Interference feature selection method based on difference of marginal discriminability

Abstract

A kind of interference feature selection method based on marginal discriminant force difference, for radar echo signal, extract seven kinds of features in time domain before and after pulse compression, frequency domain, time-frequency domain and range doppler domain;Through marginal discriminant force quantitative evaluation each domain discriminant contribution, construct the feature difference prior weight matrix for different transformation domain for different class interference score;Through the comprehensive score of whole interference domain, the optimal transformation domain combination is automatically selected for sorting, which is used to select one or more preferred transformation domain combination, so as to be used for unknown interference signal identification.The invention is closed statistical calculation, without iteration, without hyperparameter, with strong interpretability and stability, effectively improves the recognition accuracy and efficiency.

Description

Technical Field

[0001] This invention relates to the field of radar jamming signal identification technology, specifically to a differential jamming feature selection method based on marginal discriminative power, which can select the optimal transform domain combination as jamming feature and input it into a classifier for radar jamming signal identification. Background Technology

[0002] In complex electromagnetic warfare environments, active interference severely threatens the normal detection capabilities of radar systems. Effective interference type identification is a prerequisite for implementing targeted suppression measures. Before being used for interference identification, the radar received signal undergoes a series of ordered signal processing steps. The different transform domains of the outputs from different processing stages collectively constitute the multi-dimensional information source for interference feature extraction. The core process of radar signal processing is as follows: The discrete-time domain representation of the radar received signal after sampling is as follows: , ,in Let be the number of sampling points. After pulse compression, let the time-domain impulse response of the matched filter be... Then the pulse compression output is:

[0003] in The target scattering coefficient, This is the echo delay.

[0004] The signal, after pulse compression processing, enters the moving target detection module. Assuming the radar transmits within one coherent processing time... There are 1 pulse, and the pulse number is 1. After pulse compression processing, the first The pulse corresponds to the first Multiple echoes are obtained at the distance gate. , For each distance gate Windowing and performing... Point FFT yields the distance-Doppler domain output. :

[0005] in For window functions, This is the Doppler frequency index.

[0006] Existing interference identification methods are generally limited to a single transform domain in feature construction: for single-pulse signals, they are mostly concentrated in the time domain, frequency domain, or a certain domain in the time-frequency domain; for multi-pulse processing, they mainly rely on the range Doppler domain of the moving target detection output, and the joint utilization of information from multiple transform domains is obviously insufficient.

[0007] For example, Chinese invention patent application CN201911090196.7 discloses a radar jamming signal feature-level fusion identification method based on deep convolutional neural networks. This method establishes a radar jamming time-domain dataset and extracts feature vectors in two different forms and fuses them in series: First, it uses a one-dimensional convolutional neural network to extract time-domain feature vectors and fuses them in series with manually extracted expert feature vectors (including time-domain moment skewness, time-domain moment kurtosis, time-domain signal envelope undulation, time-domain mean, and variance); Second, it uses a one-dimensional convolutional neural network to extract time-domain feature vectors and fuses them in series with time-frequency domain feature vectors extracted by a two-dimensional convolutional neural network. Then, it uses PCA to reduce the dimensionality and inputs the data into a support vector machine to complete the classification, thereby realizing radar jamming signal type identification.

[0008] In the methods described above, the transform domain characteristics of the same type of interference exhibit significant differences at different processing stages, such as pulse compression and moving target detection, and the contributions of each transform domain to the discrimination of different interference categories are highly uneven. Therefore, how to fully explore the complementary potential of multiple transform domains in interference identification has become a technical problem that urgently needs to be solved by existing technologies. Summary of the Invention

[0009] The purpose of this invention is to propose a differential interference feature selection method based on marginal discriminative power, which can extract multi-transform domain features, characterize the feature differences of different interference signals, and construct different transform domain combination scores to evaluate the feature difference discrimination degree of the transform domain combination, thereby obtaining one or more combinations with the best effect on all interferences for radar interference signal identification.

