A radar jamming detection method and apparatus

Abstract

The application discloses a radar interference detection method and device, the method comprises the following steps: obtaining a radar echo signal to be detected; mapping the radar echo signal to be detected into a point in a three-dimensional feature space to obtain a three-dimensional coordinate of the point to be detected; determining an interference detection result of the radar echo signal to be detected according to a radar interference detection model trained in the three-dimensional feature space and the three-dimensional coordinate of the point to be detected; the radar interference detection model trained is obtained by unsupervised training of an initial radar interference detection model based on support vector data description constructed according to a Gaussian kernel function based on a training set; the training set is a set of multiple training sample points obtained by mapping part of echo data in non-interference radar echo data to a three-dimensional feature space. The application does not need a large number of interference echo samples, does not need to set a detection threshold, and does not need to manually label interference data, so that interference detection is performed in an unsupervised mode, and the efficiency is higher.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of radar, and particularly relates to a radar interference detection method and device. BACKGROUND

[0002] Radar plays a crucial role in the electronic battlefield today, including ranging, parameter acquisition, target detection and target recognition. In practical applications, the targets detected by the radar are often in a complex electromagnetic environment. In the radar receiving signal, all signals other than the target signal can be considered as interference. Various types of interference seriously affect the radar detection and tracking of targets and cause adverse effects on the normal operation of the radar. Therefore, when the radar processes the receiving signal, the presence of interference needs to be detected in a timely manner, and appropriate measures need to be taken to ensure that the radar can correctly detect and track targets.

[0003] In the field of radar interference detection, a series of research results have emerged in recent years. The paper "Research on multi-interference detection and recognition method based on feature extraction" proposes an interference detection and recognition algorithm based on decision tree. The algorithm extracts the spectral kurtosis, carrier factor, 0.5 times bandwidth, fractional Fourier domain maximum value and time domain moment skewness of the signal, trains the feature attributes of the sample data set, sets the corresponding feature threshold, and divides the samples into two categories at each feature attribute until all samples are recognized. However, this algorithm needs to manually set the threshold of each feature, and the threshold of each feature is highly dependent, and the recognition result shows a clear threshold effect. The document "Radio frequency interference detection based on the AC-UNet model" uses unordered convolution to replace the convolution kernel of the encoding layer and the decoding layer in the U-Net architecture, thereby segmenting the RFI in the time-frequency domain, increasing the receptive field to obtain more context information, and thereby improving the network's ability to supplement detailed information. However, the number of training samples is an important factor limiting its performance. In actual scenarios, it is difficult to obtain a large enough radio frequency interference data set for training. At the same time, the network can only handle relatively single strong interference scenarios, and the generalization ability is limited.

[0004] That is, the existing interference detection method based on feature extraction needs to set a detection threshold, and the robustness is not high. The existing interference detection method based on neural network needs a large amount of interfered echo data, and the interference data needs to be manually labeled, which has a high labor cost. SUMMARY

[0005] In order to solve the above problems existing in the prior art, the present application provides a radar interference detection method and device.

[0006] The technical problem solved by the present application is achieved by the following technical solutions:

[0007] The present application provides a radar interference detection method, comprising:

[0008] Obtaining a radar echo signal to be detected;

[0009] Mapping the radar echo signal to be detected into a point in a three-dimensional feature space to obtain a three-dimensional coordinate of the point to be detected;

[0010] According to the radar interference detection model trained in the three-dimensional feature space and the three-dimensional coordinate of the point to be detected, determining the interference detection result of the radar echo signal to be detected;

[0011] The trained radar interference detection model is obtained by unsupervised training an initial radar interference detection model based on support vector data description constructed according to a Gaussian kernel function based on a training set; the training set is a set of multiple training sample points obtained by mapping part of echo data in non-interference radar echo data to the three-dimensional feature space; and the non-interference radar echo data is obtained by inverse imaging of non-interference SAR image data.

[0012] The present application also provides a radar interference detection device, comprising a processor, a communication interface, a memory and a communication bus, the processor, the communication interface and the memory complete communication with each other through the communication bus;

[0013] The memory is used to store a computer program;

[0014] The processor is used to execute the program stored on the memory to realize the steps of the radar interference detection method.

