Electromagnetic scattering mechanism identification method

Abstract

The application provides an electromagnetic scattering mechanism identification method, and mainly solves the problem of inaccurate scattering mechanism identification. The application calculates the electromagnetic scattering field of a target by using multiple algorithms, obtains different scattering mechanisms by using the differences between different algorithms, labels the mechanisms to form a data set, and then trains the data by using a convolutional neural network, so that an electromagnetic scattering mechanism identification neural network is obtained. The scattering mechanism identification technology provided by the application can effectively identify the scattering mechanisms such as specular scattering, multiple scattering and edge diffraction, can clearly identify different components of a scattering source, can provide a basis for reconstruction or modification of an electromagnetic environment or a scattering source, can be used in target identification based on a radar image, and can significantly improve the identification accuracy.

Description

Technical Field

[0001] The present invention relates to the field of electromagnetic scattering, and mainly to a method for identifying an electromagnetic scattering mechanism. Background Art

[0002] When electromagnetic waves hit a complex target, the target's smooth surface will produce specular scattering contributions, discontinuous structures such as spires and edges will produce diffraction contributions, and concave structures will have electromagnetic coupling effects or multiple scattering contributions. In addition, there are scattering contributions from different mechanisms such as surface waves and creeping waves. Currently, existing technologies cannot completely and accurately identify the above scattering mechanisms. For example, the attribute scattering center extraction method can only identify the scattering contributions generated by specific structures. Some other existing technologies can only identify scattering contributions that have experienced different numbers of reflections or a specific scattering mechanism. Currently, existing technologies are unable to solve the problem of identifying scattering mechanisms such as specular scattering, edge diffraction, multiple scattering and creeping waves. Summary of the Invention

[0003] In order to overcome the deficiencies of the prior art and to solve the problem of electromagnetic scattering mechanism identification, the present invention provides an electromagnetic scattering mechanism identification method.

[0004] The technical solution adopted by the present invention to solve the technical problem includes the following steps:

[0005] Step 1: Obtain the scattered field of the target;

[0006] The scattering field of the target is calculated using physical optics, shooting and bouncing ray, and geometric diffraction theory. Physical optics is abbreviated as PO, ray tracing is abbreviated as SBR, and geometric diffraction theory is abbreviated as GTD.

[0007] The scattered field calculated by PO is E po ;

[0008] The scattered field calculated by SBR is E sbr ;

[0009] The scattered field calculated by GTD is E gtd ;

[0010] Step 2: Obtain the scattering mechanism of the target;

[0011] As shown in formula (1), the scattered field calculated by the physical optics method is directly used as the specular scattering mechanism E specular scattering As shown in formula (2), the scattering field calculated by the ray tracing algorithm minus the scattering field calculated by the physical light method is used to obtain the multiple scattering mechanism E Multiple scatteringAs shown in formula (3), the diffraction mechanism E is obtained by subtracting the scattered field calculated by the physical optics algorithm from the scattered field calculated by the geometric diffraction theory method. diffraction ;

[0012] E specular scattering =E po (1)

[0013] E Multiple scattering =E sbr -E po (2)

[0014] E diffraction =E gtd -E po (3)

[0015] E specular scattering 、E Multiple scattering and E diffraction All are complex numbers, including real and imaginary parts;

[0016] E specular scattering 、E Multiple scattering and E diffraction Both include vertical polarization data VP and horizontal polarization data HP;

[0017] Step 3: Obtain the scattered field of the basic target and the scattered field of the combined target;

[0018] The scattering field of the basic target is calculated using the physical optics method, ray tracing method and geometric diffraction theory method respectively; the scattering field of the combined target is calculated using the physical optics method, ray tracing method and geometric diffraction theory method respectively;

[0019] Step 4: Obtain the scattering mechanism of the basic target and the scattering mechanism of the combined target;

[0020] Using the method in step 2 to process the scattering fields calculated from the basic target and the combined target in step 3, respectively, a scattering mechanism of the basic target scattering field and a scattering mechanism of the combined target scattering field are obtained;

[0021] Step 5: Construct specular scattering mechanism dataset, multiple scattering dataset and diffraction dataset;

