A bistatic radar passive jamming identification method based on HRRP features

Abstract

The application discloses a kind of based on HRRP feature's bistatic radar passive jamming identification method;The invention is first to HRRP feature is normalized, alignment and determines support area range etc. Preprocessing operation, then extract a total of 18 kinds of features including peak point length, normalized variance and front end structure ratio, and use the feature selection algorithm based on Relief algorithm and specificity coefficient to select features with high discrimination and low redundancy to construct feature vectors, and finally use SVM to complete passive jamming identification. The application proves that the combination of bistatic radar and selected features can effectively improve the performance of jamming identification.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of radar passive jamming recognition, and particularly relates to a bistatic radar passive jamming recognition method based on HRRP features. BACKGROUND

[0002] With the continuous development of radar systems and radar technology, the bistatic radar can play an important role in electronic countermeasures due to its high concealment, strong anti-detection capability and difficulty in being attacked by anti-radiation weapons. When a radar detects or recognizes a target signal, a jamming signal can destroy its normal function. Compared with active jamming, the passive jamming represented by chaff and corner reflector has a large radar cross section (RCS) and can scatter energy in different directions, thus greatly threatening the detection performance of the bistatic radar.

[0003] In order to reduce the jamming effect of chaff and corner reflector on radar, many researchers have studied passive jamming recognition, and found that radar can use the large signal bandwidth characteristic to realize distance high-resolution imaging of the radial target scattering center, realize the representation of the fine geometric structure features and length information of the target, and has important significance in passive jamming recognition. In the aspect of passive jamming recognition based on high resolution range profile (HRRP), the main method is to extract HRRP features of different dimensions and use a classifier for recognition. However, the existing passive jamming recognition methods based on HRRP are all for monostatic radar and a single jamming type, and the research on bistatic radar HRRP feature extraction and passive jamming multi-classification is not sufficient. SUMMARY

[0004] In view of the above problems and the current research trend, the present application proposes a bistatic radar passive jamming recognition method based on HRRP features, which can comprehensively evaluate the feature discrimination and redundancy to select efficient recognition features and improve the recognition performance.

[0005] The bistatic radar passive jamming recognition method based on HRRP features proposed by the present application comprises the following steps:

[0006] Step S1, simulate and construct an HRRP dataset, and perform a series of preprocessing operations on the HRRP, aiming to make the features retain the effective information of the target as much as possible:

[0007] Step S1.1, normalize the HRRP, in order to eliminate the influence of different target distances on the HRRP amplitude;

[0008] Step S1.2, align the HRRP to move the center of gravity of the target to the center position of the window;

[0009] Step S1.3, defining the range of distance cells occupied by the target, reducing the influence of noise outside the target cell on feature extraction;

[0010] Step S2, extracting a total of 18 features including peak point length, normalized variance and front structure ratio, etc. as the basis for subsequent feature selection;

[0011] Step S3, the feature selection algorithm based on the Relief algorithm and specificity coefficient selects features with high discrimination and low redundancy to construct a feature vector;

[0012] Step S4, dividing the constructed feature vector into training set and test set, and using SVM classifier for training and testing.

[0013] Further, the preprocessing operation of the HRRP mainly includes three parts of normalization, alignment and determination of the support area range. In order to eliminate the influence of different target distances on the HRRP amplitude, the HRRP is first normalized. Let the sequence of HRRP modulus be Then the normalized processing can be represented as:

[0014] (1)

[0015] Wherein represents taking the maximum value, is the HRRP length, and the normalized HRRP sequence value range is mapped to 0 to 1.

[0016] Different targets have different positions in the distance window. The alignment of the HRRP moves the center of gravity of the target to the center position of the window. First, the center of gravity position of the sequence is calculated Then the HRRP is circularly shifted to corresponding position to the midpoint of the sequence. The center of gravity can be calculated by the following formula:

[0017] (2)

[0018] Finally, it is necessary to define the range of distance cells occupied by the target, reducing the influence of noise outside the target cell on feature extraction. In the HRRP, the radial distance interval where the target is located is the support area, and the first sampling point of the aligned sequence is regarded as noise, and its mean value is calculated and multiplied by an adjustable bias value as a threshold. The interval between the first and last sampling points higher than the threshold is regarded as the support area .

