A fast implementation method of CA-CFAR for sonar images

Abstract

The present invention belongs to the field of underwater acoustics, and specifically relates to sonar image processing. It is a fast implementation method of CA‑CFAR of sonar images, and there are six steps to calculate the CA‑CFAR result: first, the data matrix is ​​expanded in the beam dimension to solve the background estimation problem of the boundary beam; then, according to the number of beams of the background window, the expanded data is beam added, and the two-dimensional background window becomes a one-dimensional background window; then, the data is expanded in the distance dimension to solve the background estimation problem of the boundary distance point; then, the background estimation value of the first distance point of each beam is calculated; then, the background estimation value of the remaining distance points is calculated based on the background estimation value of the previous distance point; finally, the input matrix data and the background estimation value are used to calculate CA‑CFAR. The present invention can significantly reduce the running time of CA‑CFAR, improve the operation efficiency, and meet the real-time processing requirements.

Description

Technical Field

[0001] The invention relates to the field of underwater acoustics, in particular to the field of sonar image processing, and mainly provides a method for quickly realizing CA-CFAR of sonar images. Background Art

[0002] In shallow sea environments, the ocean channel is not only a multi-path channel that is randomly time-varying and space-varying, but also has multiple inhomogeneities in the seawater medium and boundaries. At the same time, the scattered sound waves that reach the receiving point are superimposed to form reverberation. How to suppress reverberation and enhance the image display effect of the echo signal is an important means to improve the detection performance of active sonar. Appropriate normalization processing methods are adopted before detection to suppress reverberation and interference, compress the dynamic range of the display data, make the background image of the sonar display more uniform, and reduce the false alarm rate, thereby effectively detecting the target. Among the many normalization methods, the two-dimensional CA-CFAR (Cell Average Constant False Alarm Rate) has a simple principle, is easy to implement, and has good results, so it has been widely used. The output value of a certain point of CA-CFAR is replaced by the ratio of its input value to the background estimate value of the point. The background estimate value of the point is the average value of the input data near it. The calculation process of CA-CFAR is shown in the attached figure. Figure 1 The CA-CFAR detection algorithm has the advantages of simple implementation, fast and efficient, strong robustness, etc. Its detection window consists of a test unit 2, a protection window 3, and a background window 1. The protection window can prevent the target power from leaking into the background window. The background window is used to count the mean of the clutter background. The background window and protection window of CA-CFAR are shown in the attached figure. Figure 2 shown. Summary of the invention

[0003] In the CA-CFAR calculation process, when the number of data points is large or the background sliding window is large, the amount of calculation increases sharply and cannot meet the requirements of real-time processing. In order to overcome this problem, the present invention provides a fast implementation method of sonar image CA-CFAR.

[0004] The object of the present invention is achieved by the following technical solution: A fast implementation method of sonar image CA-CFAR, which comprises the following steps:

[0005] Step 1: The two-dimensional sonar data is stored in the matrix A (Nb, Ds). The first and last beams of the matrix A are expanded by (LW-1) / 2 beams. The expanded two-dimensional data is defined as the matrix B (Nb+LW-1, Ds), where Nb is the number of beams, Ds is the number of range points, and LW is the number of beams in the CA-CFAR two-dimensional background window.

[0006] Step 2: Sum the matrix B(Nb+LW-1,Ds) in the beam direction according to LW beams to convert it into the matrix C(Nb,Ds);

[0007] Step 3: Expand the distance data at the head and tail of the distance dimension of the matrix C(Nb,Ds). The number of distance points expanded at the head and tail is (LR-1) / 2. The expanded data is D(Nb,DS+LR-1), where LR is the number of distance points of the background window.

[0008] Step 4: Calculate the sum of the background cells corresponding to the first point of each beam according to the matrix D, where i represents the beam number;

[0009] Step 5: Starting from the second distance point, i.e. j=2, the sum value E(i,j) of the background unit at this point is solved based on the sum value E(i,j-1) of the background unit at the j-1th distance point;

[0010] Step 6: Calculate the result value of CA-CFAR based on matrix A, matrix E and the size of the background window.

