A translationally invariant structure electromagnetic scattering feature separation method

Abstract

The application discloses a translation-invariant structure electromagnetic scattering feature separation method, constructs an electromagnetic feature scattering separation mathematical model by taking a complex target electromagnetic scattering feature complex image as basic data, realizes position estimation of each structure scattering feature in the complex target by using a target detection technology in deep learning, constructs a translation-invariant structure scattering feature separation equation by using the estimated position, and realizes electromagnetic scattering feature separation of each structure in the complex target. The method of the application uses the target detection technology in deep learning, can avoid complex operations such as large-scale matrix inversion, improves the efficiency and reliability of the algorithm, directly uses a typical structure scattering complex image of a target to construct a separation model compared with the prior art, can more accurately describe a complex target electromagnetic scattering process, improves scattering feature separation precision, and is used for complex target scattering feature measurement, radar image target fine identification and the like.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of electromagnetic scattering measurement, and particularly relates to a translation invariant structure electromagnetic scattering feature separation method. BACKGROUND

[0002] Electromagnetic scattering measurement technology is one of the important problems in the field of electromagnetic compatibility measurement, and is of great significance for improving electronic devices, antennas, and improving people's understanding of the electromagnetic scattering characteristics of complex targets. The literature "Zhang Lei. Complex target electromagnetic scattering center modeling and its application in SAR target interpretation and identification. Wuhan University, 2019." and the literature "Jiang Wen, Li Wangzhe. New method of attribute scattering center parameter estimation based on amplitude and phase separation. Journal of Radar, 2019, 8(05)" study the scattering center separation technology and apply it to the field of radar target identification

[0003] At present, the research ideas of electromagnetic scattering center parameter estimation mainly adopt the sparse reconstruction method, that is, by constructing the observation equation of the scattering measurement process, a linear equation set between the scattering center and the scattering measurement image is constructed, and the estimation of the scattering center is obtained by solving the linear equation set. However, the observation equation construction of this method is difficult, and the model solving process is based on the high-frequency scattering center model assumption, which limits the accuracy of the scattering center estimation. In addition, when dealing with larger scenes, this method needs to solve the stable inverse of large-dimensional matrix and other problems, which leads to low calculation efficiency, unstable algorithm and other problems. SUMMARY

[0004] To solve the above technical problems, the application provides a translation invariant structure electromagnetic scattering feature separation method, which constructs an electromagnetic feature scattering separation mathematical model based on the electromagnetic scattering feature complex image of a typical structural part, uses target detection technology in deep learning to estimate the position of the scattering feature of each structural part in a complex target, and uses the estimated position to construct a translation invariant structure scattering feature separation equation to separate the electromagnetic scattering features of each structural part in the complex target.

[0005] The technical scheme adopted by the application is as follows: a translation invariant structure electromagnetic scattering feature separation method, the specific steps are as follows:

[0006] Step 1, obtaining the electromagnetic scattering feature complex image of the measured part;

[0007] The electromagnetic scattering feature original complex data of the measured part at different angles is obtained by using the darkroom turntable measurement system, and the electromagnetic scattering feature original complex data of the measured part in the specified angle range is imaged by using the standard turntable imaging algorithm to obtain the electromagnetic scattering feature complex image of the measured part.

[0008] Step 2, structure component electromagnetic scattering feature library construction;

[0009] n typical structural components are placed on a turntable in a darkroom, and the electromagnetic scattering characteristic complex images P1, P2...P of the n typical structural components are obtained using the method in step 1. n By analyzing the complex images P1, P2...P of the electromagnetic scattering characteristics of typical structural components... n By performing a translation transformation to augment the data, a database S of electromagnetic scattering features of structural components can be obtained. The input variables of S are complex images P1, P2...P1 containing electromagnetic scattering features of typical structural components at different locations. n The complex image s of the electromagnetic scattering characteristics of the target under test is labeled with the center positions T1, T2...T of the electromagnetic scattering characteristics of typical structural components in the complex image of the target under test. n .

