Target electromagnetic scattering center parameter estimation method based on YOLOv8

Abstract

The invention relates to a target electromagnetic scattering center parameter estimation method based on YOLOv8, and the method comprises the steps: carrying out the fast Fourier transform of received echo data, and obtaining a standard two-dimensional complex image; using the improved YOLOv8 target detection network to predict the SAR amplitude image to obtain a position and length parameter set; intercepting an image region containing each target scattering center from the standard two-dimensional complex image based on the position and length parameter set to obtain a complex image data block; obtaining a frequency domain scattering center complex image by using a preset frequency dependence factor discrete set, and determining an optimal frequency dependence factor by calculating a root-mean-square error between the normalized complex image block and the frequency domain scattering center complex image; according to the multiple frequency domain scattering center complex images corresponding to the optimal frequency dependence factor, complex amplitude parameters are obtained through calculation, and therefore estimation of all model parameters of the target electromagnetic scattering center is completed. The method can realize joint estimation of a whole set of scattering center parameters.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of scattering feature extraction, and particularly relates to a target electromagnetic scattering center parameter estimation method based on YOLOv8. BACKGROUND

[0002] Synthetic aperture radar (SAR) is an active microwave remote sensing system that detects target characteristics by transmitting and receiving electromagnetic waves. This system has the ability to work all day and all weather, and has a long range, making it widely used in civil fields such as resource exploration, environmental monitoring, and disaster assessment. From the perspective of electromagnetic scattering mechanism, target characteristics can be divided into three regions according to the ratio of target size to radar wavelength: when the target size is much smaller than the wavelength, its scattering behavior follows the Rayleigh region rule; when the size is comparable to the wavelength, resonance characteristics are exhibited; and when the target is in the optical region, it mainly exhibits geometric optical characteristics. Most modern high-resolution radar systems operate in the optical region, and the scattering characteristics of targets in this region are closely related to their geometric structure and material properties, providing a theoretical basis for feature extraction from physical mechanisms.

[0003] With the development of imaging theory, the attribute scattering center model emerged as the times required. This model significantly enhances the characterization ability of scattering signals. It breaks through the limitations of traditional point scattering models, decomposes complex targets into a series of component-level scattering structures with clear physical meaning, and fully describes each scattering component through parameters such as amplitude, length, and azimuth dependence. This parameterized characterization method not only improves the physical interpretability of imaging results, but also provides a more discriminative feature basis for subsequent target recognition tasks.

[0004] Currently, electromagnetic feature extraction methods based on traditional mathematical modeling can be mainly divided into two categories: image domain algorithms and frequency domain algorithms. Image domain algorithms directly process radar imaging results, and after extracting scattering centers, they often use pseudo-Newton methods or image block-based methods for iterative optimization. The core idea is to approximate the maximum likelihood estimation solution through local iteration. Frequency domain algorithms focus on utilizing the sparsity of scattering centers in frequency domain data, constructing a sparse representation model, and combining Bayesian inference or matching pursuit strategies under dictionary learning to complete parameter estimation. In recent years, deep learning technology has been widely applied in radar image analysis due to its strong non-linear feature learning ability, and a series of data-driven parameter estimation methods have also emerged.

[0005] However, various methods in the prior art either are sensitive to initial values and easily fall into local optima, or have high computational complexity and are difficult to use, or are forced to simplify the model due to the difficulty in handling complex coupling between parameters, and none of them can efficiently and accurately estimate the full set of scattering center parameters including position, length, frequency dependence factor, and complex amplitude. SUMMARY

[0006] In order to solve the above problems existing in the prior art, the application provides a target electromagnetic scattering center parameter estimation method based on YOLOv8. The application provides a target electromagnetic scattering center parameter estimation method based on YOLOv8, and each model parameter of a target electromagnetic scattering center includes a position parameter, a length parameter, a frequency-dependent factor and a complex amplitude parameter. Performing fast Fourier transform on the received echo data to obtain a standard two-dimensional complex image; Using the improved YOLOv8 target detection network to predict the SAR amplitude image to obtain a position and length parameter set containing the position parameter and the length parameter of each target scattering center, wherein the SAR amplitude image is generated by extracting the amplitude information of the standard two-dimensional complex image; Based on the position and length parameter set, the image region containing the target scattering center is intercepted from the standard two-dimensional complex image to obtain a plurality of independent complex image data blocks, and each complex image data block contains a target scattering center; Inputting a preset frequency-dependent factor discrete set into an ideal scattering model to obtain a plurality of frequency domain scattering center complex images, determining a plurality of optimal frequency-dependent factors by calculating the root mean square error between a plurality of normalized complex image blocks and the plurality of frequency domain scattering center complex images, wherein each normalized complex image block is obtained by normalizing each complex image data block, and one target scattering center corresponds to one optimal frequency-dependent factor; According to the plurality of frequency domain scattering center complex images corresponding to the plurality of optimal frequency-dependent factors, a plurality of complex amplitude parameters corresponding one-to-one are calculated, so as to complete the estimation of all model parameters of the target electromagnetic scattering center.

