Radar target detection method and device based on depth time-frequency matrix manifold network

Abstract

The invention relates to a radar target detection method and device based on a depth time-frequency matrix manifold network. The method comprises the steps of using simulation target signals driven by Doppler steering vectors in sea clutter data obtained in advance to obtain echo data of simulation targets with different signal-to-noise ratios, obtaining a time-frequency characteristic pattern of the echo data by using a time-frequency analysis mode, constructing a time-frequency characteristic matrix according to the echo time-frequency characteristic pattern, and obtaining a time-frequency characteristic matrix according to the time-frequency characteristic matrix. And training a pre-constructed deep manifold network by taking the time-frequency characteristic matrix as a sample so as to enable the deep manifold network to detect a target in echo data to be detected. The method can improve the radar target detection performance.

Description

Technical Field

[0001] The present application relates to the technical field of radar target detection, and in particular to a radar target detection method and device based on a deep time-frequency matrix manifold network. Background Art

[0002] Radar systems are the core of modern surveillance and defense systems and are widely used in aviation, aerospace, and weather forecasting. The main challenge of radar target detection is to accurately detect and identify targets in complex environmental noise and interference. Clutter, that is, non-target reflected signals received by the radar system, is the main source of interference that affects radar detection performance. Different environments, such as sea, mountainous areas, and cities, will produce different types of clutter, which will affect radar target detection to varying degrees.

[0003] Traditional detection technology based on statistical characteristics has high requirements on the target signal-to-noise ratio, slow detection speed, and many false alarm and model mismatch problems. The detection performance cannot meet the needs of modern complex and diverse perception environments. Summary of the invention

[0004] Based on this, it is necessary to provide a radar target detection method and device based on a deep time-frequency matrix manifold network to address the above technical problems.

[0005] A radar target detection method based on a deep time-frequency matrix manifold network, the method comprising:

[0006] Using a simulated target signal driven by a Doppler steering vector in pre-acquired sea clutter data to obtain echo data of simulated targets with different signal-to-noise ratios; the Doppler steering vector is represented by a target signal amplitude parameter, the magnitude of which is related to an average magnitude of the sea clutter data, a signal-to-noise ratio, an initial phase, and a scattering coefficient;

[0007] Acquire a time-frequency characteristic diagram of the echo data by using a time-frequency analysis method, and construct a time-frequency characteristic matrix according to the echo time-frequency characteristic diagram;

[0008] The time-frequency feature matrix is ​​used as a sample to train a pre-constructed deep manifold network, so that the deep manifold network can detect the target in the echo data to be detected.

[0009] In one embodiment, the method further includes: using Doppler-guided vector-driven simulated target signals in pre-acquired sea clutter data to be:

[0010]

[0011] α=β×mean(S)cosθ

[0012] Where S is the sea clutter data, s is the Doppler steering vector, α is the target signal amplitude parameter, θ represents the initial phase, β represents the scattering coefficient, N represents the number of target signal pulses, and f d represents the normalized Doppler frequency.

[0013] In one of the embodiments, it also includes: respectively using the STFT time-frequency analysis method and the WVD time-frequency analysis method to obtain the echo time-frequency characteristics of the echo data.

[0014] In one of the embodiments, it further includes: constructing a time-frequency feature matrix according to the echo time-frequency feature graph:

[0015]

[0016] Among them, X0 represents the echo time-frequency characteristic diagram, X2 represents the time-frequency characteristic matrix, ε represents the correction threshold, I is the unit matrix, and U1 represents the orthogonal matrix.

[0017] In one embodiment, the method further includes: normalizing the time-frequency feature matrix using a standard score Z-score specification to:

[0018]

[0019] Among them, X3 represents the standardized image sample matrix, μ represents the mean of the original data set, and σ is the standard deviation of the original data set.

[0020] In one of the embodiments, the method further includes: taking the time-frequency feature matrix as a sample; the time-frequency feature matrix contains half of the targets and half of the non-targets;

[0021] The samples are used to design training samples, verification samples and test samples; wherein the training samples are used to perform deep manifold network propagation training; the verification samples are used to check the state and convergence of the deep manifold network; and the test samples are used to ultimately evaluate the network performance of the deep manifold network.

[0022] In one of the embodiments, it also includes: when the STFT time-frequency analysis method is used to obtain the echo time-frequency characteristics of the echo data, the detection performance of the deep manifold network improves with the increase of the number of network layers of the deep manifold network; when the WVD time-frequency analysis method is used to obtain the echo time-frequency characteristics of the echo data, the detection performance of the deep manifold network has an optimal number of network layers near a specific signal-to-noise ratio.

[0023] A radar target detection device based on a deep time-frequency matrix manifold network, the device comprising:

[0024] A data acquisition module, used for using a simulated target signal driven by a Doppler steering vector in pre-acquired sea clutter data to acquire echo data of a simulated target with different signal-to-noise ratios; the Doppler steering vector is represented by a target signal amplitude parameter, and the magnitude of the target signal amplitude parameter is related to the average magnitude of the sea clutter data, the signal-to-noise ratio, the initial phase and the scattering coefficient;

[0025] A time-frequency feature matrix construction module is used to obtain the time-frequency feature graph of the echo data by using a time-frequency analysis method, and construct a time-frequency feature matrix according to the echo time-frequency feature graph;

[0026] The target detection module is used to train a pre-constructed deep manifold network using the time-frequency feature matrix as a sample, so that the deep manifold network can detect the target in the echo data to be detected.

