Shape optimization method of complex ship target model based on SAR image

Abstract

The present invention discloses a method for optimizing the shape of a complex ship target model based on SAR images. First, the target initial model is modeled using a non-uniform rational B-spline modeling technique to obtain the coordinates of the model's control points. Then, a simulated image based on the bouncing ray method is generated for the target model. The simulated image and the measured image are compared to obtain the control points that need to be optimized, and the control points are set as optimization parameters. The optimization parameters are updated using the hummingbird optimization algorithm to generate a simulated image based on the bouncing ray method. The similarity between the simulated image and the target image is calculated as the fitness value of each individual in the hummingbird optimization algorithm, and whether the optimal fitness value meets the index is determined. If the index is met or the optimal fitness value within the population remains unchanged after several iterations, the corresponding model and control point information are output. If not, optimization is continued. Finally, a target model corresponding to the simulated image that is similar to the measured image is obtained. The present invention can be used to optimize the shape of a target model.

Description

Technical Field

[0001] The invention belongs to the technical field of signal processing, and in particular to a method for optimizing the shape of a complex ship target model based on SAR images. Background Art

[0002] Synthetic Aperture Radar (SAR), as an imaging radar, uses a small-aperture antenna in relative motion with the target to receive echoes, thereby acquiring high-resolution radar images. It can penetrate clouds and possesses global day and night sensing capabilities. Due to these outstanding advantages, SAR systems are now widely used in fields such as military reconnaissance and surface structure monitoring. Recently, methods such as deep learning have made significant progress in image recognition, providing a broad approach to interpreting SAR images. Deep learning addresses the shortcomings of traditional image recognition methods, which are unable to accurately extract features from general data. However, the performance of models built using deep learning methods is highly dependent on the amount of data. Due to time and space constraints, it is impossible to obtain massive amounts of SAR image data in the real world. Research on SAR image target recognition technology requires target sample data. Due to the unique nature of SAR image acquisition, a complete database of measured image targets is not only extremely expensive, but also often difficult to establish for non-cooperative targets.

[0003] Generating simulated SAR images is an important means to supplement the target recognition database. By accurately modeling the target geometry and then performing simulated imaging through high-frequency electromagnetic algorithms, simulated SAR images of various targets under different conditions can be easily and quickly obtained. This can cheaply and effectively establish the database required for target recognition, which has great practical significance.

[0004] There are two methods to improve image similarity: optimizing the image (Zhao Zhichun. Image processing and application based on generative adversarial network. Electronic Technology and Software Engineering 1(2021):2) and optimizing the model shape. Since the optimization of the image is not easy to explain in terms of electromagnetic scattering mechanism, the optimization of the model shape is chosen to expand the image library. The existing model shape optimization strategies include: optimizing the car shape based on genetic algorithm to improve the aerodynamic performance of the car (Chen Xiaoxiong. Car shape optimization based on genetic algorithm. Chongqing University of Technology); optimizing the aircraft shape based on Kriging proxy model to improve aerodynamic performance (Wu Gongming. Aerodynamic shape optimization of hypersonic aircraft based on Kriging proxy model. Nanjing University of Aeronautics and Astronautics); optimizing the glider shape based on proxy model to improve fluid dynamic performance and displacement volume (Ye Pengcheng, Wang Congcong, Pan Guang. Optimization design of wing-body blended underwater glider shape based on proxy model. Journal of Northwestern Polytechnical University 39.1(2021):8), etc. One method uses the Hummingbird optimization algorithm to optimize aircraft shape to reduce electromagnetic scattering (Zhao Yun. Research on Stealth Optimization of Target Coating Materials / Geometric Shapes. Nanjing University of Science and Technology, 2020. DOI: 10.27241). This method uses NURBS surface modeling technology to create a target model and then optimizes the target model shape using the Hummingbird optimization algorithm. However, this method uses radar cross section (RCS) as the optimization metric, and the resulting simulated images generated by the model cannot be used to expand the target recognition database based on measured images. Summary of the Invention

[0005] The purpose of the present invention is to provide a model shape optimization method based on SAR images.

[0006] The technical solution to achieve the purpose of the present invention is as follows: In a first aspect, the present invention provides a model shape optimization method based on SAR images, comprising the following steps:

[0007] Step 1: Model the initial model using non-uniform rational B-spline modeling technology to obtain the coordinates of the control points of the model;

[0008] Step 2: Generate a simulation image based on the bouncing ray method for the target model;

[0009] Step 3: Compare the simulation graph and the measured graph to obtain the control points that need to be optimized and set them as optimization parameters;

[0010] Step 4: Use the Hummingbird optimization algorithm to update the optimization parameters, thereby updating the shape of the target model after the change;

[0011] Step 5: Generate a simulated image based on the bouncing ray method and calculate its similarity with the target image;

[0012] Step 6: Use the similarity calculated in Step 5 as the fitness value of each individual in the Hummingbird optimization algorithm to determine whether the optimal fitness value meets the target. If the target is met or the optimal fitness value within the population remains unchanged after several iterations, output the corresponding model and control point information; otherwise, repeat Step 4.