[0010] Step 3: Construct an electromagnetic scattering feature center estimation network;

[0011] Using a standard target detection network, the electromagnetic scattering feature library S obtained in step 2 is used to train the network, resulting in a correctly trained electromagnetic scattering feature center estimation network.

[0012] The complex image of the electromagnetic scattering features of the test component from step 1 is imported into the electromagnetic scattering feature center estimation network to obtain the number n of typical structural component targets in the complex image of the test component's electromagnetic scattering features and the electromagnetic scattering feature center positions t1, t2, ... t of each typical structural component target. n .

[0013] Step 4: Construct a complex image set of electromagnetic scattering features containing the electromagnetic scattering characteristics of typical structural components;

[0014] Using the electromagnetic scattering feature complex image of the target obtained in step 3, the center positions t1, t2...t of the electromagnetic scattering features of typical structural components of the target are determined. n The original complex data D1, D2...D of the electromagnetic scattering characteristics of the corresponding typical structural components are respectively... n Perform two-dimensional convolution operations to obtain a complex image A of the electromagnetic scattering characteristics of a typical structural component. i ,Right now Then, a complex electromagnetic scattering feature image group A is obtained, which contains a complex image of electromagnetic scattering features of typical structural components.

[0015] in, This represents the convolution of a two-dimensional image.

[0016] Step 5: Construct the electromagnetic scattering feature reconstruction matrix of typical structural component targets;

[0017] Using the electromagnetic scattering characteristic complex image group A of the typical structural component electromagnetic scattering characteristic complex image obtained in step 4, all complex images A1, A2…A n in the complex image group are traversed, each complex image is multiplied by the conjugate image of all complex images in the complex image group respectively, the multiplication result is taken as an element of the electromagnetic scattering characteristic reconstruction matrix G of the typical structural component target, that is, G(i,j)=A i ×conj(A j ), and the electromagnetic scattering characteristic reconstruction matrix G of the typical structural component target is obtained.

[0018] Wherein, conj() represents the conjugate of a complex number.

[0019] Step 6, calculate the electromagnetic scattering characteristic coefficient vector of different structural components;

[0020] Using the electromagnetic scattering characteristic complex image group A of the typical structural component electromagnetic scattering characteristic complex image obtained in step 4, all complex images A1, A2…A n , in the complex image group are traversed, and the conjugate image of the electromagnetic scattering characteristic complex image s of the measured target containing the electromagnetic scattering characteristic complex image P1, P2…P n of the typical structural component electromagnetic scattering characteristic at different positions in the structural component electromagnetic scattering characteristic library S obtained in step 2 is multiplied, and the multiplication result is taken as an element of the electromagnetic scattering characteristic coefficient vector Y of different structural components, that is, Y(i)=A i ×conj(s), and the electromagnetic scattering characteristic coefficient vector Y of different structural components is obtained.

[0021] Step 7, calculate the optimization weight coefficient vector;

[0022] Using the inverse matrix G -1 of the electromagnetic scattering characteristic reconstruction matrix G of the typical structural component target obtained in step 5, the electromagnetic scattering characteristic coefficient vector Y of different structural components obtained in step 6 is multiplied, and the multiplication result is taken as the conjugate, that is, W=conj(G -1 ×Y), and the optimization weight coefficient vector W is obtained.

[0023] Step 8, electromagnetic scattering characteristic separation of different structural components;

[0024] Using the optimization weight coefficient vector W obtained in step 7, the electromagnetic scattering characteristic complex image P1, P2…P n of the typical structural component obtained in step 2 is multiplied, that is, P i ’=W(i)×P i , and the image P1’, P2’…P nThe electromagnetic scattering characteristic complex image of the typical structure component in the complex target is obtained, and electromagnetic scattering characteristic separation of the typical structure component in the complex target is realized.

[0025] The method of the present application uses the electromagnetic scattering characteristic complex image of the typical structure component as the basic data to construct the electromagnetic characteristic scattering separation mathematical model, uses the target detection technology in deep learning to realize the position estimation of the scattering characteristics of each structure component in the complex target, uses the estimated position to construct the translation invariant structure scattering characteristic separation equation, and realizes the electromagnetic scattering characteristic separation of each structure component in the complex target. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 The flow chart of the translation invariant structure electromagnetic scattering characteristic separation method of the present application.