[0022] The set of specular scattering mechanism data obtained in step 4 is used as a specular scattering mechanism data set; the set of multiple scattering data obtained in step 4 is used as a multiple scattering data set; the set of diffraction data obtained in step 4 is used as a diffraction data set;

[0023] Step 6: Construct training dataset and test dataset;

[0024] The specular scattering dataset, the multiple scattering dataset, and the diffraction dataset are combined to form a data set, and the data set is randomly split into a training set and a test set;

[0025] Step 7: Train the convolutional neural network to identify the scattering mechanism;

[0026] Import the training set and test set in step 6 into the convolutional neural network to train the convolutional neural network;

[0027] Step 8: Identification of scattering mechanism;

[0028] The unknown scattering mechanism to be identified is loaded into the convolutional neural network trained in step 7. The convolutional neural network outputs a label corresponding to the scattering mechanism to be identified, thereby identifying the type of the scattering mechanism.

[0029] Furthermore, in step 3, the basic targets include cylinders, cones, spheres and airfoils.

[0030] Furthermore, in step 3, the combined target includes a combination of at least two or more basic targets among cylinders, cones, spheres and airfoils.

[0031] Furthermore, in step 5, the specular scattering mechanism data set includes the specular scattering mechanism of a cylinder, a cone, a sphere, an airfoil and a combined target; the multiple scattering data set includes the multiple scattering mechanism of a cylinder, a cone, a sphere, an airfoil and a combined target; and the diffraction data set includes the diffraction mechanism of a cylinder, a cone, a sphere, an airfoil and a combined target.

[0032] Furthermore, in step 6, the steps of constructing the training dataset and the test dataset are:

[0033] The data labels in the specular scattering dataset are marked as 0; the data labels in the multiple scattering dataset are marked as 1, and the data labels in the diffraction dataset are marked as 2;

[0034] The specular scattering dataset, multiple scattering dataset, and diffraction dataset are merged to form a data collection, which is then randomly split into a training set and a test set with a ratio of 8:2 to provide training data and test data for deep learning.

[0035] Furthermore, in step 7, the convolutional neural network includes 6 convolutional layers Conv1D, 3 maximum pooling layers Maxpool, 1 flat layer Flatten and 2 fully connected layers Desen.

[0036] Furthermore, in step 7, the input of the convolutional neural network is the real part and the imaginary part of the scattering mechanism, and the output of the convolutional neural network is the data label of the scattering mechanism.

[0037] Furthermore, in step 7, the activation function used by the convolutional layer of the convolutional neural network is ReLU.

[0038] Furthermore, in step 7, the activation function used by the fully connected layer of the convolutional neural network is SOFTMAX.

[0039] Furthermore, in step 7, the convolutional layer, pooling layer, flattening layer and fully connected layer in the convolutional neural network use kernels of different sizes to improve E specular scattering 、E Multiple scattering and E diffraction The accuracy of identification of vertically polarized VP and horizontally polarized HP scattering mechanisms.

[0040] Furthermore, in step 7, the convolutional neural network is implemented using pytorch.

[0041] The beneficial effects of the present invention are:

[0042] (1) The present invention accurately identifies the specular scattering mechanism, the multiple scattering mechanism and the diffraction mechanism

[0043] The present invention achieved 100% accuracy for specular scattering, 100% for multiple scattering, and 95% for diffraction. Compared to existing methods, the present invention significantly improves both the types of scattering mechanisms recognized and the accuracy.

[0044] (2) Understanding of scattering mechanisms and control of scattering characteristics

[0045] The present invention can clarify the role of various mechanisms, and can also follow the laws of physics to take more targeted measures to control scattering characteristics, avoiding blind control. For example, in stealth technology, targeted stealth measures can be taken for certain scattering mechanisms to reduce the scattering contribution of specific mechanisms.