[0013] In a second aspect, the present application further provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first aspect when executing the program.

[0014] In a third aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect above.

[0015] In a fourth aspect, the present application further provides a computer program product, comprising a computer program, which implements the method described in the first aspect when executed by a processor.

[0016] Compared with the existing technology, the present invention has the following significant advantages: (1) the similarity of images is evaluated by complex wavelet structure similarity, and it is insensitive to the geometric distortion (translation, scaling and rotation) of the images; (2) the Hummingbird optimization algorithm has strong adaptability, a wide search range and a fast convergence speed for situations with multiple optimization targets and high complexity, which can balance the development and exploration capabilities of the algorithm; (3) the target is remodeled using NURBS modeling technology, and the shape of the target can be indirectly changed by changing the coordinates of the control points. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of the model shape optimization method based on SAR images of the present invention.

[0018] Figure 2 It is a schematic diagram of a simple ship model reconstructed using the non-uniform rational B-spline modeling technology in the present invention.

[0019] Figure 3 It is a schematic diagram of the slice model to be optimized in the present invention.

[0020] Figure 4 It is a schematic diagram of the slice model obtained after optimization in the present invention. DETAILED DESCRIPTION

[0021] The present invention is further described in detail below with reference to the accompanying drawings.

[0022] Combine Figure 1 A model optimization method for improving the similarity between a simulation image and a measured image of the present invention comprises the following steps:

[0023] Step 1: Model the initial model using non-uniform rational B-spline modeling technology to obtain the coordinates of the control points of the model:

[0024] Step 2: Generate a simulation image based on the bouncing ray method for the target model;

[0025] Step 3: Compare the simulation graph and the measured graph to obtain the control points that need to be optimized and set them as optimization parameters;

[0026] Step 4: Use the Hummingbird optimization algorithm to update the optimization parameters, thereby updating the shape of the target model after the change;

[0027] Step 5: Generate a simulated image based on the bouncing ray method and calculate its similarity with the target image;

[0028] Step 6: Use the similarity calculated in Step 5 as the fitness value of each individual in the Hummingbird optimization algorithm to determine whether the optimal fitness value meets the target. If the target is met or the optimal fitness value within the population remains unchanged after several iterations, output the corresponding model and control point information; otherwise, repeat Step 4.

[0029] Combine Figure 2 As shown in the ship model diagram, the geometric model of the target is established as described in step 1, as follows:

[0030] The construction of NURBS surface is inseparable from B-spline basis function. j，n (u) represents the i-th n-th order B-spline basis function about the knot vector U:

[0031]

[0032]

[0033] The shape of a NURBS surface can be expressed by controlling several control points according to a bivariate piecewise rational function. Figure 2 Given the position of the control point of the ship model to be optimized, the coordinates of any point on the NURBS surface can be expressed as:

[0034]

[0035] Among them, U and V represent the number of control points in the u and v parameter directions respectively. ij represents the control point P ij The corresponding weight factor. P ij =[P ijx , P ijy , P ijz ] represents the control point P ij Coordinates in the x, y, and z directions. N i，p(u) is the node vector U = [u1, u2, ...u n+k+1 ] According to the Cox-DeBoor recursive formula, the p-order canonical B-spline basis function is obtained. Similarly, N j，q (v) is the node vector V = [v1, v2, ...v n+k+1 ]The q-order canonical B-spline basis function is obtained according to the Cox-DeBoor recursive formula.

[0036] The Humminabirds Optimization Algorithm (HOA) described in step 4 is as follows:

[0037] Assume that there are N hummingbirds in the population, and each hummingbird corresponds to a solution. Initialize the population:

[0038] P i =ub-rand·(ub-lb)

[0039] Among them, P i is the position of the i-th hummingbird in the population, i∈{1, 2, ...N}, corresponding to the feasible solution of the optimization problem, ub and lb represent the upper and lower limits of the variables, and rand represents a random number, rand∈[0,1].