[0007] Compared with the prior art, the application has the following advantages: To address the challenge of jointly estimating all scattering center parameters using existing technologies, this invention innovatively constructs a "parameter decoupling" estimation framework and deeply integrates it with the YOLOv8 target detection network, specifically improved for electromagnetic scattering center detection, effectively resolving the core bottleneck of existing methods. Specifically, this invention decomposes the highly coupled electromagnetic scattering parameter estimation problem into two low-dimensional sub-problems: sequentially solving geometric parameter estimation and electromagnetic parameter estimation. The improved YOLOv8 network directly and quickly estimates the position and length of the scattering center from SAR amplitude images, effectively avoiding the sensitivity to initial values ​​and the computational burden of iterative optimization inherent in traditional image-domain algorithms. Furthermore, based on the estimated geometric parameters, the mixed scattering centers are separated in the image domain. For each independent scattering center, the frequency dependence factor is accurately solved by traversing and searching the discrete parameter space and matching it with the ideal scattering model, finally yielding the complex amplitude parameters. This synergistic effect of "physical model decoupling" and "deep learning perception" overcomes the problems of traditional methods being prone to getting stuck in local optima, computational complexity, or being forced to simplify the model. It effectively achieves high-precision and high-efficiency joint estimation of the entire set of scattering center parameters, significantly improving the completeness and practicality of parameter estimation. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating a target electromagnetic scattering center parameter estimation method based on YOLOv8 provided in an embodiment of the present invention; Figure 2 This is a structural example diagram of the improved YOLOv8 target detection network provided in the embodiments of the present invention; Figure 3 This is an example diagram provided by an embodiment of the present invention for calculating the root mean square error between each ideal region and the normalized region in multiple ideal regions; Figure 4 This is a schematic diagram of a typical scatterer structure provided in an embodiment of the present invention; wherein, (a) is a schematic diagram of the reflection path of electromagnetic waves when electromagnetic waves are used to irradiate a trihedral corner, (b) is a schematic diagram of the reflection path of electromagnetic waves when electromagnetic waves are used to irradiate the top of a hat, (c) is a schematic diagram of the reflection path of electromagnetic waves when electromagnetic waves are used to irradiate a sphere, (d) is a schematic diagram of the reflection path of electromagnetic waves when electromagnetic waves are used to irradiate a dihedral corner, (e) is a schematic diagram of the reflection path of electromagnetic waves when electromagnetic waves are used to irradiate a cylinder, and (f) is a schematic diagram of the reflection path of electromagnetic waves when electromagnetic waves are used to irradiate an edge. Figure 5 This is provided by the embodiments of the present invention. Figure 4 The ISAR images corresponding to each structure in (a) to (c) are shown in the figure; where (a) is... Figure 4 (a) is the ISAR image, and (b) is... Figure 4 (b) is the ISAR image, and (c) is...Figure 4 The ISAR image in (c) of the image; Figure 6 This is provided by the embodiments of the present invention. Figure 4 The ISAR images corresponding to each structure in (d)~(f) are shown in the figure; where (d) is... Figure 4 (d) is the ISAR image, and (e) is... Figure 4 The ISAR image in (e) is shown in (f). Figure 4 The ISAR image in (f) of the image; Figure 7 These are example images of ISAR imaging of multiple electromagnetic scattering centers using the method provided by the present invention, provided in an embodiment of the present invention; wherein, (a) is an ISAR image of two electromagnetic scattering centers, (b) is an ISAR image of three electromagnetic scattering centers, and (c) is an ISAR image of four electromagnetic scattering centers. Figure 8 The embodiments of this invention utilize the method provided by this invention to... Figure 7 The parameter estimation results of the electromagnetic scattering centers in each subplot are shown in the figure; where (a) is the result of the parameter estimation of the electromagnetic scattering centers in each subplot. Figure 7 (b) shows the azimuth position parameter estimation diagram of the electromagnetic scattering center in each sub-figure. Figure 5 The range-direction position parameter estimation diagrams of the electromagnetic scattering centers in each subplot, (c) is for... Figure 7 The length parameter estimation diagram of the electromagnetic scattering center in each subplot, (d) is the length parameter estimation diagram of the electromagnetic scattering center in each subplot. Figure 7 The frequency factor parameter estimation plots of the electromagnetic scattering centers in each subplot, (e) is for... Figure 7 The complex amplitude parameter estimation diagram of the electromagnetic scattering center in each sub-graph; Figure 9 This is an example diagram of ISAR imaging of a simulation model provided in an embodiment of the present invention; Figure 10 This is the embodiment of the present invention that provides the following: Figure 9 An example diagram showing the electromagnetic scattering center annotation in the simulation model; Figure 11 This is an example of estimating using the method provided by the present invention. Figure 9 Post-reconstructed electromagnetic scattering center structure diagram; Figure 12 This is the utilization provided by the embodiments of the present invention. Figure 10 and Figure 12 ISAR images generated from the residual values ​​between electromagnetic scattering centers. Detailed Implementation

[0009] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0010] A target electromagnetic scattering center parameter estimation method based on YOLOv8 is described in detail in combination with the drawings.

