Scattering center parameter extraction method and system based on structural analysis

Abstract

The invention provides a scattering center parameter extraction method and system based on structural analysis. A target is decomposed into a structural unit set through structural analysis, geometric feature parameters of curved surface structural units are extracted by using a convolutional neural network, and the structural units are mapped into scattering center parameters based on the geometric feature parameters, so that an equivalent scattering center point is extracted. The target is converted into a plurality of structural units through structural analysis, so that the processing complexity is greatly reduced, and the modeling efficiency of the scattering center is improved. The end-to-end mapping from the structural unit to the scattering center parameter is realized, and the physical meaning corresponding to the scattering center is clear and definite. The target radio frequency scene can be reconstructed through the scattering center point, and the method can be used for simulation verification of the detection and identification performance of a radar system.

Description

Technical Field

[0001] The present invention belongs to the field of target characteristic modeling and simulation, and relates to a method that can support target radio frequency scene simulation. Specifically, it relates to an equivalent scattering center parameter extraction method and system based on structural analysis, and in particular, it relates to an equivalent scattering center extraction method and system based on structural analysis and convolutional neural network. Background Art

[0002] The target scattering characteristics are the scattering forms of the target to the radar incident wave at different frequencies, angles and polarization modes. In the design, optimization and simulation process of the radar system, a large amount of target characteristic data is needed to construct a complex electromagnetic environment. In order to efficiently utilize the massive target characteristic data, it is necessary to perform parameterized digital modeling, so that the multi-dimensional target characteristics can be quickly and accurately reconstructed.

[0003] The size of most radar targets is larger than the wavelength of the detection wave, so they can be regarded as electrically large targets. At this time, high-frequency approximate theories such as geometric optics and geometric diffraction theory can be used to analyze the target scattering process. The analysis results show that the radar target can be regarded as a collection of several equivalent scattering points, so that its target characteristics can be modeled. By introducing a series of attribute parameters, the scattering center can also characterize the material, size and shape of complex radar targets. The scattering center can be divided into a local equivalent scattering center and a distributed equivalent scattering center. The local scattering center generally corresponds to structures such as edges, discontinuities, and corners. Its main feature is that it is visible in a wide angle range. The distributed scattering center generally corresponds to structures such as planes and large-area curved surfaces. Its main feature is that it is visible in a narrow angle range.

[0004] At present, the extraction method of scattering centers is mainly based on the existing target characteristic data, using heuristic algorithms such as matched orthogonal pursuit algorithm and particle filtering to estimate the number of equivalent scattering centers and corresponding attribute parameters. However, this method has the disadvantages of unclear physical meaning, large amount of calculation and too many scattering centers.

[0005] The patent document "SAR Image Feature Extraction Method Based on Convolutional Neural Network" (CN109934237A) only proposes the use of convolutional neural network for feature extraction without any preprocessing.

[0006] The information extracted by preprocessing in the patent document "SAR image target detection method based on fusion convolutional neural network" (CN110097524A) and the patent document "SAR target recognition method based on component parsing multi-scale convolutional neural network" (CN113240047A) are both electromagnetic scattering features, and the purpose is target detection.

[0007] Taking the extraction of scattering centers of ship targets as an example, the traditional scattering center extraction method solves the scattering center parameters based on scattering characteristic data such as SAR images, which is an inverse solution. In the process of solving, only the fitting error is considered, which may cause problems such as the inconsistency between the solved scattering center position and the actual scattering structure, and the inconsistency between the scattering center and the scattering mechanism. The patent document "A SAR target recognition method based on scattering center feature extraction" (CN116778349A) only makes improvements to the convolutional neural network, and does not perform preprocessing structure analysis on the obtained target.

[0008] Patent document "SAR target recognition method based on component analysis multi-scale convolutional neural network" (CN113240047A) proposes a SAR target recognition method based on multi-scale CNN. This method improves the accuracy of SAR target recognition by mining the intrinsic electromagnetic information of SAR targets. However, this method does not accurately and quantitatively analyze the mapping relationship between the three-dimensional structural information of the target and the electromagnetic characteristics, and the use of this method is limited to target recognition, and cannot be used for target scattering center extraction. In this regard, the present invention decomposes the target into a set of structural units through structural analysis, extracts the geometric feature parameters of the surface structural unit using a convolutional neural network, and maps the structural unit to the scattering center parameter based on the geometric feature parameters, thereby realizing accurate mapping and quantitative analysis of the target three-dimensional structural information and electromagnetic characteristics, and can realize scattering center extraction based on the target three-dimensional model. More specifically, the present invention performs structural analysis on the target three-dimensional model, classifies and partitions the complex structure to obtain the structural unit, and analyzes the scattering characteristics of the structural unit to obtain the independent scattering center corresponding to each part, which is mainly used for performance testing and verification of radar systems. The convolutional neural network is used to improve the estimation accuracy of the geometric characteristic parameters of the structural unit, extract the surface structural characteristic parameters, and map the structural unit to the scattering center parameters based on the geometric characteristic parameters to extract the equivalent scattering center points, which belongs to the forward modeling method. The present invention can greatly improve the efficiency and accuracy of scattering center parameter extraction, and the physical meaning corresponding to the scattering center is clear and unambiguous. Summary of the invention

[0009] In view of the defects in the prior art, the object of the present invention is to provide a method and system for extracting scattering center parameters based on structure analysis.

