阅读说明：本技术 一种区分不同脉间参差雷达数据的联合复杂度值计算方法 (Combined complexity value calculation method for distinguishing different pulse spread radar data ) 是由 余志斌 杜秋 王学文 徐晶 张译方 于 2021-08-25 设计创作，主要内容包括：本发明公开了一种区分不同脉间参差雷达数据的联合复杂度值计算方法,包括以下步骤：S1、获取原始脉间参差雷达脉冲数据,并基于原始脉间参差雷达脉冲数据构建脉冲重复间隔PRI数据；S2、对脉冲重复间隔PRI数据进行重构处理,并基于重构向量构建符号序列；S3、对重构向量计算差异等级值,并将差异等级值加入符号序列,得到编码序列；S4、根据编码序列,计算雷达脉冲数据的香农熵和雷达脉冲数据的编码类型复杂度,得到用于区分不同脉间参差雷达数据的联合复杂度值；本发明解决了现有技术提取原始脉间参差雷达脉冲数据的特征不突出,使得分类不准确的问题。(The invention discloses a combined complexity value calculation method for distinguishing different pulse spread radar data, which comprises the following steps of: s1, acquiring original pulse difference radar pulse data, and constructing Pulse Repetition Interval (PRI) data based on the original pulse difference radar pulse data; s2, reconstructing the pulse repetition interval PRI data, and constructing a symbol sequence based on the reconstructed vector; s3, calculating difference grade values of the reconstructed vectors, and adding the difference grade values into the symbol sequence to obtain a coding sequence; s4, calculating the Shannon entropy of the radar pulse data and the encoding type complexity of the radar pulse data according to the encoding sequence to obtain a combined complexity value for distinguishing different pulse spread radar data; the method solves the problem that the classification is inaccurate because the characteristics of the original pulse difference radar pulse data extracted by the prior art are not prominent.)

1. A joint complexity value calculation method for distinguishing different pulse spread radar data is characterized by comprising the following steps:

s1, acquiring original pulse difference radar pulse data, and constructing Pulse Repetition Interval (PRI) data based on the original pulse difference radar pulse data;

s2, reconstructing the pulse repetition interval PRI data, and constructing a symbol sequence based on the reconstructed vector;

s3, calculating difference grade values of the reconstructed vectors, and adding the difference grade values into the symbol sequence to obtain a coding sequence;

and S4, calculating the Shannon entropy of the radar pulse data and the encoding type complexity of the radar pulse data according to the encoding sequence to obtain a combined complexity value for distinguishing the different pulse spread radar data.

2. The method for calculating a joint complexity value for differentiating spread radar data between different pulses as claimed in claim 1, wherein said step S1 comprises the following substeps:

s11, acquiring original pulse difference radar pulse data;

s12, constructing radar pulse data containing pulse deletion and radar pulse data containing false pulses according to the original pulse-to-pulse spread radar pulse data;

s13, counting the arrival time TOA of each pulse in three types of data, namely original pulse staggered radar pulse data, radar pulse data containing pulse missing and radar pulse data containing false pulses;

s14, constructing a time of arrival (TOA) sequence according to the TOA of each pulse;

and S15, carrying out backward difference operation on the TOA sequence of the arrival time to obtain the PRI data of the pulse repetition interval.

3. The method for calculating a joint complexity value for differentiating spread radar data between different pulses as claimed in claim 1, wherein said step S2 comprises the following substeps:

s21, pulse repetition interval PRI dataSequentially dividing the data into a plurality of subsequences with equal length X (i), wherein N is the number of the subsequences obtained by division, X is pulse repetition interval PRI data, and X (i) is the ith subsequence of the pulse repetition interval PRI data;

s22, rearranging each subsequence X (i) according to the ascending order of elements to obtain a reconstructed vector;

and S23, mapping each subsequence X (i) into a symbol sequence S (i) according to the sequence relation of elements in the reconstructed vector.