[0027] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0028] Using a simulated target signal driven by a Doppler steering vector in pre-acquired sea clutter data to obtain echo data of simulated targets with different signal-to-noise ratios; the Doppler steering vector is represented by a target signal amplitude parameter, the magnitude of which is related to an average magnitude of the sea clutter data, a signal-to-noise ratio, an initial phase, and a scattering coefficient;

[0029] Acquire a time-frequency characteristic diagram of the echo data by using a time-frequency analysis method, and construct a time-frequency characteristic matrix according to the echo time-frequency characteristic diagram;

[0030] The time-frequency feature matrix is ​​used as a sample to train a pre-constructed deep manifold network, so that the deep manifold network can detect the target in the echo data to be detected.

[0031] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0032] Using a simulated target signal driven by a Doppler steering vector in pre-acquired sea clutter data to obtain echo data of simulated targets with different signal-to-noise ratios; the Doppler steering vector is represented by a target signal amplitude parameter, the magnitude of which is related to an average magnitude of the sea clutter data, a signal-to-noise ratio, an initial phase, and a scattering coefficient;

[0033] Acquire a time-frequency characteristic diagram of the echo data by using a time-frequency analysis method, and construct a time-frequency characteristic matrix according to the echo time-frequency characteristic diagram;

[0034] The time-frequency feature matrix is ​​used as a sample to train a pre-constructed deep manifold network, so that the deep manifold network can detect the target in the echo data to be detected.

[0035] The above-mentioned radar target detection method and device based on deep time-frequency matrix manifold network use Doppler steering vector-driven simulated target signals in pre-acquired sea clutter data to obtain echo data of simulated targets with different signal-to-noise ratios, and use time-frequency analysis to obtain the time-frequency feature map of the echo data. According to the echo time-frequency feature map, a time-frequency feature matrix is ​​constructed. The pre-constructed deep manifold network is trained with the time-frequency feature matrix as a sample, so that the deep manifold network detects the target in the echo data to be detected. The present invention performs positive definite processing on the time-frequency feature map containing the simulated target and constructs the time-frequency feature matrix. The deep manifold network reduces the dimension of the time-frequency feature matrix and aggregates it, which can effectively remove redundant information in the data, and is significantly better than the traditional convolutional neural network detection method in relative detection performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 1 is a flow chart of a radar target detection method based on a deep time-frequency matrix manifold network in one embodiment;

[0037] Figure 2 1 is a diagram of sea clutter data in an embodiment; wherein (a) is a diagram of nearby sea clutter data, and (b) is a diagram of selected sea clutter data;

[0038] Figure 3 1 is a signal diagram after adding a simulation target in one embodiment; wherein (a) is a signal diagram after adding a 0dB simulation target, and (b) is a signal diagram after adding a -5dB simulation target;

[0039] Figure 4 A diagram of the SPDNet structure in one embodiment;

[0040] Figure 5 : is a STFT characteristic diagram of the signal-to-noise ratio in an embodiment; wherein (a) is no target, (b) is 0 dB, (c) is -5 dB, (d) is -10 dB, (e) is -15 dB, and (f) is -20 dB;

[0041] Figure 6 is a STFT characteristic matrix of the signal-to-noise ratio in one embodiment; wherein (a) is no target, (b) is 0 dB, (c) is -5 dB, (d) is -10 dB, (e) is -15 dB, and (f) is -20 dB;

[0042] Figure 7: is a WVD characteristic diagram of the signal-to-noise ratio in an embodiment; wherein (a) is no target, (b) is 0 dB, (c) is -5 dB, (d) is -10 dB, (e) is -15 dB, and (f) is -20 dB;

[0043] Figure 8 is a WVD characteristic matrix of the signal-to-noise ratio in an embodiment; wherein (a) is no target, (b) is 0 dB, (c) is -5 dB, (d) is -10 dB, (e) is -15 dB, and (f) is -20 dB;

[0044] Fig. 9 A data input and output flow chart in one embodiment;

[0045] Fig.10 A schematic diagram of selecting a signal-to-noise ratio for a training network in one embodiment;

[0046] Fig.11 Detection probability diagrams of different networks in one embodiment; wherein (a) is the STFT feature matrix, and (b) is the WVD feature matrix;

[0047] Fig.12 Detection probability diagram of SPDNet under different learning rates in one embodiment; wherein (a) is the STFT feature matrix, and (b) is the WVD feature matrix;

[0048] Fig.13 Detection probability diagram of SPDNet under different weight attenuation in one embodiment; wherein (a) is the STFT feature matrix, and (b) is the WVD feature matrix;

[0049] Fig.14 Detection probability diagram of SPDNet under different network layers in one embodiment; wherein (a) is the STFT feature matrix, and (b) is the WVD feature matrix;

[0050] Fig.15 is a structural block diagram of a radar target detection device based on a deep time-frequency matrix manifold network in one embodiment;

[0051] Fig.16 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0053] In one embodiment, Figure 1As shown, a radar target detection method based on a deep time-frequency matrix manifold network is provided, comprising the following steps:

[0054] Step 102 : Using the simulated target signal driven by the Doppler steering vector in the pre-acquired sea clutter data, to acquire the echo data of the simulated target having different signal-to-noise ratios.

