A radar use and system joint inference method based on grey correlation hierarchical confidence

Abstract

The application discloses a radar use and system joint reasoning method based on grey correlation hierarchical confidence, first constructs a radar knowledge base containing various known radar signal parameters, then calculates the grey correlation similarity of the unknown radar signal to be measured and each record in the radar knowledge base by using a grey correlation analysis method, then fuses the similarity by a D-S evidence theory, and further obtains the accurate classification result of the radar use, then a hierarchical confidence reasoning machine is designed to accurately judge the radar system, and finally the efficient and accurate joint inference of the radar use and the radar system is realized. The method solves the problems of radar signal parameter interweaving and overlapping in a complex electromagnetic environment and low radar emitter use and system identification accuracy under a low signal-to-noise ratio condition, and improves the radar signal identification capability in the complex electromagnetic environment.

Description

Technical Field

[0001] This invention belongs to the field of radar countermeasures technology, specifically relating to a joint reasoning method for radar application and system based on gray relational hierarchical confidence. Background Technology

[0002] In recent years, the number and types of radiation sources such as radar, communication, and navigation signals in the electromagnetic space have increased dramatically, leading to an increasingly complex electromagnetic environment for electronic countermeasures systems. With the rapid development of radar technology, the models of new radar systems are unknown to our reconnaissance aircraft, making the deduction of radar purpose and system type particularly important. Radar attributes, such as purpose and system type, are crucial aspects of radar design and determine certain principles and boundaries that radar operation must follow. Radar purpose and system type deduction utilizes radar signal parameters such as arrival time, carrier frequency, pulse width, pulse repetition frequency, and angle of arrival, and is achieved through the design of specific rules. The results of purpose and system type deduction can serve as a primary basis for situation assessment and threat level determination.

[0003] Current electronic countermeasures equipment primarily uses database queries to identify the purpose and system architecture of radars. This method is effective at identifying radar targets with simple, conventional parameters, but its accuracy declines significantly with increasing parameter complexity and signal data density, and the results are heavily influenced by data quality. Neural networks, with their superior automatic feature extraction capabilities, are widely used. Many researchers construct classifiers through intelligent learning of training data and utilize these classifiers to identify relevant attributes in new radar data.

[0004] The paper "W. Zhang, H. Mou, Y. Zhao, and J. Gu, Intelligent recognition of mechanical scanned radar and electronically scanned array based on support vector machine and deep neural network, Electronic Information Warfare Technology, vol. 37, no. 2, pp. 33-37, 2022" improves the loss functions of three common deep learning networks by combining uniform distribution and cross-entropy functions to reduce the overconfidence of softmax classification. It also uses dynamic time warp (DTW) to perform preliminary screening of unlabeled data, enabling the identification of rare labeled data that is difficult to identify using conventional methods.

[0005] The paper "J. Tao and Z. Xindong, Radar emitter signal recognition method based on improved collaborative semi-supervised learning, Journal of Systems Engineering and Electronics, vol.34, no.5, pp.1182-1190, 2023" employs a deep convolutional neural network and attention mechanism, which can automatically complete feature extraction and radar scanning system recognition, and has high robustness.

[0006] The literature “L. Zhang, X. Tian, ​​S. Li, and X. Yang, 'Anti-interference recognition method of aerial infrared targets based on a spatio-temporal correlation inference network,'Infrared and Laser Engineering, vol. 51, no. 7, pp. 1-10, 2022” uses a dynamic Bayesian network to better fuse various parameters, reduce the impact of parameter overlap, and achieve the recognition of complex targets.

[0007] In summary, extensive research has demonstrated that neural networks possess excellent radar signal recognition performance. However, neural networks are essentially black boxes; that is, when using neural networks for automated intelligent analysis of radar reconnaissance intelligence, the specific classification rules cannot be seen. This hinders electronic countermeasures operators from designing and modifying the system according to actual conditions, resulting in less than ideal recognition performance in real-world applications. Summary of the Invention

[0008] To address the aforementioned technical problems, this invention provides a joint reasoning method for radar application and system based on gray relational hierarchical confidence, enabling efficient and accurate joint inference of radar application and system, thereby improving radar signal recognition capabilities in complex electromagnetic environments.