[0019] Further, the feature extraction mainly extracts HRRP scale features and structure features. The HRRP scale features mainly represent the information of the peak point position, number and interval in the distance dimension, which mainly reflects the radial dimension related properties of the target. The HRRP structure features mainly represent the proportion structure characteristics, which mainly reflects the structure characteristics such as dispersion and fluctuation of the strong scattering points of the target in the radial direction. Eighteen kinds of features are extracted, and each feature is defined as follows:

[0020] 1) Peak point length (F1): defined as the distance between the two peak points with the farthest interval in the target support area, which can be expressed as:

[0021] (3)

[0022] wherein, represents the position of the peak point in the target support area, represents the radar sampling interval, is the number of peak points.

[0023] 2) Peak point number (F2): defined as the number of peak points in the target support area, which can be expressed as:

[0024] (4)

[0025] 3) 25% length (F3): defined as the interval length in the target support area whose energy is 25% lower than the maximum peak point energy, which can be expressed as:

[0026] (5)

[0027] wherein, respectively represent the upper and lower boundaries of the sampling points whose energy is 25% lower than the maximum peak energy.

[0028] 4) 50% length (F4): defined as the interval length in the target support area whose energy is 50% lower than the maximum peak point energy, which can be expressed as:

[0029] (6)

[0030] wherein, respectively represent the upper and lower boundaries of the sampling points whose energy is 50% lower than the maximum peak energy.

[0031] 5) 25% peak point number (F5): defined as the number of peak points in the 25% length interval, which can be expressed as:

[0032] (7)

[0033] wherein Indicates the sequence length.

[0034] 6) Number of peak points at 50% length (F6): defined as the number of peak points within the 50% length interval, which can be expressed as:

[0035] (8)

[0036] 7) 25% length ratio (F7): defined as the ratio of 25% length to peak point length, which can be expressed as:

[0037] (9)

[0038] 8) 50% length ratio (F8): defined as the ratio of 50% length to peak point length, which can be expressed as:

[0039] (10)

[0040] 9) Sorted energy ratio (F9): defined as the ratio of the top 5% points in the target region to the total energy, the specific calculation process is:

[0041] a. Sort the energy of each sampling point in the target support region in descending order;

[0042] b. Sum the top 5% of the HRRP energy after descending order;

[0043] c. Calculate the ratio of the sum of the top 5% of HRRP to the total energy, which is the sorted energy ratio feature .

[0044] 10) Normalized variance (F10): defined as sliding window on all points in the support region, the maximum value of the variance divided by the mean of all windows. The specific calculation process is:

[0045] a. Extend the target HRRP support region by 3 sampling points on both sides;

[0046] b. Perform sliding window on the extended HRRP region, with a window size of 4 sampling points, calculate the mean and standard deviation of the sampling point amplitude in each window;

[0047] c. Calculate the maximum value of the ratio of the mean and standard deviation, convert to dB value, which is the normalized variance feature of the current target .

[0048] 11) Sparsity (F11): defined as the variance of the distance between all adjacent peak points within the target support region, the specific calculation process is:

[0049] a. Find all peak points in the target HRRP support region;

[0050] b. Calculate the distances between adjacent points of these peak points;

[0051] c. Calculate the variance of these distances; this value represents the sparsity feature of the current target. .

[0052] 12) Primary and Secondary Peak Ratio (F12): Defined as the ratio of the energy of the largest peak to the second largest peak in the target support zone, which can be expressed as:

[0053] (11)

[0054] in This represents the energy at its maximum peak. This indicates the energy of the second largest peak value.

[0055] 13) Front-end structure ratio (F13): Defined as the ratio of the distance from the left secondary peak to the highest peak to the length of the target support zone. The left secondary peak refers to the maximum value to the left of the highest peak of HRRP (excluding the highest peak of the target).

[0056] 14) Rear-end structure ratio (F14): Defined as the ratio of the distance from the right secondary peak to the highest peak to the length of the target support zone. The right secondary peak refers to the maximum value to the right of the highest peak of HRRP (excluding the highest peak value of the target).

[0057] 15) Symmetry (F15): Defined as the ratio of the sum of energy in the first half to the sum of energy in the second half of the target's support region, reflecting the degree of symmetry in the target's scattering cross-section distribution, and can be expressed as:

[0058] (12)

[0059] in, Indicates the first The energy value of each sampling point This indicates the number of sampling points in the support area.

[0060] 16) Dispersion (F16): Reflects the degree of dispersion of the radial amplitude distribution of the target HRRP, which can be expressed as:

[0061] (13)

[0062] in, Indicates the first The energy value of each sampling point This indicates the number of sampling points in the support area.

