A radar jamming signal type recognition method based on convolutional neural network model

Abstract

The present invention discloses a method for identifying the type of radar interference signal based on a convolutional neural network model, comprising: obtaining an echo time domain sequence and performing time-frequency conversion to obtain an echo frequency domain sequence; performing hierarchical quantization on the echo time domain sequence and the echo frequency domain sequence, respectively, and generating first and second real matrices based on the hierarchical quantization results; concatenating the first and second real matrices and converting them into grayscale images; inputting the grayscale images into a pre-trained convolutional neural network model so that the convolutional neural network model outputs a radar interference signal type identification result; wherein the convolutional neural network model is trained based on multiple sample grayscale images and the radar interference signal type corresponding to each sample grayscale image. Based on the convolutional neural network model, the present invention uses image classification and recognition methods to intelligently identify the type of radar interference signal, and has high generalization ability and recognition accuracy.

Description

Technical Field

[0001] The present invention belongs to the field of radar, and in particular relates to a method for identifying radar interference signal types based on a convolutional neural network model. Background Art

[0002] In the radar field, jamming and anti-jamming technologies engage in a mutually reinforcing, counter-attacking, and mutually reinforcing relationship. Jamming technologies are categorized as active and passive. Active jamming uses jammers that actively radiate energy to disrupt radar signals in real time; passive jamming utilizes methods such as chaff, decoys, and stealth coatings to reflect or absorb radar waves, disrupting radar detection and identification. A wide range of anti-jamming measures exist to address various jamming technologies. However, because jamming scenarios are highly variable and unpredictable, it is necessary to first identify the type of radar jamming signal and then select the appropriate anti-jamming measures.

[0003] Existing anti-interference algorithms generally only address interference based on similar principles. Specifically, they manually analyze and extract signal features based on signal characteristics. The extracted signal features are then used to determine the interference type and implement corresponding anti-interference measures. While manual signal feature extraction can accurately extract signal characteristics, changes or expansions in signal types require a redesign of the feature extraction algorithm to maintain the original recognition efficiency of the recognition algorithm. Furthermore, the manual signal feature extraction process introduces subjective factors, resulting in inaccurate recognition results, which in turn affects the effectiveness of subsequent anti-interference measures. Summary of the Invention

[0004] In order to solve the above problems existing in the prior art, the present invention provides a radar interference signal type recognition method based on a convolutional neural network model.

[0005] The technical problem to be solved by the present invention is achieved through the following technical solution: A method for identifying the type of radar interference signals based on a convolutional neural network model, comprising: a signal preprocessing step and an interference identification step;

[0006] The signal preprocessing step includes: obtaining a radar echo time domain sequence, and performing time-frequency conversion on the echo time domain sequence to obtain an echo frequency domain sequence; performing hierarchical quantization on the echo time domain sequence, and generating a first real matrix according to the hierarchical quantization result; performing hierarchical quantization on the echo frequency domain sequence, and generating a second real matrix according to the hierarchical quantization result; splicing the first real matrix and the second real matrix, and converting the spliced ​​result into a grayscale image;

[0007] The interference identification step includes: inputting the grayscale image into a pre-trained convolutional neural network model, so that the convolutional neural network model outputs a radar interference signal type identification result;

[0008] In which, the convolutional neural network model is trained based on multiple sample grayscale images and the radar interference signal type corresponding to each sample grayscale image; each of the sample grayscale images is a grayscale image obtained by processing a sample echo time domain sequence according to the signal preprocessing step.

[0009] Optionally, the step of hierarchically quantizing the echo time domain sequence and generating a first real matrix according to the hierarchical quantization result is implemented using the following formula:

[0010] a x (m,n)=0.5*sgn(m-M+1+rank x (n))+0.5,

[0011]

[0012] x max =max(x(1),x(2),…,x(N-1)),

[0013] x min =min(x(1),x(2),…,x(N-1));

[0014] Among them, a x (m,n) is the element in the mth row and nth column of the first real matrix, 0≤m≤M-1, 0≤n≤N-1, M is the number of levels, N is the length of the echo time domain sequence x; x(n) is the nth element of x; sgn(·) is the sign function.