[0011] The beneficial effects of the present invention are: compared with the traditional method, the running time of CA-CFAR can be significantly reduced, the operation efficiency can be improved, and the real-time processing requirements can be met. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is the CA-CFAR process flow chart;

[0013] Figure 2 It is a schematic diagram of the CA-CFAR background window and protection window;

[0014] Figure 3 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0015] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments:

[0016] The present invention provides a fast implementation method of sonar image CA-CFAR, which can significantly reduce the operation time compared with the traditional method. The calculation flow chart of fast CA-CFAR is shown in the attached figure. Figure 3As shown, there are six steps to calculate the CA-CFAR result: first, expand the data matrix in the beam dimension to solve the background estimation problem of the boundary beam; then, according to the number of beams in the background window, beam addition is performed on the expanded data, and the two-dimensional background window becomes a one-dimensional background window; then, the data is expanded in the distance dimension to solve the background estimation problem of the boundary distance point; then, the background estimation value of the first distance point of each beam is calculated; then, the background estimation value of the remaining distance points is calculated based on the background estimation value of the previous distance point; finally, CA-CFAR is calculated using the input matrix data and the background estimation value; specifically, the following steps are included:

[0017] Step 1: One frame of two-dimensional sonar data is stored in the matrix A (Nb, Ds), and the beam dimension is expanded (the first and last beams of matrix A are expanded by (LW-1) / 2 beams respectively). The expanded two-dimensional data is defined as B (Nb+LW-1, Ds). Nb is the number of beams, Ds is the number of range points, and LW is the number of beams in the CA-CFAR two-dimensional background window.

[0018] The sonar image data is a matrix A(Nb,Ds), with a total of Nb beams and Ds range points. Assume that the size of the CA-CFAR window is LW×LR, where LW is the number of beams in the window and LR is the number of range points in the window. The first and last beams of matrix A are expanded by (LW-1) / 2 beams, becoming B(Nb+LW-1,Ds). For a linear array, the expanded beam can be replaced by the average value of adjacent beams. For a cylindrical array or annular array, the expanded beam of the first beam is filled with the tail data; the expanded beam of the tail beam is filled with the first data.

[0019] Step 2: Sum the matrix B(Nb+LW-1,Ds) in the beam direction to convert it into matrix C(Nb,Ds).

[0020] The value of the 1# beam of matrix C is obtained by adding the corresponding distance points of the 1# beam to the LW# beam in matrix B; the value of the 2# beam of matrix C is obtained by adding the corresponding distance points of the 2# beam to the LW+1# beam in matrix B; and so on, the values ​​of the 3# to Nb# beams of matrix C are calculated. Beam addition needs to be performed Nb times in total. After beam addition, the two-dimensional CA-CFAR based on matrix B becomes the one-dimensional CA-CFAR based on matrix C.

[0021] Step 3: Perform distance data expansion on the head and tail of the distance dimension of the matrix C(Nb,Ds). The number of distance points expanded at the head and tail is (LR-1) / 2, and the expanded data is D(Nb,DS+LR-1).

[0022] The expansion rule is as follows: D(1~Nb,1~(LR-1) / 2)=mean(mean(C(1:Nb,1:(LR-1) / 2))), that is, the average value of the data at points 1~(LR-1) / 2 of beams 1~Nb of C is used to fill points 1~(LR-1) / 2 of beams 1~Nb of matrix D.

[0023] D(1:Nb,(LR-1) / 2+1:(LR-1) / 2+Ds)=C(1:Nb,1:Ds); that is, the data of the 1~Ds points of the 1~Nb beams of the matrix C are used to fill the data of the (LR-1) / 2+1~(LR-1) / 2+Ds distance points of the 1~Nb beams of the matrix D. D(1:Nb,(LR-1) / 2+Ds+1:LR+Ds)=mean(mean(C(1:Nb,Ds-(LR-1) / 2+1:Ds))), that is, the average value of the data of the Ds-(LR-1) / 2~Ds points of the 1~Nb beams of the matrix C is used to fill the data of the (LR-1) / 2+Ds+1~LR+Ds distance points of the 1~Nb beams of the matrix D.