[0015] Compared with the prior art, the present application has the following advantages:

[0016] The present application describes the signal as a point in a three-dimensional feature space, and the interference detection model used is a radar interference detection model based on support vector data description constructed according to a Gaussian kernel function in a three-dimensional feature space, so the interference detection method of the present application is suitable for interference detection of various radar platforms. In the present application, unsupervised machine learning technology is introduced when training the interference detection model, and only non-interference radar echo data is required when constructing the training set of the model, so a large number of interfered echo samples are not required, a detection threshold does not need to be set, interference data does not need to be manually labeled, the labor cost is low, the degree of realization is high, the detection performance is good, the robustness is high, and the detection efficiency is high.

[0017] The present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 is a flowchart of a radar interference detection method provided by an embodiment of the present application;

[0019] Figure 2 is a spectrum and amplitude histogram of a radar echo signal under different interference conditions provided by an embodiment of the present application;

[0020] Figure 3 is an exemplary decision boundary diagram provided by an embodiment of the present application;

[0021] Figure 4 is a visualization diagram of a detection result of a test set by a trained radar interference detection model provided by an embodiment of the present application. DETAILED DESCRIPTION

[0022] The present application will be further described in detail below with specific embodiments, but the embodiments of the present application are not limited thereto.

[0023] Figure 1 is a flowchart of a radar interference detection method provided by an embodiment of the present application, as shown in Figure 1 , the method comprises:

[0024] S101, obtaining a radar echo signal to be detected.

[0025] Here, the radar echo signal to be detected can be one or multiple. The following content takes one radar echo signal to be detected as an example for specific description.

[0026] S102, mapping the radar echo signal to be detected into a point in a three-dimensional feature space to obtain a three-dimensional coordinate of the point to be detected.

[0027] Here, the three-dimensional feature space is a feature space taking a peak value feature, a skewness feature and a kurtosis feature of a frequency domain signal as three-dimensional coordinate axes respectively.

[0028] S103, determining an interference detection result of the radar echo signal to be detected according to a trained radar interference detection model in the three-dimensional feature space and the three-dimensional coordinate of the point to be detected; wherein the trained radar interference detection model is obtained by unsupervised training an initial radar interference detection model based on support vector data description constructed according to a Gaussian kernel function based on a training set; the training set is a set composed of multiple training sample points obtained by mapping part of echo data in non-interference radar echo data into the three-dimensional feature space; and the non-interference radar echo data is obtained by inverse imaging of non-interference SAR image data.

[0029] In the present application, the above S102 can be realized by S1021-S1022:

[0030] S1021, extract the peak value feature, skewness feature and kurtosis feature of the radar echo signal to be detected.

[0031] Here, the radar echo signal to be detected can be transformed into the frequency domain, and the peak value, skewness and kurtosis of the radar echo signal to be detected in the frequency domain are extracted respectively, and the extracted peak value, skewness and kurtosis are respectively taken as the peak value feature, skewness feature and kurtosis feature of the radar echo signal to be detected.

[0032] Specifically, the radar echo signal to be detected is first subjected to Fourier transform to convert the radar echo signal to be detected into the frequency domain, and then the peak value, skewness and kurtosis are extracted from the spectrum of the radar echo signal to be detected.

[0033] Skewness is a statistical index used to measure the degree of asymmetry of data distribution. It describes the skewness or asymmetry of data distribution, i.e. the shape of the distribution around the mean. The formula is: Where μ is the mean, σ is the variance, m represents the data length, and X i represents the data value at the i-th position in the data. The skewness can be used to preliminarily judge the symmetry of the data distribution, but it cannot provide complete distribution information.

[0034] Kurtosis is a statistical quantity that describes the sharpness or flatness of the shape of a probability distribution. It measures the height of the peak of the distribution curve at its mean value, and can reflect the thickness or heavy-tailed nature of the tail. The formula is: The kurtosis of a normal distribution is 3. If the kurtosis of a certain distribution is greater than 3, it means that the peak value of the distribution is higher than that of the normal distribution, and the distribution curve is more sharp; on the contrary, if the kurtosis is less than 3, it means that the peak value of the distribution is lower than that of the normal distribution, and the distribution curve is more gentle. According to the above characteristics, kurtosis can be used to describe the shape of the data distribution, which helps to understand the peak state characteristics of the data, and also can identify whether there are abnormal values or long-tailed distributions in the data set.