[0063] 17) Descaled structural features (F17): Reflects HRRP amplitude features independent of scale transformation, and can be expressed as:

[0064] (14)

[0065] 18) fluctuation feature (F18): reflecting the strength of the target HRRP amplitude fluctuation, which can be expressed as:

[0066] (15)

[0067] wherein, represents the maximum value of the peak point energy in the support area, and respectively represent the energy and the number of all peak points whose energy is greater than , and represent the positions of the nearest peak points on the left and right of the th peak point among the peak points, and represent the respective energies thereof.

[0068] Further, the feature selection is mainly based on the relief algorithm and the specificity coefficient to select efficient recognition features with high discrimination and low redundancy to construct a feature vector. The specific method of the relief algorithm is as follows:

[0069] The relief algorithm is used to evaluate the discrimination of the features, to represent the ability of the feature to distinguish different categories. The core of the relief algorithm is to measure the distance between a sample and the near neighbor samples of the same class and the different class. When the average distance of a sample and the same class samples is greater than the average distance of the different class samples, it indicates that the feature has higher discrimination for different categories, and the weight value is also larger. The steps are as follows: first, randomly select a sample from the data set, then select samples from each category of data set, wherein the samples of the same category as sample form a set , and the samples of different categories form a set . Then calculate the sample distance to update the discrimination of each feature, as shown in the following formula:

[0070] (16)

[0071] (17)

[0072] wherein, and respectively represent the maximum and minimum feature values of the feature in all samples, is the proportion of the number of samples of category in all samples, is the number of extracted samples. representing the category to which the sample belongs.

[0073] The specificity coefficient can represent the consistency of a certain feature with other features in the sample. The more consistent the distribution, the weaker the uniqueness of the feature, and the higher the redundancy. Let the number of features in a feature set be , and the specificity coefficient of the th feature be defined as:

[0074] (18)

[0075] wherein is the number of categories, is the number of samples, is the th sample value of the th category feature in the th category, is the mean value of the th category feature in the th category. When the specificity coefficient is small, it represents that the specificity of the th feature is weak, and it can be considered as a redundant feature for elimination.

[0076] Advantages:

[0077] 1. Feature extraction for bistatic radar: The present application focuses on HRRP feature extraction for bistatic radar, and can more effectively identify various interference types in view of the insufficient research of existing methods under bistatic radar.

[0078] 2. Efficient feature selection: By combining the relief algorithm and the specificity coefficient, the present application comprehensively evaluates the discrimination and redundancy of features, thereby selecting efficient recognition features and improving the accuracy and efficiency of recognition.

[0079] 3. Verification and performance improvement: Support vector machines are used to verify the feature selection effect, further improving the recognition performance of passive interference and providing stronger technical support for practical applications. BRIEF DESCRIPTION OF DRAWINGS

[0080] Figure 1 is the overall framework diagram of the bistatic radar passive interference recognition method based on HRRP features;

[0081] Figure 2 is an HRRP support area diagram;

[0082] Figure 3 is a recognition rate change diagram under different feature numbers;

[0083] ​Figure 4 The contrast chart of single and double base recognition rate under different feature quantity and dry noise ratio;

[0084] Figure 5 The contrast chart of jamming recognition rate under different feature combination method. DETAILED DESCRIPTION

[0085] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be described below in detail with reference to the embodiments of the present application and the accompanying drawings.

[0086] The detailed process of the method in the embodiment is as follows:

[0087] Figure 1 The overall framework chart of the HRRP feature-based double base radar passive jamming recognition method is shown in the figure. The overall framework is divided into four parts: preprocessing, feature extraction, feature selection and SVM:

[0088] Step S1, simulate and construct the HRRP dataset, and perform a series of preprocessing operations on the HRRP features:

[0089] The HRRP datasets under single and double base radars are respectively constructed by simulation, containing HRRP of three types of targets: chaff cloud, corner reflector array and ship, each type having 600 samples. The HRRP of the target under different observation azimuth conditions is simulated by changing the relative position of the radar and the target. For the HRRP of the single base radar, the incident and receiving azimuth angles are the same, both being 50° to 130°. For the HRRP of the double base radar, the incident azimuth angle is 0°, and the receiving azimuth angle is 50° to 130°. The preprocessing operations of normalization, alignment and determination of the support region range are performed on all HRRPs. The support region range is shown in the figure. Figure 2

[0090] Step S2, extract a total of 18 features including peak point length, normalized variance and front end structure ratio.