[0010] To achieve this objective, the present invention adopts the following technical solution: A method for selecting differential interference features based on marginal discriminative power includes the following steps: Multi-transform domain statistical feature extraction step S110: Obtain the radar signal sample set and the total number of interference categories. Statistical features were extracted from radar interference signals in the sample set in the time domain, frequency domain, time-frequency domain, and range-Doppler domain before and after pulse compression, totaling seven categories of transform domains. These features were then combined with interference category labels. Construct a multi-dimensional, multi-transformation-domain joint feature vector ; Step S120 for constructing the marginal discriminant matrix: Scalar compression is performed on the feature subvectors of each sample in different transform domains; scalar class means are calculated for the same transform domain under the corresponding class according to different interference categories; for a certain transform domain, a globally uniform mixed intra-class variance is calculated based on all samples and all interference categories; furthermore, for that transform domain, the inter-class mean deviation is calculated and normalized based on different interference categories, finally yielding the marginal discriminant matrix. To quantitatively characterize the ability of each transform domain to distinguish various types of interference; Feature difference prior weight matrix construction step S130: All non-empty combinations of the 7 types of transform domains are enumerated; for each combination of transform domains and each type of interference, the marginal discriminative power of the transform domains contained in the combination is summed to obtain the prior discriminative score of the combination for a specific type of interference; the score is calculated by traversing all combinations of transform domains and all types of interference, and finally the feature difference prior weight matrix is ​​constructed. To quantitatively characterize the relative discriminative contribution of each transform domain combination to various types of interference; Comprehensive prior score calculation steps S140: For each transform domain combination, its prior discrimination scores across all interference categories are summed globally to obtain the comprehensive prior score of the corresponding combination. Based on the comprehensive prior score, all transform domain combinations are sorted, and the combinations with the highest scores are selected as the optimal feature input combinations to obtain the optimal feature selection result driven by marginal discrimination, which is used to complete the feature extraction and identification of unknown interference signals.

[0011] Optionally, in step S110, Pre-pulse compression temporal features extraction: mean, standard deviation, root mean square value; Pre-pulse compression frequency domain feature extraction: spectral mean, frequency centroid, and frequency standard deviation; Pre-pulse compression time-frequency domain feature extraction: time-frequency energy, spectral centroid, and spectral entropy; Extract the following time-domain features from the pulse compression analysis: mean, standard deviation, and root mean square value. Extracting frequency domain features from pulse compression: mean spectrum, centroid of frequency, and standard deviation of frequency; Post-pulse compression time-frequency domain feature extraction: time-frequency energy, spectral centroid, and spectral entropy; Range Doppler domain feature extraction: range Doppler energy and number of peaks in the signal after moving target detection.

[0012] Optionally, in step S110, Radar signal sample set is ,in The total number of samples, For the first The original received signal of each sample, Label its interference category. This represents the total number of interference categories.

[0013] Optionally, in step S110, The construction of the multi-dimensional, multi-transform domain joint feature vector specifically involves concatenating the seven transform domain feature vectors to obtain the corresponding sample. s Given the complete feature vector, perform the above concatenation process on all N samples to output a labeled sample feature set. ,in For the complete feature vector of sample s, This indicates the interference category label.

[0014] Optionally, step S120 specifically includes: For each sample Take the first Transform domain feature vectors The 2-norm serves as a scalar compression representation of this transform domain: , In the mean statistics of each transform domain, for the samples of each category, the within-class means are calculated for its seven transform domains. Specifically, for the samples belonging to the first category... All samples of class, count the number of samples. Class mean of a scalar in the transformed domain:

[0015] in For the first The set of indices of the sample class. The number of samples in this category. Mixed within-class variance statistics: For all N samples, calculate the mixed within-class variance of the i-th transform domain: , The marginal discriminant power of the i-th transform domain against the c-th type of interference is calculated using the deviation of the normalized inter-class mean: , Obtain the marginal discriminant matrix .

[0016] Optionally, step S130 specifically includes: right Enumerate all non-empty subsets of each transformation domain, and there exist... A combination of transform domains, using a 7-bit binary mask vector. , Indicates the first a combination, ,in Indicates the first The combination was included in the first Transform domain, This means it will not be included; These represent the pre-pulse compression time domain, pre-pulse compression frequency domain, pre-pulse compression time-frequency domain, post-pulse compression time domain, post-pulse compression frequency domain, post-pulse compression time-frequency domain, and range-Doppler domain, respectively. For the k-th transform domain combination and the c-th type of interference, the sum of the marginal discriminative powers of all activated transform domains in this combination is used as the prior discriminant score: , The prior discrimination scores of all 127 combinations and type C interference are summarized to form a feature difference prior matrix. .