[0035] S1022, mapping the radar echo signal to be detected to a point corresponding to the peak value feature, skewness feature and kurtosis feature in a three-dimensional feature space to obtain a detection point; wherein the peak value feature, skewness feature and kurtosis feature are respectively the three-dimensional coordinates of the detection point.

[0036] In the present application, the trained radar jamming detection model is: according to the Gaussian kernel function with trained parameters γ f , the trained regularization parameter C f, and a hypersphere constructed in the three-dimensional feature space by the support vector data description algorithm and the training set, wherein the hypersphere has a sphere center and a radius. Based on this, the S103 is implemented by S1031-S1033.

[0037] S1031, calculating a distance between the to-be-detected point and the sphere center according to the three-dimensional coordinates of the sphere center of the hypersphere, the three-dimensional coordinates of the to-be-detected point, and a Gaussian kernel function with a trained parameter γ f .

[0038] S1032, obtaining an interference detection result representing that the to-be-detected radar echo signal is not interfered when the distance is less than or equal to the radius of the hypersphere.

[0039] S1033, obtaining an interference detection result representing that the to-be-detected radar echo signal is interfered when the distance is greater than the radius of the hypersphere.

[0040] Specifically, according to the Gaussian kernel function with the trained parameter γ f , the trained regularization parameter C f , the sphere center and the radius expression of the hypersphere constructed in the three-dimensional feature space by the support vector data description algorithm and the training set, the expressions are as follows:

[0041]

[0042]

[0043] wherein R represents the radius of the hypersphere, a represents the three-dimensional coordinates of the sphere center of the hypersphere, K(.) is the Gaussian kernel function, γ f is the trained parameter of the Gaussian kernel function, x and y are two variables in the Gaussian kernel function, n represents the total number of training sample points in the training set, x i represents the i-th training sample point in the training set, x j represents the j-th training sample point in the training set, i and j are both 1 to n, α i represents the Lagrange coefficient of the i-th training sample point, α j represents the Lagrange coefficient of the j-th training sample point, x v represents any one sample point in the set SV, the set SV is composed of sample points with the Lagrange coefficient greater than 0 and less than the trained penalty parameter, and Φ(.) represents a nonlinear transformation function.

[0044] Specifically, the expression of the distance between a to-be-detected point and the above-mentioned sphere center a is as follows:

[0045]

[0046] wherein d represents the distance of the point to be detected from the center of the sphere, x test represents the point to be detected.

[0047] In the present application, before the above S103, the steps S001-S008 are further included:

[0048] S001, acquiring at least one non-interference SAR image data.

[0049] Here, one real non-interference SAR image data can be used, or multiple real non-interference SAR image data can be used, and the present application does not limit this.

[0050] S002, according to the imaging parameters, inverse CS algorithm is used to inverse image at least one non-interference SAR image data, to obtain multiple non-interference radar echo data of each non-interference SAR image data.

[0051] Here, according to the imaging parameters, after inverse CS algorithm is used to inverse image one non-interference SAR image data, a complex matrix is obtained, and each column in the complex matrix is a non-interference radar echo data corresponding to the non-interference SAR image data.

[0052] S003, using the non-interference radar echo data corresponding to at least one non-interference SAR image data to form an echo data set; each non-interference radar echo data is a sample in the echo data set.

[0053] S004, dividing the echo data set into an initial training set and an initial validation set according to a preset proportion.

[0054] In some embodiments, the echo data set can also be divided into an initial training set, an initial validation set and an initial test set according to a preset proportion. For example, the echo data set can be divided into an initial training set, an initial validation set and an initial test set according to a proportion of 6:2:2.

[0055] S005, superimposing interference on part of the samples in the initial validation set to obtain an initial validation set with added interference.

[0056] Exemplarily, the added interference is narrowband interference.

