Radar radiation source sorting method and device

Abstract

The invention relates to a radar radiation source sorting method and device, and the method comprises the steps: carrying out the matching recognition of a target radiation source pulse through a high-dimensional parameter mixed probability model corresponding to a plurality of radiation sources, and obtaining a preliminary sorting result; wherein the high-dimensional parameter mixed probability model is constructed according to multi-dimensional mixed Gaussian distribution corresponding to each CPI in each radiation source; when the preliminary sorting result contains a plurality of high-dimensional parameter mixed probability models, obtaining a decision space according to an identification result; wherein the decision space comprises a plurality of target radiation sources corresponding to the plurality of high-dimensional parameter mixed probability models; obtaining posterior probabilities of the plurality of target radiation sources in the decision space according to the Bayesian theory; and taking the target radiation source with the maximum posterior probability as a final sorting result of the target radiation source pulse. The analysis and management efficiency of the radiation source is improved by using the high-dimensional parameter mixed probability model, the influence of unknown interference pulses is reduced, and the accuracy of the obtained final sorting result is relatively high.

Description

Technical Field

[0001] The present invention relates to the field of radar technology, and in particular to a radar radiation source sorting method and device. Background Art

[0002] Radar radiation source sorting and identification refers to the analysis and processing of received radar radiation signals to achieve functions such as target feature extraction and target classification and identification. In modern multi-purpose radar systems, in order to cope with complex electromagnetic environments and diverse target characteristics, it is of great significance to accurately and quickly sort and identify different types of radiation signals. Radar radiation source sorting is mainly studied from three perspectives: inter-pulse modulation characteristics, intra-pulse modulation characteristics, and radar radiation source identification based on machine learning.

[0003] Existing radar source sorting technologies all have their own problems: (1) Template matching methods based on pulse-to-pulse modulation features are suitable for simple electromagnetic environments with a small number of early radiation sources and relatively fixed types. With the increasing richness of radar systems, the proportion of various types of modulation signals in the electromagnetic environment has gradually increased, and the template matching method has been very limited in its ability to distinguish pulses of various radiation sources. (2) Radar intra-pulse modulation features can generally be divided into intentional modulation features and unintentional modulation features according to whether the modulation method is intentional or not. The research idea of ​​source sorting based on intentional modulation features is to extract some features with high discrimination and strong sorting ability through efficient intra-pulse modulation extraction methods to complete signal sorting, but these methods have poor noise resistance and poor ability to distinguish signals of the same modulation type. Unintentional modulation features generally refer to features added to the signal introduced by non-ideal reasons inside the transmitter. Therefore, radar source signal sorting methods based on unintentional modulation features can capture inherent features of the radiation source that are difficult to modify. However, although there are many unintentional modulation features available for sorting in theory, in reality such features are usually extremely subtle and difficult to accurately extract in a low signal-to-noise ratio environment. (3) Radar emitters based on the machine learning framework can obtain regularities from sample data and use the regularities to predict unknown data. However, it is extremely unrealistic to obtain signal samples of non-cooperative emitters in advance. It will take a lot of time to label millions of samples, and the machine learning principle is not very interpretable. It is still not the preferred method for systematic use. It can be seen that in the existing technology, the efficiency of radiation source sorting in complex electromagnetic environments is low, and traditional radiation source sorting has the problems of incomplete and ambiguous sorting. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present invention provides a radar radiation source sorting method and device.

[0005] According to a first aspect of an embodiment of the present invention, a radar radiation source sorting method is provided, the method comprising:

[0006] The target radiation source pulse is matched and identified through a high-dimensional parameter mixture probability model corresponding to multiple radiation sources to obtain a preliminary sorting result; wherein the high-dimensional parameter mixture probability model is constructed according to the multi-dimensional mixed Gaussian distribution corresponding to each CPI in each radiation source;

[0007] When the preliminary sorting result includes multiple high-dimensional parameter mixed probability models, a decision space is obtained according to the recognition result; wherein the decision space includes multiple target radiation sources corresponding to the multiple high-dimensional parameter mixed probability models;

[0008] According to the Bayesian theory, the posterior probabilities of the multiple target radiation sources in the decision space are obtained;

[0009] The target radiation source with the largest posterior probability is taken as the final sorting result of the target radiation source pulses.

[0010] Optionally, after matching and identifying the target radiation source pulses through the high-dimensional parameter mixture probability model corresponding to the multiple radiation sources to obtain a preliminary sorting result, the method further includes:

[0011] When the preliminary sorting result includes a high-dimensional parameter mixture probability model, obtaining a final sorting result of the target radiation source pulse according to the radiation source corresponding to the high-dimensional parameter mixture probability model;

[0012] When the preliminary sorting result does not include a high-dimensional parameter mixed probability model, the target radiation source pulse is determined to be an unknown pulse.