[0011] Before the present embodiment is introduced, the electromagnetic scattering center is first explained.

[0012] In actual use of electromagnetic wave irradiation of complex targets, each point of the target will scatter radar waves. If described one by one, the amount of calculation is huge and unrealistic. For this, the electromagnetic scattering center (Electromagnetic Scattering Center) can be used to simplify it into a limited number of strong scattering sources, greatly reducing the complexity of analysis, and different targets correspond to different electromagnetic scattering center distributions (including number, position, intensity, etc.). By extracting and analyzing the parameters of these scattering centers, the target type can be identified.

[0013] For example, as in the dark with a flashlight (radar wave) to illuminate a complex object (such as a bicycle), it can be found that the light reflected by the whole bicycle is very complex, but the main light comes from several key parts: the bell (like a ball), the connection between the handlebar and the seat tube (like a right angle), and the spokes of the wheel (like a thin rod). These main light parts are the "electromagnetic scattering center". We do not need to describe how each inch of iron skin of the whole bicycle reflects light, but only need to describe the position, brightness and shape of these key parts (electromagnetic scattering center), which is enough to identify it as a bicycle in the dark.

[0014] Figure 1 is a flowchart of a target electromagnetic scattering center parameter estimation method based on YOLOv8 provided by the present embodiment. As shown in Figure 1 , the model parameters of each target electromagnetic scattering center include: position parameters, length parameters, frequency-dependent factors and complex amplitude parameters; based on the limitations of the joint estimation of model parameters by traditional methods, here, the parameter estimation method provided by the present embodiment decouples the model parameters into 、 and three independent parameter groups, realizing independent optimization estimation of parameters. The parameter estimation method includes: S110-S150. Specifically: S110: Fast Fourier transform is performed on the received echo data to obtain a standard two-dimensional complex image.

[0015] Specifically, the data form of the echo data is a two-dimensional complex matrix , wherein each point in the matrix represents a certain specific frequency and a certain specific elevation angle Downward, the amplitude and phase information of the electromagnetic wave reflected back from the target. The fast Fourier transform is used to first compress in the range direction, so that the echo signal of each target electromagnetic scattering center in the range direction is compressed into a sharp pulse, obtaining an intermediate data matrix in the range-time domain-azimuth frequency domain wherein, represents a range gate; then the intermediate data matrix is compressed in the azimuth direction to compress the response of each target electromagnetic scattering center in the range direction into a data point, obtaining a standard two-dimensional complex image wherein, the range direction coordinate is used to describe the vertical distance of the target from the radar, the azimuth direction coordinate is used to describe the position of the target in the direction perpendicular to the line of sight of the radar. Each pixel point in the standard two-dimensional complex image contains two kinds of information, one is amplitude and the other is phase. Specifically, the amplitude represents the intensity or brightness of the pixel point at the spatial position , and the phase is used to record the phase delay generated after the electromagnetic wave propagates to the target scattering center and returns, representing the position of the pixel point.

[0016] S120: using the improved YOLOv8 target detection network to predict the SAR amplitude image to obtain a position and length parameter set containing the position parameters and length parameters of each target scattering center; wherein the SAR amplitude image is generated by extracting the amplitude information of the standard two-dimensional complex image.

[0017] Here, the improved YOLOv8 target detection network includes a backbone network, a neck network and a head network; the backbone network includes a plurality of convolutional layers, a plurality of C2f modules and a fast spatial pyramid pooling layer, and is used to extract a plurality of feature maps of different scales from the input SAR amplitude image; the neck network includes a plurality of splicing layers, a plurality of C2f modules, a plurality of convolutional layers and a plurality of up-sampling layers, and is used to receive and fuse the plurality of feature maps output by the backbone network, and generate a feature enhancement map; the head network includes a plurality of decoupling heads, each decoupling head only retains a regression branch adopting an Anchor-free mechanism, and the head network is used to predict the feature enhancement map and output the position and length parameter set.

