' _it of the target WVV descriptor matrix (G') becomes g' (1) _{i(t+1) of the (1)} th moving WVV descriptor matrix (G' ₍₁₎ ), and the target WVV descriptor matrix g' _i360 of (G') becomes g' (1 _)i1 of the _{(1)th moving WVV descriptor matrix (G' (1)} ).

Circular shifting each row of the target WVV descriptor matrix (G') in the row direction, that is, from left to right, one cell at a time, is the remainder based on each of the N target scattering points in the target scattering point set (G). This is equivalent to rotating the target scattering points by 1°.

The shifted WVV descriptor matrix (G' ₍₁₎ ) generated by circularly shifting each row of the target WVV descriptor matrix (G') by one cell in the row direction is N target scattering points in the target scattering point set (G). All target scattering points are rotated by 1° based on the center coordinates of the points to generate a rotational target scattering point set (G _1° ), and a rotational target WVV descriptor matrix generated from the rotational target scattering point set (G _1° ) ( G' _1° ) is not equivalent. However, since the distance between the target scattering points is 1 or less, the difference in the target WVV descriptor due to the reference point difference of rotation transformation is not large. Therefore, the moving WVV descriptor matrix (G' _(k) ) and the rotational target WVV descriptor matrix (G' _k° ) may be regarded as equal to each other.

The number of shift values k may be L ₁ + L ₂ + 1. A plurality of moving WVV descriptor matrices (G' _(k) ) may be expressed as one 3-dimensional moving WVV descriptor matrix (G' _3D ). The 3D shifted WVV descriptor matrix (G' _3D ) is N consisting of the (-L ₁ ) th shifted WVV descriptor matrix (G' _(-L1) ) to the (L ₂ ) shifted WVV descriptor matrix (G' _(L2) ). It may be a × 360 × (L ₁ + L ₂ + 1) matrix.

The similarity calculation unit 122 determines the maximum shift value (k') at which the maximum similarity value (Smax(k)) between each of the plurality of moving WVV descriptor matrices (G' _(k) ) and the template WVV descriptor matrix (B') is maximized. ) can be calculated (S160).

According to an embodiment, the similarity calculation unit 122 determines between a moving WVV descriptor matrix G' _(k1), which is one of a plurality of moving WVV descriptor matrices G' _(k) , and a template WVV descriptor matrix B'. A similarity matrix S(G' _(k1) , B') may be calculated. The similarity matrix (S(G' _(k1) , B')) may be an N × M matrix consisting of s _(k1)ij (i = 1, ..., N; j = 1, ..., M) there is. At this time, s _(k1)ij can be calculated according to [Equation 6].

[Formula 6]

s _(k1)ij = (360-Σ _t │g' _(k1)it -b' _jt │ ² )/360

The similarity calculating unit 122 uses the similarity matrix S(G' _(k1) , B') to determine the maximum similarity between the moving WVV descriptor matrix G' _(k1 ) and the template WVV descriptor matrix B'. (Smax(k1)) can be calculated.

When the number (N) of target scattering points (g) in the target scattering point set (G) is greater than or equal to (M) the number (M) of template scattering points (b) in the template scattering point set (B), that is, when N≥M, The maximum similarity value Smak(k1) may be determined as the sum of maximum values of each column of the similarity matrix S(G' _(k1) , B'). That is, for example, the maximum value of column j is the maximum value of s _(k1)1j to s _(k1)Nj , and may be expressed as max({s _(k1)ij } _i ). Therefore, the maximum similarity value Smak(k1) can be expressed as [Equation 7].

[Formula 7]

Smax(k1)=Σ _j max({s _(k1)ij } _i )

When the number (N) of the target scattering points (g) in the target scattering point set (G) is less than the number (M) of the template scattering points (b) in the template scattering point set (B), that is, when N<M, The maximum similarity value Smak(k1) may be determined as the sum of the maximum values of each row of the similarity matrix S(G' _(k1) , B'). That is, for example, the maximum value of row i is the maximum value of s _(k1)i1 to s _(k1)iM , and may be expressed as max({s _(k1)ij } _j ). Therefore, the maximum similarity value (Smak(k1)) can be expressed as [Equation 8].