[0027] Figure 2 The electromagnetic scattering characteristic complex image of the typical structure component target ball 1 in the embodiment of the present application.

[0028] Figure 3 The electromagnetic scattering characteristic complex image of the typical structure component target ball 2 in the embodiment of the present application.

[0029] Figure 4 The electromagnetic scattering characteristic complex image of the typical structure component target water droplet in the embodiment of the present application.

[0030] Figure 5 The electromagnetic scattering characteristic complex image of the measured target containing the electromagnetic scattering characteristic complex image of the typical structure component target in the embodiment of the present application.

[0031] Figure 6 The electromagnetic scattering characteristic complex image of the separated typical structure component target ball 1 in the embodiment of the present application.

[0032] Figure 7 The electromagnetic scattering characteristic complex image of the separated typical structure component target ball 2 in the embodiment of the present application.

[0033] Figure 8 The electromagnetic scattering characteristic complex image of the separated typical structure component target water droplet in the embodiment of the present application. DETAILED DESCRIPTION

[0034] To facilitate the explanation of the method of this invention, some definitions involved in this invention are explained as follows:

[0035] 1. Darkroom turntable measurement system:

[0036] An anechoic chamber turntable measurement system refers to an electronic device with functions such as radio frequency front-end, digital down-conversion, and data acquisition and transmission. Its components include: a turntable, the device under test (DUT), a measuring antenna, electronic measuring equipment, and monitoring software.

[0037] The turntable consists of a mechanical system and a servo system. The servo system controls the azimuth and pitch rotation of the device under test (DUT) relative to the measuring antenna. The DUT is a target used to analyze electromagnetic characteristics, such as a metal sphere, antenna, aircraft model, or vehicle model. Based on the signal transmission direction, measuring antennas can be divided into two types: omnidirectional antennas and directional antennas. Omnidirectional antennas can transmit or receive signals in all 360 degrees, while directional antennas transmit or receive signals in only one direction. Electronic measuring equipment is used to generate test signals and collect radio frequency energy distribution data corresponding to a specific location. Monitoring software includes a human-machine interface for displaying, storing, and analyzing test results.

[0038] 2. Standard turntable imaging algorithm:

[0039] The standard turntable imaging algorithm is a method that performs coherent processing on the turntable measurement data to obtain the spatial distribution of the scattering characteristics of the target under test.

[0040] The standard turntable imaging algorithm takes the original complex data of the electromagnetic scattering characteristics of the target as input and outputs a complex image of the electromagnetic scattering characteristics of the target. The common turntable imaging algorithm is the convolution-back projection algorithm.

[0041] 3. Typical structural components:

[0042] Typical structural components refer to typical structures on the target being measured, such as: isolated scattering points, water droplet structures, angular reflection structures, spiral structures, wedges, and cavities.

[0043] 4. Data augmentation:

[0044] Data augmentation refers to the method of generating new training samples using existing data through operations such as image transformation during deep learning, such as translation augmentation and rotation augmentation.

[0045] 5. Standard target detection network:

[0046] Standard deep learning object detection algorithms consist of two stages: region proposal and feature extraction. The region proposal stage employs a selective search strategy, generating a large number of bounding boxes as candidate regions on the input image. The feature extraction stage then extracts features from each candidate region. These features are used by a classifier to determine which class the candidate region belongs to, outputting a confidence score for each class. The input is a complex image of the electromagnetic scattering features of the target object, containing a complex image of the electromagnetic scattering features of typical structural components. The output is the feature centers of the electromagnetic scattering features of the typical structural components within the complex image of the target object's electromagnetic scattering features. Common object detection networks include CNN, SPP-Net, Fast R-CNN, Faster R-CNN, R-FCN, YOLO, and SSD. Standard object detection networks require training with samples to achieve object detection.