[0055] The Doppler steering vector is represented by the target signal amplitude parameter, and the size of the target signal amplitude parameter is related to the average size of the sea clutter data, the signal-to-noise ratio, the initial phase and the scattering coefficient.

[0056] Specifically, the radar of the present invention is an X-band solid-state power amplifier surveillance / navigation radar with a transmission time of 40ns-100us and a 360° all-round scanning in the horizontal plane. The radar technical parameters are shown in Table 1.

[0057] Table 3-1 Radar parameters

[0058]

[0059]

[0060] The present invention uses the "20191012112446_01_staring.mat" in the public data to construct a network data set, which contains 10,000 pulse trains and a total of 5,250 distance unit sea clutter data. The radar echo data are all pulse compressed data. In the present invention, the 186th to 190th distance units are selected as experimental clutter sample data, such as Figure 2 (a) The distance unit near the data selected by the present invention is intercepted, wherein the data in the red frame is the experimental data selected by the present invention. Figure 2 (b) shows the sea clutter data after selecting the range unit.

[0061] In this step, the signal-to-noise ratio is a variable in the range of [0dB, -20dB], the initial phase θ of the target is 0, and the scattering coefficient β is 0.8. The N of the simulation data set is 10000, and f d Set to 0.15.

[0062] Figure 3 (a) and (b) show the signal graphs after adding 0dB and -5dB simulated targets. It can be clearly seen that the data graph changes more after adding a large signal-to-noise ratio target, but the data graph hardly changes after adding a signal with a relatively small signal-to-noise ratio. The data graphs containing clutter and target signals are almost indistinguishable from each other under naked eye observation. Since the added steering vector is a phase factor, the specific position of the signal cannot be directly observed on the signal amplitude graph.

[0063] Step 104: Obtain a time-frequency characteristic diagram of the echo data by using a time-frequency analysis method, and construct a time-frequency characteristic matrix according to the echo time-frequency characteristic diagram.

[0064] In this step, for the matrix A n×n Diagonalization and eigenvalue decomposition can be performed:

[0065] U -1 AU=Σ

[0066] A=U -1 ΣU

[0067] The eigenvalues ​​and eigenvectors of a square matrix are the essential characteristics of a matrix. m×n It is impossible to perform eigenvalue decomposition, so the singular value decomposition method is used to decompose the non-square matrix B m×n Decompose. Since the singular value decomposition originates from the nature of the matrix, it is applicable to all real matrices:

[0068] BV=UΣ

[0069] B=UΣV -1 =UΣV T

[0070] Among them, U is an m×m orthogonal matrix, Σ is an m×n orthogonal matrix, and V is an n×n orthogonal matrix. The symmetric positive definite method of any matrix X0 is to transpose the matrix X0 and its conjugate When multiplied, the generated matrix X1 must have conjugate symmetry, and then by correcting the negative eigenvalues, the time-frequency characteristic matrix is ​​obtained.

[0071] Step 106 , using the time-frequency feature matrix as a sample, training a pre-constructed deep manifold network so that the deep manifold network can detect the target in the echo data to be detected.

[0072] In the above-mentioned radar target detection method based on deep time-frequency matrix manifold network, a simulated target signal driven by a Doppler steering vector is used in the pre-acquired sea clutter data to obtain echo data of simulated targets with different signal-to-noise ratios, and a time-frequency feature map of the echo data is obtained by using a time-frequency analysis method. According to the echo time-frequency feature map, a time-frequency feature matrix is ​​constructed, and the pre-constructed deep manifold network is trained with the time-frequency feature matrix as a sample, so that the deep manifold network detects the target in the echo data to be detected. The present invention performs positive definite processing on the time-frequency feature map containing the simulated target and constructs the time-frequency feature matrix. The deep manifold network reduces the dimension and aggregates the time-frequency feature matrix, which can effectively remove redundant information in the data, and is significantly better than the traditional convolutional neural network detection method in relative detection performance.

[0073] It should be introduced that the deep manifold network is a deep learning algorithm that processes matrix manifolds. It uses matrix data as input to the neural network to perform nonlinear training on the matrix manifold. Among these networks, the deep manifold network (SPDNet) is a commonly used structure.

[0074] Figure 4 It is the structural diagram of SPDNet. The generated SPD data is sent to the input layer, and then after being processed by multiple bilinear mapping layers and feature correction layers, the dimension of the SPD matrix gradually decreases and the features gradually aggregate. Finally, these matrices are mapped to Euclidean space through logarithmic Euclidean measurement and classified in space. The number of dimensionality reduction and aggregation iteration layers of the original network in the present invention is 6, namely three bilinear mapping layers and three feature correction layers.

