Radar target recognition method, device and equipment based on prior information of scattering centers

Abstract

The present invention relates to a radar target recognition method, device, equipment and storage medium based on prior information of scattering centers. The method includes: constructing a scattering center target recognition network. The scattering center target recognition network includes: a scattering center extraction layer, an encoding network, and a classifier. The scattering-related information of the HRRP target image is extracted through the scattering center extraction layer. The scattering-related information at least includes the scattering center position and the scattering center amplitude information of the current scattering feature point in the HRRP target image. The scattering-related information is input into the encoding network for feature space mapping to obtain the discriminative feature points to be optimized. The discriminative feature points to be optimized are input into the classifier for feature classification to obtain the discriminative feature points. Gradient backpropagation training is performed according to the scattering-related information and the discriminative feature points to obtain a trained scattering center target recognition network. The present invention can improve the accuracy of HRRP target recognition.

Description

Technical Field

[0001] The present invention relates to the technical field of radar target recognition, and in particular, to a radar target recognition method, device and equipment based on prior information of scattering centers. Background Art

[0002] Radar can work stably under non-ideal detection conditions and has the advantage of non-contact. Therefore, radar has become an important sensor in the field of automatic target recognition. High resolution range profile (HRRP) can reflect the distribution of scattering points of the measured target along the radar line of sight direction and contains many physical characteristics of the measured target, such as the distribution of scattering points, the position and intensity of scattering centers, the size of the target, etc. Therefore, the radar target recognition method based on HRRP has received more and more attention from researchers in recent years and has become a hot issue in the field of radar target recognition.

[0003] With the development of deep learning technology, a large number of HRRP target recognition methods based on deep learning technology have been proposed. Based on this, HRRP target recognition methods can be roughly divided into two categories. One is the traditional HRRP target recognition method, that is, without using deep learning technology, and the other is the method based on deep learning, that is, HRRP target recognition based on deep learning technology. The traditional HRRP target recognition method can be roughly divided into three parts, namely data preprocessing, feature extraction and classifier design. The advantage of the traditional method is that the extracted features have clear physical meanings, which enables people to easily interpret the recognition results according to the extracted features. However, the shallow structure of the traditional method limits the improvement of its recognition performance, resulting in relatively low recognition accuracy. Generally, the HRRP target recognition method based on deep learning integrates data preprocessing, feature extraction and classifier design into an overall model by cleverly designing the neural network structure. Therefore, these models can make full use of big data for end-to-end training, automatically extract target features, and achieve high recognition accuracy. However, the features extracted by the HRRP target recognition method based on deep learning lack clear physical meanings, making it difficult for people to interpret its recognition results, which also makes these methods become the much-criticized "black box" models. Summary of the Invention

[0004] Based on this, it is necessary to provide a radar target recognition method, device and equipment based on prior information of scattering centers, which can improve the radar target recognition accuracy for the above technical problems.

[0005] A radar target recognition method based on prior information of scattering centers, the method includes:

[0006] Construct a scattering center target recognition network. The scattering center target recognition network includes: a scattering center extraction layer, an encoding network, and a classifier.

[0007] Extract the scattering-related information of the HRRP target image through the scattering center extraction layer. The scattering-related information includes at least the scattering center position and the scattering center amplitude information of the current scattering feature point in the HRRP target image.

[0008] Input the scattering-related information into the encoding network for feature space mapping to obtain the discriminant feature points to be optimized.

[0009] Input the discriminant feature points to be optimized into the classifier for feature classification to obtain the discriminant feature points.

[0010] Perform gradient backpropagation training based on the scattering-related information and the discriminant feature points to obtain the trained scattering center target recognition network.