[0046] (3) Diagnosis of special electromagnetic environments or scattering sources

[0047] According to the present invention, the scattering mechanism of a special electromagnetic environment or scattering source is intelligently identified, the formation mechanism of the electromagnetic environment can be determined, the different components of the scattering source can be clarified, and a basis for reconstruction or modification of the electromagnetic environment or scattering source can be provided. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a structural diagram of the overall research scheme of an electromagnetic scattering mechanism identification technology of the present invention;

[0049] Figure 2 This is a deep convolutional neural network framework diagram;

[0050] Figure 3 This is the curve of vertical polarization data set accuracy changing with training;

[0051] Figure 4 This is the curve of the accuracy of the horizontal polarization dataset changing with training. DETAILED DESCRIPTION

[0052] The technical solution adopted by the present invention to solve the technical problem is a method for identifying an electromagnetic scattering mechanism, comprising the following steps:

[0053] Step 1: Obtain the scattered field of the target;

[0054] The scattered field of the target is calculated using physical optics, shooting and bouncing ray, and geometric diffraction theory. Physical optics is abbreviated as PO, ray tracing is abbreviated as SBR, and geometric diffraction theory is abbreviated as GTD.

[0055] The scattered field calculated by PO is E po ;

[0056] The scattered field calculated by SBR is E sbr ;

[0057] The scattered field calculated by GTD is E gtd ;

[0058] Step 2: Obtain the scattering mechanism of the target;

[0059] In step 1, the scattering field calculated by PO contains only the specular scattering mechanism, the scattering field calculated by SBR contains only the specular scattering mechanism and the multiple scattering mechanism, and the scattering field calculated by GTD contains only the specular scattering mechanism and the edge / tip scattering mechanism;

[0060] As shown in formula (1), the scattered field calculated by the physical optics method is directly used as the specular scattering mechanism E specular scattering As shown in formula (2), the scattering field calculated by the ray tracing algorithm minus the scattering field calculated by the physical light method is used to obtain the multiple scattering mechanism E Multiple scattering As shown in formula (3), the diffraction mechanism E is obtained by subtracting the scattered field calculated by the physical optics algorithm from the scattered field calculated by the geometric diffraction theory method. diffraction ;

[0061] E specular scattering =E po (1)

[0062] EMultiple scattering =E sbr -E po (2)

[0063] E diffraction =E gtd -E po (3)

[0064] E specular scattering 、E Multiple scattering and E diffraction All are complex numbers, including real and imaginary parts;

[0065] E specular scattering 、E Multiple scattering and E diffraction Both include vertical polarization data VP and horizontal polarization data HP;

[0066] Step 3: Obtain the scattered field of the basic target and the scattered field of the combined target;

[0067] The scattering field of the basic target is calculated using the physical optics method, ray tracing method and geometric diffraction theory method respectively; the scattering field of the combined target is calculated using the physical optics method, ray tracing method and geometric diffraction theory method respectively;

[0068] The basic targets include cylinders, cones, spheres and airfoils;

[0069] The combined target includes a combination of at least two or more basic targets among cylinders, cones, spheres and airfoils;

[0070] Step 4: Obtain the scattering mechanism of the basic target and the scattering mechanism of the combined target;

[0071] Using the method in step 2 to process the scattered fields calculated from the basic target and the combined target in step 3, the scattering mechanisms of the scattered fields of the basic target and the combined target are obtained;

[0072] Step 5: Construct specular scattering mechanism dataset, multiple scattering dataset and diffraction dataset;

[0073] The set of specular scattering mechanism data obtained in step 4 is used as a specular scattering mechanism data set; the set of multiple scattering data obtained in step 4 is used as a multiple scattering data set; the set of diffraction data obtained in step 4 is used as a diffraction data set;

[0074] The specular scattering mechanism dataset includes the specular scattering mechanism of cylinders, cones, spheres, airfoils and combined targets; the multiple scattering mechanism dataset includes the multiple scattering mechanism of cylinders, cones, spheres, airfoils and combined targets; the diffraction mechanism dataset includes the diffraction mechanism of cylinders, cones, spheres, airfoils and combined targets;

[0075] Step 6: Build training and test datasets

[0076] The data labels in the specular scattering dataset are marked as 0; the data labels in the multiple scattering dataset are marked as 1, and the data labels in the diffraction dataset are marked as 2;

[0077] The specular scattering dataset, multiple scattering dataset, and diffraction dataset were merged to form a data set, which was then randomly split into a training set and a test set with a ratio of 8:2 to provide training data and test data for deep learning.