[0010] A scattering center parameter extraction method based on structural analysis provided by the present invention specifically includes:

[0011] Step S1: Perform structural analysis based on the selected target grid file to obtain a set of structural units;

[0012] Step S2: mapping the structural units in the structural unit set into geometric feature parameters;

[0013] Step S3: Mapping the geometric characteristic parameters of the structural unit into scattering center parameters to obtain the target scattering center.

[0014] Preferably, in step S1, the target grid file is structurally analyzed to identify edges, divide the regions according to the edges, and decompose the regions into a set of structural units consisting of plane structural units and curved structural units, including:

[0015] Step S1.1: read the target mesh file, use the triangulation information and point coordinate information in the mesh file to obtain triangular face elements, and calculate the external normal vector directions and center coordinates of all obtained triangular face elements;

[0016] Step S1.2: Determine the edge according to the direction of the external normal vector and the center coordinates of the triangle surface element, and record the edge information;

[0017] Step S1.3: According to the edge information, the target is divided into regions, and plane structure units and curved surface structure units are obtained.

[0018] Preferably, in step S1.2, the included angle of the external normal vectors of the triangular face element and its adjacent triangular face element is calculated. If the included angle of the external normal vectors is greater than a set threshold, the two face elements are on different planes, and the common edge is an edge. The edge is recorded by numbering the two endpoints of the edge.

[0019] When the angle between the external normal vectors is less than or equal to the set threshold, the two face elements are on the same plane;

[0020] The step S1.3 comprises:

[0021] When a triangular face element has no intersection with an edge, the face element is a face element on a certain face. The number of the triangle is recorded, and the numbers of the triangular face elements adjacent to it are found based on the neighborhood information of the face element, and the number of intersections between the face element and the edge is further determined.

[0022] When the number of intersections between a triangle element and an edge is 1, the triangle element has an edge on the scattering edge, and further judgment of the triangle elements adjacent to the edge is stopped;

[0023] When the intersection of a triangle element and an edge is 2, the triangle element is located at the intersection of two edges;

[0024] According to the relationship between triangular face elements and edges, the target is divided into multiple areas;

[0025] According to the flatness of the triangular face elements in each region, the region is divided into a plane region constituting a plane structural unit and a curved surface region constituting a curved surface structural unit.

[0026] Preferably, in step S2, the geometric features of the curved surface structure unit are estimated, a convolutional neural network is constructed, and the curved surface structure unit is processed by the convolutional neural network to obtain geometric feature parameters, including:

[0027] Step S2.1: construct a training data set, assign the pixel points of the surface structure unit to the normalized height value, and the height value is the height obtained by tiling on the plane according to the set observation angle;

[0028] Step S2.2: construct a convolutional neural network model, wherein the convolutional neural network comprises three cascaded convolutional layers and pooling layers and three cascaded fully connected layers, the input layer is a height map after the set of structural units is tiled, the nonlinear activation function of the convolutional layer is a ReLU function; the pooling method of the pooling layer is maximum pooling; the nonlinear activation function of the fully connected layer is a ReLU function;

[0029] Step S2.3: Generate training samples, train the convolutional neural network model, and set the loss function to the MSE error function, that is, Obtain the trained convolutional neural network model;

[0030] Among them, J represents the loss function;

[0031] MSE() represents mean square error function;

[0032] m represents the number of training samples;

[0033] y i Represents the true value of the i-th training sample;

[0034] x i represents the i-th input variable;

[0035] f() represents the prediction result of the neural network model;

[0036] Step S2.4: Use the trained convolutional neural network model to extract the geometric feature parameters of the surface structure unit. Assign the pixel points of the surface structure unit obtained by structural analysis to the normalized height value, and the height value is the height obtained by tiling on the plane according to the set observation angle, to obtain a two-dimensional array. Use the two-dimensional array as the input of the convolutional neural network, and obtain the structural feature parameters of the surface structure unit, such as curvature, normal vector, etc., through feature extraction and full connection layer operation;

[0037] Preferably, in step S3, a parameterized scattering center model is constructed according to the mapping relationship between the geometric characteristic parameters and the scattering center parameters, and the single surface parameter calculation method is: The calculation method of cylindrical single curved surface parameters is: The conical single curved surface parameters are calculated as follows: The plane parameters are calculated as follows:

[0038] Among them, E h Represents the geometric characteristic parameters of a single curved surface;

[0039] r represents the radius of curvature;

[0040] k represents the wave number;

[0041] represents the position vector of the centroid of a single curved surface;

[0042] Indicates the radar observation line of sight angle;

[0043] E cylinder Represents the geometric characteristic parameters of a cylindrical single curved surface;

[0044] λ represents wavelength;

[0045] f represents frequency;

[0046] c represents the speed of light;

[0047] Represents the direction vector of the cylinder generatrix;

[0048] l represents the busbar length;

[0049] θ represents the radar observation angle;

[0050] E conel Represents the geometric characteristic parameters of the conical single curved surface;

[0051] The subscripts max and min represent the maximum and minimum values, respectively;

[0052] γ represents the cone normal vector;

[0053] E plane Represents the geometric characteristic parameters of the plane;

[0054] N represents the number of polygon sides;

[0055] represents the centroid vector of the polygon;

[0056] It is expressed as the projection vector of the centroid vector on the polygon normal;

[0057] It is expressed as the projection vector of the centroid vector on the horizontal plane of the polygon;

[0058] represents the midpoint vector of the nth edge;

[0059] The geometric characteristic parameters of the structural unit are mapped to the scattering center parameters, the scattering fields of the target scattering centers are superimposed to obtain the overall scattering field of the target, and the target scattering centers are extracted.