[0015] Optionally, the step of hierarchically quantizing the echo frequency domain sequence and generating a second real matrix according to the hierarchical quantization result is implemented using the following formula:

[0016] a X (m,n)=0.5*sgn(m-M+1+rank X (n))+0.5,

[0017]

[0018] X max =max(X(1),X(2),…,X(N-1)),

[0019] X min =min(X(1),X(2),…,X(N-1));

[0020] Among them, a X(m,n) is the element in the mth row and nth column of the second real matrix, 0≤m≤M-1, 0≤n≤N-1, and X(n) is the nth element of the echo frequency domain sequence X.

[0021] Optionally, the step of converting the stitching result into a grayscale image includes:

[0022] Convert the value of each element in the splicing result into a grayscale value to form an initial grayscale image;

[0023] The size of the initial grayscale image is adjusted to match the input of the convolutional neural network model based on interpolation.

[0024] Optionally, the convolutional neural network model includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, a first fully connected layer, and a second fully connected layer connected in sequence;

[0025] The convolution kernel size of each convolution layer is 3*3, and the number of input channels and output channels of the convolution kernel increases along the data flow direction; the activation function of each convolution layer is ReLU;

[0026] When the convolutional neural network model is working, the grayscale image is input into the first convolutional layer, and the second fully connected layer outputs the radar interference signal type recognition result.

[0027] Optionally, the training method of the convolutional neural network model includes:

[0028] Acquire a training data set; the training data set includes a plurality of sample grayscale images, and the plurality of sample grayscale images belong to a plurality of different radar interference signal types;

[0029] The training data set is divided into k training sets and k validation sets using a k-fold cross-validation method; the k training sets and the k validation sets correspond one to one;

[0030] The convolutional neural network model is trained based on the k training sets, and the training effect of the training set is verified using the verification set corresponding to the training set.

[0031] Optionally, during the training of the convolutional neural network model, the cross entropy loss is calculated based on the radar interference signal type to which the sample grayscale image belongs and the radar interference signal type output by the convolutional neural network model, so as to adjust the network weight parameters of the convolutional neural network model based on the cross entropy loss.

[0032] Optionally, the calculation formula of the cross entropy loss is:

[0033]

[0034] Among them, the radar interference signal type output by the convolutional neural network model is an 8-dimensional vector x, x[j] means taking the j-th element of x; x[class] means taking the class-th element of x, and the value of class is determined according to the radar interference signal type to which the sample grayscale image belongs.

[0035] Optionally, the training method further includes:

[0036] Before using the k-fold cross-validation method to divide the training data set, a portion of sample grayscale images is extracted from the training data set using a stratified sampling method to form a test set, so as to subsequently use the k-fold cross-validation method to divide the remaining sample grayscale images in the training data set;

[0037] And, after obtaining the trained convolutional neural network model, the trained convolutional neural network model is tested using the test set.

[0038] Optionally, the radar jamming signal type corresponding to any of the training samples includes: intermittent sampling and forwarding jamming, targeting jamming, blocking jamming, frequency sweep jamming, range deception jamming, smart noise jamming, dense false target jamming or chaff jamming.

[0039] The radar jamming signal type identification method based on a convolutional neural network model provided by the present invention utilizes image classification and recognition methods based on the convolutional neural network model to intelligently identify the type of radar jamming signal, demonstrating high generalization capabilities. Furthermore, because the grayscale image input to the convolutional neural network model contains information in both the time and frequency domains, signal features in both the time and frequency domains can be extracted, and the correspondence between time and frequency domain signal features can be utilized. Consequently, the radar jamming signal type identification result output by the present invention is more accurate.

[0040] In summary, the radar interference signal type recognition method based on the convolutional neural network model provided by the present invention has high generalization ability and recognition accuracy, which can facilitate the radar to further take targeted anti-interference measures to achieve real-time and accurate anti-interference processing.