[0057] Exemplarily, the present application uses two single-frequency signals f1=1MHz, JSR=JSR1 and f2=5MHz, JSR=JSR2 to form narrowband interference, wherein the range of JSR1 is [5, 20]dB, and the range of JSR2 is [1, 15]dB.

[0058] For narrowband interference, its spectrum is usually concentrated in a narrow frequency range, showing a certain peak in the frequency domain. Generally, it can be regarded as the superposition of a series of single frequency signals, so the narrowband interference can be expressed as: where L represents the number of assumed single frequency interference components, A i , f i , are the amplitude, frequency and initial phase of the i-th single frequency component of the narrowband interference respectively. Therefore, the echo signal model after superimposing the interference can be expressed as: where represents fast time, represents the original echo signal, represents the narrowband interference, represents the echo signal after superimposing the interference. Let the real part and the imaginary part of the radar echo be X R and X I , which are independent and identically distributed Gaussian variables with a mean of 0. The presence of interference will cause a non-coherent complex vector to be superimposed on the complex Gaussian distribution of the useful echo. Therefore, the mean of the echo signal with interference will be shifted, and the shift amount is determined by the relative ratio of the radio frequency interference and the echo energy, that is, the jamming to signal ratio (JSR). Through the above analysis, it can be known that the presence of interference in the echo will cause the value of the statistical quantity to change, and is closely related to the energy of the interference. When there is interference in the echo, the root mean square and the standard deviation in the time domain will increase, while the skewness and kurtosis values will slightly decrease with the increase of the JSR. Compared with the time domain, the energy difference between the interference and the useful echo signal is more obvious in the frequency domain, and the Fourier transform is a linear transform, so the complex Gaussian distribution variable still obeys the complex Gaussian distribution after the Fourier linear transform. Due to the presence of narrowband interference, the spectrum presents a prominent peak, introducing non-Gaussianity. Figure 2 The spectrum and amplitude histogram of the radar echo signal under different interference conditions are compared, wherein, Figure 2 Fig. (a) is the spectrum diagram of the interference-free radar echo data; Fig. (b) is the amplitude histogram of the interference-free radar echo data in Fig. (a); Fig. (c) is the spectrum diagram of the interference-free radar echo data in Fig. (a) after superimposing JSR1=5, JSR2=10 interference on the interference-free radar echo data in Fig. (a); Fig. (d) is the amplitude histogram of the interference-free radar echo data in Fig. (a) after superimposing JSR1=5, JSR2=10 interference on the interference-free radar echo data in Fig. (a); Fig. (e) is the spectrum diagram of the interference-free radar echo data in Fig. (a) after superimposing JSR1=10, JSR2=20 interference on the interference-free radar echo data in Fig. (a); and Fig. (f) is the amplitude histogram of the interference-free radar echo data in Fig. (a) after superimposing JSR1=10, JSR2=20 interference on the interference-free radar echo data in Fig. (a). Therefore, the present application selects the peak value, skewness and kurtosis which can reflect the characteristics of signal energy or signal distribution to describe the signal.

[0059] Exemplarily, the superimposed jamming can be superimposed on 50% of the samples in the initial training set and the initial test set.

[0060] S006, each sample in the initial training set and the initial validation set with added jamming is mapped into a point in a three-dimensional feature space, obtaining a training set composed of multiple training sample points and a validation set composed of multiple validation sample points.

[0061] Specifically, the peak value, skewness and kurtosis of the frequency domain of each sample in the initial training set and the initial validation set with added jamming are extracted; the peak value, skewness and kurtosis of the frequency domain of each sample are normalized to obtain the peak value feature, skewness feature and kurtosis feature of each sample; for each training sample in the initial training set, the training sample is mapped into a point in the three-dimensional feature space corresponding to the peak value feature, skewness feature and kurtosis feature of the training sample, obtaining a training sample point, wherein the peak value feature, skewness feature and kurtosis feature of the training sample are the three-dimensional coordinates of the training sample point; for each validation sample in the initial validation set with added jamming, the validation sample is mapped into a point in the three-dimensional feature space corresponding to the peak value feature, skewness feature and kurtosis feature of the validation sample, obtaining a validation sample point, wherein the peak value feature, skewness feature and kurtosis feature of the validation sample are the three-dimensional coordinates of the validation sample point.