4. The method according to claim 3, wherein in the mapping process of step S23, when the same element exists in different positions in the subsequence X (i), its sign is mapped to the position index of the first occurrence of the element in the subsequence X (i).

5. The method for calculating a joint complexity value for differentiating spread radar data between different pulses as claimed in claim 1, wherein said step S3 comprises the following substeps:

s31, calculating the standard deviation, median and mean of elements of the reconstructed vector;

s32, calculating the difference degree according to the standard deviation, the median and the mean value;

s33, carrying out normalization processing on the difference degree, and mapping the normalized difference degree into a difference grade value;

and S34, adding the difference grade value into the symbol sequence to obtain a coding sequence.

6. The method for calculating the joint complexity value according to claim 5, wherein the formula for calculating the difference degree in step S32 is as follows:

wherein D is_{var iation}To reconstruct the disparity of the vectors, Seq_stdTo reconstruct the standard deviation of the vector, Seq_meidianTo reconstruct the median of the vector, Seq_meanIs the mean of the reconstructed vectors.

7. The method for calculating a joint complexity value for differentiating spread radar data between different pulses as claimed in claim 1, wherein said step S4 comprises the following substeps:

s41, initializing the set of S and Q strings to null, initializing j to 1, counting C_NInitialization is 0;

s42, placing the jth element in the coding sequence in a Q character string set;

s43, splicing the current Q character string set and the S character string set to obtain an SQ character string set, deleting the last element of the SQ character string set to obtain an SQ' character string set;

s44, judging whether the current Q character string set is the sub character string of the SQ' character string set, if not, C_NAdding 1 by self, updating the current S character string set into the current SQ character string set, emptying the current Q character string set, skipping to the step S42, if yes, directly skipping to the step S42 until all elements in the coding sequence are placed in the Q character string set, and obtaining the current counting quantity C_N；

S45, counting the quantity C according to the current time_NAnd the length of the coding sequence, and calculating the complexity of each coding type in the coding sequence;

s46, calculating the complexity of each coding type in all the coding sequences according to the steps S41-S45;

s47, calculating the encoding type complexity of the radar pulse data according to the encoding type complexity of each encoding type in all the encoding sequences;

s48, calculating the Shannon entropy of the radar pulse data according to the coding sequence;

and S49, adding the Shannon entropy of the radar pulse data and the encoding type complexity of the radar pulse data to obtain a combined complexity value for distinguishing the staggered radar data among different pulses.

8. The method according to claim 7, wherein the formula for calculating the complexity of each code type in the code sequence in step S45 is:

wherein, C_normComplexity of a coding type, C_NFor the counting quantity, alpha is the maximum possible value of the element type in the coding type, and n is the length of the coding sequence.

9. The method according to claim 7, wherein the formula for calculating the encoding type complexity of the radar pulse data in step S47 is as follows:

LZ_symbol＝Vector_{encoding_LZ}·Vector_{encoding_weight}

wherein, LZ_symbolFor coding type complexity of radar pulse data, Vector_{encoding_LZ}Vector, formed for each coding type complexity in all coding sequences_{encoding_weight}A vector of weights for all different coding types.

Technical Field

The invention relates to the technical field of radar countermeasure, in particular to a combined complexity value calculation method for distinguishing spread radar data among different pulses.

Background

Radar signal classification plays an important role in electronic safeguards and electronic intelligence. Classification and identification of radar signals facilitates acquisition of target radar type, carrier, purpose and threat level, thereby influencing further combat decisions. In a radar countermeasure environment, the inter-pulse spread radar signals are usually used for anti-reconnaissance tasks and have great strategic significance for correct classification, and an effective feature extraction method is established as a key step for the correct classification.

The existing method needs a large amount of manual participation and a large amount of original radar pulse data, the relevance and the dependency of the extracted radar signal characteristics on the original data are strong, the method is mainly used for radar pulse signals with characteristics of constancy, jitter and the like, and the method is an effective characteristic establishing method for radar pulse signals with staggered characteristics.