[0018] Figure 2 is a structure example diagram of the improved YOLOv8 target detection network provided by the embodiment of the present application. As Figure 2 shown, the backbone network includes 5 convolutional layers (English full name Convolutional Layer, Figure 2The core consists of four convolutional layers: 1. Convolutional layer (abbreviated as "conv", also referred to as layers P1 to P5); 2. C2f modules (the first C2f module is placed between the second and third convolutional layers, i.e., between layers P2 and P3, and the two C2f modules share one convolutional layer); and 3. A fast spatial pyramid pooling layer (full name: Spatial Pyramid Pooling – Fast). Figure 2 The abbreviation is "SPPF" in Chinese; the neck network includes four concatenation layers. Figure 2 (abbreviated as "Concat"), 2 upsampling layers ( Figure 2 The structure consists of three layers: an upsampled layer (abbreviated as "Upsample"), two convolutional layers, and four C2f modules. The first input of the first splicing layer is connected to the output of the C2f module located between layers P3 and P4 in the backbone network; the second input is connected to the output of the first upsampled layer; the input of the first upsampled layer is connected to the first output of the first C2f module in the neck network; the input of the first C2f module is connected to the output of the second splicing layer; the first input of the second splicing layer is connected to the output of the C2f module located between layers P4 and P5 in the backbone network; the second input is connected to the output of the second upsampled layer; the input of the second upsampled layer is connected to the output of the fast spatial pyramid pooling layer in the backbone network; the output of the fast spatial pyramid pooling layer in the backbone network is also connected to the first input of the third splicing layer in the neck network; the output of the first splicing layer is connected to the second C2f module in the neck network. Figure 2 The input of the P3 layer in the neck network is connected to the input of the first convolutional layer in the neck network. The output of the first convolutional layer in the neck network is connected to the first input of the fourth splicing layer. The second input of the fourth splicing layer is connected to the second output of the first C2f module. The output of the fourth splicing layer is connected to the third C2f module. Figure 2 The input of the P4 layer in the neck network is connected to the input of the third C2f module, the output of the second convolutional layer in the neck network is connected to the second input of the third splicing layer in the neck network, and the output of the third splicing layer in the neck network is connected to the second input of the fourth C2f module. Figure 2 The input terminal of the P5 layer in the middle neck network is connected.

[0019] The head network consists of three decoupling heads, the first of which is connected to... Figure 4 The P3 layer connection in the middle neck network, the second decoupling head and Figure 2 The P4 layer connection in the middle neck network, the third decoupling head and Figure 2The P5 layer connection in the middle neck network.

[0020] The anchor-free mechanism can be understood as the head network directly predicting the pixel position where each target center point in the feature enhancement map is located , and predicting the length parameter of the target from the center point . By using the regression branch of the anchor-free mechanism, the predicted center point directly corresponds to the coordinates in the image domain, and the width and height directly correspond to the size of the target electromagnetic scattering center, having a very clear mapping relationship, which eliminates the dependence on the pre-defined anchor box, thereby simplifying the model design, improving the efficiency, and being able to obtain better accuracy.

[0021] The position parameter and the length parameter in the model parameter of the target electromagnetic scattering center can be determined through S120.

[0022] S130: Based on the position and length parameter set, an image region containing each target scattering center is intercepted from the standard two-dimensional complex image, to obtain a plurality of independent complex image data blocks, and each complex image data block contains a target scattering center.

[0023] Here, the position and length parameter set contains a plurality of position and length parameters Each position and length parameter includes a position parameter for describing the center point coordinates of a target scattering center , and a length parameter for describing the length information of a target scattering center .

[0024] Further, S130 includes: for each position and length parameter, the following steps are performed: taking the position parameter in each position and length parameter as the center point of the interception frame, taking the corresponding length parameter as the side length of the interception frame, using the interception frame to intercept the corresponding region from the standard two-dimensional complex image, and setting the pixel value of the region not intercepted in the standard two-dimensional complex image to zero, to obtain a corresponding complex image data block.

[0025] S140: A preset frequency-dependent factor discrete set is input into an ideal scattering model to obtain a plurality of frequency domain scattering center complex images. A plurality of optimal frequency-dependent factors are determined by calculating the root mean square error between a plurality of normalized complex image blocks and the plurality of frequency domain scattering center complex images, wherein each normalized complex image block is obtained by normalizing each complex image data block, and one target scattering center corresponds to one optimal frequency-dependent factor.

[0026] In one possible implementation, each normalized complex image block is obtained by calculating the amplitude spectrum of each complex image data block to obtain multiple amplitude values; and by using the maximum value among the multiple amplitude values, each complex image data block is normalized to obtain the corresponding normalized complex image block.

[0027] It should be noted that the L2 norm of each complex image data block can also be used here to normalize each complex image data block.

[0028] The derivation process of the ideal model will now be explained.

[0029] First, the complete model expression (1) for the target electromagnetic scattering center is: ; in, It is the frequency of the echo data. It is the elevation angle of the echo data. These are the model parameters of the target's electromagnetic scattering center. It is the first Complex amplitude parameters of the electromagnetic scattering center of a target It is the first The azimuth parameters of the electromagnetic scattering center of a target It is the first Range parameters of the electromagnetic scattering center of a target, and Composition of the first Position parameters of the electromagnetic scattering center of a target It is the first The length parameter of the electromagnetic scattering center of a target represents the spatial extension characteristics of the scatterer. It is the first The initial pointing angle of the electromagnetic scattering center of the target. It is the first The directional dependence of the electromagnetic scattering center of a target on the elevation angle of the echo data. It is the first The frequency dependence factor of each target electromagnetic scattering center reflects the curvature correlation of the frequency response. It is the center frequency of the echo data. It's the speed of light. It is the plural part.