[0060] A scattering center parameter extraction system based on structure analysis provided by the present invention comprises:

[0061] Module M1: Perform structural analysis on the selected target grid file to obtain a set of structural units;

[0062] Module M2: Mapping the structural units in the structural unit set into geometric feature parameters;

[0063] Module M3: Map the geometric characteristic parameters of the structural unit to the scattering center parameters to obtain the target scattering center.

[0064] Preferably, in module M1, the target grid file is structurally analyzed, edges are identified, regions are divided, and the structure is decomposed into a set of structural units consisting of plane structural units and curved surface structural units, including:

[0065] Module M1.1: Read the target grid file, use the triangulation information and point coordinate information in the grid file to obtain triangular face elements, and calculate the external normal vector direction and center coordinates of all obtained triangular face elements;

[0066] Module M1.2: Determine the edge according to the direction of the external normal vector and the center coordinates of the triangle surface element, and record the edge information;

[0067] Module M1.3: Divide the target into regions based on edge information and obtain plane structure units and curved surface structure units.

[0068] Preferably, in module M1.2, the angle between the external normal vectors of a triangular face element and its adjacent triangular face element is calculated. If the angle between the external normal vectors is greater than a set threshold, the two face elements are on different planes, and the common edge is an edge. The edge is recorded by numbering the two endpoints of the edge.

[0069] If the angle between the external normal vectors is less than or equal to the set threshold, the two face elements are on the same face;

[0070] The module M1.3 includes:

[0071] When a triangular face element has no intersection with an edge, the face element is a face element on a certain face. The number of the triangle is recorded, and the numbers of the triangular face elements adjacent to it are found based on the neighborhood information of the face element, and the number of intersections between the face element and the edge is further determined.

[0072] When the number of intersections between a triangle element and an edge is 1, the triangle element has an edge on the scattering edge, and further judgment of the triangle elements adjacent to the edge is stopped;

[0073] When the intersection of a triangle element and an edge is 2, the triangle element is located at the intersection of two edges;

[0074] According to the relationship between triangular face elements and edges, the target is divided into multiple areas;

[0075] According to the flatness of the triangular face elements in each region, the region is divided into a plane region constituting a plane structural unit and a curved surface region constituting a curved surface structural unit.

[0076] Preferably, in module M2, the geometric features of the surface structural unit are estimated, a convolutional neural network is constructed, and the set of structural units is processed using the convolutional neural network to obtain geometric feature parameters, including:

[0077] Module M2.1: Construct a training data set, assign the pixel points of the surface structure unit to the normalized height value, where the height value is the height obtained by tiling on the plane according to the set observation angle;

[0078] Module M2.2: Construct a convolutional neural network model, wherein the convolutional neural network includes three cascaded convolutional layers and pooling layers and three cascaded fully connected layers, the input layer is a height map after the set of structural units is tiled, the nonlinear activation function of the convolutional layer is a ReLU function; the pooling method of the pooling layer is maximum pooling; the nonlinear activation function of the fully connected layer is a ReLU function;

[0079] Module M2.3: Generate training samples, train the convolutional neural network model, and set the loss function to the MSE error function, that is, Obtain the trained convolutional neural network model;

[0080] Among them, J represents the loss function;

[0081] MSE() represents mean square error function;

[0082] m represents the number of training samples;

[0083] y i Represents the true value of the i-th training sample;

[0084] x i represents the i-th input variable;

[0085] f() represents the prediction result of the neural network model;

[0086] Module M2.4: Use the trained convolutional neural network model to extract the geometric feature parameters of the surface structure unit. Assign the pixel points of the surface structure unit obtained by structural analysis to the normalized height value, which is the height obtained by tiling on the plane according to the set observation angle, to obtain a two-dimensional array. The two-dimensional array is used as the input of the convolutional neural network, and the structural feature parameters of the surface structure unit, such as curvature, normal vector, etc., are obtained through feature extraction and full connection layer operation.

[0087] Preferably, in the module M3, a parameterized scattering center model is constructed according to the mapping relationship between the geometric characteristic parameters and the scattering center parameters, and the single surface parameter calculation method is: The calculation method of cylindrical single curved surface parameters is: The conical single curved surface parameters are calculated as follows: The plane parameters are calculated as follows:

[0088] Among them, E h Represents the geometric characteristic parameters of a single curved surface;

[0089] r represents the radius of curvature;

[0090] k represents the wave number;

[0091] represents the position vector of the centroid of a single curved surface;

[0092] Indicates the radar observation line of sight angle;

[0093] E cylinder Represents the geometric characteristic parameters of a cylindrical single curved surface;

[0094] λ represents wavelength;

[0095] f represents frequency;

[0096] c represents the speed of light;

[0097] Represents the direction vector of the cylinder generatrix;

[0098] l represents the busbar length;

[0099] θ represents the radar observation angle;

[0100] E conel Represents the geometric characteristic parameters of the conical single curved surface;

[0101] The subscripts max and min represent the maximum and minimum values, respectively;

[0102] γ represents the cone normal vector;

[0103] E planeRepresents the geometric characteristic parameters of the plane;

[0104] N represents the number of polygon sides;

[0105] represents the centroid vector of the polygon;

[0106] Expressed as the centroid vector in the polygon normal The projection vector of

[0107] It is expressed as the projection vector of the centroid vector on the horizontal plane xyp of the polygon;

[0108] represents the midpoint vector of the nth edge;

[0109] The geometric characteristic parameters of the structural unit are mapped to the scattering center parameters. According to the scattering field superposition of the target scattering center, the overall scattering field of the target is obtained and the target scattering center is extracted.