[0009] The technical solution adopted in this invention is: a joint reasoning method for radar application and system based on gray relational hierarchical confidence, the specific steps of which are as follows:

[0010] S1. Construct a radar knowledge base containing parameters of each known radar signal;

[0011] S2. Compare the similarity between the radar signal under test and each record in the radar knowledge base obtained in step S1 using the grey relational analysis method.

[0012] S3. Using the DS evidence theory, construct an evidence fusion machine to fuse the similarity obtained in step S2 to obtain the classification result of radar use;

[0013] S4. Using expert system knowledge, design a hierarchical confidence inference engine and combine it with the radar application identified in step S3 to obtain the decision result of the radar system.

[0014] S5. Combining the algorithms of steps S2, S3 and S4, construct a framework for the joint identification of radar purpose and system, and realize the joint identification of radar purpose and system.

[0015] Furthermore, step S1 is specifically as follows:

[0016] The selected radar parameters include: bandwidth (B), carrier frequency (RF), pulse width (PW), and pulse repetition frequency (PRF). Then, a radar knowledge base of known radar signal parameters is constructed, that is, for each known radar signal, its purpose, system, and typical value range of the radar parameters are marked.

[0017] Furthermore, step S2 is specifically as follows:

[0018] The parameter feature sequence of the unknown radar signal to be identified is used as the "mother sequence", i.e., X0={X0(k)|k=1,2,...,M}; the parameter feature sequences of each known radar signal in the radar knowledge base are used as the "grey relational factor sequence", i.e., X=(X1,X2,...,X N ) T .

[0019] Among them, X i ={X i (k)|k=1,2,...,M}(i=1,2,...,N) represents a total of N samples, each sample has M features, and T represents the transpose operation.

[0020] The grey relational analysis results are transformed into probability values ​​required for the DS evidence theory input. Before the grey relational analysis, each dimension of the above data is first max-min normalized to transform it into dimensionless data, i.e.:

[0021]

[0022] Among them, X i '(k) represents the normalized data feature, min i (X i (k) represents the minimum value of the k-th dimension feature parameter among N samples, and max i (X i(k) represents the maximum value of the k-th dimension feature parameter among N samples.

[0023] Mix X0 and X i The absolute difference of the k-th feature is denoted as Δ. i (k)=|X0(k)-X i (k)|, and then the grey relational coefficient is used to measure the similarity between two samples, defined as follows:

[0024]

[0025] Where ρ∈(0,1) represents the resolution coefficient, and ρ=0.5. Represents the minimum difference between the two poles. It represents the maximum difference between the two poles.

[0026] The correlation coefficients of all features of each test sequence are represented by a single value, which is called the grey relational degree γ. k (X0,X i ), abbreviated as γ 0i (k).

[0027] Each feature is assigned a corresponding weight w(k) (k = 1, 2, ..., M) according to its importance. And w(k)≥0.

[0028] The weighted grey relational degree is defined as follows:

[0029] Where M represents the total number of sample features; X0 and X... i The grey relational sequence is r = [r1, ..., r N ].

[0030] Furthermore, step S3 is specifically as follows:

[0031] Let U = {R1, R2, ..., R} K} represents the recognition framework, R i (i = 1, 2, ..., K) represents the radar uses in the knowledge base, and there are a total of K uses. The basic probability assignment function for U is defined as m:2. U →[0,1], satisfying:

[0032]

[0033] Where Φ represents the empty set, A represents a specific event, and proposition R is a non-empty subset of U; m(R) represents the basic probability assignment function, which represents the degree of support for proposition R.