[0024] Step 4: Calculate the sum of the background units corresponding to the first point of each beam. For each beam i, solve

[0025]

[0026] Where E(i,1) is the sum of the background units of the first point of the i-th beam; m is the distance point index value; LG is the number of distance points in the protection window.

[0027] Step 5: Starting from the second distance point, i.e. j=2, solve the background unit sum E(i,j) of this point based on the background unit sum E(i,j-1) of the j-1th distance point. For the second distance point, solve the background unit sum:

[0028] E(i,2)=E(i,1)+D(i,(LR-LG) / 2+1)+D(i,LR+1)-D(i,1)-D(i,(LR+LG) / 2+1)(7)

[0029] By analogy, the sum of the background estimates for the remaining distance points is:

[0030] E(i,j)=E(i,j-1)+D(i,(LR-LG) / 2+j-1)+D(i,LR+j-1)-D(i,j-1)-D(i,(LR+LG) / 2+j-1) (8)

[0031] Where i represents the beam and j represents the distance point;

[0032] Step 6: Calculate the CA-CFAR value.

[0033] According to the matrix A, matrix E and the size of the background window, the result of CA-CFAR is:

[0034] F(i,j)=A(i,j) / (E(i,j) / ((LR-LG)*LW)) (9).

[0035] Comparative experiment between the present invention and traditional technology: On the basis of not changing the existing sonar processing, the running time of CA-CFAR can be effectively reduced and the operation efficiency can be improved. Since the amount of calculation in CA-CFAR is mainly addition, the number of additions of CA-CFAR and the fast CA-CFAR of the present invention are compared below. In traditional CA-CFAR, the number of additions required for each data point is (LR-LG)*LW. For a sonar image with Nb beams and Ds distance points, the total number of additions is:

[0036] N0=(LR-LG)*LW*Nb*Ds (1)

[0037] In the fast CA-CFAR of the present invention, the number of additions in step 2 is Nb*LW*Ds, the number of additions in step 4 is Nb*(LR-LG), and the number of additions in step 5 is Nb*(Ds-1)*4, so the total number of additions is

[0038] N1=Nb*LW*Ds+Nb*(LR-LG)+Nb*(Ds-1)*4 (2)

[0039] The ratio of the two methods is:

[0040] N0 / N1=(Nb*(LR-LG)*LW*Ds) / (Nb*LW*Ds+Nb*(LR-LG)+Nb*(Ds-1)*4) (3)

[0041] From formula (3), we can deduce:

[0042] N0 / N1=((LR-LG)*LW) / (LW+4) (4)

[0043] Since LW is generally greater than 4, equation (4) can be simplified to:

[0044] N0 / N1≥((LR-LG)*LW) / (2*LW)=(LR-LG) / 2 (5)

[0045] It can be seen from formula (5) that the running time of fast CA-CFAR is 2 / (LR-LG) of traditional CA-CFAR. Assuming LR=101 and LG=21, the running time of this method can be shortened to 1 / 40 of the traditional method. Therefore, the fast CA-CFAR of the present invention can greatly shorten the running time and improve the computing efficiency.

[0046] It is understandable that, for those skilled in the art, any equivalent replacement or change to the technical solution and inventive concept of the present invention should fall within the protection scope of the claims attached to the present invention.