[0017] Optionally, in step S140, For the k-th transform domain combination, sum the prior discriminant scores on all Class C interferences to obtain the comprehensive prior score of the combination. , .

[0018] A radar interference signal identification method utilizes the aforementioned differential interference feature selection method based on marginal discriminative power to select multiple transform domain combinations with the highest scores, and inputs the selected transform domain combinations into a classifier for radar interference signal identification.

[0019] The present invention further discloses a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described differential interference feature selection method based on marginal discriminative power.

[0020] The present invention also discloses a differential interference feature selection device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the differential interference feature selection method based on marginal discriminative power described above.

[0021] In summary, the present invention has the following advantages: 1. This invention uses marginal discriminant force to achieve quantitative and interpretable characterization of differences in features across multiple transform domains. The calculation is based on statistical closed-form derivation, which differs from black-box learning in neural networks. The weights in this invention have a clear source, rigorous theory, and higher engineering credibility.

[0022] 2. This invention uses a feature difference matrix obtained from offline sample statistics, which does not require an iterative optimization process, has high computational efficiency, and the results are reproducible. It is not affected by network initialization and parameter tuning, and has stronger stability and real-time performance in interference identification.

[0023] 3. This invention can automatically select the combination of transform domains with the strongest discrimination ability and the lowest redundancy, fully explore the complementary information of multiple domains, solve the problem of insufficient information in a single domain, and effectively improve the recognition accuracy. Attached Figure Description

[0024] Figure 1 This is a flowchart of a differential interference feature selection method based on marginal discriminative power according to a specific embodiment of the present invention. Detailed Implementation

[0025] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0026] This invention primarily involves: extracting seven types of features—time domain, frequency domain, time-frequency domain, and range-Doppler domain—before and after pulse compression of radar echo signals; quantitatively evaluating the discriminative contribution of each domain through marginal discriminative power; constructing a priori weight matrix of feature differences in scores for different types of interference across different transform domains; and automatically selecting the optimal transform domain combination based on the comprehensive score ranking across the entire interference domain. This optimal combination is then used to select one or more preferred transform domain combinations for the identification of unknown interference signals. This invention employs closed-loop statistical calculations throughout, without iteration or hyperparameters, exhibiting strong interpretability and stability, and effectively improving identification accuracy and efficiency.

[0027] See Figure 1 A flowchart of a differential interference feature selection method based on marginal discriminative power according to a specific embodiment of the present invention is disclosed, including the following steps: Multi-transform domain statistical feature extraction step S110: Obtain the radar signal sample set and the total number of interference categories. Statistical features were extracted from radar interference signals in the sample set in the time domain, frequency domain, time-frequency domain, and range-Doppler domain before and after pulse compression, totaling seven categories of transform domains. These features were then combined with interference category labels. Construct a multi-dimensional, multi-transformation-domain joint feature vector .

[0028] In one specific embodiment Pre-pulse compression temporal features extraction: mean, standard deviation, root mean square value; Pre-pulse compression frequency domain feature extraction: spectral mean, frequency centroid, and frequency standard deviation; Pre-pulse compression time-frequency domain feature extraction: time-frequency energy, spectral centroid, and spectral entropy; Extract the following time-domain features from the pulse compression analysis: mean, standard deviation, and root mean square value. Extracting frequency domain features from pulse compression: mean spectrum, centroid of frequency, and standard deviation of frequency; Post-pulse compression time-frequency domain feature extraction: time-frequency energy, spectral centroid, and spectral entropy; Range Doppler domain feature extraction: range Doppler energy and number of peaks in the signal after moving target detection.

[0029] Radar signal sample sets can be ,in The total number of samples, For the first The original received signal of each sample, Label its interference category. This represents the total number of interference categories.

[0030] The construction of the multi-dimensional, multi-transform domain joint feature vector specifically involves concatenating the seven transform domain feature vectors to obtain the corresponding sample. s Given the complete feature vector, perform the above concatenation process on all N samples to output a labeled sample feature set. ,in For the complete feature vector of sample s, This indicates the interference category label.