[0011] In one of the embodiments, it further includes: receiving the echo signal transmitted by the target to be measured in the communication baseband through a radar receiver:

[0012]

[0013] where \(l = 1, 2, 3, \cdots, L\) is the \(l\)-th individual scattering center in the \(n\)-th range cell, \(s(t)\) is the transmitted signal of the radar transmitter, \(A(l)\) is the related scattering coefficient, \(n = 1, 2, 3, \cdots, N\) is the number of range cells, \(R\) l is the radial distance between the radar and the \(l\)-th individual scattering center in the \(n\)-th range cell, is a rectangular pulse with a width of \(T\) p , \(f\) c is the radar center frequency, \(\mu\) is the chirp coefficient, and \(c\) is the propagation speed of electromagnetic waves;

[0014] Calculate the discrete frequency response of the echo signal:

[0015]

[0016] where \(r(m)\) is the \(m\)-th frequency response, \(m\in\{1, 2, 3, \cdots, M\}\) is the number of frequency responses, \(\Delta f\) is the frequency interval, and \(\omega(n)\) is the scattering coefficient of the \(n\)-th range cell. The discrete frequency response generates a scattering center extraction model through Fourier transform:

[0017]

[0018] \(\varPhi=[\varphi(r_1),\varphi(r_2),\cdots,\varphi(r N )]

[0019]

[0020] where ω * is the optimal sparse coding, ω = [ω(1), ω(2), …, ω(N)] T is the scattering coefficient, and Φ is the Fourier basis.

[0021] In one embodiment, it further includes: extracting the scattering coefficient of the strong scattering center response in the HRRP target image through the scattering center extraction model of the scattering center extraction layer, and optimizing the scattering coefficient in the scattering center extraction layer according to the soft threshold iteration method to obtain the scattering-related information corresponding to the optimal sparse coding.

[0022] In one embodiment, it further includes: inputting the scattering-related information into the encoding network, and the encoding network uses a single-channel ResNet network structure to map the scattering-related information to the feature space to be optimized, obtaining the discriminative feature points to be optimized.

[0023] In one embodiment, it further includes: inputting the discriminative feature points to be optimized into the classifier, and the classifier uses the softmax function and the preset classification loss function to perform feature classification on the discriminative feature points to be optimized, obtaining the discriminative feature points.

[0024] In one embodiment, it further includes: reconstructing the error loss function according to the scattering-related information and the discriminative feature points, and converging the scattering-related information as the prior information with the discriminative feature points, and inputting the converged discriminative feature points as the discriminative feature points to be optimized into the scattering center target recognition network to obtain the trained scattering center target recognition network.

[0025] In one embodiment, it further includes: reconstructing the extraction loss function of the scattering-related information according to the scattering-related information:

[0026]

[0027] where x i is the HRRP target image, i ∈ I is the index of the HRRP target image, I is the number of HRRP target images, Φ is the Fourier basis, and ω * is the optimal sparse coding. And reconstructing the classification loss function according to the discriminative feature points:

[0028]

[0029] where i ∈ I is the index of the HRRP target image, I is the number of HRRP target images, k = 1, 2, …, C is the category of the HRRP target image, C is the number of categories of the HRRP target image, y i is the category label of the HRRP target image x i and f ikThe HRRP target image x belonging to category k i The discriminative feature points; reconstruct the error loss function of the scattering center target recognition network according to the extraction loss function and the classification loss function:

[0030]

[0031] where x i is the HRRP target image, i ∈ I is the index of the HRRP target image, I is the number of HRRP target images, Φ is the Fourier basis, ω * is the optimal sparse coding, k = 1, 2,..., C is the category of the HRRP target image, C is the number of categories of the HRRP target image, y i is the category label of the HRRP target image x i and f ik is the discriminative feature point of the HRRP target image x i belonging to category k.

[0032] A radar target recognition device based on prior information of scattering centers, the device includes:

[0033] A network construction module, used to construct a scattering center target recognition network. The scattering center target recognition network includes: a scattering center extraction layer, an encoding network, and a classifier.

[0034] A scattering-related information extraction module, used to extract the scattering-related information of the HRRP target image through the scattering center extraction layer. The scattering-related information at least includes the scattering center position and the scattering center amplitude information of the current scattering feature point in the HRRP target image.

[0035] A module for obtaining discriminative feature points to be optimized, used to input the scattering-related information into the encoding network for feature space mapping to obtain the discriminative feature points to be optimized.