[0041] The present invention will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Flowchart of a radar interference signal type identification method based on a convolutional neural network model provided by an embodiment of the present invention;

[0043] Figure 2Schematic diagram of the training process of the convolutional neural network model used in an embodiment of the present invention;

[0044] Figure 3 Schematic diagram of the structure of a convolutional neural network model used in an embodiment of the present invention;

[0045] Figure 4(a) to Figure 4(h) shows a sample grayscale image input into the model when training the convolutional neural network model in an embodiment of the present invention;

[0046] Figure 5(a) to Figure 5(h) for Figure 3 The convolutional layers of the model shown are for Figure 4(a) to Figure 4(h) Output of the sample grayscale image shown;

[0047] Figure 6(a) shows Figure 3 The recognition accuracy curve of the convolutional neural network model shown in the figure changes with the number of training cycles during the training process;

[0048] Figure 6(b) shows Figure 3 The loss value curve of the convolutional neural network model shown in the figure changes with the number of training cycles during the training process;

[0049] Figure 7 for Figure 3 Test result recognition matrix of the convolutional neural network model shown. DETAILED DESCRIPTION

[0050] The present invention will be further described in detail below with reference to specific examples, but the embodiments of the present invention are not limited thereto.

[0051] In order to more accurately identify the type of radar interference signal, the embodiment of the present invention provides a radar interference signal type identification method based on a convolutional neural network model, see Figure 1 As shown, the method includes: a signal preprocessing step and an interference identification step.

[0052] Among them, such as Figure 1 As shown, the signal preprocessing steps include:

[0053] S11: Acquire a radar echo time domain sequence, and perform time-frequency conversion on the echo time domain sequence to obtain an echo frequency domain sequence.

[0054] Specifically, the interference signal received by the radar is sampled to obtain a one-dimensional signal sequence. The sampling process must satisfy the Nyquist sampling theorem. That is, the number of sampling points N satisfies: N = floor (T * f s ); where T is the duration of the radar interference signal to be sampled, f sis the sampling frequency, and the floor(·) operation is a rounding-down operation. In this way, the sampled echo time domain sequence contains all the information of the original radar interference signal without causing any information loss.

[0055] The collected one-dimensional signal sequence is used as the echo time domain sequence x. Then, the echo time domain sequence x is Fourier transformed to obtain the echo frequency domain sequence X.

[0056] S12: Perform hierarchical quantization on the echo time domain sequence, and generate a first real matrix according to the hierarchical quantization result.

[0057] Specifically, step S12 is implemented using the following formula:

[0058] a x (m,n)=0.5*sgn(m-M+1+rank x (n))+0.5,

[0059]

[0060] x max =max(x(1),x(2),…,x(N-1)),

[0061] x min =min(x(1),x(2),…,x(N-1));

[0062] Among them, a x (m,n) is the element in the mth row and nth column of the first real matrix, 0≤m≤M-1, 0≤n≤N-1, that is, the dimension of the first real matrix is ​​M×N, M is the number of levels, N is the length of the echo time domain sequence x; x(n) is the nth element of x; sgn(·) is the sign function. max and x min are the maximum and minimum values ​​in the echo time domain sequence x respectively. x (n) is the quantization level corresponding to the nth sampling point in the echo time domain sequence x.

[0063] S13: Perform hierarchical quantization on the echo frequency domain sequence, and generate a second real matrix according to the hierarchical quantization result.

[0064] Specifically, the step of hierarchically quantizing the echo frequency domain sequence and generating the second real matrix according to the hierarchical quantization results is implemented using the following formula:

[0065] a X (m,n)=0.5*sgn(m-M+1+rank X (n))+0.5,

[0066]

[0067] X max =max(X(1),X(2),…,X(N-1)),

[0068] X min =min(X(1),X(2),…,X(N-1));

[0069] Among them, a X (m,n) is the element in the mth row and nth column of the second real matrix, 0≤m≤M-1, 0≤n≤N-1, and X(n) is the nth element of the echo frequency domain sequence X. max and X min are the maximum and minimum values ​​in the echo time domain sequence x respectively. X (n) is the quantization level corresponding to the nth sampling point in the echo time domain sequence X.