[0110] Compared with the prior art, the present invention has the following beneficial effects:

[0111] 1. The present invention can convert the target into several structural units through structural analysis, thereby greatly reducing the processing complexity and improving the efficiency of scattering center modeling.

[0112] 2. The present invention extracts the curvature geometric characteristic parameters of the surface structural unit through a convolutional neural network, and maps them into scattering center parameters, thereby realizing an end-to-end mapping from the structural unit to the scattering center parameters.

[0113] 3. The present invention utilizes a convolutional neural network to reconstruct the target radio frequency scene through the scattering center point, which can be used to simulate and verify the detection and recognition performance of the radar system.

[0114] 4. The present invention can greatly improve the efficiency and accuracy of scattering center parameter extraction, and the physical meaning corresponding to the scattering center is clear and unambiguous. BRIEF DESCRIPTION OF THE DRAWINGS

[0115] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0116] Figure 1 Schematic diagram of the process of extracting scattering center parameters based on structure analysis.

[0117] Figure 2 Schematic diagram of scattering center parameter extraction based on structure analysis.

[0118] Figure 3 It is a schematic diagram of the structural analysis of an embodiment.

[0119] Figure 4 Schematic diagram of the convolutional neural network structure used.

[0120] Figure 5 Schematic diagram of the convolutional neural network training process used. DETAILED DESCRIPTION

[0121] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0122] The technical problem to be solved by the present invention is that the current scattering model extraction method is inefficient and the physical meaning is unclear. In order to solve the above problem, the present invention provides a scattering center parameter extraction method based on structure analysis, comprising:

[0123] Step S1: Perform structural analysis based on the target CAD structure diagram to obtain a structural unit;

[0124] Specifically, Figure 2 As shown, in step S1, the target CAD structure drawing is directly analyzed to identify edges, divide the area according to the edges, and decompose it into plane structure units and curved surface structure units.

[0125] like Figure 3 As shown, more specifically, step S1 of this embodiment first reads the mesh file of the target, and uses the triangulation information and point coordinate information of the target contained in the mesh file to calculate the direction of the external normal vector of each triangular face element and the center coordinates of the triangular face element. When calculating, it is necessary to ensure that the triangular face elements in the mesh file conform to the right-hand screw rule. Based on the triangular mesh information, the numbers of triangles with common edges are found, and then the neighborhood information of each triangle can be constructed, that is, to determine which triangles are adjacent to each triangle. Based on the above judgment, the angle of the external normal vector of each triangular face element and several adjacent triangular face elements is calculated. When the angle of the external normal vector is less than or equal to the set threshold, it can be considered that the two co-edge triangular face elements are on the same plane; when the angle of the external normal vector is greater than the set threshold, the two face elements are on different planes. Based on the above judgment, all edges in the target can be found and recorded with the two endpoints of the edge.

[0126] According to the above-mentioned edge information, the target is regionally decomposed. Specifically, starting from the first triangular face element in the model's mesh, judgments are made one by one. When a triangular face element has no intersection with an edge, the face element is a face element on a certain face, the number of the triangle is recorded, and according to the neighborhood information of the face element, the number of the triangular face element adjacent to it is found, and the number of intersections between the face element and the edge is continued to be judged. When the number of intersections between a triangular face element and an edge is 1, the triangular face element has an edge located on the scattering edge, and further judgment of the triangular face elements adjacent to the edge is stopped. When the intersection of a triangular face element and an edge is 2, the triangular face element is located at the junction of two edges, and further judgment of the triangular face elements adjacent to the two edges is stopped. When all judgments are stopped, and all triangular face elements have not been traversed once at this time, it means that the division of a region has been completed, and the same operation is performed on the remaining triangular face elements. After performing the above operations on all triangular face elements, the target is divided into multiple regions.

[0127] In a preferred embodiment, by judging the flatness of the triangular face elements in each area, the area can be divided into a plane area and a curved surface area. First, the normal vector of each triangular face element is calculated, and a set of normal vectors of all triangular face elements in the area is constructed. The standard deviation of the set is calculated. When the standard deviation of the set is greater than a set threshold, the area is determined to be a curved surface area, otherwise, it is determined to be a plane area. Due to the irregularity of the surface and the randomness of the target area division, it is difficult to extract its geometric feature parameters such as curvature. The present invention uses a convolutional neural network to extract the geometric feature parameters of the surface structure unit.

[0128] Step S2: constructing a convolutional neural network that maps surface structure units to geometric feature parameters;

[0129] Specifically, a convolutional neural network is used to process the surface structure units and estimate their geometric features such as curvature parameters.

[0130] Step S2.1: Construction of training dataset.

[0131] First, the data format needs to be determined. For each divided area, its pixel points are assigned a normalized height value. The height value refers to the height obtained by laying the decomposed structural unit on the plane according to the observation angle.

[0132] Step S2.2: Convolutional neural network model construction.