[0036] A feature point optimization module, used to input the discriminative feature points to be optimized into the classifier for feature classification to obtain the discriminative feature points.

[0037] An identification optimization training module, used to perform gradient backpropagation training according to the scattering-related information and the discriminative feature points to obtain a trained scattering center target recognition network.

[0038] A computer device, including a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0039] Construct a scattering center target recognition network. The scattering center target recognition network includes: a scattering center extraction layer, an encoding network, and a classifier.

[0040] Extract the scattering-related information of the HRRP target image through the scattering center extraction layer. The scattering-related information includes at least the scattering center position and the scattering center amplitude information of the current scattering feature point in the HRRP target image.

[0041] Input the scattering-related information into the encoding network for feature space mapping to obtain the discriminative feature points to be optimized.

[0042] Input the discriminative feature points to be optimized into the classifier for feature classification to obtain the discriminative feature points.

[0043] Perform gradient backpropagation training based on the scattering-related information and the discriminative feature points to obtain the trained scattering center target recognition network.

[0044] The above radar target recognition method, device, and equipment based on the prior information of the scattering center construct a scattering center target recognition network as a neural network. Among them, the encoding network and the classifier part achieve end-to-end training, and the scattering-related information is extracted from the HRRP target image through the scattering center extraction layer. These information are mapped through the feature space of the encoding network and finally input into the classifier for classification. Although through the extraction of scattering-related information and the design of the network structure, the physical meaning of the HRRP target image features can be extracted, which improves the interpretability of the scattering center target recognition network to a certain extent. For example, the recognition basis of the neural network for target features can be explained by analyzing the scattering center position, amplitude information, etc., reducing the negative impact of the model as a "black box". In addition, using the scattering-related information as prior information to drive the scattering center target recognition network, and performing gradient backpropagation training with the scattering-related information and the discriminative feature points can continuously adjust the network parameters to make the network adapt to the features of different targets, further improving the accuracy of radar target recognition. Description of the Drawings

[0045] Figure 1 It is an application scenario diagram of the radar target recognition method based on the prior information of the scattering center in an embodiment;

[0046] Figure 2 It is a schematic flowchart of the radar target recognition method based on the prior information of the scattering center in an embodiment;

[0047] Figure 3 It is an expanded structure diagram of the scattering center extraction layer in an embodiment;

[0048] Figure 4 It is a network structure diagram of a single-channel ResNet in an embodiment;

[0049] Figure 5 It is a structural block diagram of the radar target recognition device based on the prior information of the scattering center in an embodiment;

[0050] Figure 6 It is the internal structure diagram of a computer device in an embodiment. Specific implementation manners

[0051] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0052] The radar target recognition method based on scatter center prior information provided by the present invention can be applied to, for example, Figure 1 the scatter center target recognition network shown as follows. Among them, the entire network includes a scatter center extraction layer, an encoding network, a classifier and a loss function.

[0053] In one embodiment, as Figure 2 shown, a radar target recognition method based on scatter center prior information is provided. Taking the application of this method to the Figure 1 neural network model as an example, the method includes the following steps:

[0054] Step 202, construct a scatter center target recognition network. The scatter center target recognition network includes: a scatter center extraction layer, an encoding network and a classifier.

[0055] Step 204, extract the scatter-related information of the HRRP target image through the scatter center extraction layer. The scatter-related information includes at least the scatter center position and scatter center amplitude information of the current scatter feature point in the HRRP target image.

[0056] Specifically, first input an HRRP complex image of any length. The image is input into the scatter center extraction layer proposed in this patent to extract the position and amplitude information of the target scatter center.

[0057] Furthermore, considering that the radar emits a chirp signal, the transmitted signal s(t) can be expressed as Equation (1), where represents a rectangular pulse with a width of T p , f c represents the radar center frequency, μ represents the chirp coefficient, represents the envelope of the transmitted signal.