[0034] The obtained grey relational degree sequence is rearranged according to the radar application, that is, the grey relational degree sequence r = [r1,...,rN The factors in [ ] are integrated according to their uses in the corresponding grey relational factor sequence, and the basic probability assignment function expression of the grey relational degree sequence is defined as follows:

[0035]

[0036] Where V represents the event that the radar signal to be measured is unknown, m j (V) represents the probability assignment function where the radar signal to be measured is unknown, m j (R i ) indicates that the radar signal under test belongs to the reference sequence X. i The probability assignment function is given by: i = 1, 2, ..., K, j = 1, 2, ..., t, where t represents the number of columns in the completed grey relational matrix, i.e., the number of pieces of evidence for each purpose.

[0037] Then, the rearranged grey relational similarity is transformed into a probability assignment function, as shown in the following expression:

[0038]

[0039] for The trust function m(R) represents the total degree of trust in proposition R, and is defined as follows:

[0040]

[0041] The conflict coefficient is defined as follows:

[0042]

[0043] By fusing basic probability assignment functions of the same purpose using the above method, we obtain the fused probability assignment function [m]. 1×2...×t (R1),m 1×2...×t (R2),…,m 1×2...×t (R K ),m 1×2...×t (V)].

[0044] Using a decision-making approach based on basic probability assignment, let... The following conditions must be met:

[0045]

[0046] If:

[0047]

[0048] R1 is the decision result, that is, the radar use classification result, where ε1 and ε2 represent preset thresholds, and ε1 = 0.2 and ε2 = 0.08.

[0049] Furthermore, step S4 is specifically as follows:

[0050] The matching degree is calculated based on the specific values ​​of the characteristic parameters of the radar signal under test and each record in the expert knowledge base.

[0051] If a radar radiation source has multiple bandwidths B, carrier frequencies RF, pulse widths PW, and pulse repetition frequencies PRF, then when matching the radar signal under test with the parameters currently recorded in the expert database, the matching degree λ(q) is calculated as follows:

[0052]

[0053] Where q = 1 to 4 represent the bandwidth B, carrier frequency RF, pulse width PW, and pulse repetition frequency PRF characteristics, respectively. a represents the number of characteristic parameters of the radar radiation source to be identified that match the characteristic parameters in the current record of the expert system database; b represents the total number of characteristic parameters in the current record of the expert database.

[0054] For radar application parameters, there are only two cases: matching and non-matching. The expression is as follows:

[0055]

[0056] "Match" means that the purpose of the radar signal under test, inferred by the classification algorithm based on feature subset filtering, is the same as the purpose in the current record of the expert system database, while "mismatch" means the opposite.

[0057] The confidence level of the corresponding feature parameter is recorded using d(q) (q = 1 to 5), and its definition expression is as follows:

[0058] d(q) = λ(q)·w(q)

[0059] Among them, weights w(q) (q=1,2,...,5) are introduced. And w(q)≥0.

[0060] The total confidence level D after fusion is calculated using the following expression:

[0061] D = 1 - π q (1-d(q))

[0062] The radar system with the highest fusion confidence level is selected as the decision result.

[0063] Furthermore, step S5 is specifically as follows:

[0064] The output of step S2 is rearranged and used as the input of step S3. The radar purpose inferred in step S3 is used as the auxiliary input of step S4 to achieve joint identification of radar purpose and system, thus obtaining the joint identification algorithm for radar purpose and system.

[0065] The beneficial effects of this invention are as follows: First, the method of this invention constructs a radar knowledge base containing various known radar signal parameters. Then, using grey relational analysis, it calculates the grey relational similarity between the unknown radar signal to be tested and each record in the radar knowledge base. Next, it fuses the similarities using DS evidence theory to obtain a precise classification result for the radar's application. Furthermore, a hierarchical confidence inference engine is designed to accurately determine the radar system type. Finally, it achieves efficient and accurate joint inference of radar application and radar system type. This invention solves the problem of low accuracy in identifying the application and system type of radar radiation sources under low signal-to-noise ratio conditions due to the interweaving and overlapping of radar signal parameters in complex electromagnetic environments, thus improving the radar signal identification capability in complex electromagnetic environments. Attached Figure Description

[0066] Figure 1 This is a flowchart of a radar application and system joint reasoning method based on grey relational hierarchical confidence according to the present invention.