Disclosure of Invention

Aiming at the defects in the prior art, the combined complexity value calculation method for distinguishing the different pulse difference radar data solves the problem that the classification is inaccurate because the characteristics of the original pulse difference radar pulse data extracted in the prior art are not prominent.

In order to achieve the purpose of the invention, the invention adopts the technical scheme that: a combined complexity value calculation method for distinguishing different pulse spread radar data comprises the following steps:

s1, acquiring original pulse difference radar pulse data, and constructing Pulse Repetition Interval (PRI) data based on the original pulse difference radar pulse data;

s2, reconstructing the pulse repetition interval PRI data, and constructing a symbol sequence based on the reconstructed vector;

s3, calculating difference grade values of the reconstructed vectors, and adding the difference grade values into the symbol sequence to obtain a coding sequence;

and S4, calculating the Shannon entropy of the radar pulse data and the encoding type complexity of the radar pulse data according to the encoding sequence to obtain a combined complexity value for distinguishing the different pulse spread radar data.

The invention has the beneficial effects that: according to the method, the Shannon entropy of the radar pulse data and the coding type complexity of the radar pulse data are calculated to obtain the combined complexity value for distinguishing the different pulse-width spread radar data, the original complex original pulse-width spread radar pulse data are converted into the combined complexity value capable of representing the complexity degree and the characteristics of the original pulse-width spread radar pulse data, the characteristics of the original radar signals are represented through the combined complexity value, the radar signal classification is not based on a large amount of original radar data, the complexity and the workload of feature extraction in the radar signal classification process are reduced, the combined complexity value obtained by combining the Shannon entropy of the radar pulse data and the coding type complexity of the radar pulse data can effectively reflect the difference of the different pulse-width spread radar signals, and therefore the high accuracy of signal classification is achieved.

Further, the step S1 includes the following sub-steps:

s11, acquiring original pulse difference radar pulse data;

s12, constructing radar pulse data containing pulse deletion and radar pulse data containing false pulses according to the original pulse-to-pulse spread radar pulse data;

s13, counting the arrival time TOA of each pulse in three types of data, namely original pulse staggered radar pulse data, radar pulse data containing pulse missing and radar pulse data containing false pulses;

s14, constructing a time of arrival (TOA) sequence according to the TOA of each pulse;

and S15, carrying out backward difference operation on the TOA sequence of the arrival time to obtain the PRI data of the pulse repetition interval.

The beneficial effects of the above further scheme are: the pulse repetition interval PRI data obtained by preprocessing the original pulse spread radar pulse can well reflect the characteristics of the pulse spread radar pulse signal, and the extracted characteristics are more suitable for classifying the pulse spread radar pulse signal by extracting the characteristics of the PRI data.

Further, the step S2 includes the following sub-steps:

s21, pulse repetition interval PRI dataSequentially dividing the data into a plurality of subsequences with equal length X (i), wherein N is the number of the subsequences obtained by division, X is pulse repetition interval PRI data, and X (i) is the ith subsequence of the pulse repetition interval PRI data;

s22, rearranging each subsequence X (i) according to the ascending order of elements to obtain a reconstructed vector;

and S23, mapping each subsequence X (i) into a symbol sequence S (i) according to the sequence relation of elements in the reconstructed vector.

The beneficial effects of the above further scheme are: the pulse repetition interval PRI data X is cut and divided, which is a key step of data compression, and the divided sequences are rearranged according to the magnitude relation of element values, so that the conversion from a data sequence to a symbol sequence is realized. The process not only greatly reduces the complexity of the data, but also keeps the change rule and characteristics inside the data.

Further, in the mapping process of step S23, when there is a same element in different positions in the sub-sequence x (i), its sign is mapped to the position index of the first occurrence of the element in the sub-sequence x (i).

The beneficial effects of the above further scheme are: the conversion from data to symbols is realized, the condition that the same elements are coded into different symbols only depending on the magnitude of the sequence relation is avoided, the coding type types are enriched, and the capability of representing the characteristics of the data is further improved.