[0075] Similar to the classic convolutional neural network, SPDNet contains two key functional layers: Bilinear Map Layer (BiMap Layer) and Eigenvalue Rectification Layer (ReEig Layer). After processing by these layers, the original SPD matrix will be converted into a new matrix with smaller dimensions and more significant features. This conversion enhances the distinction between targets and clutter on the matrix manifold.

[0076] The purpose of the bilinear mapping layer is to make the input SPD matrix smaller and more distinguishable. Through the bilinear mapping layer, the input on the original SPD manifold is converted into a new input to form a new SPD manifold. In other words, the data space on each bilinear mapping layer corresponds to an SPD manifold. The calculation formula of the bilinear mapping layer is defined as follows:

[0077]

[0078] Among them, X k-1 is the input matrix of the k-1th layer, X k is the output matrix of the kth layer, W k is a weight matrix with full row rank.

[0079] The purpose of the feature correction layer is to correct the input SPD matrix, correcting the negative eigenvalue to a positive number, ensuring the symmetric positive definiteness of the matrix, and thus retaining more significant features. The calculation formula of the eigenvalue correction layer is defined as follows:

[0080]

[0081] Among them, U k-1 and Σ k-1 By eigenvalue decomposition Obtained, ε is the correction threshold, and I is the identity matrix.

[0082] After passing through multiple bilinear mapping layers and eigenvalue correction layers, the SPD matrix enters the Log-Euclidean calculation layer (Eigenvalue Logarithm, LogEig Layer). The logarithmic Euclidean metric gives the SPD matrix manifold a Lie group structure that follows the characteristics of the Riemann manifold, and expands the SPD matrix manifold into a flat Euclidean space, forming a vector

[48] The calculation formula of the Log-Euclidean calculation layer is as follows:

[0083]

[0084] Among them, log(Σ k-1 ) is a diagonal matrix of logarithmic eigenvalues. After logarithmic Euclidean measurement processing, the input SPD sample matrix is ​​classified using a linear classification method.

[0085] In classification problems, the final output layer is generally a Softmax layer. The vectors that pass through the LogEig Layer are converted into a probability distribution in the range of [0,1] and a sum of 1 by the Softmax function. The calculation formula of Softmax is as follows:

[0086]

[0087] Among them, x i is the output of the ith node, C is the number of output nodes, that is, the number of classifications, where is the probability of having a target or not having a target, and the final detection probability is obtained from the classification result.

[0088] In addition to the forward propagation process, the back propagation process of SPDNet uses the Stochastic Gradient Descent (SGD) algorithm, which is a simple but very effective method that can be used for both classification and regression calculations. SGD randomly extracts a group from the samples, updates them once according to the gradient after training, and then extracts another group and updates them again. The randomness here means that during each iteration, the samples are randomly shuffled to effectively reduce the parameter update offset problem caused by the samples. The calculation formula is as follows:

[0089]

[0090] Among them, α is the network learning rate, θ j is the jth coefficient of the prediction function, x(i) is the input data, y(i) is the actual value of each sample in the data set, and h θ is the prediction function, and the predicted y(i) value is calculated according to θ for each input x(i).

[0091] In one embodiment, the simulated target signal using Doppler steering vector drive in pre-acquired sea clutter data is:

[0092]

[0093] α=β×mean(S)cosθ

[0094] Where S is the sea clutter data, s is the Doppler steering vector, α is the target signal amplitude parameter, θ represents the initial phase, β represents the scattering coefficient, N represents the number of target signal pulses, and f d represents the normalized Doppler frequency. The signal-to-noise ratio is defined as: SCR = 10log10 (α 2 / σ 2 ), σ 2 is the variance of the clutter data.

[0095] In another embodiment, the STFT time-frequency analysis method and the WVD time-frequency analysis method are respectively used to obtain the echo time-frequency characteristics of the echo data.

[0096] STFT time-frequency analysis method:

[0097] Specifically, after superimposing the simulated targets with different signal-to-noise ratios on the radar echo data of each range unit, a Hamming window with a length of 25, an overlap length of 24, and a sampling point number of 100 is used to perform short-time Fourier transform, and the size of the generated time-frequency graph is 100×76. Randomly taking the 77th time-frequency graph of the 187th range unit as an example, the following is obtained: Figure 5 Feature map of the signal-to-noise ratio of the parts without and with targets.

[0098] Observe the time-frequency characteristic diagram with and without targets. From the perspective of amplitude, the clutter is concentrated in the low-frequency part, and the clutter spectrum line distribution amplitude is large. The target is distributed at the frequency f d = 0.15, the amplitude decreases as the signal-to-noise ratio decreases. From the graphical characteristics, the clutter is irregularly distributed, with all frequencies distributed, while the time-frequency distribution containing the target is a straight line. In the STFT experimental results, the target and clutter frequency distributions show a corresponding relationship.

[0099] For the feature graphs with different signal-to-noise ratios, as the signal-to-noise ratio decreases, the target's time-frequency feature amplitude decreases significantly. When the signal-to-noise ratio drops to -10dB, only a small part of the signal can be seen, and if the signal-to-noise ratio continues to decrease, it is no longer possible to distinguish the target from the clutter in the image with the naked eye.