[0067] Figure 2 This is a graph showing the change in application identification accuracy as a function of database size in an embodiment of the present invention.

[0068] Figure 3 This is a graph showing the change in system identification accuracy as a function of database size in an embodiment of the present invention.

[0069] Figure 4 This is a graph showing the variation of the accuracy of application identification with the range of data fluctuation in an embodiment of the present invention.

[0070] Figure 5 This is a graph showing the variation of system identification accuracy with data fluctuation range in an embodiment of the present invention. Detailed Implementation

[0071] The method of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0072] like Figure 1 The flowchart of a radar application and system joint reasoning method based on grey relational hierarchical confidence according to the present invention is shown below. The specific steps are as follows:

[0073] S1. Construct a radar knowledge base containing parameters of each known radar signal;

[0074] S2. Compare the similarity between the radar signal under test and each record in the radar knowledge base obtained in step S1 using the grey relational analysis method.

[0075] S3. Using the DS evidence theory, construct an evidence fusion machine to fuse the similarity obtained in step S2 to obtain the classification result of radar use;

[0076] S4. Using expert system knowledge, design a hierarchical confidence inference engine and combine it with the radar application identified in step S3 to obtain the decision result of the radar system.

[0077] S5. Combining the algorithms of steps S2, S3 and S4, construct a framework for the joint identification of radar purpose and system, and realize the joint identification of radar purpose and system.

[0078] In this embodiment, step S1 is specifically as follows:

[0079] The selected radar parameters include: bandwidth B, carrier frequency RF, pulse width PW, and pulse repetition frequency PRF. Then, a radar knowledge base of known radar signal parameters is constructed, that is, for each known radar signal, its purpose, system, and typical value range of the radar parameters are marked.

[0080] In this embodiment, step S2 is specifically as follows:

[0081] The parameter feature sequence of the unknown radar signal to be identified is used as the "mother sequence", i.e., X0={X0(k)|k=1,2,...,M}; the parameter feature sequences of each known radar signal in the radar knowledge base are used as the "grey relational factor sequence", i.e., X=(X1,X2,...,X N ) T .

[0082] Among them, X i ={X i (k)|k=1,2,...,M}(i=1,2,...,N) represents a total of N samples, each sample has M features, and T represents the transpose operation.

[0083] The grey relational analysis results are transformed into probability values ​​required for the DS evidence theory input. Before the grey relational analysis, each dimension of the above data is first max-min normalized to transform it into dimensionless data, i.e.:

[0084]

[0085] Among them, X i '(k) represents the normalized data feature, min i (X i (k) represents the minimum value of the k-th dimension feature parameter among N samples, and max i (X i (k) represents the maximum value of the k-th dimension feature parameter among N samples.

[0086] Mix X0 and X i The absolute difference of the k-th feature is denoted as Δ. i (k)=|X0(k)-X i (k)|, and then the grey relational coefficient is used to measure the similarity between two samples, defined as follows:

[0087]

[0088] Where ρ∈(0,1) represents the resolution coefficient. The smaller ρ is, the stronger the resolution. In this embodiment, a fixed value of ρ = 0.5 is used. This represents the minimum difference between two levels. This represents the maximum difference between the two levels. The larger the grey relational coefficient, the stronger X becomes. i The more similar the data of the two sequences (k) and X0(k), the better.

[0089] It can be seen that the grey relational coefficient calculates the correlation values ​​between each test sequence and the reference sequence on each dimension of features. The resulting data is numerous, leading to overly scattered information and making comparison difficult. Therefore, it is necessary to concentrate the correlation coefficients of all features of each test sequence into a single value, which is the grey relational coefficient γ. k (X0,X i ), abbreviated as γ 0i (k).