[0013] Optionally, the high-dimensional parameter mixture probability model is represented by reference as follows:

[0014]

[0015] Among them, p R represents the high-dimensional parameter mixture probability model, π i represents the probability of the i-th CPI pattern appearing in the radiation source R, I is the number of CPI patterns, Represents the multidimensional mixed Gaussian distribution of the i-th CPI pattern in the radiation source R.

[0016] Optionally, the multidimensional mixed Gaussian distribution reference is expressed as follows:

[0017]

[0018] Where K represents the number of sub-distributions within the i-th CPI pattern in the radiation source R, n krepresents the probability of the kth sub-distribution occurring in the ith CPI pattern in the radiation source R, p k A high-dimensional Gaussian model representing the k-th subdistribution within the i-th CPI pattern in the radiation source R.

[0019] Optionally, the high-dimensional Gaussian model reference of the k-th sub-distribution in the i-th CPI pattern in the radiation source R is expressed as follows:

[0020]

[0021] Where m represents the number of radar parameters of the radiation source R, C i represents the variance sequence of the ith CPI pattern in the radiation source R, μ i represents the radar parameter value sequence in the i-th CPI pattern in the radiation source R, exp is an exponential function, x represents other radar parameter vectors of the target radiation source pulse, and the other radar parameter vectors are composed of radar parameters except the radar pulse repetition period, and T represents a transpose operation.

[0022] Optionally, the posterior probabilities of the multiple target radiation sources in the decision space are obtained according to the following formula:

[0023]

[0024] Among them, P(R l ∣TOA i' ) represents the lth target radiation source R l The posterior probability, TOA i' represents the pulse arrival time of the i'th pulse in the target radiation source pulse after sorting by TOA, L is the number of target radiation sources in the decision space, P(TOA i' ∣R l ) indicates that the identification result is the target radiation source R l When TOA i' Conditional probability of occurrence, P(TOA i' ∣R k' ) indicates that the identification result is the target radiation source R k' When TOA i' The conditional probability of occurrence, P(Rl) represents the target radiation source R l The corresponding prior probability, P(Rk') represents the target radiation source R k' The corresponding prior probability.

[0025] Optionally, when the identification result is a radiation source R l When TOA i' The conditional probability of occurrence is expressed as follows:

[0026]

[0027] in, Indicates that according to the target radiation source R l The likelihood value of other radar parameter vector x of the target radiation source pulse obtained by the corresponding high-dimensional parameter mixture probability model, Indicates that according to the target radiation source R l The likelihood value of the pulse repetition period of the target radiation source pulse is obtained by the corresponding high-dimensional parameter mixture probability model.

[0028] According to a second aspect of an embodiment of the present invention, a radar radiation source sorting device is provided, the device comprising:

[0029] A preliminary identification module is used to match and identify the target radiation source pulses through a high-dimensional parameter mixture probability model corresponding to multiple radiation sources to obtain a preliminary sorting result; wherein the high-dimensional parameter mixture probability model is constructed according to the multi-dimensional mixed Gaussian distribution corresponding to each CPI in each radiation source;

[0030] A decision space acquisition module, used for obtaining a decision space according to the recognition result when the preliminary sorting result includes multiple high-dimensional parameter mixed probability models; wherein the decision space includes multiple target radiation sources corresponding to the multiple high-dimensional parameter mixed probability models;

[0031] A posterior probability calculation module, used to obtain the posterior probabilities of the multiple target radiation sources in the decision space according to the Bayesian theory;

[0032] The final sorting module is used to take the target radiation source with the largest posterior probability as the final sorting result of the target radiation source pulses.

[0033] The technical solution provided by the present invention may include the following beneficial effects:

[0034] Through the above technical scheme, the high-dimensional parameter mixed probability model corresponding to each radiation source is used to improve the analysis and management efficiency of the radiation source, so as to facilitate the subsequent sorting of the radiation sources; and the prior probability and the posterior probability are linked by Bayesian theory, and the posterior probability is obtained according to the preliminary sorting result, and the mixed radiation source is supplemented with reasoning and judgment, so as to improve the recognition accuracy of the mixed radiation source; and the preliminary sorting result of the high-dimensional parameter mixed probability model is also used as a reference to reduce the influence of unknown interference pulses on the target radiation source pulses during the radiation source sorting, so that the final sorting result obtained has a higher accuracy.

[0035] Other features and advantages of the present invention will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present invention but do not constitute a limitation of the present invention. In the accompanying drawings:

[0037] Figure 1 is a flow chart of a radar radiation source sorting method according to an exemplary embodiment.