[0047] 6. Two-dimensional convolution algorithm:

[0048] Two-dimensional convolution operations are commonly used in computer vision and image processing. By performing convolution operations with different convolution kernels, different feature information of the original image can be extracted.

[0049] In image processing tasks, convolutional layers extract specific feature information by sliding a small matrix called a kernel or filter across the input image. In a two-dimensional convolutional layer, the kernel is a two-dimensional matrix that is multiplied and summed element-wise with the input image to obtain the output feature map.

[0050] The method of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0051] like Figure 1 The flowchart of the translation-invariant structure electromagnetic scattering feature separation method of the present invention is shown below, and the specific steps are as follows:

[0052] Step 1: Obtain a complex image of the electromagnetic scattering characteristics of the test piece;

[0053] In this embodiment, the original complex data of electromagnetic scattering characteristics of the test piece at the horizontal angle is obtained using a darkroom turntable measurement system. The original complex data of electromagnetic scattering characteristics of the test piece at the horizontal angle is then imaged using a standard turntable imaging algorithm to obtain a complex image of the electromagnetic scattering characteristics of the test piece.

[0054] Step 2: Establish a library of electromagnetic scattering characteristics of structural components;

[0055] like Figure 2 , Figure 3 , Figure 4As shown, in this embodiment, typical structural components sphere 1, sphere 2, and a water droplet are placed on a turntable in a darkroom. Using the method in step 1, complex images P1, P2, and P3 of the electromagnetic scattering characteristics of the typical structural components are obtained. Data amplification is performed by shifting these complex images P1, P2, and P3 to obtain a library S of the electromagnetic scattering characteristics of the structural components. Figure 5 As shown, the input variable of S is the electromagnetic scattering feature complex image s of P1, P2, and P3 containing typical structural components sphere 1, sphere 2, and water droplet at different locations, and the label is the position T1, T2, T3 of the electromagnetic scattering feature of the typical structural components in the electromagnetic scattering feature complex image of the target under test.

[0056] Step 3: Construct an electromagnetic scattering feature center estimation network;

[0057] Using a standard target detection network, the electromagnetic scattering feature library S obtained in step 2 is used to train the network, resulting in a correctly trained electromagnetic scattering feature center estimation network.

[0058] The complex image of electromagnetic scattering features of the test component in step 1 is imported into the electromagnetic scattering feature center estimation network to obtain the number 3 of typical structural component targets in the complex image of electromagnetic scattering features of the test component and the electromagnetic scattering feature center positions t1, t2, t3 of each typical structural component target.

[0059] Step 4: Construct a complex image set of electromagnetic scattering features containing the electromagnetic scattering characteristics of typical structural components;

[0060] Using the electromagnetic scattering feature center positions t1, t2, and t3 of the typical structural components sphere 1, sphere 2, and water droplet in the electromagnetic scattering feature complex image of the target obtained in step 3, perform two-dimensional convolution operations with the original complex data D1, D2, and D3 of the electromagnetic scattering characteristics of the corresponding typical structural components sphere 1, sphere 2, and water droplet to obtain the electromagnetic scattering feature complex image A of the typical structural components. i ,Right now Then, a complex image group A containing the electromagnetic scattering features of typical structural components is obtained. A contains three complex images A1, A2, and A3.

[0061] in, This represents the convolution of a two-dimensional image.

[0062] Step 5: Construct the electromagnetic scattering feature reconstruction matrix of typical structural component targets;

[0063] Using the complex electromagnetic scattering feature image group A obtained in step 4, which contains complex images of electromagnetic scattering features of typical structural components, all complex images A1, A2, and A3 in complex image group A are traversed. Each complex image is multiplied by the conjugate image of all complex images in complex image group A. The result of the multiplication is used as an element of the electromagnetic scattering feature reconstruction matrix G of the typical structural component target, that is, G(1,1) = A1 × conj(A1), G(1,2) = A1 × conj(A2), G(1,3) = A1 × conj(A3), G(2,1) = A2 × conj(A1)……G(3,3) = A3 × conj(A3), thus obtaining the electromagnetic scattering feature reconstruction matrix G of the typical structural component target.