Further, the step S3 includes the following sub-steps:

s31, calculating the standard deviation, median and mean of elements of the reconstructed vector;

s32, calculating the difference degree according to the standard deviation, the median and the mean value;

s33, carrying out normalization processing on the difference degree, and mapping the normalized difference degree into a difference grade value;

and S34, adding the difference grade value into the symbol sequence to obtain a coding sequence.

The beneficial effects of the above further scheme are: the addition of the difference grade value enables the coding types to be richer, enables the coding sequence not only to comprise the size relation of the data but also to comprise the change condition of the data, and improves the data representation capability of the coding sequence.

Further, the formula for calculating the difference degree in step S32 is as follows:

wherein D is_variationTo reconstruct the disparity of the vectors, Seq_stdTo reconstruct the standard deviation of the vector, Seq_meidianTo reconstruct the median of the vector, Seq_meanIs the mean of the reconstructed vectors.

The beneficial effects of the above further scheme are: the influence of the mean value and the median on the data is comprehensively considered, so that the calculation of the difference degree is more accurate.

Further, the step S4 includes the following sub-steps:

s41, initializing the set of S and Q strings to null, initializing j to 1, counting C_NInitialization is 0;

s42, placing the jth element in the coding sequence in a Q character string set;

s43, splicing the current Q character string set and the S character string set to obtain an SQ character string set, deleting the last element of the SQ character string set to obtain an SQ' character string set;

s44, judging whether the current Q character string set is the sub character string of the SQ' character string set, if not, C_NAdding 1 by self, updating the current S character string set into the current SQ character string set, emptying the current Q character string set, skipping to the step S42, if yes, directly skipping to the step S42 until all elements in the coding sequence are placed in the Q character string set, and obtaining the current counting quantity C_N；

S45, counting the quantity C according to the current time_NAnd the length of the coding sequence, and calculating the complexity of each coding type in the coding sequence;

s46, calculating the complexity of each coding type in all the coding sequences according to the steps S41-S45;

s47, calculating the encoding type complexity of the radar pulse data according to the encoding type complexity of each encoding type in all the encoding sequences;

s48, calculating the Shannon entropy of the radar pulse data according to the coding sequence;

and S49, adding the Shannon entropy of the radar pulse data and the encoding type complexity of the radar pulse data to obtain a combined complexity value for distinguishing the staggered radar data among different pulses.

Further, the formula for calculating the complexity of each coding type in the coding sequence in step S45 is as follows:

wherein, C_normComplexity of a coding type, C_NFor the counting quantity, alpha is the maximum possible value of the element type in the coding type, and n is the length of the coding sequence.

The beneficial effects of the above further scheme are: the complexity measure for all elements in the coding sequence is achieved by the complexity of the coding type.

Further, the formula for calculating the encoding type complexity of the radar pulse data in step S47 is as follows:

LZ_symbol＝Vector_{encoding_LZ}·Vector_{encoding_weight}

wherein, LZ_symbolFor coding type complexity of radar pulse data, Vector_{encoding_LZ}Vector, formed for each coding type complexity in all coding sequences_{encoding_weight}A vector of weights for all different coding types.

The beneficial effects of the above further scheme are: the complexity of the coding type effectively represents the complexity of the inside of the coding sequence, the complexity of the coding type of the radar pulse data after coding weighting also takes into account the complexity of all the coding types, and the Shannon entropy and the complexity of the coding weighting jointly realize the multi-angle representation of the data, so that the data characteristics are more comprehensively reflected.

Drawings

FIG. 1 is a flow chart of a joint complexity value calculation method for differentiating spread radar data between different pulses.

FIG. 2 is a diagram of raw inter-pulse stagger radar pulse data;

FIG. 3 is a graph of a 3-stagger PRI signal;

FIG. 4 is a graph of the relationship between four different inter-pulse spread radar signals and the joint complexity value;

FIG. 5 is a graph comparing the classification accuracy of conventional features (such as pulse width PW and carrier frequency RF) and the joint complexity on different pulse-to-pulse spread radar signal classification tasks.