[0038] Figure 2 The figure is a schematic diagram showing changes in PRI, carrier frequency and pulse width parameters according to an exemplary embodiment.

[0039] Figure 3 It is a schematic diagram of a one-dimensional Gaussian mixture probability density function according to an exemplary embodiment.

[0040] Figure 4 It is a schematic diagram showing a preliminary sorting result of an aliased pulse after being identified by a high-dimensional parameter mixture probability model according to an exemplary embodiment.

[0041] Figure 5 is a pulse sorting accuracy comparison diagram shown according to an exemplary embodiment.

[0042] Figure 6 The figure is a block diagram of a radar radiation source sorting device according to an exemplary embodiment. DETAILED DESCRIPTION

[0043] Figure 1 is a flow chart of a radar radiation source sorting method according to an exemplary embodiment. Figure 1 As shown, the method comprises the following steps:

[0044] S1. Match and identify the target radiation source pulses through a high-dimensional parameter mixed probability model corresponding to multiple radiation sources to obtain preliminary sorting results; wherein the high-dimensional parameter mixed probability model is constructed based on the multi-dimensional mixed Gaussian distribution corresponding to each CPI in each radiation source.

[0045] It can be understood that the preliminary sorting result is the result obtained after comparing the likelihood value obtained by the high-dimensional parameter mixed probability model with the decision threshold. For example, if there are three radiation sources R1, R2, and R3, the parameters of the target radiation source pulse are substituted into the model calculation. If the likelihood value of R1 and R2 exceeds the decision threshold, it is considered that the target radiation source pulse is associated with R1 and R2, and the preliminary result is R1 and R2; if the likelihood value of R3 does not exceed the decision threshold, it is considered that the association fails, so the preliminary result does not include R3.

[0046] It can be understood that the parameters of each dimension in the parameter space are described in a formula based on the radar parameters of the radiation source and then modeled in a unified form, so as to achieve the purpose of intuitive and centralized management of the parameters of each dimension of the radiation source. The modeling of the radar parameters of the radiation source can refer to the following steps:

[0047] (1) First, by conducting an in-depth analysis of the radar parameters of the radiation source, a unified model is constructed to manage parameters of various dimensions, and each radar parameter in the parameter space is uniformly modeled as a binary group Θ.

[0048]

[0049] in, is the type space of radar parameters, M represents the specific type of radar parameters, represents the general type of radar parameter, while θ consists of the specific value and variance of the radar parameter.

[0050] (2) By uniformly modeling the radar parameters of the radiation source and comprehensively considering the relationship between the parameters in each dimension, a high-dimensional parameter mixed probability model is created to more accurately describe the relationship between different CPI (Coherent Processing Interval) styles and parameters.

[0051] Assume that a radar radiation source has Q characteristic parameters X = [x1, x2, ..., x Q ] T , then the characteristic parameter of the dth CPI style of the radiation source is The dth CPI style of the radiation source is modeled as:

[0052]

[0053] in, The model represents the Qth radar parameter of the dth CPI style of the emitter, and has:

[0054]

[0055] Among them, the parameter value sequence of the Qth radar parameter under the dth CPI style of a certain radiation source is recorded as The variance series is recorded as The number of repetitions is recorded as The parameter types are documented as express The M'th descriptor sequence in Respectively The mth subsequence in variance value, number of pulses, parameter type, etc.

[0056] In one embodiment, Figure 2 This is a schematic diagram of the change of PRI, carrier frequency, and pulse width parameters according to an exemplary embodiment. For example, it is assumed that radar R1 has two CPI styles. Carrier frequency style: The carrier frequency of CPI1 is 12000kHz, and the carrier frequency of CPI2 is 12800kHz. The carrier frequency parameter types are all fixed frequencies, and the CPI style is switched every 2500 pulses. PRI style: Under the two CPI styles, the PRI change method is the same, and the styles are all staggered re-frequency groups, with a total of 6 groups (34 / 37 / 41 / 47 / 53 / 59), which appear randomly, and the PRI residence time is 10000us. Pulse width style: Under the two CPI styles, the pulse width change method is the same, both are fixed pulse widths, such as Figure 2 .