[0091] Step S3, select the features with large discrimination and low redundancy to construct the feature vector based on the feature selection algorithm of Relief algorithm and specificity coefficient. First, the Relief algorithm is used to sort the features of the HRRP from high to low according to the discrimination, then the specificity coefficient value is calculated to evaluate the redundancy of each feature, and the features are removed according to the number of features. The remaining features are selected according to the sorting to construct the feature vector.

[0092] ​The HRRP features of the monostatic radar and the bistatic radar are respectively selected based on the method, and the results are shown in Table 1 and Table 2. For the HRRP features of the monostatic radar, the 50% length (F4), the 25% length ratio (F7) and the sorted energy ratio (F9) have high discrimination for different types of targets, and the high discrimination features of the bistatic radar HRRP mainly include the fluctuation structure feature (F18), the sorted energy ratio (F9) and the peak point number (F2). It is shown that the recognition features of the passive jamming HRRP under the monostatic radar and the bistatic radar are not completely the same.

[0093] Table 1

[0094] ;

[0095] Table 2

[0096] ;

[0097] In step S4, the constructed feature vectors are divided into a training set and a test set according to 8:2, and the SVM classifier is used for training and testing:

[0098] The training set samples are input into the SVM classifier for training, and after the training is completed, the test set is extracted according to the feature selection results to construct a feature vector, and is input into the trained classifier to complete the passive jamming recognition. Since the SVM is a two-class algorithm, and the passive jamming recognition of the present application is a three-class problem, a two-stage classification method is needed: first, a classifier is trained to classify “class A” and “class B+C”, and then another classifier is trained to further classify all “class B+C” samples output by the first stage into “class B” and “class C”.

[0099] Without loss of generality, the accuracy index is used to evaluate the recognition performance.

[0100] Figure 3 The recognition rate changes under different feature numbers are shown in the figure, and the HRRP features obtained by sorting the discrimination under the bistatic radar are selected as the feature vectors and the SVM is used to detect the recognition performance.

[0101] It can be seen from Figure 3 that when the highest discrimination feature F18 is used alone as the input, the jamming recognition rate is only 62.3%. It is shown that it is difficult to distinguish different classes by considering only a single feature. With the increase of the number of features, the jamming recognition rate shows a trend of first increasing and then decreasing, and when the number of input features is 6-7, the jamming recognition rate is highest, which is 92.7%. When all 18 features are input, the jamming recognition rate is 77.3%, which shows that the features with low discrimination have a negative impact on the construction of the SVM hyperplane because the feature value distributions of different targets are similar or overlap greatly, so that the model performance is reduced.​Figure 3 and Table 2, when the number of features is increased from 4 to 10, more redundant features are increased, the interference recognition performance changes little and more calculation amount is consumed. These results prove that the feature selection method proposed can comprehensively represent the feature effectiveness, and provide key information for constructing the feature vector.

[0102] Then, the interference recognition performance of monostatic and bistatic radars is compared, the same feature selection number of 4, 8, 16 is set respectively, and the recognition rates of the two are compared under different JNR conditions, and the results are shown in Figure 4 It can be seen that the recognition effect is not good under different feature numbers at low JNR, because the strong noise covers most of the HRRP features with distinguishability.

[0103] The average recognition rate performance comparison is shown in Table 3. The average recognition rate of the bistatic radar under different JNR is higher than that of the monostatic radar by 2.4% to 5.4%, and the feature number required by the bistatic radar to achieve the highest average recognition rate is less than that of the monostatic radar. It shows that the bistatic radar is a better passive interference recognition system by introducing the receiving direction freedom.

[0104] Table 3

[0105] ;

[0106] Based on the bistatic radar, the feature vector constructed by the feature selection method proposed in this paper is compared with the recognition performance of the dimension, feature length and distribution entropy features of the equivalent scattering center under different JNR. Among them, the feature selection number of the application is set to 3, and the results are shown in Figure 5 The recognition rate of the method proposed in this application is higher than that of the comparison method under different JNR conditions, which proves that the feature selection method proposed has better recognition performance.

[0107] The above description shows and describes several preferred embodiments of the application, but as mentioned above, it should be understood that the application is not limited to the forms disclosed herein, and should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be modified within the scope of the application concept described herein by the above teaching or related art or knowledge. Any modification and change made by those skilled in the art without departing from the spirit and scope of the application shall be within the protection scope of the claims of the application.