[0058]

[0059] The high-resolution range response of the measured target can be represented by the sum of the responses of the strong scatter centers of the target along the radar line of sight. In a wideband radar system, the HRRP of the measured target consists of multiple range cells, and the range resolution ΔR = c / 2B, B = μT pLet \(B\) denote the radar bandwidth and \(c\) denote the propagation speed of electromagnetic waves. Therefore, the echo signal of a range cell of the target under test can be regarded as the coherent summation of the echo signals of individual scatterers in this range cell. After the dechirp processing, the received echo signal of the \(n\)-th range cell in the baseband can be expressed as Equation (2).

[0060]

[0061] where \(l = 1, 2, 3, \cdots, L\) represents the \(l\)-th individual scatterer in the \(n\)-th range cell, \(L\) represents the total number of scatterers in the \(n\)-th range cell, \(n = 1, 2, 3, \cdots, N\), \(N\) represents the total number of range cells, \(R_{ln}\ l denotes the radial distance between the radar and the \(l\)-th individual scatterer in the \(n\)-th range cell, and \(A(l)\) represents the associated scattering coefficient. After Fourier transform, Equation (2) can be written as Equation (3).

[0062]

[0063] where \(F(\cdot)\) represents the Fourier transform operation, and \(x(f)=F(x(t))\) represents the Fourier transform of the signal \(x(t)\).

[0064] Through the pulse compression technique, the frequency response can be expressed as Equation (4).

[0065]

[0066] where \(x\ * (f)\) represents the conjugate form of \(x(f)\), and \(P(f)=x(f)x\ * (f)\) represents the power spectrum of the signal \(x(t)\).

[0067] For a chirp signal, its power spectrum Therefore, Equation (4) can be expressed as Equation (5), where

[0068]

[0069] The discrete frequency response of Equation (5) can be expressed as Equation (6).

[0070]

[0071] where \(r(m)\) represents the \(m\)-th frequency response, where \(m\in\{1, 2, 3, \cdots, M\}\), \(M\) represents the total number of frequency responses, and \(\Delta f\) represents the frequency interval. Assuming that \(R\ l is an integer multiple of the range resolution \(\Delta R\), then Equation (6) can be written as Equation (7).

[0072]

[0073] where represents the positions of N range cells, ω(n) represents the scattering coefficient of the n-th range cell, and let

[0074]

[0075] Φ = [φ(r1), φ(r2), …, φ(r N )] (9)

[0076] Considering the influence of actual noise, the signal model represented by (6) can be rewritten in the vector-matrix form

[0077] r = Φω + η (10)

[0078] where Φ represents the Fourier basis, ω = [ω(1), ω(2), …, ω(N)] T represents the scattering coefficients, and η = [η(1), η(2), …, η(N)] T represents the noise

[0079] According to the above scattering center model, the HRRP of the target can be represented by the responses of the strong scattering centers with sparse distribution of the target. This sparsity is reflected in the sparsity of ω. Moreover, the scattering center coefficient ω contains the position and amplitude information of the scattering centers. Therefore, for an input HRRP complex data, the function of the scattering center extraction layer is defined as outputting the optimal sparse coding ω * . Further, it is transformed into the problem of finding the optimal solution ω by solving the Lasso shown in equation (11) * :

[0080]

[0081] The implicit neural network layer defined by formula (11) is called the scattering center extraction layer because the corresponding problem is based on the scattering center model. The soft threshold iteration method is used for the forward propagation of the neural network, that is, to solve the optimization problem defined by equation (11). At the same time, the expanded network structure is designed according to its solution process as Figure 3 shown, and the position and amplitude information of the scattering centers in the target HRRP image are extracted through automatic gradient derivation and backpropagation

[0082] Step 206, input the scattering-related information into the encoding network for feature space mapping to obtain the discriminative feature points to be optimized

[0083] Specifically, the encoding network maps the target scattering center position and amplitude information extracted by the scattering center extraction layer into a more distinguishable feature space, thereby obtaining discriminative features that are more conducive to classification. The encoding network uses a single-channel ResNet because the HRRP data is single-channel. The structure of the single-channel ResNet network used is as Figure 4 shown. Among them, "1×3conv, 64" represents a convolutional layer with 64 convolutional kernels and a convolutional kernel size of 1×3, "BatchNorm" represents a batch normalization layer, "Conv Block" represents a convolutional module, "Avgpool" represents an adaptive pooling layer, "ReLU" represents a ReLu activation function, and "Shortcut" represents a skip connection.