[0057] The high-dimensional parameter mixed probability model of the first parameter carrier frequency f of radar R1 under CPI1 and CPI2 styles is expressed as and

[0058]

[0059] The high-dimensional parameter mixed probability model of the second parameter pulse width PW of radar R1 under CPI1 and CPI2 styles is expressed as and

[0060]

[0061] The high-dimensional parameter mixed probability model of the third parameter PRI of radar R1 under CPI1 and CPI2 styles is expressed as and

[0062]

[0063] Then the high-dimensional parameter mixed probability model of radar R1 is expressed as:

[0064]

[0065] Optionally, since each dimensional characteristic parameter of the radiation source is modeled as a scalar or discrete sequence, and each value has a corresponding variance. Therefore, a high-dimensional Gaussian mixture probability model is considered for modeling. The high-dimensional Gaussian mixture probability model can combine multiple Gaussian distributions into an overall model to adapt to the multidimensional characteristics of complex distributions. For the multidimensional parameters of the radiation source, each parameter can be regarded as a dimension of the Gaussian distribution, and all dimensions can be combined into a mixed model in a high-dimensional space. Figure 3 is a schematic diagram of a one-dimensional Gaussian mixture probability density function according to an exemplary embodiment, such as Figure 3 By introducing appropriate weights, the joint distribution of radiation source parameters can be established according to the weights of parameters of different dimensions.

[0066] According to the above model, the high-dimensional Gaussian model of the kth sub-distribution in the i-th CPI pattern in the radiation source R can be expressed as follows:

[0067]

[0068] Where m represents the number of radar parameters of the radiation source R, C i represents the variance sequence of the i-th CPI pattern in the radiation source R, that is, μ i represents the radar parameter value sequence in the i-th CPI style of the radiation source R, that is, exp is an exponential function, x represents other radar parameter vectors of the target radiation source pulse, the other radar parameter vectors are composed of radar parameters except the radar pulse repetition period, and T represents a transposition operation.

[0069] For example, let the three parameters of the first CPI style of the radiation source R1 be The three parameters of the second CPI style are:

[0070]

[0071] The carrier frequency and pulse width joint hybrid model of the radiation source R1 after the first and second CPI style modeling is as follows:

[0072]

[0073] Among them, the mean vector μ1=[12019.348,1], μ2=[12819.348,1], and the covariance matrix C1=C2=[3278.428,0.004].

[0074] The PRI Gaussian mixture model of the radiation source R1 after the first and second CPI style modeling is as follows:

[0075]

[0076] Where x is the PRI parameter of the radiation source, the mean sequence μ = [34, 37, 41, 47, 53, 59, 10000], and the variance sequence σ = [0.002, 0.002, 0.002, 0.002, 0.002, 0.002, 0.002, 0.002].

[0077] Then, under the three dimensions of the above three parameters, the mixed probability model of each CPI style of the radiation source R1 is

[0078]

[0079] Optionally, a multidimensional mixed Gaussian distribution can be obtained according to the high-dimensional Gaussian model of the k-th sub-distribution in the i-th CPI pattern in the radiation source R, as shown below:

[0080]

[0081] Where K represents the number of sub-distributions within the i-th CPI pattern in the radiation source R, n k represents the probability of the kth sub-distribution occurring in the ith CPI pattern in the radiation source R, p k A high-dimensional Gaussian model representing the k-th subdistribution within the i-th CPI pattern in the radiation source R.

[0082] Optionally, there are multiple CPI styles in each radiation source, and each CPI style is described by a different multidimensional mixed Gaussian distribution. Therefore, multiple multidimensional Gaussian distributions of different CPI styles constitute a high-dimensional parameter mixed probability model describing the CPI as a unit, which is represented as follows:

[0083]

[0084] Among them, p R represents a high-dimensional parameter mixture probability model, π i represents the probability of the i-th CPI pattern appearing in the radiation source R, I is the number of CPI patterns, Represents the multidimensional mixed Gaussian distribution of the i-th CPI pattern in the radiation source R.

[0085] S2. When the preliminary sorting result includes multiple high-dimensional parameter mixed probability models, a decision space is obtained according to the recognition result; wherein the decision space includes multiple target radiation sources corresponding to the multiple high-dimensional parameter mixed probability models.

[0086] It can be understood that the decision space is obtained according to the preliminary sorting results, and the decision space can be represented as follows:

[0087] Ω={R1,R2,…,R L};

[0088] Where Ω represents the decision space, R L represents the Lth target radiation source;

[0089] The pulse sequence of the received target radiation source pulses arranged according to TOA (pulse arrival time) is:

[0090]

[0091] Where T represents the pulse sequence, TOAN Represents the TOA of the Nth target radiation source pulse, where N is the total number of pulses;

[0092] S3. According to Bayesian theory, the posterior probabilities of multiple target radiation sources in the decision space are obtained.