[Formula 8]

Smax(k1)=Σ _i max({s _(k1)ij } _j )

The similarity calculating unit 122 may calculate a maximum similarity value Smax(k) between each of the plurality of moving WVV descriptor matrices G′ _(k) and the template WVV descriptor matrix B′. That is, between each of the (-L ₁ )-th moving WVV descriptor matrix (G' _(-L1) ) to (L ₂ )-th moving WVV descriptor matrix (G' _(L2) ) and the template WVV descriptor matrix (B') ( -L ₁ ) Maximum similarity value (Smax(-L ₁ )) to (L ₂ ) maximum similarity value (Smax(L ₂ )) may be calculated. The similarity calculating unit 122 calculates a maximum shift having the largest maximum similarity value among the (−L ₁ )th maximum similarity value (Smax(−L ₁ )) to (L ₂ )th maximum similarity value (Smax(L ₂ )). A value (k') can be determined. The maximum shift value (k') can be expressed as [Equation 9].

[Formula 9]

k'= arg max Smax(k)

It indicates that the degree of similarity with the template information 132 (eg, the template image 132b) is highest when the target image 110a is rotated at an angle corresponding to the maximum shift value k'.

The similarity calculation unit 122 may determine the final similarity (S') based on the maximum similarity value (Smax(k')) corresponding to the maximum shift value (k') (S170). The final similarity (S') may be expressed as S'=Smax(k') using the maximum shift value (k') calculated in step S160. The target identification device 120, eg, the target identification unit 124, calculates the final similarity (S) between the target image 110a calculated by the similarity calculation unit 122 and the template information 132 (eg, the template image 132b). '), the target of the target image 110a may be identified.

According to another embodiment, in order to directly compare the final similarity (S') with various template information 132, the final similarity (S') is S' = Smax (k') / min (M, N) can be calculated by For example, when N≥M, the final similarity (S') is calculated by S'=Smax(k')/M, and when N<M, the final similarity (S') is S'=Smax It can be calculated by (k')/N.

7 is a flowchart illustrating a method for calculating similarity between targets according to an embodiment of the present invention.

Referring to FIG. 7 , the target identification device 120 receives a target image (eg, 110a of FIG. 2a) (S210). The target image 110a may be a part corresponding to a target of interest among SAR images.

The target identification device 120 receives a plurality of template information 132 from the database 130 (S220). Each of the plurality of template information 132 may include a template target type 132a and a template image 132b of the template target. Below, it is assumed that the plurality of template information 132 includes first template information 132-1 to T-th template information 132-T.

The target identification unit 124 may calculate a final similarity (S') between the target image 110a and the plurality of template information 132, respectively (S230). The target identification unit 124 may calculate a final similarity (S′) between the target image 110a and each template information 132 using the target similarity calculation method described above with reference to FIG. 5 . The t final similarity (S' _t ) between the target image 110a and the t th template information 132-t is the maximum similarity value corresponding to the maximum shift value (k' _t ) calculated in step S160 of FIG. Based on (Smax(k' _t )), (S' _t )=Smax(k' _t ) /min(M, N). Accordingly, even if the number (M) of template scattering points extracted from the template information 132 is different, they can be compared based on the final similarity value (S').

For example, the target identification unit 124 calculates a first final similarity (S′ ₁ ) between the target image 110a and the first template information 132-1, and calculates the target image 110a and the second template information. A second final similarity (S′ ₂ ) between (132-2) may be calculated. In this way, the target identification unit 124 may calculate the Tth final similarity (S′ _T ) between the target image 110a and the Tth template information 132-T.

The target identification unit 124 may identify the target of the target image 110a based on the template information having the highest final similarity among the first to Tth final similarities (S' ₁ to S' _T ) (S240). . For example, if the t'th final similarity (S't' ) is the highest among the first to Tth final similarities (S' ₁ to S' _{T )} , the target of the target image 110a is the t'th template information It can be identified as the same type as the template target corresponding to (132). For example, if the template target corresponding to the t' template information 132 is a 'K-1 tank', the target of the target image 110a can also be identified as a K-1 tank.

In order to show that the method for calculating the similarity of a target according to an embodiment of the present invention is robust to the effect of the rotational transformation of the target scattering point, the scattering extracted from the target image provided in MSTAR (Moving and Stationary Target Acquisition and Recognition) data point was used.

A template scatter point set (B) and a template scatter point set used for calculating similarity between the target scatter point set (G) extracted from the target image and the target WVV descriptor matrix (G') generated from the target scatter point set (G) The template WVV descriptor matrix (B') generated from (B) was generated through the following simulation experiment.

1) A template scattering point set (B) was created by removing some randomly selected target scattering points ( _gi ) (i=1, ..., N) in the target scattering point set (G). For an environment that does not correspond one-to-one (ie, the number of scattering points is not the same), only the minimum N / 2 and maximum N target scattering points are randomly extracted from the target scattering points (g _i ) to set the template scattering points ( B) was created. The template scattering point set (B) includes template scattering points (b _j ) (i=1, ..., M), where M is greater than or equal to N/2 and less than or equal to N.