[0030] Based on differences in azimuth-dependent characteristics, electromagnetic scattering centers can be divided into two main categories: localized and distributed. In typical target structures, the former mainly exhibits physical phenomena such as trihedral reflection, cap effect, and spherical scattering, while the latter encompasses scattering mechanisms such as dihedral response, cylindrical reflection, and edge diffraction. The length parameter of a localized scattering center... and initial pointing angle parameters The value is zero, which is the dependence factor of the azimuth angle. Typically, a very small value is taken. However, the distributed scattering center... Other parameters are not zero. Based on this, it is set here. The value is 0, and thus the complete model expression of the target electromagnetic scattering center is simplified to expression (2), which is as follows: .

[0031] It should be noted that in the settings When the value is 0 and the electromagnetic scattering center is localized, the length parameter in expression (2) is... and initial pointing angle parameters All values ​​are 0.

[0032] Since the frequency dependence factor of a single scattering center is estimated using a preset discrete set of frequency dependence factors, the expression (2) contains... Replace with The summation operation in expression (2) is removed, and the amplitude parameter in expression (2) is also removed to remove the influence of the amplitude parameter, thus obtaining the expression (3) of the ideal scattering model, which is as follows: ; in, It is the frequency of the echo data. It is the elevation angle of the echo data. It is the first The azimuth parameters of the electromagnetic scattering center of a target It is the first Range parameters of the electromagnetic scattering center of a target, and Composition of the first Position parameters of the electromagnetic scattering center of a target It is the first The length parameter of the electromagnetic scattering center of a target. It is the first The initial pointing angle of the electromagnetic scattering center of the target. It is a preset set of discrete values ​​of frequency dependence factors used for the first An ergodic value of the electromagnetic scattering centers of a target. It is the maximum frequency of the echo data. It's the speed of light. It is the plural part.

[0033] Figure 3 This is an example diagram illustrating the calculation of the root mean square error between each ideal region and the normalized region in multiple ideal regions, provided by an embodiment of the present invention. For example...Figure 3 As shown, each frequency-dependent factor in the preset discrete set of frequency-dependent factors is input into the expression (3) of the ideal scattering model with the position parameter and the length parameter corresponding to the current normalized complex image block, to generate a plurality of ideal echo data. The plurality of ideal echo data is converted to a frequency domain image space by Fourier transform, to generate a plurality of corresponding frequency domain scattering center complex images .

[0034] Further, the root mean square error between the plurality of frequency domain scattering center complex images and the current normalized complex image block is calculated, to obtain a plurality of error values, including: taking the position parameter corresponding to the current normalized complex image block as the center point of the positioning frame, and taking the length parameter corresponding to the current normalized complex image block as the side length of the positioning frame, and using the positioning frame to locate the same region on each frequency domain scattering center complex image and the current normalized complex image block , to obtain a plurality of ideal regions and a normalized region ; the root mean square error between each ideal region in the plurality of ideal regions and the normalized region is calculated , to obtain a plurality of error values.

[0035] S150: According to the plurality of frequency domain scattering center complex images corresponding to the plurality of optimal frequency-dependent factors, a plurality of complex amplitude parameters are calculated one by one, to complete the estimation of all model parameters of the target electromagnetic scattering center.

[0036] Specifically, S150 includes: for the frequency domain scattering center complex image corresponding to each frequency domain scattering center complex image, the following steps are performed: calculate the scaling factor between the frequency domain scattering center complex image corresponding to the current optimal frequency-dependent factor and the complex image data block corresponding thereto, to obtain a complex amplitude scaling factor; multiply the complex amplitude scaling factor with the pixel value of the frequency domain scattering center complex image corresponding to the current optimal frequency-dependent factor, to obtain the corresponding complex amplitude parameter.

[0037] Here, after obtaining all the model parameters, the parameter estimation method provided by the present application constructs a radar echo signal according to all the estimated model parameters; calculates the residual error between the radar echo signal and the echo data, to evaluate the estimation accuracy.

[0038] The above is a detailed introduction to the target electromagnetic scattering center parameter estimation method based on YOLOv8 provided by the embodiments of the present application. Now, a data set is made using simulation software, and the performance of the method is evaluated.

[0039] Before conducting the evaluation, the types of electromagnetic scattering centers are introduced to demonstrate that different target structures correspond to electromagnetic scattering centers with different shapes and intensities, which can accurately reflect the structure of the target being detected. Figure 4 This is a schematic diagram of a typical scatterer structure provided in an embodiment of the present invention. Figure 5 This is provided by the embodiments of the present invention. Figure 4 The ISAR images corresponding to each structure in (a) to (c) are shown below. Figure 6 This is provided by the embodiments of the present invention. Figure 4 The ISAR images corresponding to each structure in (d)~(f) are shown.