[0093] Optionally, the posterior probabilities of multiple target radiation sources in the decision space are obtained according to the following formula:

[0094]

[0095] Among them, P(R l ∣TOA i' ) represents the lth target radiation source R l The posterior probability, TOA i' represents the pulse arrival time of the i'th pulse in the target radiation source pulse after sorting by TOA, L is the number of target radiation sources in the decision space, P(TOA i' ∣R l ) indicates that the identification result is the target radiation source R l When TOA i' Conditional probability of occurrence, P(TOA i' ∣R k' ) indicates that the identification result is the target radiation source R k' When TOA i' The conditional probability of occurrence, P(Rl) represents the target radiation source R l The corresponding prior probability, P(Rk') represents the target radiation source R k' The corresponding prior probability.

[0096] Calculate the PRI (pulse repetition period) likelihood value of each target radiation source pulse. For example, the radiation source label sequence corresponding to the first 100 pulses in the target radiation source pulse is D = (…, R3, R1, R1, R2, R1, R2), then the PRI parameter likelihood value of the 101st pulse can be expressed as:

[0097]

[0098] in, Indicates that TOA 101 -TOA 99 The likelihood value after substituting into the mixed probability density function of the PRI parameters of radar R1.

[0099] Optionally, when the identification result is the radiation source R l When TOA i' The conditional probability of occurrence is shown in the following table

[0100]

[0101] in, Indicates that according to the target radiation source R l The likelihood value of other radar parameter vector x of the target radiation source pulse obtained by the corresponding high-dimensional parameter mixture probability model, Indicates that according to the target radiation source R l The likelihood value of the pulse repetition period of the target radiation source pulse is obtained by the corresponding high-dimensional parameter mixture probability model. According to the likelihood value set obtained from the initial sorting results,

[0102]

[0103] S4. The target radiation source with the largest posterior probability is taken as the final sorting result of the target radiation source pulse.

[0104] In one embodiment, Figure 4 FIG. 1 is a schematic diagram of a preliminary sorting result of an aliased pulse after being identified by a high-dimensional parameter mixture probability model according to an exemplary embodiment. For example, when an interceptor receives a target radiation source pulse mixed with many radiation sources, due to the possible parameter aliasing phenomenon of radars R1 and R2, some pulses are matched to two radiation sources in the radar library at the same time, thereby generating D n = 0, such as Figure 4 As shown. After being processed by the high-dimensional parameter mixture probability model, D n = 0 pulses is reduced to {R1, R2}, and the preliminary sorting results of the high-dimensional parameter mixture probability model are used to assist in calculating the Likelihood value, and then get the conditional probability P(TOA i' ∣R l ).

[0105] Optionally, after S2, the method may further include:

[0106] When the preliminary sorting result includes a high-dimensional parameter mixture probability model, the final sorting result of the target radiation source pulse is obtained according to the radiation source corresponding to the high-dimensional parameter mixture probability model;

[0107] When the preliminary sorting result does not contain a high-dimensional parameter mixed probability model, the target radiation source pulse is determined to be an unknown pulse.

[0108] In one embodiment, a simulation experiment is conducted according to the present invention, and the simulation conditions are: the training data is composed of 7 known radar radiation source signal samples, and are represented by class1-7. 5,000 pulses are taken from each radar for modeling and creating a radar library, totaling 35,000 sample pulses, of which many radars have complex CPI patterns. The data samples of test set 1 are all composed of 7 radiation sources of known patterns in the radar library. The test set pulse stream has a total of 25,462 pulses, with a total duration of 0.009999966 to 6.070732365s, and an average of more than 4,200 pulses per second. The data samples of test set 2 are based on the simulation 2 test set, and the unknown pulse samples class99 of the proportion are added around the parameter values ​​of each dimension of the known radiation source pulse class1-7 according to different proportions, and are used to represent the percentage of unknown pulses.

[0109] Simulation content: The present invention is used to establish a parameterized model for 7 known radiation sources in CPI style. Based on the established parameter model, the method is used to sort and identify the aliased pulse streams in the complex electromagnetic environment simulated by test set 1 and test set 2, and the effect is compared and analyzed with machine learning methods such as KNN, NBC, SVM, and CART.

[0110] Simulation results analysis:

[0111] Table 1 shows that the present invention uses the radiation source multi-dimensional parameter modeling method to model 7 known radiation sources in CPI style. It can be seen from Table 1 that under this simulation condition, the density of signal pulses is very high, and pulses from different radiation sources and different CPI styles will overlap with each other during the duration, resulting in overlap of various parameter dimensions.