Claims

1. A method for identifying passive jamming of a bistatic radar based on HRRP features, characterized in that, The method comprises the following steps: Step S1, simulating and constructing an HRRP dataset, performing a series of preprocessing operations such as normalization, alignment and determination of the range of the support area on the HRRP; Step S2, extracting a total of 18 features including peak point length, normalized variance and front-end structure ratio; Step S3, selecting features with large discrimination and low redundancy to construct a feature vector based on a Relief algorithm and a specificity coefficient feature selection algorithm; The Relief algorithm is used to evaluate the discrimination of features, to represent the ability of the features to distinguish different categories. The steps are as follows: first, a sample is randomly selected from the data set , and then samples are selected from each category of data set, wherein the samples of the same category as the sample form a set , and the samples of different categories form a set ; then, the distance of the sample is calculated to update the discrimination of each feature, as shown in the following formula: ； ； wherein, and respectively represent the maximum and minimum feature values of the feature in all samples, is the proportion of the number of samples of the category in all samples, is the number of times of extracting samples; represents the category to which the sample belongs. The specificity coefficient is used to evaluate the redundancy of the features, and the specificity coefficient of the i-th feature in the feature set is defined as: ​ ； wherein F is the number of features, is the number of classes, is the number of samples, is the value of the th feature of the th sample in the th class, is the mean value of the th feature in the th class; Step S4, dividing the constructed feature vector into a training set and a test set, and training and testing by using an SVM classifier.

2. The method of claim 1, wherein the HRRP features are determined by, The step S1 comprises the following steps: Step S1.1, performing normalization processing on the HRRP, in order to eliminate the influence of different target distances on the HRRP amplitude; Step S1.2, performing alignment processing on the HRRP to move the center of gravity of the target to the center position of the window; Step S1.3, limiting the range of distance units occupied by the target, and reducing the influence of noise outside the target unit on feature extraction.

3. The method for identifying passive interference in bistatic radar based on HRRP features as described in claim 1, characterized in that, The method for performing normalization processing on the HRRP in step S1 is as follows: Let the sequence of HRRP module values be After normalization, it can be expressed as: ； wherein denotes taking the maximum value, is the HRRP length, the normalized HRRP sequence value range is mapped to 0 to 1.

4. The method of claim 1, wherein the method further comprises: The center of gravity in step S1 The calculation method is: ； wherein, is a sequence of HRRP magnitudes, is a HRRP length.

5. The method of claim 1, wherein the method further comprises: The radial distance interval of the target position in the HRRP in step S1 is the support area, and the sequence before alignment is regarded as noise The mean value is calculated and multiplied by an adjustable bias value as a threshold, and the interval between the first and last sampling points higher than the threshold is the support area .

6. The method of claim 1, wherein the method further comprises: The 18 features extracted in step S2 are: peak point length (F1), peak point number (F2), 25% length (F3), 50% length (F4), 25% peak point number (F5), 50% peak point number (F6), 25% length ratio (F7), 50% length ratio (F8), sorted energy proportion (F9), normalized variance (F10), sparsity (F11), primary and secondary peak value ratio (F12), front-end structure ratio (F13), rear-end structure ratio (F14), symmetry (F15), dispersion (F16), descaling structure feature (F17), and fluctuation structure feature (F18).

7. The method of claim 1, wherein the method further comprises: In step S3, when the specificity coefficient When smaller, it represents the first The specificity of each feature is relatively weak, and it can be considered a redundant feature and removed. The specific method is as follows: First, sort the HRRP features from high to low according to their discriminative power using the Relief algorithm. Then, calculate the specificity coefficient value to evaluate the redundancy of each feature and remove features according to the number of features selected. After selecting a certain number of the remaining features according to the sorting, construct a feature vector.

8. The method of claim 1, wherein the method further comprises: The specific method of step S4 is as follows: The constructed feature vector is divided into a training set and a test set according to 8:2, the training set samples are input into the SVM classifier for training, after the training is completed, the test set is extracted according to the feature selection result to construct a feature vector, and is sent into the trained classifier to complete the passive interference recognition.

9. The method of claim 1, wherein the method further comprises: In step S4, the SVM is used for two-level classification: a classifier is first trained to classify "category A" and "category B+C", and then another classifier is trained to further classify all "category B+C" samples output by the previous stage into "category B" and "category C".

Images