[0040] like Figure 4 As shown, Figure 4 In the diagram, (a) shows the reflection path of an electromagnetic wave when it shines on a trihedral corner; (b) shows the reflection path of an electromagnetic wave when it shines on the top of a hat; (c) shows the reflection path of an electromagnetic wave when it shines on a sphere; (d) shows the reflection path of an electromagnetic wave when it shines on a dihedral corner; (e) shows the reflection path of an electromagnetic wave when it shines on a cylinder; and (f) shows the reflection path of an electromagnetic wave when it shines on an edge. Among these, (a) through (c) represent localized scattering centers and possess... Features of =0, subgraphs The values ​​are 1, 0.5, and 0 respectively; (d)~(f) correspond to the distributed scattering centers, and their length parameters are... >0, corresponding to the frequency dependence factor The values ​​1, 0.5, and 0 represent the electromagnetic scattering characteristics of different curvature features, respectively.

[0041] Figure 5 The subgraphs in the diagram are obtained by keeping other parameters constant and only changing... In different cases, Substituting expression (2) in step S140 above, and performing ISAR imaging processing, yields the result. Specifically, Figure 5 (a) in the middle is Figure 4 The ISAR image in (a) of the image, Figure 5 (b) in the middle is Figure 4 The ISAR image in (b) of the image, Figure 5 (c) in the middle is Figure 4 The ISAR image in (c) of the image, Figure 6 (d) in the middle is Figure 4 The ISAR image in (d) of the image. Figure 6 (e) in the middle is Figure 4 The ISAR image in (e) of the image. Figure 6 (f) in the middle is Figure 4 The ISAR image in (f) is shown.Figures 5-6 Each subgraph in FIG. 1 can obtain that when the frequency-dependent factor and the length parameter take different values, several basic scattering types can be effectively distinguished under the condition that the radar parameters and other related parameters remain unchanged.

[0042] Embodiment one Here, ISAR simulation work is carried out by means of CST electromagnetic simulation software. Simulation parameters are set as follows: the radar center frequency is set to 10 GHz, the bandwidth is 0.5 GHz, the azimuth and range resolutions are both 0.3 m, and data sets containing two, three and four electromagnetic scattering centers respectively are constructed, wherein the data sets cover local and distributed scattering centers, and each scattering center is randomly distributed. Figure 7 is an example diagram of ISAR imaging of multiple electromagnetic scattering centers by means of the method provided in the embodiment of the present application. In FIG. 1, Figure 7 (a) in FIG. 1 is an ISAR image of two electromagnetic scattering centers, Figure 7 (b) in FIG. 1 is an ISAR image of three electromagnetic scattering centers, Figure 7 (c) in FIG. 1 is an ISAR image of four electromagnetic scattering centers.

[0043] Figure 8 is an example diagram of ISAR imaging of multiple electromagnetic scattering centers by means of the method provided in the embodiment of the present application. In FIG. 1, Figure 7 is a parameter estimation result diagram of electromagnetic scattering centers in each subgraph in FIG. 1. In FIG. 1, Figure 8 (a) in FIG. 1 is an azimuth position parameter estimation diagram of electromagnetic scattering centers in Figure 7 (b) in FIG. 1 is a range position parameter estimation diagram of electromagnetic scattering centers in Figure 8 (c) in FIG. 1 is a length parameter estimation diagram of electromagnetic scattering centers in Figure 7 (d) in FIG. 1 is a frequency factor parameter estimation diagram of electromagnetic scattering centers in Figure 8 (e) in FIG. 1 is a complex amplitude parameter estimation diagram of electromagnetic scattering centers in Figure 7 (f) in FIG. 1 is a frequency-dependent factor parameter estimation diagram of electromagnetic scattering centers in Figure 8 (g) in FIG. 1 is a complex amplitude parameter estimation diagram of electromagnetic scattering centers in Figure 7 (h) in FIG. 1 is a frequency-dependent factor parameter estimation diagram of electromagnetic scattering centers in Figure 8 (i) in FIG. 1 is a complex amplitude parameter estimation diagram of electromagnetic scattering centers in Figure 8 (j) in FIG. 1 is a frequency-dependent factor parameter estimation diagram of electromagnetic scattering centers in Figure 8 (a)-(c) in FIG. 1 can obtain that the method has good robustness in parameter estimation of multiple scattering centers, the mean square error between the true value and the predicted value is low, and the subsequent parameter estimation is not affected. Figure 9 (d) in FIG. 1 can obtain that the estimation effect of the frequency-dependent factor is significantly improved, and the correct prediction probability is more than 75%; when the signal-to-noise ratio reaches 90 dB, the correct prediction probability of two scattering centers is more than 85%. Figure 10From (e), we can see that as the number of parameters increases, mutual interference increases, the prediction difficulty increases, and random prediction bias occurs. For example, when the signal-to-noise ratio increases, the prediction effect of four scattering features is better than that of two scattering centers. This is due to the uncertainty of the randomly generated dataset itself. Each experimental curve follows the rule that the prediction effect continuously optimizes as the signal-to-noise ratio increases.