[0070] S14: Concatenate the first real matrix and the second real matrix, and convert the concatenated result into a grayscale image.

[0071] Specifically, the first real matrix and the second real matrix are merged according to the target dimension, and the resulting structure is converted into a grayscale matrix. The grayscale matrix is ​​then output as a grayscale image. The target dimension can be either a row dimension or a column dimension. If the target dimension is a row dimension, the resulting matrix has dimensions of M × (2N), while if the target dimension is a row dimension, the resulting matrix has dimensions of (2M) × N.

[0072] Among them, the process of converting the splicing result into a grayscale matrix can be expressed by the following formula:

[0073] a=(2 8 -1)·[a x ,a X ];

[0074] In this formula, a x represents the first real matrix, a X represents the second real matrix, [a x ,a X ] means a x with a X The concatenated matrix.

[0075] Alternatively, in an optional implementation, if the dimension of the spliced ​​matrix does not match the input of the convolutional neural network model, the value of each element in the splicing result can be converted into a grayscale value to form an initial grayscale image; then, the size of the initial grayscale image is adjusted to match the input of the convolutional neural network model based on the interpolation method, thereby obtaining a grayscale image to be input into the convolutional neural network for interference recognition.

[0076] Continue to see Figure 1 As shown, the above interference identification steps include:

[0077] S20: Input the grayscale image into the pre-trained convolutional neural network model, so that the convolutional neural network model outputs the radar interference signal type recognition result.

[0078] Among them, the convolutional neural network model is trained based on multiple sample grayscale images and the radar interference signal type corresponding to each sample grayscale image. Each sample grayscale image is a grayscale image obtained by processing a sample echo time domain sequence according to the signal preprocessing steps.

[0079] For example, see Figure 2 As shown, the convolutional neural network model may include a first convolutional layer conv1, a second convolutional layer conv2, a third convolutional layer conv3, a fourth convolutional layer conv4, a first fully connected layer mlp1 and a second fully connected layer mlp2 connected in sequence.

[0080] The convolution kernel size of each convolutional layer is 3*3, and the number of input and output channels of the convolution kernel increases along the data flow. The activation function of each convolutional layer is ReLU. Each convolutional layer is connected through convolution operations on the convolution kernel, and fully connected layers are connected to adjacent layers through a fully connected method.

[0081] When the convolutional neural network model is working, the grayscale image is input into the first convolutional layer conv1, and the second fully connected layer mlp2 outputs the radar interference signal type recognition result.

[0082] It can be seen that the convolutional neural network model contains 4 cascaded convolutional layers, which can more deeply mine the signal features in both time domain and frequency domain from the grayscale image, as well as the correspondence between these two features, making the final recognition result more accurate.

[0083] In a specific example, the model structure parameters of the above convolutional neural network model are as follows:

[0084]

[0085]

[0086] It is understandable that deep convolutional neural network models of other structural forms can also be used as the convolutional neural network model in the embodiment of the present invention, that is, the structure of the convolutional neural network model used in the embodiment of the present invention is not limited to the above examples.

[0087] The training methods of the above convolutional neural network model include:

[0088] (1) obtaining a training data set; the training data set includes a plurality of sample grayscale images, and the plurality of sample grayscale images belong to a plurality of different radar interference signal types;

[0089] (2) Use the k-fold cross-validation method to divide the training data set into k training sets and k validation sets; the k training sets and k validation sets correspond one to one;

[0090] (3) The convolutional neural network model is trained based on k training sets, and the validation set corresponding to each training set is used to verify the training effect based on the training set.