[0062] Here, through normalization processing, the influence of the large difference between different feature values on the subsequent detection performance can be avoided. Exemplarily, the maximum-minimum value normalization method can be used for normalization processing, and the formula of the maximum-minimum value normalization is: wherein x min is the minimum value of the feature x, and x max is the maximum value of the feature x.

[0063] Here, when the initial test set exists, the initial test set with added jamming is also normalized and mapped into the three-dimensional feature space according to the principle of S006.

[0064] S007, in the three-dimensional feature space, an initial radar jamming detection model based on support vector data description is constructed according to the Gaussian kernel function and the training set.

[0065] Here, the initial radar jamming detection model is an initial hypersphere constructed in the three-dimensional feature space according to the Gaussian kernel function with initial parameters γ 0 , the initial regularization parameter C 0 , the support vector data description algorithm and the training set. The principle of constructing the initial hypersphere is as follows:

[0066] When the training set is V n×dwhere n is the number of training sample points, d is the dimension of features, x i j are the ith and jth training sample points, respectively. First, the data is mapped from the original space to a three-dimensional feature space by a nonlinear transformation function Φ: x i → F, and then a hyper-sphere with the minimum volume is sought in the three-dimensional feature space. In order to construct such a hyper-sphere with the minimum volume, the support vector data description algorithm needs to solve the following optimization problem:

[0067]

[0068] s.t. ||Φ(x i )-a||2 2 ≤R 2 +ξ i , ξ i ≥ 0, i = 1, 2,..., n.

[0069] where R is the radius of the hyper-sphere, a is the center of the hyper-sphere, ξ is the slack factor, and C is a regularization parameter, which is a penalty parameter used to balance the volume of the hyper-sphere and the misclassification rate. Using the Lagrange multiplier method, the dual problem of the original problem is obtained:

[0070]

[0071]

[0072] where K(.) is a Gaussian kernel function. Normally, the data does not present a spherical distribution, so it is necessary to use the kernel function method to improve the expression capacity of the model. The core idea of the Gaussian kernel function is to map each training sample point to an infinite-dimensional feature space, so that the originally linearly inseparable data becomes linearly separable. The formula of the Gaussian kernel function is: where γ is the hyper-parameter of the Gaussian kernel function. After solving the dual problem, the Lagrange coefficients corresponding to all the training sample points in the training set can be obtained. Among all the training sample points, the sample whose Lagrange coefficient satisfies 0 < α i <C is called a support vector. Assuming that the set of training sample points in the training set that belong to the support vector is SV, the calculation formulas of the center and the radius of the hyper-sphere are:

[0073]

[0074]

[0075] where α i is the Lagrange coefficient of the ith training sample point, and x v ∈ SV. When the above γ is the initial parameter γ 0 ​The regularization parameter C is the initial regularization parameter C. 0 If so, an initial hypersphere can be constructed using the above principles.

[0076] Here, C and γ are two common hyperparameters in support vector data description algorithms, and both C and γ are greater than 0. γ controls the range of influence of a single training sample on the model boundary; the larger γ is, the larger the model boundary, which can lead to overfitting. C controls the model's tolerance for outliers in the training set. When C is small, outlier training samples are allowed outside the model boundary; when C is large, the model will fully consider outliers, which are generally within the model boundary.

[0077] S008. The initial radar interference detection model is trained and validated using the training set and validation set to obtain a trained radar interference detection model.

[0078] Specifically, we can first fix the parameters of the Gaussian kernel function γ as the initial parameters γ. 0 The regularization parameter is continuously adjusted based on the training set to ensure that outlier samples in the training set are located outside the obtained hypersphere, thus obtaining the trained regularization parameter C. f Then, the regularization parameter C is fixed to the trained regularization parameter C0. f The parameters of the Gaussian kernel function are continuously adjusted based on the validation and training sets to maximize the accuracy of the obtained hypersphere in validating the validation set, while ensuring that outlier points in the training set are located outside the obtained hypersphere. This yields the parameters γ of the trained Gaussian kernel function. f And a trained radar interference detection model.