[0090] Each feature has a different level of importance for radar application identification and classification. Each feature is assigned a corresponding weight w(k) (k = 1, 2, ..., M) according to its importance. And w(k)≥0.

[0091] The weighted grey relational degree is defined as follows:

[0092] Where M represents the total number of sample features; X0 and X... i The grey relational sequence is r = [r1, ..., r N ].

[0093] In this embodiment, step S3 is specifically as follows:

[0094] Let U = {R1, R2, ..., R} K} represents the recognition framework, R i (i = 1, 2, ..., K) represents the radar uses in the knowledge base, and there are a total of K uses. The basic probability assignment function for U is defined as m:2. U →[0,1], satisfying:

[0095]

[0096] Where Φ represents the empty set, A represents a specific event, and proposition R is a non-empty subset of U; m(R) represents the basic probability assignment function, which represents the degree of support for proposition R.

[0097] The obtained grey relational degree sequence is rearranged according to the radar application, that is, the grey relational degree sequence r = [r1, ..., r N The factors in (N represents the number of sample entries in the knowledge base) are integrated according to their uses in the corresponding grey relational factor sequence. The rearrangement results in this embodiment are shown in Table 1.

[0098] Table 1

[0099] Application 2 <![CDATA[r4 r5]]> ... ... Application K <![CDATA[r N-2 r N-1 r N ]]>

[0100] Then, the basic probability assignment function expression for the grey relational degree sequence is defined as follows:

[0101]

[0102] Where V represents the event that the radar signal to be measured is unknown, m j (V) represents the probability assignment function where the radar signal to be measured is unknown, m j (R i ) indicates that the radar signal under test belongs to the reference sequence X. i The probability assignment function is given by i = 1, 2, ..., K, j = 1, 2, ..., t, where t represents the number of columns in the completed grey relational matrix, i.e., the number of pieces of evidence for each purpose.

[0103] Then, the rearranged grey relational similarity is transformed into a probability assignment function, as shown in the following expression:

[0104]

[0105] In the process of multi-evidence information fusion, the degree of support for the hypothesis provided by different samples in the database varies, and therefore the basic probability assignments for each piece of evidence also differ. If m1,…m t It is the basic probability assignment function of t independent evidences on the same identification framework U. Using Dempster's combination rule, the total trust function of the evidence after fusion processing can be calculated, and then the trust function and likelihood function of each proposition after fusion processing can be calculated. Based on certain decision rules, the hypothesis with the highest support is selected as the final radar application reasoning result.

[0106] for The trust function m(R) represents the total degree of trust in proposition R, and is defined as follows:

[0107]

[0108] The conflict coefficient is defined as follows:

[0109]

[0110] By fusing basic probability assignment functions of the same purpose using the above method, we obtain the fused probability assignment function [m]. 1×2...×t (R1),m 1×2...×t (R2),…,m 1×2...×t (R K ),m 1×2...×t (V)].

[0111] How to make decisions after combining evidence using evidence theory is a problem closely related to specific applications. This embodiment adopts decision-making based on basic probability assignment. Let... The following conditions must be met:

[0112]

[0113] If:

[0114]

[0115] R1 is the decision result, that is, the radar application classification result, where ε1 and ε2 represent preset thresholds. In this embodiment, they are fixed values: ε1 = 0.2 and ε2 = 0.08.

[0116] In this embodiment, step S4 is specifically as follows:

[0117] Hierarchical confidence-based radar system inference involves calculating the confidence level based on the parameters of the radar signal under test, combined with the inference results of the radar's intended use, using expert system methods to infer the most likely radar system.

[0118] The matching degree is calculated based on the specific values ​​of the characteristic parameters of the radar signal under test and each record in the expert knowledge base. The bandwidth B, carrier frequency RF, pulse width PW, and pulse repetition frequency PRF differ from the way they are expressed for different radar applications; therefore, different methods are required for calculating the matching degree.