[0091] Exemplarily, in one implementation, a radar jamming signal set can be used to construct a training data set to train the convolutional neural network model: the data set contains multiple radar jamming signals, which belong to eight types of radar jamming signals: intermittent sampling and forwarding jamming, aiming jamming, blocking jamming, frequency sweep jamming, range deception jamming, smart noise jamming, dense false target jamming, and chaff jamming. There are 20,000 data of each type, for a total of 16,000 data. Each of the radar jamming signals is processed according to the above-mentioned signal preprocessing steps to obtain a training data set. Thus, in the constructed training data set, the radar jamming signal type corresponding to any training sample includes: intermittent sampling and forwarding jamming, aiming jamming, blocking jamming, frequency sweep jamming, range deception jamming, smart noise jamming, dense false target jamming, or chaff jamming. Figure 4(a) to Figure 4(h) The grayscale images of the samples obtained according to the above 8 radar interference signals are shown respectively. Figure 5(a) to Figure 5(h) It is shown accordingly Figure 3 The convolutional layers of the model shown are for Figure 4(a) to Figure 4(h) Output for the sample grayscale image shown.

[0092] In addition, optionally, in order to test the training effect, before using the k-fold cross-validation method to divide the training data set, a stratified sampling method can be used to extract a portion (e.g., 20%) of the sample grayscale images from the training data set to form a test set, so that the k-fold cross-validation method is used to divide the remaining sample grayscale images in step (2). Thus, k training sets, k validation sets, and 1 test set can be formed based on the training data set. The division results are shown as follows:

[0093]

[0094] In another implementation, various types of interference signals may be actually emitted outward by the radar, and the training data set may be formed by collecting echo signals. This is also possible.

[0095] In the process of training the convolutional neural network model in step (3), the cross entropy loss is calculated based on the radar interference signal type to which the sample grayscale image belongs and the radar interference signal type output by the convolutional neural network model, so as to evaluate the training effect of the convolutional neural network model based on the cross entropy loss.

[0096] Here, the cross entropy loss is calculated as:

[0097]

[0098] Among them, the radar interference signal type output by the convolutional neural network model is an 8-dimensional vector x, x[j] means taking the j-th element of x; x[class] means taking the class-th element of x, and the value of class is determined according to the radar interference signal type to which the sample grayscale image belongs.

[0099] Specifically, in the actual training process, see Figure 2 As shown, k training sets are used to train the convolutional neural network model. Each time a training sample is input into the model, a cross-entropy loss is calculated, so that the model weight parameters are adjusted according to the statistical cross-entropy loss value. After each p training sets (epochs), the corresponding validation set is used to verify the recognition effect. If the validation result shows poor recognition effect, the remaining training sets are used for training. If the recognition effect is good, the model is tested using the entire training dataset. If the test result shows that the accuracy rate does not meet the standard (for example, less than 97%), it indicates that the model is underfitting, and training continues using the remaining training sets. If the test result shows that the accuracy rate meets the standard, the trained convolutional neural network model is tested specifically using the test set. If the test passes, the model can be put into use. If the test fails, it indicates that the model is overfitting, and methods such as dataset augmentation and dropout can be used to retrain.

[0100] The model training parameters of the convolutional neural network model during the training process can be seen as follows, but are of course not limited to these:

[0101] Maximum number of training cycles 5000 Number of reads per training batch 64 Learning rate 0.001*power(0.76^epoch) Is the training dataset shuffled? yes Test whether the data set is shuffled no

[0102] In this table, power is a function that returns the power of a given number, epoch represents a training set, and each completion of training for a training set is an epoch.

[0103] In practical applications, the construction and training of the above convolutional neural network model can be implemented in Python software.

[0104] Figures 6(a) and 6(b) show the recognition accuracy and loss curves, respectively, over epochs during the training process of the convolutional neural network model. On the validation set, the loss curve of the convolutional neural network model remained stable and relatively low after 500 epochs of training. The convolutional neural network model performed well on both the training and validation sets, with no underfitting observed. This indicates that the network structure is sufficient to fit the radar jamming signal data. Ultimately, the convolutional neural network model achieved a recognition accuracy of 99.1% on the training set and 97.6% on the validation set, demonstrating excellent performance.