Detailed Description

The following description of the embodiments of the present invention is provided to facilitate the understanding of the present invention by those skilled in the art, but it should be understood that the present invention is not limited to the scope of the embodiments, and it will be apparent to those skilled in the art that various changes may be made without departing from the spirit and scope of the invention as defined and defined in the appended claims, and all matters produced by the invention using the inventive concept are protected.

As shown in fig. 1, a joint complexity calculation method for distinguishing spread radar data between different pulses includes the following steps:

s1, acquiring original pulse difference radar pulse data, and constructing Pulse Repetition Interval (PRI) data based on the original pulse difference radar pulse data;

s2, reconstructing the pulse repetition interval PRI data, and constructing a symbol sequence based on the reconstructed vector;

s3, calculating difference grade values of the reconstructed vectors, and adding the difference grade values into the symbol sequence to obtain a coding sequence;

and S4, calculating the Shannon entropy of the radar pulse data and the encoding type complexity of the radar pulse data according to the encoding sequence to obtain a combined complexity value for distinguishing the different pulse spread radar data.

The step S1 includes the following sub-steps:

s11, acquiring original pulse difference radar pulse data;

s12, constructing radar pulse data containing pulse deletion and radar pulse data containing false pulses according to the original pulse-to-pulse spread radar pulse data;

s13, counting the arrival time TOA of each pulse in three types of data, namely original pulse staggered radar pulse data, radar pulse data containing pulse missing and radar pulse data containing false pulses;

s14, constructing a time of arrival (TOA) sequence according to the TOA of each pulse;

and S15, carrying out backward difference operation on the TOA sequence of the arrival time to obtain the PRI data of the pulse repetition interval.

Fig. 2 is a schematic diagram of raw pulse-to-pulse spread radar pulse data, and it can be seen that the intervals of pulses are different but regular, namely pulse repetition intervals PRI, and the pulses of fig. 2 appear in three PRI cycles, which is also a significant characteristic of such pulse-to-pulse spread radar pulse signals. These PRI data are extracted to construct a 3-staggered PRI signal (i.e., a 3-staggered PRI signal, i.e., pulse repetition interval PRI data with a 3-staggered parameter) as shown in fig. 3, which lays a foundation for subsequent calculation of the joint complexity value.

The step S2 includes the following sub-steps:

s21, pulse repetition interval PRI dataSequentially dividing the data into a plurality of subsequences with equal length X (i), wherein N is the number of the subsequences obtained by division, X is pulse repetition interval PRI data, and X (i) is the ith subsequence of the pulse repetition interval PRI data;

where X (i) is the element of the pulse repetition interval PRI data X, τ is the delay amount, τ is typically 1, i.e. the data in each sub-sequence is continuous, and m is the number of columns.

In step S21, the pulse repetition interval PRI data is divided into N m-dimensional subsequences, so that the joint complexity value calculation can be performed based on the original whole data, but expanded according to the subsequences, thereby reducing the data processing amount to a certain extent.

S22, rearranging each subsequence X (i) according to the ascending order of elements to obtain a reconstructed vector

Reconstructing a vectorThe size relationship of the elements of (1) satisfies:

x[i+(η₁-1)τ]≤x[i+(η₂-1)τ]≤…≤x[i+(η_m-1)τ]

wherein eta is₁、η₂、…、η_mTo reconstruct the vectorThe position index of (1).

And S23, mapping each subsequence X (i) into a symbol sequence S (i) according to the sequence relation of elements in the reconstructed vector.

S(i)＝[η₁,η₂,η₃,…,η_m]

In the mapping process of step S23, when there are identical elements in different positions in the sub-sequence x (i), the symbols thereof are mapped to the position index of the first occurrence of the element in the sub-sequence x (i).

In this step, the problem of how to select the symbols of the same element is solved, for example, two sequences [ 1.31.61.32.3 ] and [ 1.31.61.42.3 ], if the symbolization completely depends on the respective index positions of the elements, the symbolization results of both sequences are [ 1324 ], obviously this method maps different sequences to the same symbol sequence, and if the method of step S23 is adopted, then [ 1124 ] and [ 1324 ] are obtained, and the symbolized results still represent the difference of the original sequences.