[0044] Example 2 The simulation parameters are set as follows: center frequency is set to 10GHz, bandwidth is 0.5GHz, sampling interval is 3.9MHz; observation angle range is defined as 88.57° to 91.43°, sampling interval is 0.022°, observation angle is 90°, and elevation angle is 30°. Figure 9 This is an example diagram of ISAR imaging of a simulation model provided in an embodiment of the present invention. Figure 10 This is the embodiment of the present invention that provides the following: Figure 9 An example diagram of electromagnetic scattering center annotation in the simulation model, where... Figure 11 Seven electromagnetic scattering centers were identified; they are listed here. Figure 9 The actual parameter data of the electromagnetic scattering center of the simulation model are shown in Table 1. Figure 12 This is an example of estimating using the method provided by the present invention. Figure 10 Post-reconstructed electromagnetic scattering center structure diagram, Figure 11 This is the utilization provided by the embodiments of the present invention. Figure 11 and Figures 10-12 ISAR images generated from the residual values ​​between electromagnetic scattering centers.

[0045] Depend on Figure 11 It can be seen that only the extraction of scattering center number 7 has a certain deviation, while the extraction results of the remaining scattering centers are basically accurate. Based on Figure 10 We can obtain: Figure 11 The reconstruction results are basically the same as ​ The original results are the same. Now extract... ​ The parameters of each electromagnetic scattering center are listed in Table 2. Based on the comparison between Table 1 and Table 2, it can be concluded that when the prediction accuracy of the center position and length parameters is at the centimeter level, the extracted results are basically consistent with the true values. Regarding the prediction of the initial pointing angle parameter, only the predictions of scattering centers 1 and 7 showed deviations, but the errors were still within acceptable limits. The frequency dependence factor of scattering center 5, which should have been 1, was extracted as 0, causing the corresponding scattering type to be misjudged from dihedral reflection as edge lateral reflection. The predictions of the other scattering center types were all correct, confirming the effectiveness of the algorithm proposed in this chapter in predicting targets with complex combined attribute scattering centers.

[0046] Table 1

[0047] Table 2

[0048] In order to solve the problem that the joint estimation of a complete set of scattering center parameters is difficult to achieve in the prior art, an innovative "parameter decoupling" estimation framework is constructed, and a YOLOv8 target detection network specially improved for electromagnetic scattering center detection is deeply integrated, effectively solving the core bottleneck of existing methods. Specifically, by using the parameter decoupling strategy, the highly coupled electromagnetic scattering parameter estimation problem is decomposed into two low-dimensional sub-problems of geometric parameter estimation and electromagnetic parameter estimation which are solved sequentially. The improved YOLOv8 network is used to directly and quickly estimate the position and length of the scattering center and other geometric parameters from the SAR amplitude image, effectively avoiding the sensitivity of traditional image domain algorithms to initial values and the computational burden of iterative optimization. Then, based on the estimated geometric parameters, the mixed scattering center is separated in the image domain, and for each independent scattering center, the frequency-dependent factor is accurately solved by searching and matching with the ideal scattering model in the discrete parameter space, and finally the complex amplitude parameter is calculated. The synergistic effect of this "physical model decoupling" and "deep learning perception" overcomes the problems of traditional methods being easily trapped in local optimum, high computational complexity or being forced to simplify the model, effectively realizing high-precision and high-efficiency joint estimation of a complete set of scattering center parameters, and significantly improving the integrity and practicality of parameter estimation.

[0049] The above is a further detailed description of the present application in combination with specific preferred embodiments, and the specific implementation of the present application cannot be limited to these descriptions. For ordinary skilled persons in the technical field to which the present application belongs, several simple deductions or substitutions can be made without departing from the concept of the present application, and all of them should be regarded as falling within the protection scope of the present application.

Claims

1. A YOLOv8-based target electromagnetic scattering center parameter estimation method, characterized in that, The model parameters of each target electromagnetic scattering center include a position parameter, a length parameter, a frequency-dependent factor and a complex amplitude parameter; the parameter estimation method includes: performing fast Fourier transform on the received echo data to obtain a standard two-dimensional complex image; using the improved YOLOv8 target detection network to predict the SAR amplitude image to obtain a position and length parameter set containing the position parameters and length parameters of each target scattering center; wherein the SAR amplitude image is generated by extracting the amplitude information of the standard two-dimensional complex image; based on the position and length parameter set, the image region containing the target scattering centers is cut from the standard two-dimensional complex image to obtain independent multiple complex image data blocks, each of which contains a target scattering center; inputting a preset frequency-dependent factor discrete set into an ideal scattering model to obtain multiple frequency domain scattering center complex images, and determining multiple optimal frequency-dependent factors by calculating the root mean square error between multiple normalized complex image blocks and the multiple frequency domain scattering center complex images, wherein each normalized complex image block is obtained by normalizing each complex image data block, and one target scattering center corresponds to one optimal frequency-dependent factor; according to the multiple frequency domain scattering center complex images corresponding to the multiple optimal frequency-dependent factors, a one-to-one corresponding multiple complex amplitude parameters are calculated to complete the estimation of all model parameters of the target electromagnetic scattering center.