[0112] Table 1

[0113]

[0114]

[0115] Table 2 is a comparison table of the accuracy results of KNN, NBC, SVM, CART and the method of the present invention for sorting aliased pulse streams using the pulse samples of the test set 1. It can be seen from Table 2 that the sorting method of the present invention shows a high accuracy when classifying class2, class4, class5, class6 and class7, especially in the classification of class4 and class6, the accuracy is significantly higher than other classifications, which shows that the method has a high recognition accuracy for these two types of radiation sources and is relatively less likely to make misjudgments. However, in the classification of class1 and class3, the accuracy is low, indicating that the method may make misjudgments when identifying these two types of radiation source pulses. In addition, the accuracy of SVM in the classification of class1 and class3 is NaN, which shows that the SVM classifier failed to correctly identify the pulses marked as class1 or class3 in the test set, showing that it has a very poor effect on the classification of these two types of radiation sources. In addition, in the other three methods except SVM and the method of the present invention, the classification accuracy for class5 is only about 30%, showing that SVM and the method of the present invention have a lower risk of misjudgment when processing class5 than other methods.

[0116] Table 2

[0117]

[0118]

[0119] Table 3 is a comparison table of the recall rate results of KNN, NBC, SVM, CART and the method of the present invention for sorting aliased pulse streams using pulse samples from the test set 1. As can be seen from Table 3, the sorting method of the present invention exhibits a high recall rate when classifying class1, class2, class3 and class7, all exceeding 90%. In particular, in the classification of class1 and class3, the recall rate is significantly better than other classification methods, showing that the method of the present invention has an excellent recall effect on these two types of radiation sources. On the contrary, in the classification of class4, the recall rate is lower than that of other methods, indicating that the method of the present invention may have a certain degree of missed detection when identifying class4. In addition, the recall rate of SVM in the classification of class1 and class3 is 0, indicating that the SVM classifier failed to correctly identify pulses belonging to class1 or class3 in the test set. However, KNN, NBC, SVM and CART have large deviations in the recall effect of class1-7. For example, the recall rates of KNN and NBC for class2, SVM for class1 and class3, and CART for class1, class2 and class3 are all lower than 50%, and even the recall rates of some categories are less than 10%. In comparison, except for the recall rate of class4, which is about 50%, the recall rates of the method of the present invention for other categories are all over 80%, showing that it has a more balanced effect in recalling all categories.

[0120] Table 3

[0121] class KNN NBC SVM CART The present invention class1 91.52％ 99.46％ 0.00％ 6.68％ 99.88％ class2 19.50％ 33.06％ 95.56％ 19.50％ 91.20％ class3 61.67％ 96.55％ 0.00％ 20.39％ 99.17％ class4 100.00％ 97.76％ 100.00％ 94.30％ 50.91％ class5 90.88％ 83.24％ 91.83％ 90.84％ 82.34％ class6 100.00％ 100.00％ 100.00％ 100.00％ 87.30％ class7 100.00％ 100.00％ 100.00％ 100.00％ 100.00％

[0122] Table 4 is a comparison table of the accuracy results of KNN, NBC, SVM, CART and the method of the present invention for sorting aliased pulse streams using the pulse samples of test set 1. According to the data in Table 4, the accuracy of the present invention is 88.12%, close to 90%, which is higher than the accuracy of the other four methods. Except that the accuracy of the NBC method is not much different from that of the method of the present invention, the accuracy of the other three methods is less than 80%, which shows that in the complex environment of dense pulse flow, the method of the present invention has a high accuracy in sorting aliased pulses.

[0123] Table 4

[0124] class KNN NBC SVM CART The present invention Accuracy 76.77％ 84.44％ 65.95％ 54.12％ 88.12％

[0125] Table 5 is the confusion matrix after the CART method uses test set 1 for sorting and identification. Through the confusion matrix, we can intuitively measure the performance of the algorithm and measure the verification results of the predicted value and the actual value. From Table 5, it can be seen that the CART method has serious confusion when identifying class1, class2, and class3. 93% of the pulses of class1 are misclassified as class4, 80% of the pulses of class2 are misclassified as class5, and 79% of the pulses of class3 are misclassified as class6. In addition, the algorithm has a low misclassification rate for class5, and there is no confusion in the identification of class6 and class7.

[0126] Table 5

[0127]

[0128] Table 6 is the confusion matrix after the KNN method uses test set 1 for sorting and identification. The data in Table 6 shows that the KNN method has serious confusion when identifying class 2, and 80% of pulses in class 2 are misclassified as class 5. In addition, 38% of pulses in class 3 are misclassified as class 6, 8% of pulses in class 1 are misclassified as class 4, and 9% of pulses in class 5 are misclassified as class 2. There is no confusion in the identification of class 6 and class 7.

[0129] Table 6

[0130]

[0131]

[0132] Table 7 is the confusion matrix after the NBC method uses test set 1 for sorting and identification. From the data in Table 7, it can be seen that the NBC method has serious confusion when identifying class 1 and class 3. All pulses of class 1 are misclassified as class 4, and all pulses of class 3 are misclassified as class 6. In addition, class 2 and class 5 also have different degrees of confusion. 0.4% of pulses of class 2 are misclassified as class 5, and 8% of pulses of class 5 are misclassified as class 2. There is no confusion in the identification of class 4, class 6, and class 7.