[0040] In the self-search phase, if hummingbird i finds a better position This means that the current search area can continue to search. The new location of the Hummingbird i is as follows:

[0041]

[0042] are the positions of the i-th hummingbird at the t-th and t+1-th iterations respectively. If the fitness value of is better, it means that the algorithm has found a better solution, then the position is updated; otherwise, the position before this iteration remains unchanged.

[0043] If hummingbird i searches for several times in a row and fails to find a better food source This means that based on experience, there is no better location in the current area, so the search direction is randomly changed:

[0044]

[0045]

[0046] Where α=0.01,β=1.5, represents the optimal solution of the tth iteration, and both μ and ν obey the normal distribution.

[0047] In the guided search phase, the individual that occupies the best position is called the leader bird, and the other individuals are called followers. The leader bird flies continuously around the territory:

[0048] P T，t+1 =P T,t +r d ·λ

[0049] Among them, P T，t is the position of the leader bird at the tth iteration, r d is a random number between -1 and 1, λ=0.1(ub-lb), if P T ，t+1 If the position represented by P is better, T，t+1 Replace P T，t .

[0050] Different from the leader bird, the following bird's movement can be divided into two states. If the leader bird does not find the follower bird j, the follower bird j approaches the leader bird area:

[0051]

[0052] MF=round[1+rand(0,1){2-1}]

[0053] in is the position of the jth following bird after the tth iteration. MF is a mutation factor that determines whether to change the position of the next bird. The round function represents rounding, and rand represents a random number.

[0054] If follower bird j is discovered by the leader bird, it needs to select one of the other follower birds, labeled k. If the position of follower bird k is better than that of follower bird j, follower bird j will fly towards follower bird k, otherwise follower bird j will fly away from follower bird k:

[0055]

[0056] Where j, k∈{1, 2, ...N-1}. If Better than Use replace

[0057] The image similarity evaluation index described in step 5 is as follows:

[0058] Structural Similarity (SSIM) defines structural information as an attribute that reflects the target structure in the scene independent of brightness and contrast from the perspective of image composition, and characterizes distortion as a combination of three different factors: brightness, contrast, and structure.

[0059] The SSIM index is as follows:

[0060] SSIM(x, y) = l(x, y) α1 c(x, y) β1 s(x, y) γ1

[0061] Let α1=β1=γ1=1, C3=C2 / 2, then SSIM simplifies to the following formula

[0062]

[0063] where μ x and μ y are the average pixel intensity of images x and y, respectively, x and σ y are the standard deviations of images x and y, σ xy Represents the covariance of images x and y. Let C1=(k1*L) 2 , C2=(k2*L) 2 , usually k1 = 0.01, k2 = 0.03. Where L represents the value range of the image pixel value, and in this embodiment, the value is 255.

[0064] Complex Wavelet SSIM (CW-SSIM) is an extension of SSIM in the complex wavelet domain. The basic idea of ​​CW-SSIM is that small geometric distortions of an image cause consistent phase changes in local wavelet coefficients, and consistent phase changes in wavelet coefficients do not change the structural content of the image.

[0065] The mother wavelet signal is a low-pass filter where g(u) represents a slowly changing symmetric function, ω c Represents the center frequency of the modulated bandpass filter. The mother wavelet generates a wavelet cluster through various scaling and translation:

[0066]

[0067] Where ω(·) represents the wavelet transform form of the signal, s∈R + is the scale factor and p∈R is the translation factor.

[0068] The continuous wavelet transform of the real signal x(u) is:

[0069]

[0070] Where X(ω) and G(ω) represent the Fourier transform results of x(u) and g(u) respectively. Discrete wavelet coefficients are the sampling form of continuous wavelet transform.

[0071] In the complex wavelet transform domain, the two sets of coefficients extracted from the same spatial position in the same wavelet subband of the two compared images are c x ={c x,i |i=1,2,…N} and c y ={c y,i |i=1,2,…N}. CW-SSIM is defined as:

[0072]

[0073] Here c * represents the complex conjugate of c, and K is a small positive constant. Its main purpose is to improve the robustness of the CW-SSIM measure when the local signal-to-noise ratio is low. In order to use CW-SSIM to compare the similarity between two images, the images are first decomposed using a multi-scale, multi-directional Steerable Pyramid (SP). Next, a sliding window is used to traverse each wavelet subband to calculate the local CW-SSIM and obtain the CW-SSIM mapping matrix. Finally, the mapping matrix is ​​weighted and summed to generate the final CW-SSIM. The weight factors are obtained according to a Gaussian distribution, and the standard deviation of the function is equal to 1 / 4 the size of the image at the bottom layer of the pyramid.

[0074] The present invention uses the similarity between the simulation image and the measured image as the optimization index. The optimized model can use SBR rapid imaging to generate a large number of simulation images, which can be used to expand the database based on the measured images.