2. The YOLOv8-based target electromagnetic scattering center parameter estimation method according to claim 1, characterized in that, The position and length parameter set contains multiple position and length parameters, each of which includes a position parameter for describing the center point coordinates of the target scattering center and a length parameter for describing the length information of the target scattering center. Based on the position and length parameter set, the image region containing the target scattering centers is cut from the standard two-dimensional complex image to obtain independent multiple complex image data blocks, which includes: for each position and length parameter, the following steps are performed: using the position parameter in each position and length parameter as the center point of the cutting frame and the corresponding length parameter as the side length of the cutting frame, the corresponding region is cut from the standard two-dimensional complex image using the cutting frame, and the pixel values of the region not cut in the standard two-dimensional complex image are set to zero to obtain a corresponding complex image data block.

3. The YOLOv8-based target electromagnetic scattering center parameter estimation method according to claim 2, characterized in that, The preset frequency-dependent factor discrete set is input into the ideal scattering model to obtain multiple frequency domain scattering center complex images, and multiple optimal frequency-dependent factors are determined by calculating the root mean square error between multiple normalized complex image blocks and the multiple frequency domain scattering center complex images, which includes: for each normalized complex image block in the multiple normalized complex image blocks, the following steps are performed: input each frequency-dependent factor in the preset frequency-dependent factor discrete set and the position and length parameters corresponding to the current normalized complex image block into the ideal scattering model to generate multiple ideal echo data; performing Fourier transform on the multiple ideal echo data to generate one-to-one corresponding multiple frequency domain scattering center complex images; calculating root mean square errors between the plurality of frequency domain scattering center complex images and the current normalized complex image block to obtain a plurality of error values; screening a minimum value in the plurality of error values as an optimal frequency-dependent factor of a target scattering center in the current normalized complex image block, and taking a frequency domain scattering center complex image corresponding to the minimum value as a contrast image.

4. The YOLOv8-based target electromagnetic scattering center parameter estimation method according to claim 3, characterized in that, The calculating root mean square errors between the plurality of frequency domain scattering center complex images and the current normalized complex image block to obtain a plurality of error values comprises: taking a position parameter corresponding to the current normalized complex image block as a center point of a positioning frame, and taking a length parameter corresponding to the current normalized complex image block as a side length of the positioning frame, and using the positioning frame to locate the same region on each frequency domain scattering center complex image and the current normalized complex image block respectively to obtain a plurality of ideal regions and a normalized region; calculating root mean square errors between each ideal region in the plurality of ideal regions and the normalized region to obtain the plurality of error values.

5. The YOLOv8-based target electromagnetic scattering center parameter estimation method according to claim 3, characterized in that, The calculating a plurality of complex amplitude parameters one by one from the frequency domain scattering center complex images corresponding to the plurality of optimal frequency-dependent factors comprises: for the frequency domain scattering center complex image corresponding to each frequency domain scattering center complex image, the following steps are performed: calculating a scaling factor between the frequency domain scattering center complex image corresponding to the current optimal frequency-dependent factor and the corresponding complex image data block to obtain a complex amplitude scaling factor; multiplying the complex amplitude scaling factor and a pixel value of the frequency domain scattering center complex image corresponding to the current optimal frequency-dependent factor to obtain a corresponding complex amplitude parameter.

6. The YOLOv8-based target electromagnetic scattering center parameter estimation method according to claim 1, characterized in that, The expression of the ideal scattering model is: ； wherein, is a frequency of the echo data, is an elevation angle of the echo data, is an azimuth parameter of the first target electromagnetic scattering center, is a range parameter of the first target electromagnetic scattering center, and comprise a position parameter of the first target electromagnetic scattering center, is a length parameter of the first target electromagnetic scattering center, is an initial pointing angle of the first target electromagnetic scattering center, is one of the preset frequency-dependent factor discrete value sets for the first target electromagnetic scattering center, is a maximum frequency of the echo data, is a speed of light, is a complex part.

7. The YOLOv8-based target electromagnetic scattering center parameter estimation method according to claim 1, characterized in that, The improved YOLOv8 target detection network comprises a backbone network, a neck network and a head network. The backbone network comprises a plurality of convolutional layers, a plurality of C2f modules and a fast spatial pyramid pooling layer, and is used to extract a plurality of feature maps of different scales from an input SAR amplitude image; The neck network comprises a plurality of splicing layers, a plurality of C2f modules, a plurality of convolutional layers and a plurality of up-sampling layers, and is used to receive and fuse the plurality of feature maps output by the backbone network and generate a feature enhancement map; The head network comprises a plurality of decoupling heads with the same structure, each decoupling head only retains a regression branch adopting an Anchor-free mechanism, and the head network is used to predict the feature enhancement map and output the set of position and length parameters.

8. The YOLOv8-based target electromagnetic scattering center parameter estimation method according to claim 1, characterized in that, Each normalized complex image block is obtained by: calculating an amplitude spectrum of each complex image data block to obtain a plurality of amplitude values; normalizing each complex image data block by using a maximum value in the plurality of amplitude values to obtain a corresponding normalized complex image block.

Images