[0084] Step 208: Input the discriminative feature points to be optimized into the classifier for feature classification to obtain discriminative feature points.

[0085] Step 210: Perform gradient backpropagation training based on the scattering-related information and the discriminative feature points to obtain a trained radar target recognition network for scattering centers.

[0086] In the above radar target recognition method based on prior information of scattering centers, the constructed radar target recognition network for scattering centers is a neural network. Among them, the encoding network and the classifier part achieve end-to-end training, and scattering-related information is extracted from the HRRP target image through the scattering center extraction layer. These information are mapped through the feature space of the encoding network and finally input into the classifier for classification. Although the physical meaning of the HRRP target image features can be extracted through the extraction of scattering-related information and the design of the network structure, to a certain extent, the interpretability of the radar target recognition network for scattering centers is improved. For example, the recognition basis of the neural network for target features can be explained by analyzing the scattering center position, amplitude information, etc., reducing the negative impact of the model as a "black box". In addition, by using the scattering-related information as prior information to drive the radar target recognition network for scattering centers, and performing gradient backpropagation training using the scattering-related information and the discriminative feature points, the network parameters can be continuously adjusted to make the network adapt to the features of different targets, further improving the accuracy of radar target recognition.

[0087] In one of the embodiments, it further includes: receiving the echo signal transmitted by the target to be measured in the communication baseband through a radar receiver:

[0088]

[0089] where \(l = 1, 2, 3, \cdots, L\) is the \(l\)th individual scattering center in the \(n\)th range cell, is the transmitted signal of the radar transmitter, \(A(l)\) is the relevant scattering coefficient, \(n = 1, 2, 3, \cdots, N\) is the number of range cells, \(R\) lis the radial distance between the radar and the l-th individual scattering center in the n-th range cell, is a rectangular pulse with width T p , f c is the radar center frequency, μ is the chirp coefficient, and c is the propagation speed of electromagnetic waves;

[0090] Calculate the discrete frequency response of the echo signal:

[0091]

[0092] where r(m) is the m-th frequency response, m ∈ {1, 2, 3, …, M} is the number of frequency responses, Δf is the frequency interval, and ω(n) is the scattering coefficient of the n-th range cell. The discrete frequency response is Fourier-transformed to generate the scattering center extraction model:

[0093]

[0094] Φ = [φ(r1), φ(r2), …, φ(r N )]

[0095]

[0096] where ω * is the optimal sparse coding, ω = [ω(1), ω(2), …, ω(N)] T is the scattering coefficient, and Φ is the Fourier basis.

[0097] In one embodiment, the scattering coefficient of the strong scattering center response in the HRRP target image is extracted by the scattering center extraction model of the scattering center extraction layer, and the scattering coefficient is optimized in the scattering center extraction layer according to the soft threshold iteration method to obtain the scattering-related information corresponding to the optimal sparse coding.

[0098] In one embodiment, the scattering-related information is input into the coding network, and the coding network uses a single-channel ResNet network structure to map the scattering-related information to the feature space to be optimized, obtaining the discriminative feature points to be optimized.

[0099] In one embodiment, the discriminative feature points to be optimized are input into the classifier, and the classifier uses the softmax function and a preset classification loss function to perform feature classification on the discriminative feature points to be optimized, obtaining the discriminative feature points.

[0100] It should be noted that the classifier uses a common softmax classifier, whose structure is a fully connected layer followed by a softmax layer. The dimension of the fully connected layer is M×C, where M represents the dimension of the input data and C represents the dimension of the output data, which is the same as the total number of categories. The classification loss function uses a common cross-entropy loss function, and its formula can be expressed as:

[0101]

[0102] where i represents the index of the HRRP target image, N represents the total number of HRRP target images, k represents the category index, C represents the total number of categories, yi represents the label of the HRRP target image, and f ik represents the probability value that the HRRP target image i is considered to be of category k after being processed by the classifier module, c represents the category index, and f ic represents the probability value that the HRRP target image i is considered to be of category c after being processed by the classifier module.