[0119] If a radar radiation source has multiple bandwidths B, carrier frequencies RF, pulse widths PW, and pulse repetition frequencies PRF, then when matching the radar signal under test with the parameters currently recorded in the expert database, the matching degree λ(q) is calculated as follows:

[0120]

[0121] Where q = 1 to 4 represent the bandwidth B, carrier frequency RF, pulse width PW, and pulse repetition frequency PRF characteristics, respectively. a represents the number of characteristic parameters of the radar radiation source to be identified that match the characteristic parameters in the current record of the expert system database; b represents the total number of characteristic parameters in the current record of the expert database.

[0122] For radar application parameters, there are only two cases: matching and non-matching. The expression is as follows:

[0123]

[0124] "Match" means that the purpose of the radar signal under test, inferred by the classification algorithm based on feature subset filtering, is the same as the purpose in the current record of the expert system database, while "mismatch" means the opposite.

[0125] The confidence level of the corresponding feature parameter is recorded using d(q) (q = 1 to 5), and its definition expression is as follows:

[0126] d(q) = λ(q)·w(q)

[0127] Among them, weights w(q) (q=1,2,...,5) are introduced. And w(q)≥0.

[0128] The total confidence level D after fusion is calculated using the following expression:

[0129] D = 1 - π q (1-d(q))

[0130] The radar system with the highest fusion confidence level is selected as the decision result.

[0131] In this embodiment, step S5 is specifically as follows:

[0132] The output of step S2 is rearranged and used as the input of step S3. The radar purpose inferred in step S3 is used as the auxiliary input of step S4 to achieve joint identification of radar purpose and system, thus obtaining the joint identification algorithm of radar purpose and system.

[0133] In this embodiment, the method of the present invention further includes: performing a recognition performance test, as detailed below:

[0134] Considering the different parameters of various radars for different purposes and systems, this embodiment simulates and generates 50 different radar signals, combining 20 radar purposes and 7 radar systems. Each radar signal has one purpose and system. Based on the typical waveform parameters corresponding to different radar purposes and the survey results, the parameter settings for different radar purposes are shown in Table 2, and the parameter settings for different radar systems are shown in Table 3. Random errors are introduced into each radar signal to generate 4 / 8 / 12 / 15 samples respectively, thus constructing databases of four sizes: 200 / 400 / 600 / 750 for the above 50 radar signals.

[0135] Table 2

[0136]

[0137] Table 3

[0138]

[0139] To verify the feasibility of the method of this invention, simulations of radar application and system identification were performed on datasets of different sizes using the algorithm proposed in this invention, the Support Vector Machine (SVM) algorithm, and the Convolutional Neural Network (CONV). The experimental results are as follows: Figure 2 and Figure 3 As shown, the accuracy of the three methods decreases with the reduction of data size, while the accuracy of the method proposed in this invention decreases more slowly. Even when the data size is only 200, the accuracy of the method proposed in this invention can still be maintained above 90%, while the accuracy of the neural network is less than 40% at this time. This demonstrates that the method of this invention outperforms support vector machines and convolutional neural networks in recognizing smaller sample sizes. This is because the method of this invention does not require a large amount of data for structural training, thus ensuring high recognition performance even on a smaller scale. Furthermore, for the method of this invention, changes in dataset size do not affect the algorithm itself or its internal parameters; only the size of the gray relational matrix changes, without affecting the similarity comparison calculation. Therefore, the accuracy does not increase dramatically with increasing dataset size, nor does it decrease significantly with decreasing dataset size. Simultaneously, since there are fewer radar system types compared to application types, the accuracy of support vector machines and convolutional neural networks decreases significantly faster when directly inferring the system from parameters compared to the application type. However, due to the two-stage inference mechanism employed in the method of this invention, when the application inference result has a high accuracy, the application can be used as a parameter to assist in inferring the system, thereby improving the accuracy of system inference.