[0064] Here, conj() represents the conjugate of a complex number.

[0065] Step 6: Calculate the electromagnetic scattering characteristic coefficient vectors of different structural components;

[0066] Using the electromagnetic scattering feature complex image group A of the typical structural component electromagnetic scattering feature complex image obtained in step 4, traverse all complex images A1, A2, A3 in the complex image group, and multiply them with the conjugate images of the electromagnetic scattering feature complex images s of the target electromagnetic scattering feature complex images P1, P2, P3 of the typical structural component electromagnetic scattering feature complex images at different positions in the structural component electromagnetic scattering feature library S obtained in step 2. The result of the multiplication is used as an element of the electromagnetic scattering feature coefficient vector Y of different structural components, that is, Y(1)=A1×conj(s), Y(2)=A2×conj(s), Y(3)=A3×conj(s), to obtain the electromagnetic scattering feature coefficient vector Y of different structural components.

[0067] Step 7: Calculate the optimal weight coefficient vector;

[0068] The inverse matrix G of the reconstruction matrix G using the electromagnetic scattering characteristics of typical structural components obtained in step 5 is... -1 Multiply the result by the electromagnetic scattering characteristic coefficient vector Y of different structural components obtained in step 6, and take the conjugate of the result, i.e., W = conj(G -1 ×Y), to obtain the optimal weight coefficient vector W.

[0069] Step 8: Separation of electromagnetic scattering characteristics of different structural components

[0070] like Figure 6 , Figure 7 , Figure 8 As shown, the optimized weight coefficient vector W obtained in step 7 is multiplied by the complex electromagnetic scattering feature images P1, P2, and P3 of the typical structural components obtained in step 2, i.e., P i '=W(i)×P iImages P1', P2', and P3' of targets with different typical structural components are obtained, and electromagnetic scattering characteristic intensity images of typical structural components in complex targets are obtained, thus realizing the separation of typical electromagnetic scattering characteristics of complex targets.

[0071] In summary, the method of this invention utilizes target detection technology in deep learning to estimate the location of scattering features and combines it with sparse reconstruction methods to separate multiple scattering features. Employing target detection technology in deep learning avoids complex operations such as large-scale matrix inversion, improving the efficiency and reliability of the algorithm. Compared with existing technologies, directly using the complex scattering image of a typical target structure to construct a separation model can more accurately describe the electromagnetic scattering process of complex targets, improve the accuracy of scattering feature separation, and is applicable to tasks such as complex target scattering feature measurement and refined target identification in radar images.

[0072] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Various modifications and variations can be made to the invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the scope of the claims of the invention.