[0100] The obtained time-frequency graph is symmetrically positively defined to convert it into a time-frequency feature matrix with a size of 100×100, such as Figure 6 shown.

[0101] It can be seen from the STFT feature matrix that on the matrix diagonal, the center point is low-frequency clutter, and the upper left point is the target. Corresponding to the feature map, when the signal-to-noise ratio is reduced to -10dB, the target can still be distinguished from the image color depth by the naked eye, but when the signal-to-noise ratio is further reduced, it is difficult to distinguish the target from the clutter from the feature matrix.

[0102] WVD time-frequency analysis method:

[0103] Similar to the short-time Fourier transform, the echo data of each range unit are superimposed with simulated targets of different signal-to-noise ratios to perform Wigner-Ville distribution calculation, and the size of the generated time-frequency diagram is 100×76. Randomly taking the 77th time-frequency diagram of the 187th range unit as an example, the following is obtained: Figure 7 Feature map of the signal-to-noise ratio of the parts without and with targets.

[0104] Observe the time-frequency characteristic diagram with and without targets. From the energy perspective, the clutter is concentrated near the zero frequency, and the clutter spectrum distribution amplitude is large. The target is distributed in f d =0.3 spectrum line, the intensity decreases as the signal-to-noise ratio decreases. From the graphical features, the clutter still presents an irregular distribution, while the time-frequency distribution containing the target is a continuous straight line.

[0105] For feature maps with different signal-to-noise ratios, as the signal-to-noise ratio decreases, the time-frequency feature strength of the target decreases. When the signal-to-noise ratio is -10dB, it is basically impossible to distinguish the target from the clutter in the feature map.

[0106] It is worth noting that in the WVD feature map, the targets are distributed in f d =0.3 The reason for the spectrum line is that there is a delay before and after the calculation of the radar echo WVD time-frequency characteristics. This delay will result in the remaining twice the frequency of the corresponding signal after the calculation of the conjugate convolution, forming the target f d = 0.3. And the middle f d =0.15 The spectrum line is the cross term between sea clutter and signal. Although the cross term will affect the resolution, for target detection in image and matrix learning networks, the target and clutter influence each other, which helps to enhance the strength of the cross term between weak targets and strong clutter, so that the target can still be detected through the cross term when the signal-to-noise ratio is smaller.

[0107] Then the obtained time-frequency graph is symmetrically positive-definited to convert it into a time-frequency feature matrix with a size of 100×100, such as Figure 8 shown.

[0108] It can be seen from the WVD feature matrix that the center point on the matrix diagonal is low-frequency clutter, the upper left point is the target, and the middle point is the cross term of the target and clutter. Corresponding to the feature graph, when the signal-to-noise ratio drops to -10dB, the target is still indistinguishable by the naked eye, but due to the existence of the cross term, a weak target can be barely distinguished from the cross term, proving that under the same conditions, the cross term is conducive to better target detection, so the influence of the cross term on the resolution can be ignored in the present invention.

[0109] In one embodiment, according to the echo time-frequency feature diagram, a time-frequency feature matrix is ​​constructed as follows:

[0110]

[0111] Among them, X0 represents the echo time-frequency characteristic diagram, X2 represents the time-frequency characteristic matrix, ε represents the correction threshold, I is the unit matrix, and U1 represents the orthogonal matrix.

[0112] To ensure that each input image sample matrix has the same scale, in another embodiment, the time-frequency feature matrix is ​​normalized using the standard score Z-score specification as follows:

[0113]

[0114] Among them, X3 represents the standardized image sample matrix, μ represents the mean of the original data set, and σ is the standard deviation of the original data set.

[0115] In one of the embodiments, a time-frequency feature matrix is ​​used as a sample; the time-frequency feature matrix contains half the number of targets and half the number of non-targets; training samples, verification samples and test samples are designed based on the samples; wherein the training samples are used for deep manifold network propagation training; the verification samples are used to check the state and convergence of the deep manifold network; and the test samples are used to finally evaluate the network performance of the deep manifold network.

[0116] In another embodiment, when the STFT time-frequency analysis method is used to obtain the echo time-frequency characteristics of the echo data, the detection performance of the deep manifold network improves with the increase of the number of network layers of the deep manifold network. When the WVD time-frequency analysis method is used to obtain the echo time-frequency characteristics of the echo data, the detection performance of the deep manifold network has an optimal number of network layers near a specific signal-to-noise ratio.

[0117] In summary, based on the above method, the results are analyzed:

[0118] The obtained STFT and WVD time-frequency feature maps are put into the classification output of the fully connected layer adjusted by AlexNet, GoogleNet and DarkNet pre-trained by ImageNet for transfer learning. The STFT and WVD time-frequency feature matrices are put into the deep manifold network SPDNet for learning. The number of targets and non-targets in all feature maps and feature matrices accounts for 50% each. The sample data includes training samples, verification samples and test samples. The training samples are used for network propagation training in the data set to determine the weights and biases of the network model; the verification samples are used to check the network status and convergence; the test samples are used to finally evaluate the network performance. The data input and output process framework is as follows: Fig. 9 shown.