[0078] Step 7: Train the convolutional neural network to identify the scattering mechanism;

[0079] Import the training set and test set in step 6 into the convolutional neural network and train the convolutional neural network; the input of the convolutional neural network is the real part and imaginary part of the scattering mechanism, and the output of the convolutional neural network is the data label of the scattering mechanism;

[0080] The convolutional neural network consists of 6 convolutional layers Conv1D, 3 maximum pooling layers Maxpool, 1 flat layer Flatten and 2 fully connected layers Desen;

[0081] The activation function used in the convolutional layer is ReLU;

[0082] The activation function used in the fully connected layer is SOFTMAX;

[0083] The convolutional layers, pooling layers, flattening layers, and fully connected layers in the convolutional neural network use kernels of different sizes to improve E specular scattering 、E Multiple scattering and E diffraction The accuracy of identifying the vertically polarized VP and horizontally polarized HP scattering mechanisms;

[0084] Convolutional neural networks are implemented using pytorch;

[0085] Step 8: Identification of scattering mechanism;

[0086] The unknown scattering mechanism to be identified is loaded into the convolutional neural network trained in step 7. The convolutional neural network outputs a label corresponding to the scattering mechanism to be identified and the type of the scattering mechanism is identified.

[0087] Figure 3 This is the curve of vertical polarization data set accuracy changing with training. Figure 4 is the curve of the accuracy of the horizontal polarization data set changing with training, Figure 3 and Figure 4The horizontal axis epoch represents the number of training iterations, and the vertical axis accuracy represents the accuracy. The Training and Validation in the figure represent the training data set and test data respectively. Figure 3 The curve in the figure converges to around 0.97 after 40 iterations of training curve and around 0.95 after testing curve. Figure 4 The curve in the figure converges to around 0.95 after 80 iterations of training and testing curves. Figure 3 and Figure 4 It shows that more than 95% of the scattering mechanisms in the data set can be accurately identified by the patented method, proving the effectiveness of the present invention.

[0088] The present invention will be further described below with reference to the accompanying drawings and examples.

[0089] The purpose of the present invention is to realize the identification of various scattering mechanisms, construct a scattering mechanism identification method based on artificial intelligence, and form an intelligent technology for electromagnetic scattering mechanism identification.

[0090] The three methods of physical optics, ray tracing, and geometric diffraction theory are very mature and well-known. Their basic principles are as follows:

[0091] Physical optics (PO) is an algorithm that uses the approximate integral of the induced electric field to solve scattering problems. It does not calculate the interactions between the induced currents in different parts of the target surface. Instead, it considers only the incidence and reflection phenomena at local locations on the target to independently solve for the approximate surface induced current. Therefore, PO only considers the electromagnetic field reflected from the mirror.

[0092] Shooting and bouncing ray (SBR) is a hybrid method combining geometric and physical optics to solve electromagnetic models with multiple reflections that physical optics cannot. The SBR algorithm expands an incident uniform plane wave into a dense array of ray tubes parallel to the incident direction. Rays are then traced until they leave the surface of the object and travel to infinity. Finally, an integral surface is selected and the far-field radiation field is solved according to the principles of physical optics. The SBR algorithm accounts for multiple scattering effects between different parts of the target. The scattered field calculated by the bouncing ray method incorporates both multiple and specular scattering.

[0093] The geometric theory of diffraction (GTD) addresses the problem that geometric optics only studies direct radiation, reflection, and refraction, and cannot account for the diffraction of electromagnetic waves. The diffraction field propagates along diffraction rays. The trajectories of these rays are determined by the generalized Fermat principle, and diffraction is determined solely by the properties and geometry of the region surrounding the diffraction point on the scatterer. After leaving the diffraction point, the diffraction ray still obeys the laws of geometric optics, namely, energy conservation within the diffraction ray tube, assuming that the phase delay along the path is equal to the product of the wave number of the medium and the distance. The scattered field calculated by the geometric theory of diffraction includes edge / tip diffraction and specular scattering.