Claims

1. A method for separating electromagnetic scattering features of translation-invariant structures, the specific steps of which are as follows: Step 1: Obtain a complex image of the electromagnetic scattering characteristics of the test piece; The original complex data of electromagnetic scattering characteristics of the test piece at different angles are obtained by using a darkroom turntable measurement system. The original complex data of electromagnetic scattering characteristics of the test piece within a specified angle range are imaged using a standard turntable imaging algorithm to obtain a complex image of electromagnetic scattering characteristics of the test piece. in, The standard turntable imaging algorithm is a method of obtaining the spatial distribution of the scattering characteristics of the target by coherently processing the turntable measurement data. The input of the standard turntable imaging algorithm is the original complex data of the electromagnetic scattering characteristics of the target, and the output is the complex image of the electromagnetic scattering characteristics of the target. The turntable imaging algorithm is a convolution-back projection algorithm. Step 2: Establish a database of electromagnetic scattering characteristics for structural components; n typical structural components are placed on a turntable in a darkroom, and the electromagnetic scattering characteristic complex images P1, P2...P of the n typical structural components are obtained using the method in step 1. n By analyzing the complex images P1, P2...P of the electromagnetic scattering characteristics of typical structural components... n By performing a translation transformation to augment the data, we can obtain the electromagnetic scattering feature library S of the structural components. The input variables of S are the complex images P1, P2...P1 of the electromagnetic scattering features of typical structural components at different locations. n The complex image s of the electromagnetic scattering characteristics of the target under test is labeled with the center positions T1, T2...T of the electromagnetic scattering characteristics of typical structural components in the complex image of the target under test. n ; Step 3: Construct an electromagnetic scattering feature center estimation network; A standard target detection network is used, and the electromagnetic scattering feature library S of the structural components obtained in step 2 is used to train the network to obtain a correctly trained electromagnetic scattering feature center estimation network. The standard target detection network takes as input a complex image of the electromagnetic scattering features of the target object, which includes a complex image of the electromagnetic scattering features of the target object with typical structural components, and outputs as the feature centers of the electromagnetic scattering features of the target object in the complex image of the electromagnetic scattering features of the target object with typical structural components. Target detection networks include: CNN, SPP-Net, Fast R-CNN, Faster R-CNN, R-FCN, YOLO, and SSD. The complex image of the electromagnetic scattering features of the test component from step 1 is imported into the electromagnetic scattering feature center estimation network to obtain the number n of typical structural component targets in the complex image of the test component's electromagnetic scattering features and the electromagnetic scattering feature center positions t1, t2, ... t of each typical structural component target. n ; Step 4: Construct a complex image set of electromagnetic scattering features containing the electromagnetic scattering characteristics of typical structural components; Using the electromagnetic scattering feature complex image of the target obtained in step 3, the center positions t1, t2...t of the electromagnetic scattering features of typical structural components of the target are determined. n The original complex data D1, D2...D of the electromagnetic scattering characteristics of the corresponding typical structural components are respectively... n Perform two-dimensional convolution operations to obtain a complex image A of the electromagnetic scattering characteristics of a typical structural component. i ,Right now Then, a complex image group A of electromagnetic scattering features containing typical structural components is obtained; in, Represents two-dimensional image convolution; Step 5: Construct the electromagnetic scattering feature reconstruction matrix of typical structural component targets; Using the complex image group A containing the electromagnetic scattering feature complex images of typical structural components obtained in step 4, traverse all complex images A1, A2...A1 in the complex image group. n Each complex image is multiplied by the conjugate image of all complex images in the complex image group. The result of the multiplication is used as an element of the electromagnetic scattering feature reconstruction matrix G of the typical structural component target, i.e., G(i,j)=A. i ×conj(A j ), thus obtaining the electromagnetic scattering feature reconstruction matrix G of a typical structural component target; Where conj() represents the conjugate of a complex number; Step 6: Calculate the electromagnetic scattering characteristic coefficient vectors of different structural components; Using the complex electromagnetic scattering feature image group A of the typical structural component obtained in step 4, traverse all complex images A1, A2...A1 in the complex image group. n The complex images P1, P2...P of typical structural component electromagnetic scattering features at different locations in the electromagnetic scattering feature library S obtained in step 2 are compared with those obtained in step 2. n The conjugate image of the complex electromagnetic scattering feature image s of the target under test is multiplied, and the result of the multiplication is used as an element of the electromagnetic scattering feature coefficient vector Y of different structural components, i.e., Y(i) = A i ×conj(s) yields the electromagnetic scattering characteristic coefficient vector Y for different structural components; Step 7: Calculate the optimal weight coefficient vector; The inverse matrix G of the reconstruction matrix G using the electromagnetic scattering characteristics of typical structural components obtained in step 5 is... -1 Multiply the result by the electromagnetic scattering characteristic coefficient vector Y of the different structural components obtained in step 6, and take the conjugate of the result, i.e., W = conj(G -1 ×Y), to obtain the optimal weight coefficient vector W; Step 8: Separation of electromagnetic scattering characteristics of different structural components; Using the optimized weight coefficient vector W obtained in step 7, and the complex electromagnetic scattering characteristic images P1, P2...P of typical structural components obtained in step 2, n Multiplication, i.e., P i =W(i)×P i Images of different typical structural components, P1', P2'...P, are obtained. n This method obtains electromagnetic scattering intensity images of typical structural components in complex targets, enabling the separation of electromagnetic scattering features of typical structural components in complex targets.

Images