[0105] Figure 7 The test results of the test set on the convolutional neural network model are shown below. The ordinate of the matrix represents the actual label of the data, and the abscissa represents the recognition result output by the convolutional neural network model. The labels "0-7" correspond to eight types of interference signals. The matrix value represents the proportion of cases in the total test data set that were identified as data with the ordinate label, indicating the frequency of each situation in the recognition results. Only the diagonal elements of the matrix represent correctly identified situations; the remaining matrix elements represent incorrectly identified situations.

[0106] The following details the convolutional neural network model's recognition accuracy for eight types of radar jamming signals. It can be seen that the average jamming recognition accuracy on the test set reached 97.92%, demonstrating excellent performance in identifying radar jamming signals.

[0107] Interference Type Test set data volume Number of correctly identified samples Recognition accuracy Intermittent sampling and forwarding interference 4000 3925 99.64％ Targeted jamming 4000 4000 100.00％ Blocking interference 4000 3999 99.99％ Sweep frequency interference 4000 3984 99.60％ Distance deception jamming 4000 4000 100.00％ Smart noise jamming 4000 3925 98.12％ Dense false target interference 4000 3906 97.65％ Chaff jamming 4000 3533 88.33％ total 32000 / 97.92％

[0108] In the radar interference signal type identification method based on a convolutional neural network model provided by the embodiment of the present invention, based on the convolutional neural network model, an image classification and recognition method is used to intelligently identify the radar interference signal type. Compared with the working mode of traditional anti-interference algorithms that can only perform anti-interference processing for interference with similar principles and perform specific analysis for specific problems, the embodiment of the present invention has a higher model generalization capability. In addition, because the grayscale image input to the convolutional neural network model in the embodiment of the present invention contains information in both the time domain and the frequency domain, signal features in both the time domain and the frequency domain can be extracted, and the correspondence between the time domain signal features and the frequency domain signal features can be utilized. Therefore, the radar interference signal type identification result output by the embodiment of the present invention is more accurate and applicable to complex electromagnetic environments. It can also perform interference identification in the face of unpredictable signal types or complex types, thereby facilitating the radar to further take targeted anti-interference measures to achieve real-time and accurate anti-interference processing.

[0109] The method provided in the embodiments of the present invention can be applied to electronic devices. Specifically, the electronic devices can be radars, desktop computers, portable computers, servers, etc. This is not limited here; any electronic device that can implement the present invention falls within the scope of protection of the present invention.

[0110] It should be noted that the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Therefore, features defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0111] In the description of this specification, the reference terms "one embodiment," "some embodiments," "example," "specific example," or "some examples" mean 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 invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described can be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification.

[0112] Although the present application is described herein in conjunction with various embodiments, in the process of implementing the claimed application, those skilled in the art can understand and implement other variations of the disclosed embodiments by reviewing the drawings, the disclosure, and the appended claims.

[0113] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. For those skilled in the art of the present invention, without departing from the concept of the present invention, several simple deductions or substitutions can be made, which should be considered to fall within the scope of protection of the present invention.

Claims

1. A radar interference signal type recognition method based on a convolutional neural network model, characterized in that: include: Signal pre-processing step and interference identification step; The signal preprocessing step includes: obtaining a radar echo time domain sequence, and performing time-frequency conversion on the echo time domain sequence to obtain an echo frequency domain sequence; performing hierarchical quantization on the echo time domain sequence, and generating a first real matrix according to the hierarchical quantization result; performing hierarchical quantization on the echo frequency domain sequence, and generating a second real matrix according to the hierarchical quantization result; splicing the first real matrix and the second real matrix, and converting the spliced ​​result into a grayscale image; The interference identification step includes: inputting the grayscale image into a pre-trained convolutional neural network model, so that the convolutional neural network model outputs a radar interference signal type identification result; In which, the convolutional neural network model is trained based on multiple sample grayscale images and the radar interference signal type corresponding to each sample grayscale image; each of the sample grayscale images is a grayscale image obtained by processing a sample echo time domain sequence according to the signal preprocessing step.