[0103] In one embodiment, a reconstruction error loss function for discriminative feature points is reconstructed based on scattering-related information, and the scattering-related information is used as prior information to converge with the discriminative feature points. The converged discriminative feature points are used as the discriminative feature points to be optimized and input into the scattering center target recognition network to obtain a trained scattering center target recognition network.

[0104] It should be noted that by combining the prior knowledge of the target scattering center with the neural network to construct a scattering center extraction layer, the position and amplitude information of the target scattering center can be effectively extracted, the features extracted by the neural network layer are given clear physical meanings, and at the same time, the radar HRRP target recognition accuracy is improved.

[0105] In one embodiment, an extraction loss function for reconstructing scattering-related information is reconstructed based on scattering-related information:

[0106]

[0107] where x i is the HRRP target image, i∈I is the index of the HRRP target image, I is the number of HRRP target images, Φ is the Fourier basis, and ω * is the optimal sparse coding. And a classification loss function for reconstructing discriminative feature points is:

[0108]

[0109] where i∈I is the index of the HRRP target image, I is the number of HRRP target images, k = 1, 2,..., C is the category of the HRRP target image, C is the number of categories of the HRRP target image, and y iFor the HRRP target image x i is the class label, and f ik is the discriminative feature point of the HRRP target image x belonging to class k i Reconstruct the error loss function of the scattering center target recognition network according to the extraction loss function and the classification loss function:

[0110]

[0111] where x i is the HRRP target image, i ∈ I is the index of the HRRP target image, I is the number of HRRP target images, Φ is the Fourier basis, ω * is the optimal sparse coding, k = 1, 2,..., C is the class of the HRRP target image, C is the number of classes of the HRRP target image, and y i is the class label of the HRRP target image x i and f ik is the discriminative feature point of the HRRP target image x belonging to class k i .

[0112] It should be understood that although Figure 2 the steps in the flowchart are shown in sequence according to the indication of the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise clearly stated in this article, the execution of these steps has no strict order restriction, and these steps can be executed in other orders. Moreover, Figure 2 at least a part of the steps in

[0113] In one embodiment, as Figure 5 shown, a radar target recognition device based on scattering center prior information is provided, including: a network construction module 502, a scattering-related information extraction module 504, a discriminative feature point to be optimized acquisition module 506, a feature point optimization module 508, and an identification optimization training module 510, where:

[0114] The network construction module 502 is used to construct a scattering center target recognition network. The scattering center target recognition network includes: a scattering center extraction layer, an encoding network, and a classifier.

[0115] The scattering-related information extraction module 504 is used to extract the scattering-related information of the HRRP target image through the scattering center extraction layer. The scattering-related information at least includes the scattering center position and the scattering center amplitude information of the current scattering feature point in the HRRP target image.

[0116] The to-be-optimized discrimination feature point acquisition module 506 is used to input the scattering-related information into the encoding network for feature space mapping to obtain the to-be-optimized discrimination feature points.

[0117] The feature point optimization module 508 is used to input the to-be-optimized discrimination feature points into the classifier for feature classification to obtain the discrimination feature points.

[0118] The recognition optimization training module 510 is used to perform gradient backpropagation training based on the scattering-related information and the discrimination feature points to obtain a trained scattering center target recognition network.