Claims

1. A fast implementation method of sonar image CA-CFAR, characterized by: It contains the following steps: Step 1: The two-dimensional sonar data is stored in the matrix A (Nb, Ds). The first and last beams of the matrix A are expanded by (LW-1) / 2 beams. The expanded two-dimensional data is defined as the matrix B (Nb+LW-1, Ds), where Nb is the number of beams, Ds is the number of range points, and LW is the number of beams in the CA-CFAR two-dimensional background window. Step 2: Sum the matrix B(Nb+LW-1,Ds) in the beam direction according to LW beams to convert it into the matrix C(Nb,Ds); Step 3: Expand the distance data at the head and tail of the distance dimension of the matrix C(Nb,Ds). The number of distance points expanded at the head and tail is (LR-1) / 2. The expanded data is D(Nb,DS+LR-1), where LR is the number of distance points of the background window. Step 4: Calculate the sum of the background cells corresponding to the first point of each beam according to the matrix D, where i represents the beam number; Step 5: Starting from the second distance point, i.e. j=2, the sum value E(i,j) of the background unit at this point is solved based on the sum value E(i,j-1) of the background unit at the j-1th distance point; Step 6: Calculate the result value of CA-CFAR according to the matrix A, matrix E and the size of the background window; In step 1, for a linear array, the extended beam is replaced by the average value of adjacent beams. For a cylindrical array or annular array, the extended beam of the head end beam is filled with the tail end data; the extended beam of the tail end beam is filled with the head end data. In step 2, the value of beam 1# of matrix C is obtained by adding the corresponding distance points of beam 1# to beam LW# in matrix B; the value of beam 2# of matrix C is obtained by adding the corresponding distance points of beam 2# to beam LW+1# in matrix B; and so on, the values ​​of beams 3# to Nb# of matrix C are calculated, and beam addition needs to be performed Nb times in total. After beam addition, the two-dimensional CA-CFAR based on matrix B becomes the one-dimensional CA-CFAR based on matrix C; In step 3, the expansion rule is as follows: D(1~Nb,1~(LR-1) / 2)=mean(mean(C(1:Nb,1:(LR-1) / 2))), that is, the average value of the data of the 1~(LR-1) / 2 points of the 1~Nb beams of C is used to fill the 1~(LR-1) / 2 points of the 1~Nb beams of the matrix D; D(1:Nb,(LR-1) / 2+1:(LR-1) / 2+Ds)=C(1:Nb,1:Ds), i.e., the data of the 1~Ds points of the 1~Nb beams of the matrix C are used to fill the data of the (LR-1) / 2+1~(LR-1) / 2+Ds distance points of the 1~Nb beams of the matrix D; D(1:Nb,(LR-1) / 2+Ds+1:LR+Ds)=mean(mean(C(1:Nb,Ds-(LR-1) / 2+1:Ds))), i.e., the average value of the data of the Ds-(LR-1) / 2~Ds points of the 1~Nb beams of the matrix C is used to fill the data of the (LR-1) / 2+Ds+1~LR+Ds distance points of the 1~Nb beams of the matrix D; In step 4, the sum of the background units corresponding to the first point of each beam is calculated. For each beam i, solve Where E(i,1) is the sum of the background units of the first point of the i-th beam; m is the distance point index value; LG is the number of distance points in the protection window; In step 5, starting from the second distance point, i.e. j=2, the sum of the background units at this point E(i,j) is solved based on the sum of the background units at the j-1th distance point E(i,j-1); for the second distance point, the sum of the background units is solved: E(i,2)=E(i,1)+D(i,(LR-LG) / 2+1)+D(i,LR+1)-D(i,1)-D(i,(LR+LG) / 2+1) (7) By analogy, the sum of the background estimates for the remaining distance points is: E(i,j)=E(i,j-1)+D(i,(LR-LG) / 2+j-1)+D(i,LR+j-1)-D(i,j-1)-D(i,(LR+LG) / 2+j-1) (8) Where i represents the beam and j represents the distance point; In step 6, the CA-CFAR value is calculated. According to the matrix A, matrix E and the size of the background window, the result value of CA-CFAR is: F(i,j)=A(i,j) / (E(i,j) / ((LR-LG)*LW)) (9).

Images