[0133] Table 7

[0134]

[0135] Table 8 is the confusion matrix after the SVM method uses test set 1 for sorting and identification. The data in Table 8 shows that the SVM method has serious confusion when identifying class2, and 67% of pulses in class2 are misclassified as class5. In addition, class1, class3, class4, and class5 also have different degrees of confusion. 0.5% of pulses in class1 are misclassified as class4, 3% of pulses in class3 are misclassified as class6, 2% of pulses in class4 are misclassified as class1, and 16% of pulses in class5 are misclassified as class2. There is no confusion in the identification of class6 and class7.

[0136] Table 8

[0137]

[0138]

[0139] Table 9 is the confusion matrix after the method of the present invention uses test set 1 for sorting and identification. From the data in Table 9, it can be seen that the method of the present invention only has a relatively serious confusion when identifying class4, and 49% of the pulses of class4 are misjudged as class5. In addition, class1, class2, class3, class5, and class6 have different degrees of confusion. Most of the pulses of class1 are correctly identified, and only 0.1% of the pulses are misjudged as class4; 4% of the pulses of class2 are misjudged as class5 and unknown pulses respectively; 17% of class5 are misjudged as class2; 14% of the pulses of class6 are misjudged as class3; there is no confusion in the identification of lass7. It can be seen that the method of the present invention has the best sorting and identification effect for class1, class2, and class7, especially the misjudgment rate of class1 is much lower than other methods. Moreover, compared with other methods, the average misjudgment rate of the method of the present invention for misjudged radiation sources is also low, and no more than 50% of the pulses of each radiation source are misjudged.

[0140] Table 9

[0141]

[0142] Figure 5 is a pulse sorting accuracy comparison diagram according to an exemplary embodiment. Figure 5 Table 10 and Table 11 are respectively the accuracy table and accuracy change curve of the results of the present invention and other methods after the test set 2 containing different proportions of unknown pulses is sorted, and η is used to represent the percentage of unknown pulses. From the data in Table 10 and Figure 5It can be seen from the changing trend that with the increase of η, the number of unknown pulses in the aliased pulse stream gradually increases, and the sorting accuracy of various methods decreases significantly when dealing with the increasing number of unknown situations. Compared with other sorting methods, the present invention can still maintain a relatively high level of sorting accuracy and stability when facing different percentages of unknown pulses, which shows that the method of the present invention has strong stability and robustness when dealing with aliased pulses.

[0143] The above technical solutions and examples illustrate that the present invention combines a high-dimensional parameter mixed probability model with Bayesian theory, effectively solving the problems of low efficiency and high misjudgment rate in radar radiation source identification and sorting in a complex electromagnetic environment, and can identify incomplete pulses and unknown radiation sources through Bayesian reasoning, thereby solving the problems of incomplete and ambiguous sorting.

[0144] Figure 6 is a block diagram of a radar radiation source sorting device according to an exemplary embodiment. Figure 6 As shown, the apparatus 600 may include:

[0145] The preliminary identification module 601 is used to match and identify the target radiation source pulses through the high-dimensional parameter mixture probability model corresponding to multiple radiation sources to obtain a preliminary sorting result; wherein the high-dimensional parameter mixture probability model is constructed according to the multi-dimensional mixed Gaussian distribution corresponding to each CPI in each radiation source;

[0146] A decision space acquisition module 602 is used to obtain a decision space according to the recognition result when the preliminary sorting result includes multiple high-dimensional parameter mixed probability models; wherein the decision space includes multiple target radiation sources corresponding to the multiple high-dimensional parameter mixed probability models;

[0147] The posterior probability calculation module 603 is used to obtain the posterior probabilities of multiple target radiation sources in the decision space according to the Bayesian theory;

[0148] The final sorting module 604 is used to take the target radiation source with the largest posterior probability as the final sorting result of the target radiation source pulses.

[0149] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0150] The preferred embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, a variety of simple modifications can be made to the technical solution of the present invention, and these simple modifications all belong to the protection scope of the present invention.

[0151] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

[0152] In addition, various embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the present invention, they should also be regarded as the contents disclosed by the present invention.