[0119] In one embodiment, the recognition optimization training module is further used to reconstruct the extraction loss function of the scattering-related information:

[0120]

[0121] where x i is the HRRP target image, i ∈ I is the index of the HRRP target image, I is the number of HRRP target images, Φ is the Fourier basis, and ω * is the optimal sparse coding. And reconstruct the classification loss function according to the discrimination feature points:

[0122]

[0123] where i ∈ I is the index of the HRRP target image, I is the number of HRRP target images, k = 1, 2,..., C is the category of the HRRP target image, C is the number of categories of the HRRP target image, y i is the category label of the HRRP target image x i , f ik is the discrimination feature point of the HRRP target image x i belonging to category k; reconstruct the error loss function of the scattering center target recognition network according to the extraction loss function and the classification loss function:

[0124]

[0125] where x i is the HRRP target image, i ∈ I is the index of the HRRP target image, I is the number of HRRP target images, Φ is the Fourier basis, and ω *For the optimal sparse coding, k = 1, 2, …, C are the categories of HRRP target images, C is the number of categories of HRRP target images, and y i is the HRRP target image x i 's class label, and f ik is the discriminative feature point of the HRRP target image x i belonging to category k.

[0126] For the specific limitations on the radar target recognition device based on prior information of scattering centers, reference can be made to the limitations on the radar target recognition method based on prior information of scattering centers in the above text, which will not be elaborated here. Each module in the above radar target recognition device based on prior information of scattering centers can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in the processor of the computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above modules.

[0127] In one embodiment, a computer device is provided. The computer device can be a terminal, and its internal structure diagram can be as Figure 6 shown. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a radar target recognition method based on prior information of scattering centers. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the shell of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0128] Those skilled in the art can understand that Figures 5 - 6 the structure shown in

[0129] is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have a different component layout.

[0130] Construct a scattering center target recognition network. The scattering center target recognition network includes: a scattering center extraction layer, an encoding network, and a classifier.

[0131] Extract the scattering-related information of the HRRP target image through the scattering center extraction layer. The scattering-related information at least includes the scattering center position and the scattering center amplitude information of the current scattering feature point in the HRRP target image.

[0132] Input the scattering-related information into the encoding network for feature space mapping to obtain the discriminative feature points to be optimized.

[0133] Input the discriminative feature points to be optimized into the classifier for feature classification to obtain the discriminative feature points.

[0134] Perform gradient backpropagation training based on the scattering-related information and the discriminative feature points to obtain a trained scattering center target recognition network.

[0135] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided by the present invention can include non-volatile and / or volatile memories. Non-volatile memories can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0136] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0137] The above-described embodiments merely represent several implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the appended claims.

Claims

1. A radar target recognition method based on prior information of scattering centers, characterized in that The method includes: Constructing a scattering center target recognition network; the scattering center target recognition network includes: a scattering center extraction layer, an encoding network, and a classifier; Extracting scattering-related information of the HRRP target image through the scattering center extraction layer; the scattering-related information at least includes the scattering center position and scattering center amplitude information of the current scattering feature point in the HRRP target image; Inputting the scattering-related information into the encoding network for feature space mapping to obtain a discriminative feature point to be optimized; Inputting the discriminative feature point to be optimized into the classifier for feature classification to obtain a discriminative feature point; Performing gradient backpropagation training based on the scattering-related information and the discriminative feature point, reconstructing an error loss function based on the scattering-related information and the discriminative feature point, and converging the scattering-related information as prior information with the discriminative feature point. Taking the converged discriminative feature point as the discriminative feature point to be optimized and inputting it into the scattering center target recognition network to obtain a trained scattering center target recognition network.

2. The method according to claim 1, wherein Before the step of constructing the scattering center target recognition network, it includes: Receiving the echo signal transmitted by the target to be measured in the communication baseband through a radar receiver: Among them, is the th individual scattering center in the th range cell, is the transmitted signal of the radar transmitter, is the associated scattering coefficient, is the number of range cells, is the radial distance between the radar and the th individual scattering center in the th range cell, is a rectangular pulse with a width of , is the radar center frequency, is the chirp coefficient, is the propagation speed of electromagnetic waves; Calculating the discrete frequency response of the echo signal: Among them, is the th frequency response, is the number of frequency responses, is the frequency interval, is the th scattering coefficient of the distance unit; Generating a scattering center extraction model by Fourier transform of the discrete frequency response: wherein, is the optimal sparse coding, is the scattering coefficient, is the Fourier basis.