2. The method according to claim 1, characterized in that The step of hierarchically quantizing the echo time domain sequence and generating a first real matrix according to the hierarchical quantization results is implemented using the following formula: a x (m,n)=0.5*sgn(m-M+1+rank x (n))+0.5, x max =max(x(1),x(2),…,x(N-1)), x min =min(x(1),x(2),…,x(N-1)); Among them, a x (m,n) is the element in the mth row and nth column of the first real matrix, 0≤m≤M-1, 0≤n≤N-1, M is the number of levels, N is the length of the echo time domain sequence x; x(n) is the nth element of x; sgn(·) is the sign function.

3. The method according to claim 1, characterized in that The step of hierarchically quantizing the echo frequency domain sequence and generating a second real matrix according to the hierarchical quantization results is implemented using the following formula: a X (m,n)=0.5*sgn(m-M+1+rank X (n))+0.5, X max =max(X(1),X(2),…,X(N-1)), X min =min(X(1),X(2),…,X(N-1)); Among them, a X (m,n) is the element in the mth row and nth column of the second real matrix, 0≤m≤M-1, 0≤n≤N-1, and X(n) is the nth element of the echo frequency domain sequence X.

4. The method according to claim 1, wherein The step of converting the stitching result into a grayscale image comprises: Convert the value of each element in the splicing result into a grayscale value to form an initial grayscale image; The size of the initial grayscale image is adjusted to match the input of the convolutional neural network model based on interpolation.

5. The method according to claim 1, wherein The convolutional neural network model includes a first convolutional layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, a first fully connected layer, and a second fully connected layer connected in sequence; The convolution kernel size of each convolution layer is 3*3, and the number of input channels and output channels of the convolution kernel increases along the data flow direction; the activation function of each convolution layer is ReLU; When the convolutional neural network model is working, the grayscale image is input into the first convolutional layer, and the second fully connected layer outputs the radar interference signal type recognition result.

6. The method according to claim 1, characterized in that The training method of the convolutional neural network model includes: Acquire a training data set; the training data set includes a plurality of sample grayscale images, and the plurality of sample grayscale images belong to a plurality of different radar interference signal types; The training data set is divided into k training sets and k validation sets using a k-fold cross-validation method; the k training sets and the k validation sets correspond one to one; The convolutional neural network model is trained based on the k training sets, and the training effect of the training set is verified using the verification set corresponding to the training set.

7. The method according to claim 6, characterized in that During the training of the convolutional neural network model, the cross entropy loss is calculated based on the radar interference signal type to which the sample grayscale image belongs and the radar interference signal type output by the convolutional neural network model, so as to adjust the network weight parameters of the convolutional neural network model based on the cross entropy loss.

8. The method according to claim 7, characterized in that The calculation formula of the cross entropy loss is: Among them, the radar interference signal type output by the convolutional neural network model is an 8-dimensional vector x, x[j] means taking the j-th element of x; x[class] means taking the class-th element of x, and the value of class is determined according to the radar interference signal type to which the sample grayscale image belongs.

9. The method according to claim 7, characterized in that The training method also includes: Before using the k-fold cross-validation method to divide the training data set, a portion of sample grayscale images is extracted from the training data set using a stratified sampling method to form a test set, so as to subsequently use the k-fold cross-validation method to divide the remaining sample grayscale images in the training data set; And, after obtaining the trained convolutional neural network model, the trained convolutional neural network model is tested using the test set.

10. The method according to claim 1, characterized in that The radar jamming signal types corresponding to any of the training samples include: intermittent sampling and forwarding jamming, aiming jamming, blocking jamming, frequency sweep jamming, range deception jamming, smart noise jamming, dense false target jamming or chaff jamming.

Images