[0119] After training on multiple signal-to-noise ratio data, the SPDNet trained on the target and clutter dataset with a signal-to-noise ratio of -5dB has the best detection performance. Fig.10 As shown in the figure, the sample data used for network training are all the -5dB time-frequency feature maps and time-frequency feature matrices extracted by STFT and WVD. Because when the signal-to-noise ratio is -6dB, SPDNet trained at any learning rate cannot converge, so the network sample data involved in the selection is only -5dB.

[0120] For AlexNet, GoogleNet, and DarkNet, the signal-to-noise ratio should be consistent with SPDNet's -5dB. 70% of them are randomly selected as training samples and 30% as verification samples. Finally, all the time-frequency feature maps of the trained network are tested and compared with the actual presence or absence of the target to obtain the detection probability distribution under different signal-to-noise ratios. For SPDNet, the sample data is the -5dB time-frequency feature matrix after the time-frequency feature map is symmetric and positively defined. 40% of them are randomly selected as training samples and 20% as verification samples. The remaining 40% of the test samples of the time-frequency feature matrix of all signal-to-noise ratios participate in the detection probability distribution of the test network for different signal-to-noise ratios. The corresponding number of samples is shown in Table 2.

[0121] After network learning of the training samples and validation samples, the trained SPDNet processes the test samples under different signal-to-noise ratios to obtain the corresponding detection probabilities. d ) is the ratio of the number of samples containing the target that are judged as targets by each network. The detection performance of the network is judged according to the size of the detection probability. The calculation formula of the detection probability is as follows:

[0122]

[0123] Among them, TP is the number of positive samples predicted as positive, that is, the number of target samples judged as having targets, FN is the number of positive samples predicted as negative, that is, the number of target samples judged as having no targets, P d It can reflect the model's ability to correctly predict the full range of positive samples.

[0124] Table 2 Number of samples of various types in SPDNet

[0125]

[0126] Performance Analysis:

[0127] The hyperparameters of each network training are shown in Table 3. Batchsize is the number of samples that the network enters training at a time to update the network weights, which directly affects the memory consumption, speed and stability of model training; the learning rate controls the speed of weight update; weight decay controls the data perturbation of weight update; the number of iterations refers to the number of times the entire training data is traversed; the number of network layers refers to the total number of bilinear mapping layers and feature correction layers in SPDNet.

[0128] Table 3 Network model training hyper parameters

[0129]

[0130] from Fig.11 In the data, when the input is the STFT feature matrix, SPDNet reaches a detection probability of 70% at around -8.12dB, while ALexNet reaches a detection probability of 70% at around -7.58dB, GoogleNet reaches a detection probability of 70% at around -7.42dB, and DarkNet reaches a detection probability of 70% at around -7.61dB. When the input is the WVD feature matrix, SPDNet reaches a detection probability of 70% at around -8.51dB, while ALexNet reaches a detection probability of 70% at around -7.97dB, GoogleNet reaches a detection probability of 70% at around -7.74dB, and DarkNet reaches a detection probability of 70% at around -8.09dB. Therefore, under the same conditions, the performance of SPDNet with STFT samples and WVD samples as input is at least 0.51dB and 0.54dB higher than that of convolutional neural networks, respectively.

[0131] The experimental results show that SPDNet can effectively remove redundant information in the data by reducing the dimension of the matrix, while the convolutional neural network has no dimensionality reduction effect. The research method of the present invention is superior to the convolutional neural network in terms of detection performance.

[0132] Hyperparameter Analysis:

[0133] The performance of deep learning networks depends largely on the choice of hyperparameters, and for most neural networks, changing the learning rate and weight decay will have a certain substantial impact on network training. In addition, the initial training can only meet the convergence of network training, but it is impossible to evaluate whether the converged network can achieve the optimal detection performance. Therefore, in-depth analysis of learning rate and weight decay is valuable. The discussion of the number of network layers is to determine whether the dimensionality reduction process of SPDNet is too fast or too slow, thereby affecting the information loss or redundancy in the matrix. The following will study the impact of learning rate, weight decay and number of network layers on the deep manifold network SPDNet.

[0134] Learning Rate:

[0135] During the training process, the learning rate determines the speed and direction of the model weight update. Specifically, the learning rate is the amplitude of the direction of parameter change adjustment. If the learning rate is set too high, it may cause the model loss function to fluctuate over a large range or even diverge; on the contrary, if the learning rate is too low, it may fall into a local minimum and fail to reach the global optimal solution. The present invention selects learning rates of 0.01, 0.001, and 0.0001 for comparative experiments to study the ideal learning rate of SPDNet.

[0136] Fig.12 The detection probability curve of SPDNet under different learning rates. It can be seen that as the signal-to-noise ratio gradually decreases, the smaller the learning rate, the smoother the detection probability curve. When the input is the STFT feature matrix, when the signal-to-noise ratio is greater than -8.4dB, the network detection performance with a learning rate of 0.01 is the best, and when the signal-to-noise ratio is less than -8.4dB, the network detection performance with a learning rate of 0.0001 is the best. When the input is the WVD feature matrix, when the signal-to-noise ratio is greater than -8.39dB, the network detection performance with a learning rate of 0.01 is the best; when the signal-to-noise ratio range is -9.99dB--8.39dB, the network detection performance with a learning rate of 0.001 is the best; when the signal-to-noise ratio is less than -9.99dB, the network detection performance with a learning rate of 0.0001 is the best.