[0094] Step 1: Use physical optics, ray tracing and geometric diffraction theory to calculate Figure 1 The scattered field of the target shown in . The physical optics (PO) method is an algorithm that uses the approximate integral of the induced electric field to solve the scattering problem. Compared with high-precision algorithms such as the method of moments, PO does not calculate the interaction between the induced currents on different parts of the target surface, and thus independently solves the approximate surface induced current. The shot ray (SBR) method is a high-frequency asymptotic method that combines geometric optics (GO) and PO to solve scattering problems. It obtains more accurate results by considering scattering caused by multiple interactions. The SBR method is used to generate a scattered field that includes multiple scattering. The geometric theory of diffraction (GTD) is a generalization of the theory of geometric optics. It is based on the exact solution of the peak diffraction field and solves the diffraction field problem through the linear correlation between the diffraction coefficient and the incident field.

[0095] Step 2: If Figure 1 As shown in , the total scattering fields calculated by different algorithms in step 1 are subtracted to obtain different scattering mechanisms. First, the scattering field calculated by the physical optics method is directly used as the mirror scattering mechanism. Then, the scattering field calculated by the ray tracing algorithm is subtracted from the scattering field calculated by the physical optics method to obtain the multiple scattering mechanism. Finally, the scattering field calculated by the geometric diffraction theory is subtracted from the scattering field calculated by the physical optics algorithm to obtain the edge diffraction mechanism.

[0096] Step 3: Use the method in step 2 to process the scattered field calculated in step 3 to obtain the scattering mechanism in the scattered field of the cylinder, cone, sphere, airfoil and combined target.

[0097] Step 4: The scattering mechanism data in step 3 are formed into a data set. The specular scattering mechanism data set includes the specular scattering mechanism of cylinders, cones, spheres, airfoils, and combined targets; the multiple scattering data set includes the multiple scattering mechanism of combined targets; and the diffraction data set includes the diffraction mechanism of cylinders, cones, spheres, airfoils, and combined targets.

[0098] Step 5: Mark the data in the specular scattering dataset, multiple scattering dataset, and diffraction dataset as 0, 1, and 2, respectively, combine these data to form a dataset, and then randomly split the dataset into a training set and a test set in a ratio of 8:2 to provide training data and test data for deep learning, respectively.

[0099] Step 6: The structure of the deep convolutional neural network is as follows Figure 2 As shown in , the dataset from step 5 is imported into a convolutional neural network. This learning network consists of six convolutional layers (Conv1D), three maximum pooling layers (Maxpool), one flattening layer (Flatten), and two fully connected layers (Desen). The activation function used in the convolutional layers is ReLU, and the activation function used in the fully connected layers is SOFTMAX. The network input is the real and imaginary parts of the scattering data, and the network output is the type of scattering mechanism. Because the scattering mechanism is sensitive to the polarization characteristics of the incident field, the convolutional layers, pooling layers, flattening layers, and fully connected layers in the convolutional neural network use kernels of different sizes for vertical polarization (VP) and horizontal polarization (HP) to more accurately identify different scattering mechanisms.