2) The coordinates (x _j , y _j ) of the template scattering points (b _j ) were randomly moved. The coordinates (x _j , y _j ) of the template scattering points (b _j ) were shifted by values following a normal distribution N(0, σ ² ) having the same variance on the x-axis and y-axis, respectively.

3) The template scattering points (b _j ) were rotated at 1° intervals between [-10°, 10°] arbitrarily set based on the center coordinates.

4) The above process was repeated more than 100 times for each target.

The degree of error of the rotation conversion angle estimated by the similarity calculation method of the target of the present invention is the degree of random movement along the normal distribution N (0, σ ² ) of the scattering point coordinates (x _j , y _j ), and [-10° .

8 shows the degree of error of the rotation conversion angle estimated by the method for calculating the similarity of targets according to the degree of random movement of scattering point coordinates (x _j , y _j ) according to the random movement amount.

Referring to FIG. 8, according to the degree of random movement along the normal distribution N(0, σ ² ) of scattering point coordinates (x _j , y _j ), a circular movement is applied according to the similarity calculation method of the target according to the present invention. The result of obtaining the degree of error of the rotation conversion angle is shown. The rotation conversion angle estimated by the target similarity calculation method of the present invention means the maximum shift value k′ calculated in step S160 of FIG. 5 .

The standard deviation of the rotation transform estimation error increases according to the standard deviation (σ) used for the random movement of the scattering point coordinates (x _j , y _j ). When the standard deviation (σ) is 0.1m, the rotation conversion estimation error is about 1.5°, and when the standard deviation (σ) is 0.2m, the rotation conversion estimation error is about 2.0°.

9 shows the error degree of the rotation conversion angle estimated by the target similarity calculation method according to the rotation conversion angle between [-10°, 10°].

Referring to FIG. 9, according to the rotation conversion angle between [-10°, 10°], the result of calculating the degree of error of the rotation conversion angle estimated by applying circular movement according to the target similarity calculation method according to the present invention is shown. . The rotation conversion angle estimated by the target similarity calculation method of the present invention means the maximum shift value k′ calculated in step S160 of FIG. 5 .

There is almost no effect depending on the angle of rotation applied to the scattering point, and an error in estimation of rotation transformation of about 1.5° occurred at all angles of rotation transformation.

10 shows a result of comparing the final similarity estimated by the method of calculating the similarity of a target according to the rotation conversion angle between [-10°, 10°] with the similarity estimated by the conventional method.

Referring to FIG. 10 , a reference template scattering point set (B _ref ) was created by extracting only a random number of scattering points and randomly moving the coordinates of the extracted scattering points without applying rotation transformation. A target WVV descriptor matrix (G′) was generated from the target scatter point set (G), and a reference template WVV descriptor matrix (B′ _ref ) was generated from the reference template scatter point set (B _ref ). A reference similarity matrix (S ref (G', B')) between the target WVV descriptor matrix (G') and the reference template WVV descriptor matrix (B' _ref ) is generated, and the similarity _{matrix (S ref (} G', B' _ref )), a reference similarity value (S _ref ) was calculated.

According to the method of calculating the similarity of the target of the present invention, the shift value (k) is applied to generate the similarity matrix (S(G' _(k) , B')), and according to the method of calculating the similarity of the target of the present invention, the similarity matrix ( The final similarity (S') was calculated from S(G' _(k) , B')). The solid line in FIG. 10 represents the difference between the final similarity (S′) estimated by the target similarity calculation method of the present invention and the reference similarity value (S _ref ) according to the rotation conversion angle between [-10°, 10°]. . B' extracts only a random number of scattering points among the target scattering points g _i (i=1, ..., N) in the target scattering point set G, and randomly moves the coordinates of the extracted scattering points, It shows the template WVV descriptor matrix (B') generated from the template scattering point set (B) generated by applying rotation transformation according to the rotation transformation angle between [-10°, 10°].

The conventional method generates a similarity matrix (S(G', B')) between the target WVV descriptor matrix (G') and the template WVV descriptor matrix (B') without applying the shift value (k), and the target of the present invention The similarity (S) is calculated from the similarity matrix (S(G', B')) according to the similarity calculation method of The broken line in FIG. 10 represents the difference between the similarity (S) estimated by the conventional method and the reference similarity value (S _ref ) according to the rotation conversion angle between [-10°, 10°].

As can be seen through the broken line in FIG. 10 , the similarity (S) estimated by the conventional method gradually becomes smaller than the reference similarity value (S _ref ) as the rotation conversion angle moves away from 0. In addition, the standard deviation also increased as the absolute value of the rotation conversion angle became larger than zero.