[0137] Weight decay:

[0138] Weight decay is a regularization method that reduces the model complexity by reducing the impact of data disturbances, making the model smoother, and suppressing model overfitting, thereby improving the generalization of the model. The present invention selects weight decays of 0.005, 0.0005, and 0.00005 for comparative experiments to study the ideal weight decay of SPDNet.

[0139] from Fig.13It can be seen that no matter the input is STFT or WVD time-frequency feature matrix, the detection probability does not change with the change of weight decay. From the effect of weight decay, the training process of SPDNet does not produce overfitting problems, the loss function can converge normally, and the training effect of the network is good.

[0140] Network layers:

[0141] Generally speaking, for deep learning networks, increasing the number of network layers helps to improve the recognition of target features, thereby improving the detection performance of the network. However, simply increasing the number of network layers not only increases the training time, but may also cause overfitting. At the same time, excessive matrix aggregation will also lead to loss of information features. However, reducing the number of network layers will lead to insufficient network feature extraction, and too large a dimensionality reduction span will also lead to loss of information features. The present invention selects comparative experiments with 4, 6, 8, and 10 network layers to study the ideal number of network layers of SPDNet, ensuring that the final matrix size of the dimensionality reduction aggregation of all network layers is the same, and the dimensionality reduction matrix size corresponding to the number of network layers is shown in Table 4.

[0142] Table 4 Network layer number reduction matrix

[0143]

[0144] like Fig.14 , when the input is the STFT feature matrix, the more network layers there are, the better the detection performance of SPDNet. When the input is the WVD feature matrix, the performance difference of different network layers is small. However, when the signal-to-noise ratio is greater than -9.10dB, the performance of the network with 8 layers is the best, and when the signal-to-noise ratio is less than -9.10dB, the performance of the network with 4 layers is the best. For larger targets, the features are more prominent, and it is necessary to slowly reduce the dimension and aggregate step by step to avoid losing information features as much as possible; for weak targets, the features are not prominent, and the dimensionality can be quickly reduced and aggregated without causing a large impact, so the above experimental situation occurs.

[0145] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence according to the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in the present invention, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0146] In one embodiment, Fig.15 As shown, a radar target detection device based on a deep time-frequency matrix manifold network is provided, comprising: a data acquisition module 1502, a time-frequency feature matrix construction module 1504 and a target detection module 1506, wherein:

[0147] The data acquisition module 1502 is used to use the simulated target signal driven by the Doppler steering vector in the pre-acquired sea clutter data to acquire the echo data of the simulated target with different signal-to-noise ratios; the Doppler steering vector is represented by the target signal amplitude parameter, and the magnitude of the target signal amplitude parameter is related to the average magnitude of the sea clutter data, the signal-to-noise ratio, the initial phase and the scattering coefficient;

[0148] The time-frequency feature matrix construction module 1504 is used to obtain the time-frequency feature graph of the echo data by using the time-frequency analysis method, and construct the time-frequency feature matrix according to the echo time-frequency feature graph;

[0149] The target detection module 1506 is used to train a pre-constructed deep manifold network using the time-frequency feature matrix as a sample, so that the deep manifold network can detect the target in the echo data to be detected.

[0150] In one embodiment, the data acquisition module 1502 is further configured to use the simulated target signal driven by the Doppler steering vector in the pre-acquired sea clutter data as:

[0151]

[0152] α=β×mean(S)cosθ

[0153] Where S is the sea clutter data, s is the Doppler steering vector, α is the target signal amplitude parameter, θ represents the initial phase, β represents the scattering coefficient, N represents the number of target signal pulses, and f d represents the normalized Doppler frequency.

[0154] In one embodiment, the time-frequency feature matrix construction module 1504 is further used to respectively adopt the STFT time-frequency analysis method and the WVD time-frequency analysis method to obtain the echo time-frequency features of the echo data.

[0155] In one embodiment, the time-frequency feature matrix construction module 1504 is further used to construct a time-frequency feature matrix according to the echo time-frequency feature graph:

[0156]

[0157] Among them, X0 represents the echo time-frequency characteristic diagram, X2 represents the time-frequency characteristic matrix, ε represents the correction threshold, I is the unit matrix, and U1 represents the orthogonal matrix.

[0158] In one embodiment, the time-frequency feature matrix construction module 1504 is further configured to normalize the time-frequency feature matrix using a standard score Z-score specification as follows:

[0159]

[0160] Among them, X3 represents the standardized image sample matrix, μ represents the mean of the original data set, and σ is the standard deviation of the original data set.

[0161] In one embodiment, the target detection module 1506 is also used to use the time-frequency feature matrix as a sample; the time-frequency feature matrix contains half of the targets and half of the non-targets; the samples are used to design training samples, verification samples and test samples; wherein the training samples are used for deep manifold network propagation training; the verification samples are used to check the state and convergence of the deep manifold network; and the test samples are used to finally evaluate the network performance of the deep manifold network.