[0133] like Figure 4As shown in the figure, the convolutional neural network contains three cascaded convolutional layers and pooling layers and three cascaded fully connected layers. The convolutional layer is used to extract features from two-dimensional images. The fully connected layer mainly maps feature vectors to geometric parameters. The input layer of the convolutional neural network is a height map after the two-dimensional surface is tiled, which has only one channel. The nonlinear activation function of the convolutional layer (i.e. Figure 4 The activation function in ): ReLU, the pooling method of the pooling layer is maximum pooling, and the nonlinear activation function of the fully connected layer is ReLU.

[0134] Step S2.3: Convolutional neural network model training.

[0135] like Figure 5 As shown in the figure, the training flow chart of the convolutional neural network is given. First, the training samples are generated. This generation step can be regarded as the reverse process of the convolutional neural network, setting the geometric parameters of the structural unit, such as the curvature of the surface, and then generating the corresponding structural unit according to the curvature. In order to ensure that the convolutional neural network can obtain sufficient training sample data, it is also necessary to perform data enhancement operations such as rotation, translation, and scaling on the generated structural units, and finally convert them into a two-dimensional image form to meet the format requirements of the convolutional neural network. Since the purpose of this convolutional neural network is to estimate the attribute scattering center parameters, the loss function is set to the MSE error function, as shown below:

[0136]

[0137] Among them, J represents the loss function;

[0138] MSE() represents mean square error function;

[0139] m represents the number of training samples;

[0140] y i Represents the true value of the i-th training sample;

[0141] x i represents the i-th input variable;

[0142] f() represents the prediction result of the neural network model;

[0143] The current training algorithm for convolutional neural networks is mainly the gradient descent method, which can be subdivided into the batch gradient descent (BGD), stochastic gradient descent (SGD), mini batch gradient descent (MBGD), and adaptive momentum (Adam) algorithm. Among them, the Adam algorithm designs a special adaptive learning rate for different parameters through the first-order moment estimation and second-order moment estimation of the gradient, achieving better training results, and is widely used as an optimization algorithm for deep learning.

[0144] Step S2.4: Use the trained convolutional neural network model to extract the geometric feature parameters of the surface structure unit. Assign the pixel points of the surface structure unit obtained by structural analysis to the normalized height value, and the height value is the height obtained by tiling on the plane according to the set observation angle, to obtain a two-dimensional array. Use the two-dimensional array as the input of the convolutional neural network, and obtain the structural feature parameters of the surface structure unit, such as curvature, normal vector, etc., through feature extraction and full connection layer operation;

[0145] Step S3: constructing a mapping relationship between the structural unit geometric characteristic parameter mapping and the scattering center parameters.

[0146] Specifically, the geometric characteristic parameters of the structural unit are mapped to the scattering center parameters, so as to finally extract the target scattering center.

[0147] like Figure 2 As shown, after obtaining the geometric characteristic parameters of the structural unit, a parameterized scattering center model can be constructed based on the mapping relationship between the geometric characteristic parameters and the scattering center parameters. The single curved surface can be calculated by the following formula:

[0148]

[0149] Among them, E h Represents the geometric characteristic parameters of a single curved surface;

[0150] r represents the radius of curvature;

[0151] k represents the wave number;

[0152] represents the position vector of the centroid of a single curved surface;

[0153] Indicates the radar observation line of sight angle.

[0154] For a cylindrical single curved surface, it can be calculated by the following formula:

[0155]

[0156] Among them, E cylinder Represents the geometric characteristic parameters of a cylindrical single curved surface;

[0157] λ represents wavelength;

[0158] f represents frequency;

[0159] c represents the speed of light;

[0160] Represents the direction vector of the cylinder generatrix;

[0161] l represents the busbar length;

[0162] θ represents the radar observation angle.

[0163] For a conical single curved surface, it can be calculated by the following formula:

[0164]

[0165] Among them, E conel Represents the geometric characteristic parameters of the conical single curved surface;

[0166] The subscripts max and min represent the maximum and minimum values, respectively;

[0167] γ represents the cone normal vector.

[0168] For a plane, it can be calculated by the following formula:

[0169]

[0170] Among them, E plane Represents the geometric characteristic parameters of the plane;

[0171] N represents the number of polygon sides;

[0172] Represented as the centroid vector of the polygon;

[0173] Expressed as the centroid vector in the polygon normal The projection vector of

[0174] It is expressed as the projection vector of the centroid vector on the horizontal plane xyp of the polygon;

[0175] Represents the midpoint vector of the nth edge.

[0176] After obtaining the scattering fields of all scattering centers of the target, the overall scattering field of the target can be obtained by superposition.

[0177] The present invention also provides a scattering center parameter extraction system based on structure analysis. The scattering center parameter extraction system based on structure analysis can be implemented by executing the process steps of the scattering center parameter extraction method based on structure analysis, that is, those skilled in the art can understand the scattering center parameter extraction method based on structure analysis as a preferred implementation of the scattering center parameter extraction system based on structure analysis.

[0178] Those skilled in the art know that, in addition to realizing the system and its various devices, modules, and units provided by the present invention in a purely computer-readable program code, it is entirely possible to realize the same functions in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered as a hardware component, and the devices, modules, and units included therein for realizing various functions can also be regarded as structures within the hardware component; the devices, modules, and units for realizing various functions can also be regarded as both software modules for realizing the method and structures within the hardware component.