[0075] In order to verify the effectiveness and correctness of the method of the present invention, the simulation imaging parameters of the optimized slice model in the MSTAR database and the CW-SSIM at various angles are given below.

[0076] The similarity between the simulated images at six different angles and the corresponding measured images in the MSTAR database was compared, and the average value was taken as the fitness value of the individuals in the HOA population. To suppress the influence of rough surfaces in the measured images, rectangular frames were added to the simulated and measured images, retaining only the target area. The parameters for SBR rapid imaging are shown in Table 1.

[0077] Table 1

[0078] Frequency scanning range 9.3GHz~9.9GHz Angle scanning range 0.0625rad Range resolution 0.25m Azimuth resolution 0.25m Distance to scene -7.95m~7.95m Azimuth scene -7.95m~7.95m

[0079] The number of individuals in the HOA population was set to 10, and the number of optimizations was set to 2000. The coordinates of the control points on all surfaces of the slice model, except the bottom, were set as optimization parameters, and the model structure was required to remain unchanged during the optimization process. The CW-SSIM was used as the image similarity evaluation metric. The optimization algorithm exited the loop and output the generated model when the fitness value reached 0.8. To improve the fitness of the HOA, a weighted sum of the CW-SSIM values ​​of the simulated and measured images at multiple angles was considered as the fitness value of the individuals in the population. The angle with the minimum CW-SSIM value between the simulated image and the measured image of the initial model was assigned a weight of 0.4, the next smallest was assigned to 0.2, and the remaining four angles were assigned a weight of 0.1. The similarity of the simulated images of the slice model after weighting the six angles using the Hummingbird optimization algorithm is shown in Table 2.

[0080] Table 2

[0081]

[0082] From the above simulation and result analysis, we can see that when taking the weighted sum of CW-SSIM under multiple angles, the fitness value of HOA increases from 0.41 to 0.821, and the similarity between the simulated image and the measured image is generally improved. Figure 3 、 Figure 4 shown.

[0083] In summary, the present invention makes full use of the insensitivity of complex wavelet structure similarity to image geometric distortion, while considering the convenience of changing the model when modeling with non-uniform rational B-spline technology, and adopts the Hummingbird optimization algorithm to optimize the model shape based on SAR images, thereby improving the similarity between simulated images and measured images.

[0084] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A method for optimizing the shape of a complex ship target model based on SAR images, characterized in that: The following steps are involved: Step 1: Model the initial model using non-uniform rational B-spline modeling technology to obtain the coordinates of the control points of the model; Step 2: Generate a simulation image based on the bouncing ray method for the target model; Step 3: Compare the simulation graph and the measured graph to obtain the control points that need to be optimized and set them as optimization parameters; Step 4: Use the Hummingbird optimization algorithm to update the optimization parameters, thereby updating the shape of the target model after the change; Step 5: Generate a simulated image based on the bouncing ray method and calculate its similarity with the target image; Image similarity evaluation indicators are as follows: Structural similarity, or SSIM, defines structural information from the perspective of image composition as a property that reflects the structure of objects in the scene, independent of brightness and contrast, and characterizes distortion as a combination of three different factors: brightness, contrast, and structure; The SSIM index is as follows: SSIM(x,y)=l(x,y) α1 ·c(x,y) β1 ·s(x,y) γ1 Let α1=β1=γ1=1, C3=C2 / 2, then SSIM simplifies to the following formula where μ x and μ y are the average pixel intensity of images x and y, respectively, x and σ y are the standard deviations of images x and y, σ xy Represents the covariance of images x and y; let C1=(k1*L) 2 , C2=(k2*L) 2 , where L represents the value range of the image pixel value; Complex wavelet structural similarity is an extension of SSIM in the complex wavelet domain; The mother wavelet signal is a low-pass filter modulation; where g(u) represents a slowly varying symmetric function, ω c Represents the center frequency of the modulated bandpass filter; the mother wavelet generates a wavelet cluster after various scaling and translation: Where ω(·) represents the wavelet transform form of the signal, s∈R + is the scale factor, p∈R is the translation factor; The continuous wavelet transform of the real signal x(u) is: Where X(ω) and G(ω) represent the Fourier transform results of x(u) and g(u) respectively; discrete wavelet coefficients are the sampling form of continuous wavelet transform; In the complex wavelet transform domain, the two sets of coefficients extracted from the same spatial position in the same wavelet subband of the two compared images are c x ={c x,i |i=1,2,…N} and c y ={c y,i |i=1,2,…N}; CW-SSIM is defined as: To use CW-SSIM to compare the similarity between two images, we first decompose the images using a multi-scale, multi-directional variable pyramid. Next, we use a sliding window to traverse each wavelet subband and calculate the local CW-SSIM to obtain the CW-SSIM mapping matrix. Finally, we generate the final CW-SSIM by weighted summation of the mapping matrix. The weight factor is obtained according to the Gaussian distribution, and the standard deviation of the function is equal to 1 / 4 of the size of the bottom image of the pyramid; Step 6: Use the similarity calculated in step 5 as the fitness value of each individual in the hummingbird optimization algorithm to determine whether the optimal fitness value meets the index; if the index is met or the optimal fitness value in the population remains unchanged after several iterations, output the corresponding model and control point information; otherwise, repeat step 4.