[0079] Here, with the Gaussian kernel function fixed at γ... 0 When adjusting the regularization parameter C, it can be based on the Gaussian kernel function γ. 0 Using the above principle, a hypersphere is constructed with the current regularization parameters. The decision boundary map is obtained by plotting this hypersphere and all training sample points in the training set in a three-dimensional feature space. (For example,...) Figure 3 This is a decision boundary map, where each point in the map represents a training sample point, and the gray 3D space region in the map represents a constructed hypersphere. The decision boundary map determines whether training sample points in the training set that are far from the region where most training sample points are clustered (e.g., the distance is greater than a preset distance) are within the hypersphere. If such training sample points are within the hypersphere, then C can be reduced. Afterwards, the Gaussian kernel function γ is used... 0 Using the above principle, a hypersphere is constructed with the current regularization parameters, and further judgments are made to ensure that the training sample points are outside the boundary of the constructed hypersphere. Thus, the trained regularization parameters C are obtained. f .

[0080] Here, the fixed regularization parameter C is the trained regularization parameter C f When adjusting γ according to the validation set and the training set, on the one hand, the hyper-sphere can be used to verify the validation sample points in the validation set, and the correct rate of verification can be obtained, and on the other hand, a hyper-sphere can be constructed according to the Gaussian kernel function γ 0 and the current regularization parameter by using the above principle, and a decision boundary diagram can be obtained by plotting the hyper-sphere and all training sample points in the three-dimensional feature space, so as to determine whether the boundary of the constructed hyper-sphere is too large or too small (for example, when the number of training sample points located outside the boundary of the hyper-sphere exceeds a certain number, it is considered that the boundary of the hyper-sphere is too small, and when the hyper-sphere contains training sample points far away from the region where most of the training sample points in the training set are gathered, it is considered that the boundary of the hyper-sphere is too large); when the correct rate is low and the boundary is too small, γ can be reduced, and when the correct rate is low and the boundary is too large, γ can be increased, and then a hyper-sphere can be constructed according to the current γ and the trained regularization parameter C f by using the above principle, and the judgment is continued to make the correct rate of the constructed hyper-sphere for the validation set verification maximum, and the training sample points far away from the region where most of the training sample points in the training set are gathered are located outside the constructed hyper-sphere, so that the trained parameter γ f and a trained radar jamming detection model can be obtained. It should be noted that during the adjustment process, the correct rate of the constructed hyper-sphere for the validation set verification is given priority, and the distribution space of the hyper-sphere and the training sample points in the training set is as close as possible.

[0081] In some embodiments, when calculating the correct rate of the constructed hyper-sphere for the validation set verification, the following formula can be used for calculation:

[0082]

[0083] Wherein, TP represents the total number of test sample points in the validation set without interference and detected as no interference, FN represents the total number of test sample points in the validation set without interference but detected as interference, FP represents the total number of test sample points in the validation set with interference but detected as no interference, and TN represents the total number of test sample points in the validation set with interference and detected as interference.

[0084] For example, when C f is 0.05 and γ f is 50, the detection correct rate of the trained radar jamming detection model for the validation set verification is 99.9%, and the model has good performance.

[0085] When the trained radar interference detection model is obtained, the generalization ability of the model can be verified by using the test set, and the detection performance of the detection model can be evaluated according to the detection accuracy. For example, after verification, when C f is 0.05, and gamma f is 50, the detection accuracy of the trained radar interference detection model on the test set is 97.5%, so the model can well fit the training data, has good generalization ability, and shows good detection performance, and the visual image of the detection result of the model on the test set is as shown in Figure 4 .

[0086] The present application extracts the frequency domain peak value, skewness and kurtosis, and constructs a radar interference detection model based on unsupervised learning. The model achieves good interference detection performance. Compared with the traditional detection method, the present application does not need a large number of interference echo samples, does not need to set a detection threshold, and does not need to manually label interference data. The interference detection is performed in an unsupervised mode, and the efficiency is higher.

[0087] The present application also provides a radar interference detection device, which comprises a processor, a communication interface, a memory and a communication bus. The processor, the communication interface and the memory complete communication with each other through the communication bus. The memory is used to store a computer program. The processor is used to execute the program stored on the memory to realize the steps of the radar interference detection method.