[0179] A scattering center parameter extraction system based on structure analysis provided by the present invention comprises:

[0180] Module M1: Perform structural analysis on the selected target grid file to obtain a set of structural units;

[0181] Module M2: Mapping the structural units in the structural unit set into geometric feature parameters;

[0182] Module M3: Map the geometric characteristic parameters of the structural unit to the scattering center parameters to obtain the target scattering center.

[0183] Specifically, in module M1, the target grid file is structurally analyzed, edges are identified, regions are divided, and the structure is decomposed into a set of structural units consisting of plane structural units and curved surface structural units, including:

[0184] Module M1.1: Read the target grid file, use the triangulation information and point coordinate information in the grid file to obtain triangular face elements, and calculate the external normal vector direction and center coordinates of all obtained triangular face elements;

[0185] Module M1.2: Determine the edge according to the direction of the external normal vector and the center coordinates of the triangle surface element, and record the edge information;

[0186] Module M1.3: Divide the target into regions based on edge information and obtain plane structure units and curved surface structure units.

[0187] Specifically, in module M1.2, the angle between the external normal vectors of a triangle element and its adjacent triangle element is calculated. If the angle between the external normal vectors is greater than a set threshold, the two elements are on different planes and the common edge is an edge. The edge is recorded by using the two endpoints of the edge.

[0188] If the angle between the external normal vectors is less than or equal to the set threshold, the two face elements are on the same face;

[0189] The module M1.3 includes:

[0190] When a triangular face element has no intersection with an edge, the face element is a face element on a certain face. The number of the triangle is recorded, and the numbers of the triangular face elements adjacent to it are found based on the neighborhood information of the face element, and the number of intersections between the face element and the edge is further determined.

[0191] When the number of intersections between a triangle element and an edge is 1, the triangle element has an edge on the scattering edge, and further judgment of the triangle elements adjacent to the edge is stopped;

[0192] When the intersection of a triangle element and an edge is 2, the triangle element is located at the intersection of two edges;

[0193] According to the relationship between triangular face elements and edges, the target is divided into multiple areas;

[0194] According to the flatness of the triangular face elements in each region, the region is divided into a plane region constituting a plane structural unit and a curved surface region constituting a curved surface structural unit.

[0195] Specifically, in module M2, the geometric features of the surface structural unit are estimated, a convolutional neural network is constructed, and the convolutional neural network is used to process the set of structural units to obtain geometric feature parameters, including:

[0196] Module M2.1: Construct a training data set, assign the pixel points of the surface structure unit to the normalized height value, where the height value is the height obtained by tiling on the plane according to the set observation angle;

[0197] Module M2.2: Construct a convolutional neural network model, wherein the convolutional neural network includes three cascaded convolutional layers and pooling layers and three cascaded fully connected layers, the input layer is a height map after the set of structural units is tiled, the nonlinear activation function of the convolutional layer is a ReLU function; the pooling method of the pooling layer is maximum pooling; the nonlinear activation function of the fully connected layer is a ReLU function;

[0198] Module M2.3: Generate training samples, train the convolutional neural network model, and set the loss function to the MSE error function, that is, Obtain the trained convolutional neural network model;

[0199] Among them, J represents the loss function;

[0200] MSE() represents mean square error function;

[0201] m represents the number of training samples;

[0202] y i Represents the true value of the i-th training sample;

[0203] x i represents the i-th input variable;

[0204] f() represents the prediction result of the neural network model;

[0205] Module M2.4: Use the trained convolutional neural network model to extract the geometric feature parameters of the surface structure unit. Assign the pixel points of the surface structure unit obtained by structural analysis to the normalized height value, which is the height obtained by tiling on the plane according to the set observation angle, to obtain a two-dimensional array. Use this two-dimensional array as the input of the convolutional neural network, and obtain the structural parameters of the surface structure unit, such as curvature, normal vector, etc., through feature extraction and full connection layer operation;

[0206] Specifically, in the module M3, a parameterized scattering center model is constructed based on the mapping relationship between the geometric characteristic parameters and the scattering center parameters, and the single surface parameter calculation method is: The calculation method of cylindrical single curved surface parameters is: The conical single curved surface parameters are calculated as follows: The plane parameters are calculated as follows:

[0207] Among them, E h Represents the geometric characteristic parameters of a single curved surface;

[0208] r represents the radius of curvature;

[0209] k represents the wave number;

[0210] represents the position vector of the centroid of a single curved surface;

[0211] Indicates the radar observation line of sight angle;

[0212] E cylinder Represents the geometric characteristic parameters of a cylindrical single curved surface;

[0213] λ represents wavelength;

[0214] f represents frequency;

[0215] c represents the speed of light;

[0216] Represents the direction vector of the cylinder generatrix;

[0217] l represents the busbar length;

[0218] θ represents the radar observation angle;

[0219] E conel Represents the geometric characteristic parameters of the conical single curved surface;

[0220] The subscripts max and min represent the maximum and minimum values, respectively;

[0221] γ represents the cone normal vector;

[0222] E plane Represents the geometric characteristic parameters of the plane;

[0223] N represents the number of polygon sides;

[0224] represents the centroid vector of the polygon;

[0225] Expressed as the centroid vector in the polygon normal The projection vector of

[0226] It is expressed as the projection vector of the centroid vector on the horizontal plane xyp of the polygon;

[0227] represents the midpoint vector of the nth edge;

[0228] The geometric characteristic parameters of the structural unit are mapped to the scattering center parameters. According to the scattering field superposition of the target scattering center, the overall scattering field of the target is obtained and the target scattering center is extracted.