[0031] In one specific embodiment, the process is as follows: Constructing a signal sample set ,in The total number of samples, For the first The original received signal of each sample, Label its interference category. This represents the total number of interference categories.

[0032] Pre-pulse compression temporal feature extraction:

[0033] in, These represent the mean, standard deviation, and root mean square value of the pre-pulse compression signal, respectively.

[0034] Extraction of pre-pulse compression frequency domain features:

[0035] in, These represent the spectrum, centroid, and standard deviation frequency of the pre-pulse signal, respectively.

[0036] Pre-pulse compression time-frequency domain feature extraction:

[0037] in, These are the time-frequency energy, spectral centroid, and spectral entropy of the pre-pulse compression signal, respectively.

[0038] Extracting post-pulse compression temporal features:

[0039] in, These represent the mean, standard deviation, and root mean square value of the signal after pulse compression, respectively.

[0040] Extracting frequency domain features after pulse compression:

[0041] in, These are the frequency spectrum, frequency centroid, and standard deviation frequency of the signal after pulse compression.

[0042] Post-pulse compression time-frequency domain feature extraction:

[0043] in, These are the time-frequency energy, spectral centroid, and spectral entropy of the pulse-compressed signal, respectively.

[0044] Feature extraction in the range-Doppler domain:

[0045] in These represent the range Doppler energy and the number of peaks in the signal after moving target detection, respectively.

[0046] The multi-dimensional, multi-transformation-domain joint feature vector is obtained by concatenation: By concatenating the seven transform domain feature vectors, we obtain the complete feature vector of the corresponding sample s:

[0047] For all Perform the above concatenation process on each sample to output a labeled sample feature set. ,in For the complete feature vector of sample s, This indicates the interference category label.

[0048] Step S120 for constructing the marginal discriminant matrix: Scalar compression is performed on the feature subvectors of each sample in different transform domains; scalar class means are calculated for the same transform domain under the corresponding class according to different interference categories; for a certain transform domain, a globally uniform mixed intra-class variance is calculated based on all samples and all interference categories; furthermore, for that transform domain, the inter-class mean deviation is calculated and normalized based on different interference categories, finally yielding the marginal discriminant matrix. This is to quantitatively characterize the ability of each transform domain to distinguish various types of interference.

[0049] In one specific embodiment For each sample Take the first Transform domain feature vectors The 2-norm serves as a scalar compression representation of this transform domain: , In the mean statistics of each transform domain, for the samples of each category, the within-class means are calculated for its seven transform domains. Specifically, for the samples belonging to the first category... All samples of class, count the number of samples. Class mean of a scalar in the transformed domain:

[0050] in For the first The set of indices of the sample class. This represents the number of samples in this category.

[0051] Mixed within-class variance statistics: For all N samples, calculate the mixed within-class variance of the i-th transform domain:

[0052] The marginal discriminant power of the i-th transform domain against the c-th type of interference is calculated using the deviation of the normalized inter-class mean:

[0053] Obtain the marginal discriminant matrix .

[0054] Feature difference prior weight matrix construction step S130: All non-empty combinations of the 7 types of transform domains are enumerated; for each combination of transform domains and each type of interference, the marginal discriminative power of the transform domains contained in the combination is summed to obtain the prior discriminative score of the combination for a specific type of interference; the score is calculated by traversing all combinations of transform domains and all types of interference, and finally the feature difference prior weight matrix is ​​constructed. This is to quantitatively characterize the relative discriminative contribution of each transform domain combination to various types of interference.

[0055] In one specific embodiment First, perform a combination enumeration of transformation fields: right Enumerate all non-empty subsets of each transformation domain, and there exist... A combination of transform domains, using a 7-bit binary mask vector. , Indicates the first a combination, ,in Indicates the first The combination was included in the first Transform domain, This means it will not be included; These represent the pre-pulse compression time domain, pre-pulse compression frequency domain, pre-pulse compression time-frequency domain, post-pulse compression time domain, post-pulse compression frequency domain, post-pulse compression time-frequency domain, and range-Doppler domain, respectively.