The step S3 includes the following sub-steps:

s31, calculating the standard deviation, median and mean of elements of the reconstructed vector;

s32, calculating the three-value difference degree according to the standard deviation, the median and the mean value;

the formula for calculating the difference degree in step S32 is:

wherein D is_{var iation}To reconstruct the disparity of the vectors, Seq_stdTo reconstruct the standard deviation of the vector, Seq_meidianTo reconstruct the median of the vector, Seq_meanIs the mean of the reconstructed vectors.

S33, normalizing the difference, mapping the normalized difference into a difference grade value, and mapping the difference into a difference grade value, thereby realizing symbolization of the difference value;

the mapping formula in step S33 is:

D_{var iation-level}＝floor(10*D_{var iation-normalized})

wherein D is_{var iation-level}For differential level values, floor (. about.). about._{var iation-normalized}In order to normalize the difference, the difference is mapped into symbols by numerical values in the calculation of the difference grade values, and the minimum difference grade value is 0 and the maximum difference grade value is 10, and 11 grades are counted.

And S34, adding the difference grade value into the symbol sequence to obtain a coding sequence.

The coding sequence is: s (i)_new＝[η₁,η₂,η₃,…,η_m,D_{var iation-kevel}]。

The step S4 includes the following sub-steps:

s41, initializing the set of S and Q strings to null, initializing j to 1, counting C_NInitialization is 0;

s42, placing the jth element in the coding sequence in a Q character string set;

s43, splicing the current Q character string set and the S character string set to obtain an SQ character string set, deleting the last element of the SQ character string set to obtain an SQ^′A set of character strings;

s44, judging whether the current Q character string set is SQ^′Sub-strings of the string set, if not, C_NAdding 1 by self, updating the current S character string set into the current SQ character string set, emptying the current Q character string set, skipping to the step S42, if yes, directly skipping to the step S42 until all elements in the coding sequence are placed in the Q character string set, and obtaining the current counting quantity C_N；

The complexity calculation in this section is based on the final encoding result S (i) [. eta. ]of S2₁,η₂,η₃,…,η_m]Counting the types of S (i), i.e. the types of the coding types, then calculating the complexity of each coding type, calculating the weight according to the occurrence times of each coding type, and finally solving the weighted complexity.

If the current set of Q strings is not a substring of the SQ' string set, then the current set of Q strings is a newly occurring string, execute C_NAdding 1, considering the current Q as the new mode, counting C_NOne needs to be added.

S45, counting the quantity C according to the current time_NAnd the length of the coding sequence, and calculating the complexity of each coding type in the coding sequence;

the formula for calculating the complexity of each coding type in the coding sequence in step S45 is as follows:

wherein, C_normComplexity of a coding type, C_NFor the counting, α is the maximum possible value of the element type in the code type, e.g. the code type is only '1', '2', '3', α is 3, and n is the length of the code sequence.

S46, calculating the complexity of each coding type in all the coding sequences according to the steps S41-S45;

s47, calculating the encoding type complexity of the radar pulse data according to the encoding type complexity of each encoding type in all the encoding sequences;

the formula for calculating the encoding type complexity of the radar pulse data in step S47 is as follows:

LZ_symbol＝Vector_{encoding_LZ}·Vector_{encoding_weight}

wherein, LZ_symbolFor coding type complexity of radar pulse data, Vector_{encoding_LZ}Vector, formed for each coding type complexity in all coding sequences_{encoding_weight}A vector of weights for all different coding types.

Counting the frequency of different coding types in a coding sequence, dividing the frequency by the length of the coding sequence to obtain the weight of each coding type in the coding sequence, and constructing the weights of all the coding types in all the coding sequences as Vector vectors_{encoding_weight}。

The frequency of occurrence of different coding types is different, the coding types with more frequency can better represent the characteristics of the whole data, and the different coding types are weighted, so that the different coding types have different weight proportions, and the complexity of the coding types of the radar pulse data can more represent the characteristics of the original radar signal.