[0229] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A method for extracting scattering center parameters based on structural analysis, characterized in that: include: Step S1: Perform structural analysis based on the selected target grid file to obtain a set of structural units; Step S2: mapping the structural units in the structural unit set into geometric feature parameters; Step S3: Mapping the geometric characteristic parameters of the structural unit into scattering center parameters to obtain the target scattering center.

2. The method for extracting scattering center parameters based on structural analysis according to claim 1, characterized in that: In step S1, the target grid file is structurally analyzed to identify edges, divide the area according to the edges, and decompose it into a set of structural units consisting of plane structural units and curved surface structural units, including: Step S1.1: read the target mesh file, use the triangulation information and point coordinate information in the mesh file to obtain triangular face elements, and calculate the external normal vector directions and center coordinates of all obtained triangular face elements; Step S1.2: Determine the edge according to the direction of the external normal vector and the center coordinates of the triangle surface element, and record the edge information; Step S1.3: According to the edge information, the target is divided into regions, and plane structure units and curved surface structure units are obtained.

3. The method for extracting scattering center parameters based on structural analysis according to claim 2, characterized in that: In step S1.2, the angle between the external normal vectors of the triangular face element and its adjacent triangular face element is calculated. If the angle between the external normal vectors is greater than a set threshold, the two face elements are on different planes and the common edge is an edge. The edge is recorded by numbering the two endpoints of the edge. When the angle between the external normal vectors is less than or equal to the set threshold, the two face elements are on the same plane; The step S1.3 comprises: When a triangular face element has no intersection with an edge, the face element is a face element on a certain face. The number of the triangle is recorded, and the numbers of the triangular face elements adjacent to it are found based on the neighborhood information of the face element, and the number of intersections between the face element and the edge is further determined. When the number of intersections between a triangle element and an edge is 1, the triangle element has an edge on the scattering edge, and further judgment of the triangle elements adjacent to the edge is stopped; When the intersection of a triangle element and an edge is 2, the triangle element is located at the intersection of two edges; According to the relationship between triangular face elements and edges, the target is divided into multiple areas; According to the flatness of the triangular face elements in each region, the region is divided into a plane region constituting a plane structural unit and a curved surface region constituting a curved surface structural unit.

4. The method for extracting scattering center parameters based on structural analysis according to claim 1, characterized in that: In step S2, the geometric features of the surface structure unit are estimated, a convolutional neural network is constructed, and the surface structure unit is processed using the convolutional neural network to obtain geometric feature parameters, including: Step S2.1: construct a training data set, assign the pixel points of the surface structure unit to the normalized height value, and the height value is the height obtained by tiling on the plane according to the set observation angle; Step S2.2: construct a convolutional neural network model, wherein the convolutional neural network comprises three cascaded convolutional layers and pooling layers and three cascaded fully connected layers, the input layer is a height map after the set of structural units is tiled, the nonlinear activation function of the convolutional layer is a ReLU function; the pooling method of the pooling layer is maximum pooling; the nonlinear activation function of the fully connected layer is a ReLU function; Step S2.3: Generate training samples, train the convolutional neural network model, and set the loss function to the MSE error function, that is, Obtain the trained convolutional neural network model; Among them, J represents the loss function; MSE( ) represents mean square error function; m represents the number of training samples; y i Represents the true value of the i-th training sample; x i represents the i-th input variable; f( ) represents the prediction result of the neural network model; Step S2.4: Use the trained convolutional neural network model to extract the geometric feature parameters of the surface structure unit; assign the pixel points of the surface structure unit obtained by structural analysis to the normalized height value, and the height value is the height obtained by tiling on the plane according to the set observation angle, to obtain a two-dimensional array. The two-dimensional array is used as the input of the convolutional neural network, and the structural feature parameters of the surface structure unit are obtained through feature extraction and full connection layer operation.

5. The method for extracting scattering center parameters based on structural analysis according to claim 1, characterized in that: In step S3, a parameterized scattering center model is constructed based on the mapping relationship between the geometric feature parameters and the scattering center parameters. The single surface parameter calculation method is: The calculation method of cylindrical single curved surface parameters is: The conical single surface parameters are calculated as follows: The plane parameters are calculated as follows: Among them, E h Represents the geometric characteristic parameters of a single curved surface; r represents the radius of curvature; k represents the wave number; represents the position vector of the centroid of a single curved surface; Indicates the radar observation line of sight angle; E cylinder Represents the geometric characteristic parameters of a cylindrical single curved surface; λ represents wavelength; f represents frequency; c represents the speed of light; Represents the direction vector of the cylinder generatrix; l represents the busbar length; θ represents the radar observation angle; E conel Represents the geometric characteristic parameters of the conical single curved surface; The subscripts max and min represent the maximum and minimum values, respectively; γ represents the cone normal vector; E plane Represents the geometric characteristic parameters of the plane; N represents the number of polygon sides; represents the centroid vector of the polygon; Expressed as the centroid vector in the polygon normal The projection vector of It is expressed as the projection vector of the centroid vector on the horizontal plane xyp of the polygon; represents the midpoint vector of the nth edge; The geometric characteristic parameters of the structural unit are mapped to the scattering center parameters, the scattering fields of the target scattering centers are superimposed to obtain the overall scattering field of the target, and the target scattering centers are extracted.