[0056] Then, the prior discriminant scores for each combination are calculated: For the k-th transform domain combination and the c-th type of interference, use all activated transform domains in this combination (i.e. The sum of the marginal discriminative powers of each digit is used as the prior discriminant score:

[0057] The prior discrimination scores of all 127 combinations and type C interference are summarized to form a feature difference prior matrix. .

[0058] Comprehensive prior score calculation steps S140: For each transform domain combination, its prior discrimination scores across all interference categories are summed globally to obtain the comprehensive prior score of the corresponding combination. Based on the comprehensive prior score, all transform domain combinations are sorted, and the combinations with the highest scores are selected as the optimal feature input combinations to obtain the optimal feature selection result driven by marginal discrimination, which is used to complete the feature extraction and identification of unknown interference signals.

[0059] In one specific embodiment For the k-th transform domain combination, sum the prior discriminant scores on all Class C interferences to obtain the comprehensive prior score of the combination. , .

[0060] Based on the ranking of scores, one or more transform domain combinations with the highest scores are selected as the optimal feature input combination to obtain the optimal feature selection result of the difference driven by marginal discriminative power, which is used to complete the feature extraction and recognition of unknown interference signals.

[0061] The present invention further discloses a radar interference signal identification method, characterized in that, by using the aforementioned differential interference feature selection method based on marginal discriminative power, multiple transform domain combinations with the highest scores are selected, and the selected transform domain combinations are input into a classifier for radar interference signal identification.

[0062] The present invention further discloses a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described differential interference feature selection method based on marginal discriminative power.

[0063] The present invention also discloses a differential interference feature selection device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the differential interference feature selection method based on marginal discriminative power described above.

[0064] In summary, the present invention has the following advantages: 1. This invention uses marginal discriminant force to achieve quantitative and interpretable characterization of differences in features across multiple transform domains. The calculation is based on statistical closed-form derivation, which differs from black-box learning in neural networks. The weights in this invention have a clear source, rigorous theory, and higher engineering credibility.

[0065] 2. This invention uses a feature difference matrix obtained from offline sample statistics, which does not require an iterative optimization process, has high computational efficiency, and the results are reproducible. It is not affected by network initialization and parameter tuning, and has stronger stability and real-time performance in interference identification.

[0066] 3. This invention can automatically select the combination of transform domains with the strongest discrimination ability and the lowest redundancy, fully explore the complementary information of multiple domains, solve the problem of insufficient information in a single domain, and effectively improve the recognition accuracy.

[0067] Obviously, those skilled in the art will understand that the various units or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device, or optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computing device. Alternatively, they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0068] The above description is a further detailed explanation of the present invention in conjunction with specific preferred embodiments. It should not be considered that the specific embodiments of the present invention are limited to this. For those skilled in the art, several simple deductions or substitutions can be made without departing from the concept of the present invention, and all such deductions or substitutions should be considered to fall within the scope of protection of the present invention as defined by the submitted claims.

Claims

1. A method for selecting differential interference features based on marginal discriminative power, characterized in that, Includes the following steps: Multi-transform domain statistical feature extraction step S110: Obtain the radar signal sample set and the total number of interference categories. Statistical features were extracted from radar interference signals in the sample set in the time domain, frequency domain, time-frequency domain, and range-Doppler domain before and after pulse compression, totaling seven categories of transform domains. These features were then combined with interference category labels. Construct a multi-dimensional, multi-transformation-domain joint feature vector ; Step S120 for constructing the marginal discriminant matrix: Scalar compression is performed on the feature subvectors of each sample in different transform domains; according to different interference categories, the scalar class mean of the same transform domain under the corresponding category is calculated; For a given transform domain, a globally uniform mixed within-class variance is calculated based on all samples and all interference categories. Then, for the same transform domain, the inter-class mean deviation is calculated and normalized for each interference category, ultimately yielding the marginal discriminant matrix. To quantitatively characterize the ability of each transform domain to distinguish various types of interference; Feature difference prior weight matrix construction step S130: All non-empty combinations of the 7 types of transform domains are enumerated; for each combination of transform domains and each type of interference, the marginal discriminative power of the transform domains contained in the combination is summed to obtain the prior discriminative score of the combination for a specific type of interference; the score is calculated by traversing all combinations of transform domains and all types of interference, and finally the feature difference prior weight matrix is ​​constructed. To quantitatively characterize the relative discriminative contribution of each transform domain combination to various types of interference; Comprehensive prior score calculation steps S140: For each transform domain combination, its prior discrimination scores across all interference categories are summed globally to obtain the comprehensive prior score of the corresponding combination. Based on the comprehensive prior score, all transform domain combinations are sorted, and the combinations with the highest scores are selected as the optimal feature input combinations to obtain the optimal feature selection result driven by marginal discrimination, which is used to complete the feature extraction and identification of unknown interference signals.