Claims

1. A method for identifying electromagnetic scattering mechanisms, characterized in that: The steps include: Step 1: Obtain the scattered field of the target; The scattering field of the target is calculated using physical optics, shooting and bouncing ray and geometric diffraction theory. The physical optics method is abbreviated as PO, the ray tracing method is abbreviated as SBR, and the geometric diffraction theory method is abbreviated as GTD; The scattered field calculated by PO is E po ; The scattered field calculated by SBR is E sbr ; The scattered field calculated by GTD is E gtd ; Step 2: Obtain the scattering mechanism of the target; As shown in formula (1), the scattered field calculated by the physical optics method is directly used as the specular scattering mechanism E specularscattering As shown in formula (2), the scattering field calculated by the ray tracing algorithm minus the scattering field calculated by the physical light method is used to obtain the multiple scattering mechanism E Multiplescattering As shown in formula (3), the diffraction mechanism E is obtained by subtracting the scattered field calculated by the physical optics algorithm from the scattered field calculated by the geometric diffraction theory method. diffraction ; AND specular scattering =And po (1) AND Multiplescattering =And sbr -AND po (2) AND diffraction =And gtd -AND po (3) E specularscattering 、E Multiplescattering and E diffraction All are complex numbers, including real and imaginary parts; E specular scattering 、E Multiple scattering and E diffraction Both include vertical polarization data VP and horizontal polarization data HP; Step 3: Obtain the scattered field of the basic target and the scattered field of the combined target; The scattered fields of the base targets are calculated using physical optics, ray tracing and geometric diffraction theory methods respectively. The scattered fields of the combined target are calculated using physical optics, ray tracing and geometric diffraction theory methods respectively. Step 4: Obtain the scattering mechanism of the basic target and the scattering mechanism of the combined target; Using the method in step 2 to process the scattering fields calculated from the basic target and the combined target in step 3, respectively, a scattering mechanism of the basic target scattering field and a scattering mechanism of the combined target scattering field are obtained; Step 5: Construct specular scattering mechanism dataset, multiple scattering dataset and diffraction dataset; The set of specular scattering mechanism data obtained in step 4 is used as a specular scattering mechanism data set; The set of multiple scattering data obtained in step 4 is used as a multiple scattering data set; The set of diffraction data obtained in step 4 is used as a diffraction data set; Step 6: Construct training dataset and test dataset; The specular scattering dataset, the multiple scattering dataset, and the diffraction dataset are combined to form a data set, and the data set is randomly split into a training set and a test set; Step 7: Train the convolutional neural network to identify the scattering mechanism; Import the training set and test set in step 6 into the convolutional neural network to train the convolutional neural network; Step 8: Identification of scattering mechanism; The unknown scattering mechanism to be identified is loaded into the convolutional neural network trained in step 7. The convolutional neural network outputs a label corresponding to the scattering mechanism to be identified, thereby identifying the type of the scattering mechanism.

2. The electromagnetic scattering mechanism identification method according to claim 1, characterized in that: In step 3, the basic targets include cylinders, cones, spheres and airfoils.

3. The electromagnetic scattering mechanism identification method according to claim 1, characterized in that: In step 3, the combined target includes a combination of at least two or more basic targets among cylinders, cones, spheres and airfoils.

4. The electromagnetic scattering mechanism identification method according to claim 1, characterized in that: In step 5, the specular scattering mechanism data set includes the specular scattering mechanism of a cylinder, a cone, a sphere, an airfoil and a combined target; the multiple scattering data set includes the multiple scattering mechanism of a cylinder, a cone, a sphere, an airfoil and a combined target; and the diffraction data set includes the diffraction mechanism of a cylinder, a cone, a sphere, an airfoil and a combined target.

5. The electromagnetic scattering mechanism identification method according to claim 1, characterized in that: In step 6, the steps of constructing the training data set and the test data set are: The data labels in the specular scattering dataset are marked as 0; the data labels in the multiple scattering dataset are marked as 1, and the data labels in the diffraction dataset are marked as 2; The specular scattering dataset, multiple scattering dataset, and diffraction dataset are merged to form a data collection, which is then randomly split into a training set and a test set with a ratio of 8:2 to provide training data and test data for deep learning.

6. The electromagnetic scattering mechanism identification method according to claim 1, characterized in that: In step 7, the convolutional neural network includes 6 convolutional layers Conv1D, 3 maximum pooling layers Maxpool, 1 flat layer Flatten and 2 fully connected layers Desen; the convolutional neural network is implemented using pytorch.

7. The electromagnetic scattering mechanism identification method according to claim 1, characterized in that: In step 7, the input of the convolutional neural network is the real part and the imaginary part of the scattering mechanism, and the output of the convolutional neural network is the data label of the scattering mechanism.

8. The electromagnetic scattering mechanism identification method according to claim 1, characterized in that: In step 7, the activation function used by the convolutional layer of the convolutional neural network is ReLU.

9. The electromagnetic scattering mechanism identification method according to claim 1, characterized in that: In step 7, the activation function used by the fully connected layer of the convolutional neural network is SOFTMAX.

10. The electromagnetic scattering mechanism identification method according to claim 1, characterized in that: In step 7, the convolutional layer, pooling layer, flattening layer and fully connected layer in the convolutional neural network use kernels of different sizes to improve E specular scattering 、E Multiple scattering and E diffraction The accuracy of identification of vertically polarized VP and horizontally polarized HP scattering mechanisms.

Images