Claims

1. A radar radiation source sorting method, characterized in that: The method comprises: The target radiation source pulse is matched and identified through a high-dimensional parameter mixture probability model corresponding to multiple radiation sources to obtain a preliminary sorting result; wherein the high-dimensional parameter mixture probability model is constructed according to the multi-dimensional mixed Gaussian distribution corresponding to each CPI in each radiation source; When the preliminary sorting result includes multiple high-dimensional parameter mixed probability models, a decision space is obtained according to the recognition result; wherein the decision space includes multiple target radiation sources corresponding to the multiple high-dimensional parameter mixed probability models; According to the Bayesian theory, the posterior probabilities of the multiple target radiation sources in the decision space are obtained; The target radiation source with the largest posterior probability is taken as the final sorting result of the target radiation source pulses.

2. The radar radiation source sorting method according to claim 1, characterized in that: After matching and identifying the target radiation source pulses through the high-dimensional parameter mixed probability model corresponding to the multiple radiation sources to obtain a preliminary sorting result, the method further includes: When the preliminary sorting result includes a high-dimensional parameter mixture probability model, obtaining a final sorting result of the target radiation source pulse according to the radiation source corresponding to the high-dimensional parameter mixture probability model; When the preliminary sorting result does not include a high-dimensional parameter mixed probability model, the target radiation source pulse is determined to be an unknown pulse.

3. The radar radiation source sorting method according to claim 1, characterized in that: The high-dimensional parameter mixed probability model is represented as follows: Among them, p R represents the high-dimensional parameter mixture probability model, π i represents the probability of the i-th CPI pattern appearing in the radiation source R, I is the number of CPI patterns, Represents the multidimensional mixed Gaussian distribution of the i-th CPI pattern in the radiation source R.

4. The radar radiation source sorting method according to claim 3, characterized in that: The multidimensional mixed Gaussian distribution reference is expressed as follows: Where K represents the number of sub-distributions within the i-th CPI pattern in the radiation source R, n k represents the probability of the kth sub-distribution occurring in the ith CPI pattern in the radiation source R, p k A high-dimensional Gaussian model representing the k-th subdistribution within the i-th CPI pattern in the radiation source R.

5. The radar radiation source sorting method according to claim 4, characterized in that: The high-dimensional Gaussian model of the k-th sub-distribution in the i-th CPI pattern in the radiation source R is represented as follows: Where m represents the number of radar parameters of the radiation source R, C i represents the variance sequence of the ith CPI pattern in the radiation source R, μ i represents the radar parameter value sequence in the i-th CPI pattern in the radiation source R, exp is an exponential function, x represents other radar parameter vectors of the target radiation source pulse, and the other radar parameter vectors are composed of radar parameters except the radar pulse repetition period, and T represents a transpose operation.

6. The radar radiation source sorting method according to claim 5, characterized in that: The posterior probabilities of the multiple target radiation sources in the decision space are obtained according to the following formula: Among them, P(R l ∣TOA i' ) represents the lth target radiation source R l The posterior probability, TOA i' represents the pulse arrival time of the i'th pulse in the target radiation source pulse after sorting by TOA, L is the number of target radiation sources in the decision space, P(TOA i' ∣R l ) indicates that the identification result is the target radiation source R l When TOA i' Conditional probability of occurrence, P(TOA i' ∣R k' ) indicates that the identification result is the target radiation source R k' When TOA i' The conditional probability of occurrence, P(Rl) represents the target radiation source R l The corresponding prior probability, P(Rk') represents the target radiation source R k' The corresponding prior probability.

7. The radar radiation source sorting method according to claim 6, characterized in that: The identification result is the radiation source R l When TOA i' The conditional probability of occurrence is expressed as follows: in, Indicates that according to the target radiation source R l The likelihood value of other radar parameter vectors x of the target radiation source pulse obtained by the corresponding high-dimensional parameter mixture probability model, p Rl (PRI) represents the target radiation source R l The likelihood value of the pulse repetition period of the target radiation source pulse is obtained by the corresponding high-dimensional parameter mixture probability model.

8. A radar radiation source sorting device, characterized in that: The device comprises: A preliminary identification module is used to match and identify the target radiation source pulses through a high-dimensional parameter mixture probability model corresponding to multiple radiation sources to obtain a preliminary sorting result; wherein the high-dimensional parameter mixture probability model is constructed according to the multi-dimensional mixed Gaussian distribution corresponding to each CPI in each radiation source; A decision space acquisition module, used for obtaining a decision space according to the recognition result when the preliminary sorting result includes multiple high-dimensional parameter mixed probability models; wherein the decision space includes multiple target radiation sources corresponding to the multiple high-dimensional parameter mixed probability models; A posterior probability calculation module, used to obtain the posterior probabilities of the multiple target radiation sources in the decision space according to the Bayesian theory; The final sorting module is used to take the target radiation source with the largest posterior probability as the final sorting result of the target radiation source pulses.

Images