2. The method according to claim 1, characterized in that: In step S110, Pre-pulse compression temporal features extraction: mean, standard deviation, root mean square value; Pre-pulse compression frequency domain feature extraction: spectral mean, frequency centroid, and frequency standard deviation; Pre-pulse compression time-frequency domain feature extraction: time-frequency energy, spectral centroid, and spectral entropy; Extract the following time-domain features from the pulse compression analysis: mean, standard deviation, and root mean square value. Extracting frequency domain features from pulse compression: mean spectrum, centroid of frequency, and standard deviation of frequency; Post-pulse compression time-frequency domain feature extraction: time-frequency energy, spectral centroid, and spectral entropy; Range Doppler domain feature extraction: range Doppler energy and number of peaks in the signal after moving target detection.

3. The method according to claim 2, characterized in that: In step S110, Radar signal sample set is ,in The total number of samples, For the first The original received signal of each sample, Label its interference category. This represents the total number of interference categories.

4. The method according to claim 3, characterized in that: In step S110, The construction of the multi-dimensional, multi-transform domain joint feature vector specifically involves concatenating the seven transform domain feature vectors to obtain the corresponding sample. s Given the complete feature vector, perform the above concatenation process on all N samples to output a labeled sample feature set. ,in For the complete feature vector of sample s, This indicates the interference category label.

5. The method according to claim 4, characterized in that: Step S120 is as follows: For each sample Take the first Transform domain feature vectors The 2-norm serves as a scalar compression representation of this transform domain: ， In the mean statistics of each transform domain, for the samples of each category, the within-class means are calculated for its seven transform domains. Specifically, for the samples belonging to the first category... All samples of class, count the number of samples. Class mean of a scalar in the transformed domain: in For the first The set of indices of the sample class. The number of samples in this category. Mixed within-class variance statistics: For all N samples, calculate the mixed within-class variance of the i-th transform domain: ， The marginal discriminant power of the i-th transform domain against the c-th type of interference is calculated using the deviation of the normalized inter-class mean: ， Obtain the marginal discriminant matrix .

6. The method according to claim 5, characterized in that: Step S130 is as follows: right Enumerate all non-empty subsets of each transformation domain, and there exist... A combination of transform domains, using a 7-bit binary mask vector. , Indicates the first a combination, ,in Indicates the first The combination was included in the first Transform domain, This means it will not be included; These represent the pre-pulse compression time domain, pre-pulse compression frequency domain, pre-pulse compression time-frequency domain, post-pulse compression time domain, post-pulse compression frequency domain, post-pulse compression time-frequency domain, and range-Doppler domain, respectively. For the k-th transform domain combination and the c-th type of interference, the sum of the marginal discriminative powers of all activated transform domains in this combination is used as the prior discriminative score: ， The prior discrimination scores of all 127 combinations and type C interference are summarized to form a feature difference prior matrix. .

7. The method according to claim 1, characterized in that: In step S140, For the k-th transform domain combination, sum the prior discriminant scores on all Class C interferences to obtain the comprehensive prior score of the combination. , 。 8. A method for identifying radar interference signals, characterized in that, Using the differential interference feature selection method based on marginal discriminative power as described in any one of claims 1-7, multiple combinations of transform domains with the highest scores are selected, and the selected combinations of transform domains are input into a classifier for radar interference signal identification.

9. A computer-readable storage medium having a computer program stored thereon, When the computer program is executed by the processor, it implements the steps of the differential interference feature selection method based on marginal discriminative power as described in any one of claims 1-7.

10. A differential interference feature selection device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the differential interference feature selection method based on marginal discriminative power as described in any one of claims 1-7.

Images