[0140] To verify the robustness of the method of this invention under different parameter fluctuation ranges, a certain degree of error was added to the parameters of each radar signal sample. After adding different degrees of error jitter to the sample parameters with a data scale of 600, the accuracy rates of the three methods for inferring application and system type are as follows: Figure 4 and Figure 5 As shown, when the fluctuation range of sample parameters in the dataset reaches 50% of the nominal value, the accuracy of the system and purpose inference method of this invention can still be above 85%, while the inference accuracy of the convolutional neural network method is less than 70%. This shows that the method of this invention has better recognition performance than support vector machines and convolutional neural networks when the sample parameters fluctuate greatly. This is because the method of this invention first extracts feature values ​​from the sample parameters and then performs grey relational analysis on the feature values, thereby mitigating the impact of parameter fluctuations. Furthermore, the method of this invention treats the parameters as a sequence as a whole rather than separate elements, while support vector machines and convolutional neural network methods treat the parameters as elements, making them more sensitive to parameter fluctuations. Comparing the changes in the accuracy of system and purpose inference, it can be seen that when the fluctuation increases, the accuracy of system inference decreases faster. This is because when the parameter fluctuation increases, it not only affects the accuracy analysis of the expert system, but also the accuracy of grey relational analysis, and indirectly introduces errors into the system inference through the purpose inference results.

[0141] In summary, the method of this invention obtains the grey relational matrix between the features of the radar signal under test and the features of each record in the radar knowledge base through grey relational analysis. Then, the importance of each feature is evaluated to obtain feature weights, and a combined grey relational similarity is obtained through weighted fusion. Next, the combined grey relational similarity is fused using DS evidence theory, and a classification result for radar application is obtained using a basic probability assignment function. Finally, the radar application classification result is added as a new feature to the input, and a hierarchical confidence inference engine is designed. By deriving the confidence of multi-feature fusion, the decision result of radar system is obtained. Experimental results in this embodiment show that, even with a small dataset and large parameter fluctuations, the method of this invention has superior performance in radar application and system identification.

[0142] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Various modifications and variations can be made to the invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the scope of the claims of the invention.

Claims

1. A joint reasoning method for radar application and system based on grey relational hierarchical confidence, the specific steps of which are as follows: S1. Construct a radar knowledge base containing parameters of each known radar signal; S2. Compare the similarity between the radar signal under test and each record in the radar knowledge base obtained in step S1 using the grey relational analysis method. S3. Using the DS evidence theory, construct an evidence fusion machine to fuse the similarity obtained in step S2 to obtain the classification result of radar use; S4. Using expert system knowledge, design a hierarchical confidence inference engine and combine it with the radar application identified in step S3 to obtain the decision result of the radar system. Step S4 is as follows: The matching degree is calculated based on the specific values ​​of the characteristic parameters of the radar signal under test and each record in the expert knowledge base. If a radar radiation source has multiple bandwidths (B), carrier frequencies (RF), pulse widths (PW), and pulse repetition frequencies (PRF), then when the radar signal under test is matched with the parameters currently recorded in the expert database, the matching degree is... The calculation expression is as follows: ； in, These represent the characteristics of bandwidth B, carrier frequency RF, pulse width PW, and pulse repetition frequency PRF, respectively; a represents the number of characteristic parameters of the radar radiation source to be identified that match the characteristic parameters in the current record of the expert system database; b represents the total number of characteristic parameters in the current record of the expert database. For radar application parameters, there are only two cases: matching and non-matching. The expression is as follows: ； Here, "match" means that the purpose of the radar signal under test inferred by the classification algorithm based on feature subset filtering is the same as the purpose in the current record of the expert system database, while "mismatch" means the opposite. use The confidence level of the corresponding feature parameter is recorded, and its definition expression is as follows: ； Among them, weights are introduced. and ; Then the total confidence level after fusion Calculated using the following expression: ； The radar system with the highest fusion confidence level is selected as the decision result; S5. Combining the algorithms of steps S2, S3 and S4, construct a framework for the joint identification of radar purpose and system, and realize the joint identification of radar purpose and system.