[0088] It should be noted that the terms "first", "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined as "first", "second" can explicitly or implicitly include one or more features. In the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise specifically limited.

[0089] In the description of the present application, the description of the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" means that the specific features or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In the present application, the illustrative description of the above terms is not necessarily for the same embodiment or example. Moreover, the specific features or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in the present application.

[0090] In the description, the word "comprising" does not exclude other components or steps, and the word "a" or "an" does not exclude a plurality. The mere fact that certain measures are recited in mutually different embodiments does not indicate that these measures cannot be combined in one embodiment.

[0091] The above description is further to the application in conjunction with specific preferred embodiments, and cannot be deemed to limit the specific implementation of the application to these descriptions. For those skilled in the art to which the application belongs, without departing from the concept of the application, a number of simple deductions or replacements can be made, which should be regarded as falling within the protection scope of the application.

Claims

1. A radar jamming detection method, characterized by, The method comprises: obtaining radar echo signals to be detected; mapping the radar echo signals to be detected into points in a three-dimensional feature space to obtain three-dimensional coordinates of the points to be detected; determining an interference detection result of the radar echo signals to be detected according to a radar interference detection model trained in the three-dimensional feature space and the three-dimensional coordinates of the points to be detected; wherein the trained radar interference detection model is obtained by unsupervised training of an initial radar interference detection model based on support vector data description constructed according to a Gaussian kernel function based on a training set; the training set is a set of multiple training sample points obtained by mapping part of echo data in non-interference radar echo data to the three-dimensional feature space; the non-interference radar echo data is obtained by inverse imaging of non-interference SAR image data; wherein mapping the radar echo signals to be detected into points in a three-dimensional feature space to obtain three-dimensional coordinates of the points to be detected comprises: extracting peak value features, skewness features and kurtosis features of a frequency domain of the radar echo signals to be detected; mapping the radar echo signals to be detected into points in a three-dimensional feature space corresponding to the peak value features, the skewness features and the kurtosis features to obtain the points to be detected; wherein the peak value features, the skewness features and the kurtosis features are respectively three-dimensional coordinates of the points to be detected.

2. The radar jamming detection method of claim 1, wherein, The trained radar interference detection model comprises a Gaussian kernel function with trained parameters , a trained regularization parameter , a support vector data description algorithm and a hypersphere constructed in the three-dimensional feature space by the training set, the hypersphere having a sphere center and a radius; and determining the interference detection result of the radar echo signal to be detected according to the trained radar interference detection model in the three-dimensional feature space and the three-dimensional coordinates of the point to be detected. According to the three-dimensional coordinates of the center of the hypersphere, the three-dimensional coordinates of the point to be detected, and the Gaussian kernel function with trained parameters , the distance of the point to be detected from the center of the hypersphere is calculated; when the distance is less than or equal to the radius of the hypersphere, an interference detection result representing that the radar echo signals to be detected are non-interference is obtained; when the distance is greater than the radius of the hypersphere, an interference detection result representing that the radar echo signals to be detected are interference is obtained.

3. The radar jamming detection method of claim 2, wherein, The expression of the radius of the hypersphere is as follows: ； The expression of the center of the hypersphere is as follows: ； wherein, denotes a radius of the hypersphere, denotes a three-dimensional coordinate of a center of the hypersphere, is a Gaussian kernel function, , is the trained parameter of the Gaussian kernel function, are two variables in the Gaussian kernel function, denotes a total number of training sample points in the training set, denotes an i-th training sample point in the training set, denotes a j-th training sample point in the training set, i and j are both 1 to , denotes a Lagrange coefficient of the i-th training sample point, denotes a Lagrange coefficient of the j-th training sample point, denotes any one sample point in a set SV, the set SV is composed of sample points whose Lagrange coefficients are greater than 0 and less than a trained penalty parameter, denotes a nonlinear transformation function.