6. A scattering center parameter extraction system based on structural analysis, characterized in that: include: Module M1: Perform structural analysis on the selected target grid file to obtain a set of structural units; Module M2: Mapping the structural units in the structural unit set into geometric feature parameters; Module M3: Map the geometric characteristic parameters of the structural unit to the scattering center parameters to obtain the target scattering center.

7. The scattering center parameter extraction system based on structure analysis according to claim 6, characterized in that: In module M1, the target grid file is structurally analyzed, edges are identified, regions are divided, and decomposed into a set of structural units consisting of plane structural units and curved surface structural units, including: Module M1.1: Read the target grid file, use the triangulation information and point coordinate information in the grid file to obtain triangular face elements, and calculate the external normal vector direction and center coordinates of all obtained triangular face elements; Module M1.2: Determine the edge according to the direction of the external normal vector and the center coordinates of the triangle surface element, and record the edge information; Module M1.3: Divide the target into regions based on edge information and obtain plane structure units and curved surface structure units.

8. The method for extracting scattering center parameters based on structural analysis according to claim 7, characterized in that: In module M1.2, the angle between the external normal vectors of a triangle element and its adjacent triangle element is calculated. If the angle between the external normal vectors is greater than a set threshold, the two elements are on different planes and the common edge is an edge. The edge is recorded by using the two endpoints of the edge. If the angle between the external normal vectors is less than or equal to the set threshold, the two face elements are on the same face; The module M1.3 includes: When a triangular face element has no intersection with an edge, the face element is a face element on a certain face. The number of the triangle is recorded, and the numbers of the triangular face elements adjacent to it are found based on the neighborhood information of the face element, and the number of intersections between the face element and the edge is further determined. When the number of intersections between a triangle element and an edge is 1, the triangle element has an edge on the scattering edge, and further judgment of the triangle elements adjacent to the edge is stopped; When the intersection of a triangle element and an edge is 2, the triangle element is located at the intersection of two edges; According to the relationship between triangular face elements and edges, the target is divided into multiple areas; According to the flatness of the triangular face elements in each region, the region is divided into a plane region constituting a plane structural unit and a curved surface region constituting a curved surface structural unit.

9. The method for extracting scattering center parameters based on structural analysis according to claim 6, characterized in that: In module M2, the geometric features of the surface structural unit are estimated, a convolutional neural network is constructed, and the convolutional neural network is used to process the set of structural units to obtain geometric feature parameters, including: Module M2.1: Construct a training data set, assign the pixel points of the surface structure unit to the normalized height value, where the height value is the height obtained by tiling on the plane according to the set observation angle; Module M2.2: Construct a convolutional neural network model, wherein the convolutional neural network includes three cascaded convolutional layers and pooling layers and three cascaded fully connected layers, the input layer is a height map after the set of structural units is tiled, the nonlinear activation function of the convolutional layer is a ReLU function; the pooling method of the pooling layer is maximum pooling; the nonlinear activation function of the fully connected layer is a ReLU function; Module M2.3: Generate training samples, train the convolutional neural network model, and set the loss function to the MSE error function, that is, Obtain the trained convolutional neural network model; Among them, J represents the loss function; MSE() represents mean square error function; m represents the number of training samples; y i Represents the true value of the i-th training sample; x i represents the i-th input variable; f() represents the prediction result of the neural network model; Module M2.4: Use the trained convolutional neural network model to extract the geometric feature parameters of the surface structure unit; assign the pixel points of the surface structure unit obtained by structural analysis to the normalized height value, which is the height obtained by tiling on the plane according to the set observation angle, to obtain a two-dimensional array. The two-dimensional array is used as the input of the convolutional neural network, and the structural feature parameters of the surface structure unit are obtained through feature extraction and full connection layer operation.

10. The method for extracting scattering center parameters based on structural analysis according to claim 6, characterized in that: In the module M3, a parameterized scattering center model is constructed based on the mapping relationship between geometric feature parameters and scattering center parameters. The single surface parameter calculation method is: The calculation method of cylindrical single curved surface parameters is: The conical single surface parameters are calculated as follows: The plane parameters are calculated as follows: Among them, E h Represents the geometric characteristic parameters of a single curved surface; r represents the radius of curvature; k represents the wave number; represents the position vector of the centroid of a single curved surface; Indicates the radar observation line of sight angle; E cylinder Represents the geometric characteristic parameters of a cylindrical single curved surface; λ represents wavelength; f represents frequency; c represents the speed of light; Represents the direction vector of the cylinder generatrix; l represents the busbar length; θ represents the radar observation angle; E conel Represents the geometric characteristic parameters of the conical single curved surface; The subscripts max and min represent the maximum and minimum values, respectively; γ represents the cone normal vector; E plane Represents the geometric characteristic parameters of the plane; N represents the number of polygon sides; represents the centroid vector of the polygon; Expressed as the centroid vector in the polygon normal The projection vector of It is expressed as the projection vector of the centroid vector on the horizontal plane xyp of the polygon; represents the midpoint vector of the nth edge; The geometric characteristic parameters of the structural unit are mapped to the scattering center parameters. According to the scattering field superposition of the target scattering center, the overall scattering field of the target is obtained and the target scattering center is extracted.

Images