On the other hand, as can be seen from the solid line in FIG. 10, the difference between the final similarity (S′) estimated by the target similarity calculation method of the present invention and the reference similarity value (S _ref ) is very small regardless of the rotation conversion angle. and the standard deviation was also low.

11 shows a result of comparing the final similarity estimated by the method of calculating the similarity of a target according to the random movement amount of the scattering point coordinates (x _j , y _j ) with the similarity estimated by the conventional method.

Referring to FIG. 11, the solid line indicates the final similarity (S′) estimated by the target similarity calculation method according to the random movement amount of the scattering point coordinates (x _j , y _j ) and the reference similarity value (S _ref ). indicate the difference. The broken line represents the difference between the similarity (S) estimated by the conventional method and the reference similarity value (S _ref ) according to the amount of random movement of the scattering point coordinates (x _j , y _j ).

As can be seen through the solid line, the final similarity (S′) estimated by the method of calculating the similarity of targets of the present invention is the same as the reference similarity value (S _ref ), regardless of the amount of random movement of the scattering point coordinates (x _j , y _j ). There was little difference. On the other hand, as can be seen through the broken line, the similarity (S) estimated by the conventional method was significantly lower than the reference similarity value (S _ref ) as a whole.

Compared to the similarity (S) estimated by the conventional method, the final similarity (S') estimated by the method of calculating the similarity of targets of the present invention is hardly affected by the rotational transformation of the scattering points.

On the other hand, those skilled in the art related to the present embodiment will be able to understand that it can be implemented in a modified form within a range that does not deviate from the essential characteristics of the present invention. Therefore, the disclosed methods are to be considered in an illustrative rather than a limiting sense. The scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the equivalent scope will be construed as being included in the present invention.

The devices described above may be implemented as hardware components, software components, and/or a combination of hardware components and software components. For example, devices and components described in the embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA) , a programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. The processing device may run an operating system (OS) and one or more software applications running on the operating system. A processing device may also access, store, manipulate, process, and generate data in response to execution of software. For convenience of understanding, there are cases in which one processing device is used, but those skilled in the art will understand that the processing device includes a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that it can include. For example, a processing device may include a plurality of processors or a processor and a controller. Other processing configurations are also possible, such as parallel processors.

Software may include a computer program, code, instructions, or a combination of one or more of the foregoing, which configures a processing device to operate as desired or processes independently or collectively. You can command the device. Software and/or data may be any tangible machine, component, physical device, virtual equipment, computer storage medium or device, intended to be interpreted by or provide instructions or data to a processing device. , or may be permanently or temporarily embodied in a transmitted signal wave. Software may be distributed on networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer readable media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer readable medium. Computer readable media may include program instructions, data files, data structures, etc. alone or in combination. Program commands recorded on the medium may be specially designed and configured for the embodiment or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. - includes hardware devices specially configured to store and execute program instructions, such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include high-level language codes that can be executed by a computer using an interpreter, as well as machine language codes such as those produced by a compiler. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

As described above, although the embodiments have been described with limited examples and drawings, those skilled in the art can make various modifications and variations from the above description. For example, the described techniques may be performed in an order different from the method described, and/or components of the described system, structure, device, circuit, etc. may be combined or combined in a different form than the method described, or other components may be used. Or even if it is replaced or substituted by equivalents, appropriate results can be achieved.

Therefore, other implementations, other embodiments, and equivalents of the claims are within the scope of the following claims.

Claims (12)