4. The radar jamming detection method of claim 2, wherein, The expression of the distance between the point to be detected and the center of the hypersphere is as follows: ； wherein, denotes the distance of the point to be detected from the center of the ball, is a Gaussian kernel function, , is the trained parameter of the Gaussian kernel function, are two variables in the Gaussian kernel function, denotes the total number of training sample points in the training set, denotes the i-th training sample point in the training set, denotes the j-th training sample point in the training set, i and j are both 1 to , denotes the Lagrange coefficient of the i-th training sample point, denotes the Lagrange coefficient of the j-th training sample point, denotes the point to be detected.

5. The radar jamming detection method of claim 1, wherein, Before the step of determining the interference detection result of the radar echo signals to be detected according to the radar interference detection model trained in the three-dimensional feature space and the three-dimensional coordinates of the points to be detected, the method further comprises: obtaining at least one non-interference SAR image data; performing inverse CS algorithm on the at least one non-interference SAR image data according to imaging parameters to obtain multiple non-interference radar echo data of each non-interference SAR image data; constructing an echo data set by using the non-interference radar echo data corresponding to the at least one non-interference SAR image data; each non-interference radar echo data is a sample in the echo data set; dividing the echo data set into an initial training set and an initial validation set according to a preset proportion; adding interference to part of the samples in the initial validation set to obtain an initial validation set with added interference; mapping each sample in the initial training set and the initial validation set with added interference into a point in the three-dimensional feature space to obtain the training set composed of multiple training sample points and the validation set composed of multiple validation sample points; constructing an initial radar interference detection model based on support vector data description in the three-dimensional feature space according to a Gaussian kernel function and the training set. The initial radar interference detection model is trained and verified by using the training set and the verification set, and a trained radar interference detection model is obtained.

6. The radar jamming detection method of claim 5, wherein, The initial training set and each sample in the initial verification set with added interference are respectively mapped to a point in the three-dimensional feature space, and the training set composed of multiple training sample points and the verification set composed of multiple verification sample points are obtained, including: Peak values, skewness and kurtosis of a frequency domain of each sample in the initial training set and the initial verification set with added interference are extracted; The peak values, skewness and kurtosis of the frequency domain of each sample are normalized to obtain the peak value feature, the skewness feature and the kurtosis feature of each sample; For each training sample in the initial training set, the training sample is mapped to a point in the three-dimensional feature space corresponding to the peak value feature, the skewness feature and the kurtosis feature of the training sample, and a training sample point is obtained, wherein the peak value feature, the skewness feature and the kurtosis feature of the training sample are three-dimensional coordinates of the training sample point. For each verification sample in the initial verification set with added interference, the verification sample is mapped to a point in the three-dimensional feature space corresponding to the peak value feature, the skewness feature and the kurtosis feature of the verification sample, and a verification sample point is obtained, wherein the peak value feature, the skewness feature and the kurtosis feature of the verification sample are three-dimensional coordinates of the verification sample point.

7. The radar jamming detection method of claim 5, wherein, The superimposed interference is narrowband interference.

8. The radar jamming detection method of claim 5, wherein, The initial radar jamming detection model is constructed according to a Gaussian kernel function with initial parameters , an initial regularization parameter , a support vector data description algorithm and an initial hypersphere constructed in the three-dimensional feature space according to the training set. The initial radar interference detection model is trained and verified by using the training set and the verification set, and a trained radar interference detection model is obtained, including: Fixed gaussian kernel function The parameters of the fixed gaussian kernel function are initial parameters The regularization parameter is continuously adjusted according to the training set, so that the outlier sample points in the training set are located outside the obtained hyper-sphere, and a trained regularization parameter is obtained ; Fixed regularization parameter For the trained regularization parameter , the parameters of the Gaussian kernel function are continuously adjusted according to the validation set and the training set, so that the correct rate of the obtained hyper-sphere to the validation set is the highest, and the outlier sample points in the training set are located outside the obtained hyper-sphere, to obtain the trained parameters of the Gaussian kernel function And the trained radar jamming detection model.

9. A radar jamming detection device comprising a processor, a communication interface, a memory and a communication bus, characterized in that, The processor, the communication interface and the memory complete mutual communication through the communication bus; The memory is used to store a computer program; The processor is used to execute the program stored on the memory, and the method steps of any one of claims 1-8 are implemented.

Images