2. The method for optimizing the shape of a complex ship target model based on SAR images according to claim 1, characterized in that: In step 1, the geometric model of the target is established as follows: Use N j,n (u) represents the j-th n-th order B-spline basis function about the knot vector U: The shape of a NURBS surface is controlled by several control points according to a two-variable piecewise rational function; the coordinates of any point on the NURBS surface are expressed as: Among them, U and V represent the number of control points in the u and v parameter directions respectively; w ij represents the control point P ij The corresponding weight factor; P ij =[P ijx ,P ijy ,P ijz ] represents the control point P ij Coordinates in the x, y, z directions; N i,p (u) is the node vector U=[u1,u2,…u n+k+1 ] According to the Cox-DeBoor recursive formula, the p-order canonical B-spline basis function is obtained. Similarly, N j,q (v) is the node vector V = [v1, v2, ... v n+k+1 ]The q-order canonical B-spline basis function is obtained according to the Cox-DeBoor recursive formula.

3. The method for optimizing the shape of a complex ship target model based on SAR images according to claim 2, characterized in that: The hummingbird optimization algorithm described in step 4 is as follows: Assume that there are N hummingbirds in the population, and each hummingbird corresponds to a solution. Initialize the population: P i =ub-rand·(ub-lb) Among them, P i is the position of the i-th hummingbird in the population, i∈{1,2,…N}, corresponding to the feasible solution of the optimization problem, ub and lb represent the upper and lower limits of the variables, and rand represents a random number, rand∈[0,1]; In the self-search phase, if hummingbird i finds a better position, P i t ≠P i t-1 , it means that the current search area can continue to search; the new location of Hummingbird i is as follows: P i t+1 =P i t +rand·(P i t -P i t-1 ) P i t 、P i t+1 are the positions of the i-th hummingbird at the t-th and t+1-th iterations respectively; if P i t+1 If the fitness value of is better, it means that the algorithm has found a better solution, then the position is updated; otherwise, the position before this iteration remains unchanged; If hummingbird i searches for several times in a row and does not find a better food source, P i t =P i t-1 , which means that based on experience there is no better location in the current area, the search direction is changed randomly: in represents the optimal solution of the tth iteration, μ and ν both obey the normal distribution; In the guided search phase, the individual occupying the best position is called the leader bird, and the other individuals are called followers. The leader bird flies continuously around the territory: P T,t+1 =P T,t +r d ·l Among them, P T,t is the position of the leader bird at the tth iteration, r d is a random number between -1 and 1, λ=0.1(ub-lb), if P T,t+1 If the position represented by P is better, T,t+1 Replace P T,t ; Different from the leader bird, the movement of the follower bird can be divided into two states; if the leader bird does not find the follower bird j, the follower bird j approaches the leader bird area: P j F,t+1 =P j F,t +rand·(P T,t -MF·P j F,t ) MF=round[1+rand(0,1){2-1}] Among them, P j F,t is the position of the jth following bird after the tth iteration; MF is a mutation factor, round function indicates rounding, and rand indicates a random number; If follower bird j is discovered by the leader bird, it needs to select one of the other follower birds, labeled k. If the position of follower bird k is better than that of follower bird j, follower bird j will fly towards follower bird k, otherwise follower bird j will fly away from follower bird k: Where j, k∈{1,2,…N-1}; if Better than P j F,t , then use Replace P j F,t .

4. The method for optimizing the shape of a complex ship target model based on SAR images according to claim 3, characterized in that: α＝0.01，β＝1.5。 5. The method for optimizing the shape of a complex ship target model based on SAR images according to claim 1, characterized in that: Take k1=0.01, k2=0.

03.

6. The method for optimizing the shape of a complex ship target model based on SAR images according to claim 1, characterized in that: The value of L is 255.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method according to any one of claims 1 to 6 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images