receiving a target image;
extracting target scattering points (g) from the target image and generating a target scattering point set (G) based on the positions of the target scattering points (g);
generating a target WVV descriptor matrix (G′) from the target scattering point set (G) using a World View Vector (WVV);
Obtaining a template WVV descriptor matrix (B') from template information;
Generating a plurality of shifted WVV descriptor matrices G' (k) by circularly moving the target WVV descriptor matrix G' in a row direction by a plurality of shift values _k within a preset range, respectively; and
Calculating a maximum shift value ( _{k') at which a maximum similarity value (Smax(k)) between each of the plurality of shifted WVV descriptor matrices (G' (k} )) and the template WVV descriptor matrix (B') is maximized; A method for calculating the similarity of a target, comprising determining the final similarity (S') based on the maximum similarity value (Smax(k')) at this time. According to claim 1,
The number of target scattering points g in the target scattering point set G is N,
The target WVV descriptor matrix (G') is an N × 360 matrix,
The values of the ith row of the target WVV descriptor matrix (G′) are the position of the ith target scattering point ( _gi ) as the origin and the remaining target scattering points (g ₁ to g _i-1 , g _i+1 to g A method for calculating the similarity of a target, characterized in that it is determined based on the distance and angle to _N ). According to claim 1,
Obtaining the template WVV descriptor matrix (B'),
Receiving a template image from a database;
extracting template scattering points (b) from the template image and generating a template scattering point set (B) based on the locations of the template scattering points (b); and
and generating a template WVV descriptor matrix (B′) from the template scattering point set (B) using the WVV. According to claim 1,
The similarity matrix (S(G' _{(k1 ) ,} B')), further comprising the step of calculating
The moving WVV descriptor matrix (G' _(k1) ) is an N × 360 matrix consisting of g' _(k1)it (i = 1, ..., N; t = 1, 2, ..., 360),
The template WVV descriptor matrix (B') is an M × 360 matrix consisting of b' _jt (j = 1, ..., M; t = 1, 2, ..., 360),
The similarity matrix (S(G' _(k1) , B')) is an N × M matrix consisting of s _(k1)ij (i = 1, ..., N; j = 1, ..., M) ,
s _(k1)ij is calculated as s _(k1)ij = (360-Σ _t │g' _(k1)it -b' _jt │ ² )/360. According to claim 4,
The maximum similarity value (Smax _(k1 )) between the moving WVV descriptor matrix (G' _{(k1)) and the template WVV descriptor matrix (B') using the similarity matrix (S(G' (k1),} B')) ), further comprising the step of calculating
When N≥M, the maximum similarity value (Smak(k1)) is calculated by Smax(k1)=Σ _j max({s _(k1)ij } _i ),
When N<M, the maximum similarity value (Smak(k1)) is calculated by Smax(k1)=Σ _i max({s _(k1)ij } _j ). According to claim 1,
The maximum shift value (k') is calculated by k' = arg max Smax (k),
The final similarity (S') is calculated by S' = Smax (k'). receiving a target image;
extracting target scattering points (g) from the target image and generating a target scattering point set (G) based on the positions of the target scattering points (g);
generating a target WVV descriptor matrix (G′) from the target scattering point set (G) using a World View Vector (WVV);
Generating a plurality of shifted WVV descriptor matrices G' (k) by circularly moving the target WVV descriptor matrix G' in a row direction by a plurality of shift values _k within a preset range, respectively;
obtaining a template WVV descriptor matrix (B') from each of a plurality of template information stored in a database;
Calculating a maximum shift value ( _{k') at which a maximum similarity value (Smax(k)) between each of the plurality of shifted WVV descriptor matrices (G' (k} )) and the template WVV descriptor matrix (B') is maximized; determining the final similarity (S') based on the maximum similarity value (Smax(k')) at this time; and
Identifying a target of the target image based on a template WVV descriptor matrix having the highest final similarity (S′) to the target image among a plurality of template WVV descriptor matrices obtained from the plurality of template information, respectively. method. According to claim 7,
Each of the plurality of template information includes a type of template target and a template image of the template target. According to claim 7,
The similarity matrix (S(G' _{(k1 ) ,} B')), further comprising the step of calculating
The moving WVV descriptor matrix (G' _(k1) ) is an N × 360 matrix consisting of g' _(k1)it (i = 1, ..., N; t = 1, 2, ..., 360),
The template WVV descriptor matrix (B') is an M × 360 matrix consisting of b' _jt (j = 1, ..., M; t = 1, 2, ..., 360),
The similarity matrix (S(G' _(k1) , B')) is an N × M matrix consisting of s _(k1)ij (i = 1, ..., N; j = 1, ..., M) ,
s _(k1)ij is calculated as s _(k1)ij = (360-Σ _t │g' _(k1)it -b' _jt │ ² )/360. According to claim 9,
The maximum similarity value (Smax _(k1 )) between the moving WVV descriptor matrix (G' _{(k1)) and the template WVV descriptor matrix (B') using the similarity matrix (S(G' (k1),} B')) ) Further comprising the step of calculating
When N≥M, the maximum similarity value (Smak(k1)) is calculated by Smax(k1)=Σ _j max({s _(k1)ij } _i ),
When N<M, the maximum similarity value Smak(k1) is calculated by Smax(k1)=Σ _i max({s _(k1)ij } _j ). According to claim 9,
When N≥M, the final similarity (S') is calculated by S'=Smax(k')/M,
When N<M, the final similarity (S') is calculated by S'=Smax(k')/N. A computer program stored in a medium to execute the method of any one of claims 1 to 11 using a computing device.

Images