2. The radar application and system joint reasoning method based on grey relational hierarchical confidence as described in claim 1, characterized in that, The specific steps of S1 are as follows: The selected radar parameters include: bandwidth (B), carrier frequency (RF), pulse width (PW), and pulse repetition frequency (PRF). Then, a radar knowledge base of known radar signal parameters is constructed, that is, for each known radar signal, its purpose, system, and typical value range of the radar parameters are marked.

3. The radar application and system joint reasoning method based on grey relational hierarchical confidence as described in claim 1, characterized in that, Step S2 is as follows: The sequence of unknown radar signal parameters to be identified is used as the "mother sequence", that is... The parameter feature sequences of each known radar signal in the radar knowledge base are used as "grey relational factor sequences", that is... ; in, This indicates that there are N samples in total, each sample has M features, and T represents the transpose operation; The grey relational analysis results are transformed into probability values ​​required for the DS evidence theory input. Before the grey relational analysis, each dimension of the above data is first max-min normalized to transform it into dimensionless data, i.e.: ； in, This represents the characteristics of the normalized data. Let represent the minimum value of the k-th dimension feature parameter among N samples. This represents the maximum value of the k-th dimension feature parameter among N samples; Will and The absolute difference of the kth feature is denoted as Then, the grey relational coefficient is used to measure the similarity between two samples, and the expression is defined as follows: ； in, Represents the resolution coefficient, and ; Represents the minimum difference between the two poles. Indicates the maximum difference between the two poles; The correlation coefficients of all features of each test sequence are represented by a single value, which is called the grey relational degree. abbreviated as ; Each feature is assigned a corresponding weight based on its importance. and ; The weighted grey relational degree is defined as follows: ; Where M represents the total number of sample features; definition and The grey relational sequence is .

4. The method for joint reasoning of radar application and system based on grey relational hierarchical confidence as described in claim 1, characterized in that, Step S3 is as follows: set up Represents the recognition framework. This represents the radar uses in the knowledge base, and there are K uses in total; define... The basic probability assignment function is ,satisfy: ； in, Let A represent the empty set, and let A represent a specific event. The proposition... for The non-empty subset; This represents the basic probability assignment function, which represents the probability assignment of a proposition. The level of support; The obtained grey relational sequence is rearranged according to the radar application, that is, the grey relational sequence is... The factors in the data are integrated according to their uses in the corresponding grey relational factor sequence, and the basic probability assignment function expression of the grey relational degree sequence is defined as follows: ； Here, V represents the event that the radar signal to be measured is unknown. This represents the probability assignment function when the radar signal to be measured is unknown. This indicates that the radar signal under test belongs to the reference sequence. The probability assignment function; , ,and This indicates the number of columns in the completed grey relational matrix, i.e., the number of pieces of evidence for each purpose; Then, the rearranged grey relational similarity is transformed into a probability assignment function, as shown in the following expression: ； for Trust function Indicates the proposition The overall level of trust is defined by the following expression: ； The conflict coefficient is defined as follows: ； By fusing basic probability assignment functions of the same purpose using the above method, we obtain a fused probability assignment function. ; Using a decision-making approach based on basic probability assignment, let... The following conditions must be met: ； If: ； but The judgment result, that is, the classification result of radar use, is as follows: This indicates a preset threshold, and .

5. The radar application and system joint reasoning method based on grey relational hierarchical confidence as described in claim 1, characterized in that, Step S5 is as follows: The output of step S2 is rearranged and used as the input of step S3. The radar purpose inferred in step S3 is used as the auxiliary input of step S4 to achieve joint identification of radar purpose and system, thus obtaining the joint identification algorithm of radar purpose and system.

Images