3. The method according to claim 2, wherein Extracting scattering-related information of the HRRP target image through the scattering center extraction layer, including: Extracting the scattering coefficient of the strong scattering center response in the HRRP target image through the scattering center extraction model of the scattering center extraction layer. The scattering coefficient is optimized in the scattering center extraction layer according to the soft threshold iteration method to obtain the scattering-related information corresponding to the optimal sparse coding.

4. The method according to claim 3, wherein Inputting the scattering-related information into the encoding network for feature space mapping to obtain a discriminative feature point to be optimized, including: Inputting the scattering-related information into the encoding network. The encoding network uses a single-channel ResNet network structure to map the scattering-related information to the feature space to be optimized to obtain a discriminative feature point to be optimized.

5. The method according to claim 4, wherein Inputting the discriminative feature point to be optimized into the classifier for feature classification to obtain a discriminative feature point, including: Inputting the discriminative feature point to be optimized into the classifier. The classifier uses the softmax function and a preset classification loss function to perform feature classification on the discriminative feature point to be optimized to obtain a discriminative feature point.

6. The method according to claim 5, characterized in that Reconstructing an error loss function based on the scattering-related information and the discriminative feature point, including: Reconstructing an extraction loss function of the scattering-related information based on the scattering-related information: Among them, is the HRRP target image, is the index of the HRRP target image, is the number of the HRRP target images, is the Fourier basis, is the optimal sparse coding; And reconstructing a classification loss function based on the discriminative feature point: Among them, is the index of the HRRP target image, is the number of the HRRP target images, is the category of the HRRP target image, is the number of categories of the HRRP target images, is the HRRP target image 's category label, is the discriminative feature point of the HRRP target image belonging to the category ; Reconstructing the error loss function of the scattering center target recognition network based on the extraction loss function and the classification loss function: Among them, is the HRRP target image, is the index of the HRRP target image, is the number of the HRRP target images, is the Fourier basis, is the optimal sparse coding, is the category of the HRRP target image, is the number of categories of the HRRP target image, is the HRRP target image 's category label, belongs to the category of the HRRP target image 's discriminative feature points.

7. A radar target recognition device based on prior information of scattering centers, characterized in that, The device includes: A network construction module for constructing a scattering center target recognition network; the scattering center target recognition network includes: a scattering center extraction layer, an encoding network, and a classifier; A scattering-related information extraction module, which is used to extract the scattering-related information of the HRRP target image through the scattering center extraction layer; the scattering-related information at least includes the scattering center position and the scattering center amplitude information of the current scattering feature point in the HRRP target image; An identification feature point to be optimized acquisition module, which is used to input the scattering-related information into the encoding network for feature space mapping to obtain the identification feature point to be optimized; A feature point optimization module, which is used to input the identification feature point to be optimized into the classifier for feature classification to obtain the identification feature point; An identification optimization training module, which is used to perform gradient backpropagation training according to the scattering-related information and the identification feature point, reconstruct an error loss function according to the scattering-related information and the identification feature point, and converge the scattering-related information as prior information with the identification feature point, and input the converged identification feature point into the scattering center target recognition network as the identification feature point to be optimized to obtain a trained scattering center target recognition network.

8. The device according to claim 7, wherein The identification optimization training module is further used to reconstruct an extraction loss function of the scattering-related information according to the scattering-related information: wherein, is the HRRP target image, is the index of the HRRP target image, is the number of the HRRP target images, is the Fourier basis, is the optimal sparse coding; And reconstruct a classification loss function according to the identification feature point: Among them, is the index of the HRRP target image, is the number of the HRRP target images, is the category of the HRRP target image, is the number of categories of the HRRP target images, is the HRRP target image 's category label, is the HRRP target image belonging to the category 's discriminative feature points; Reconstruct an error loss function of the scattering center target recognition network according to the extraction loss function and the classification loss function: Among them, is the HRRP target image, is the index of the HRRP target image, is the number of the HRRP target images, is the Fourier basis, is the optimal sparse coding, is the category of the HRRP target image, is the number of categories of the HRRP target image, is the HRRP target image 's category label, is the one belonging to the category of the HRRP target image 's discriminative feature points.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

Images