[0162] In one of the embodiments, when the STFT time-frequency analysis method is used to obtain the echo time-frequency characteristics of the echo data, the detection performance of the deep manifold network improves as the number of network layers of the deep manifold network increases. When the WVD time-frequency analysis method is used to obtain the echo time-frequency characteristics of the echo data, the detection performance of the deep manifold network has an optimal number of network layers near a specific signal-to-noise ratio.

[0163] For the specific limitations of the radar target detection device based on the deep time-frequency matrix manifold network, please refer to the limitations of the radar target detection method based on the deep time-frequency matrix manifold network mentioned above, which will not be repeated here. Each module in the above-mentioned radar target detection device based on the deep time-frequency matrix manifold network can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0164] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig.16As 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, a radar target detection method based on a deep time-frequency matrix manifold network is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0165] Those skilled in the art will understand that Fig.16 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0166] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method in the above embodiment when executing the computer program.

[0167] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method in the above embodiment are implemented.

[0168] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and 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-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), 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).

[0169] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, 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, they should be considered to be within the scope of this specification.

[0170] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A radar target detection method based on deep time-frequency matrix manifold network, It is characterized in that The method comprises: Using a simulated target signal driven by a Doppler steering vector in pre-acquired sea clutter data to acquire echo data of simulated targets with different signal-to-noise ratios; the Doppler steering vector is represented by a target signal amplitude parameter, the magnitude of which is related to an average magnitude of the sea clutter data, a signal-to-noise ratio, an initial phase, and a scattering coefficient; Acquire a time-frequency characteristic diagram of the echo data by using a time-frequency analysis method, and construct a time-frequency characteristic matrix according to the echo time-frequency characteristic diagram; The time-frequency feature matrix is ​​used as a sample to train a pre-constructed deep manifold network, so that the deep manifold network can detect the target in the echo data to be detected.

2. The method according to claim 1, It is characterized in that Simulated target signals driven using Doppler-steering vectors in pre-acquired sea clutter data, including: The simulated target signal using Doppler-steering vector drive in pre-acquired sea clutter data is: α=β×mean(S)cosθ Where S is the sea clutter data, s is the Doppler steering vector, α is the target signal amplitude parameter, θ represents the initial phase, β represents the scattering coefficient, N represents the number of target signal pulses, and f d represents the normalized Doppler frequency.

3. The method according to claim 1, It is characterized in that The method of obtaining the echo time-frequency characteristics of the echo data by using the time-frequency analysis method includes: The STFT time-frequency analysis method and WVD time-frequency analysis method are used to obtain the echo time-frequency characteristics of the echo data.

4. The method according to claim 1, It is characterized in that According to the echo time-frequency feature graph, a time-frequency feature matrix is ​​constructed, including: According to the echo time-frequency feature diagram, the time-frequency feature matrix is ​​constructed as follows: Among them, X 0 Represents the echo time-frequency characteristic diagram, X 2 represents the time-frequency feature matrix, ε represents the correction threshold, I is the unit matrix, U 1 represents an orthogonal matrix.

5. The method according to claim 4, It is characterized in that The method further comprises: The time-frequency feature matrix is ​​normalized using the standard score Z-score specification as follows: Among them, X 3 represents the normalized image sample matrix, μ represents the mean of the original data set, and σ is the standard deviation of the original data set.

6. The method according to claim 1, It is characterized in that Taking the time-frequency feature matrix as a sample, training a pre-built deep manifold network includes: The time-frequency feature matrix is ​​used as a sample; the time-frequency feature matrix contains half of the targets and half of the non-targets; The samples are used to design training samples, verification samples and test samples; wherein the training samples are used to perform deep manifold network propagation training; the verification samples are used to check the state and convergence of the deep manifold network; and the test samples are used to ultimately evaluate the network performance of the deep manifold network.

7. The method according to claim 3, It is characterized in that When the STFT time-frequency analysis method is used to obtain the echo time-frequency characteristics of the echo data, the detection performance of the deep manifold network improves with the increase of the number of network layers of the deep manifold network. When the WVD time-frequency analysis method is used to obtain the echo time-frequency characteristics of the echo data, the detection performance of the deep manifold network has an optimal number of network layers near a specific signal-to-noise ratio.

8. A radar target detection device based on deep time-frequency matrix manifold network, It is characterized in that The device comprises: A data acquisition module, used for using a simulated target signal driven by a Doppler steering vector in pre-acquired sea clutter data to acquire echo data of a simulated target with different signal-to-noise ratios; the Doppler steering vector is represented by a target signal amplitude parameter, and the magnitude of the target signal amplitude parameter is related to the average magnitude of the sea clutter data, the signal-to-noise ratio, the initial phase and the scattering coefficient; A time-frequency feature matrix construction module is used to obtain the time-frequency feature graph of the echo data by using a time-frequency analysis method, and construct a time-frequency feature matrix according to the echo time-frequency feature graph; The target detection module is used to train a pre-constructed deep manifold network using the time-frequency feature matrix as a sample, so that the deep manifold network can detect the target in the echo data to be detected.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program. It is characterized in that When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images