S48, calculating the Shannon entropy of the radar pulse data according to the coding sequence;

the calculation formula of the shannon entropy of the coding sequence in the step S48 is as follows:

H＝-∑P_*ln(P_*)

wherein H is the sum of the shannon entropies of all coding sequences, P_*Is a coding sequence S (i)_newProbability of occurrence in the entire coding sequence.

The Shannon entropy reflects the regularity of the coding sequence, the smaller the entropy value is, the simpler the data is, and the larger the entropy value is, the more complex the data is.

And S49, adding the Shannon entropy of the radar pulse data and the encoding type complexity of the radar pulse data to obtain a combined complexity value for distinguishing the staggered radar data among different pulses.

The formula for obtaining the joint complexity value for distinguishing the different pulse-to-pulse spread radar data in step S49 is as follows:

C_S-LZ＝H+LZ_symbol

wherein, C_S-LZFor joint complexity values for distinguishing different inter-pulse spread radar data, H is the sum of the Shannon entropy of all coding sequences, i.e. the Shannon entropy, LZ of the radar pulse data_symbolIs the encoding type complexity of the radar pulse data.

The method of using the joint complexity value of the present invention is:

and B1, dividing the joint complexity values corresponding to the pulse spread radar signals into a training set and a test set, training a Support Vector Machine (SVM) by using the training set to obtain a classification model, and verifying by using the test set.

B2, calculating the combined complexity value of the pulse difference radar signals to be classified by using the method of the invention, inputting the combined complexity value to the verified classification model, and obtaining a classification result, namely the classification result of the pulse difference radar signals to be classified.

FIG. 4 is a graph of the correlation between the combined complexity values and four different pulse-to-pulse spread radar signals, where A1 in FIG. 4 indicates: 2 pulse repetition interval PRI data; a2 denotes: 3 pulse repetition interval PRI data; a3 denotes: 4 pulse repetition interval PRI data; a4 denotes: 6 pulse repetition interval PRI data; where the numbers 2, 3, 4 and 6 before the pulse repetition interval PRI data represent the parameter numbers of the PRI data.

As can be seen from fig. 4, the joint complexity values of the inter-pulse difference radar signals with different inter-pulse difference numbers are increased, which are in accordance with the actual complexity of the four signals, and the distribution of the joint complexity values also shows that the inter-pulse difference radar signals have good intra-class aggregation and inter-class dispersion.

Experiment: training and test set division are carried out on original pulse difference radar pulse data and different types of pulse difference radar data under the three conditions that missing pulses and false pulses exist in different proportions, an SVM is trained by using the training set to obtain a classification model, the trained classification model is tested by using the test set, and finally a comparison experiment for carrying out accuracy on different pulse difference radar signal classification tasks by using traditional characteristics and joint complexity values is set.

In addition, pulse loss and false pulses with different proportions are considered in an experiment to verify whether the effectiveness of characteristics in a classification task of a pulse-to-pulse spread radar signal is easily influenced by the pulse loss and the false pulses, and the table 1 shows that the classification accuracy of four kinds of pulse repetition interval PRI data under the conditions of pulse loss and the false pulses with different proportions can be seen, and a better classification result can be achieved by using a combined complexity value.

TABLE 1

Table 2 shows the classification accuracy of four inter-pulse spread radar signals with 10% pulse loss, which shows that the result is still considerable; the experiment also compares the joint complexity value with the conventional features (pulse width PW and carrier frequency RF) to check the superiority of the joint complexity value.

TABLE 2

Fig. 5 is a classification result of a comparison experiment, and it is obvious that the classification accuracy using the joint complexity value is relatively stable and maintains a relatively high value, and it can be seen that the feature of the joint complexity value is obviously superior to other two traditional features in the task of classifying the radar signals with the pulse width varying.