A radiation source pulse sorting method, program, device, and storage medium

Abstract

This invention belongs to the field of radiation source pulse sorting technology, specifically relating to a radiation source pulse sorting method, program, device, and storage medium. This invention models the determination process of the neighborhood radius and minimum number of neighborhood points for DBSCAN clustering in the pre-sorting of radiation source pulses as a multi-objective optimization problem, designs a multi-objective hoarfrost ice crystal optimization algorithm, and introduces multiple clustering performance indicators as optimization objectives under a unified evaluation framework to jointly constrain multiple structural performance aspects of the pre-sorting results. Simultaneously, an ice crystal subpopulation structure is introduced into the hoarfrost algorithm, integrating water molecule adsorption and soft frost search mechanisms, enabling the parameter search process to maintain both global search capability and good local search capability, ensuring the basic consistency of the pulse sequence structure in the pre-sorting results. This provides reliable input for the subsequent main sorting of each radiation source pulse sequence, reducing the problem of mis-batch sorting in radiation source pulses.

Description

Technical Field

[0001] This invention belongs to the field of radiation source pulse sorting technology, specifically relating to a radiation source pulse sorting method, program, device and storage medium. Background Technology

[0002] The core objective of radiation source pulse sorting is to progressively separate the pulse sequence structures corresponding to each radiation source within a mixed radiation source pulse sequence. A complete radiation source pulse sorting process typically includes two stages: pulse pre-sorting and main pulse sorting. Pre-sorting primarily serves to initially dilute and structurally separate the high-density pulse stream, aiming to provide a pulse sequence with basic consistency for the main sorting stage. If the pre-sorting results lack basic structural consistency, it can easily lead to difficulties in effectively sorting the pulse sequences of each radiation source during the main sorting stage, resulting in mis-sorting issues in the final sorting results.

[0003] Existing methods for pre-sorting radiation source pulses typically employ a combined intra-pulse and inter-pulse parameter clustering approach. Commonly used methods include partitioning clustering methods such as K-means and density-based clustering methods such as DBSCAN. Among these, density-based clustering methods have certain advantages in radiation source pulse pre-sorting because they do not require pre-defined cluster centers. However, the clustering effect of this type of method is highly sensitive to key parameters such as neighborhood radius and minimum number of neighborhood points. When these parameters are not selected properly, it is easy for a large number of pulses from different radiation sources to be incorrectly assigned to the same pulse sequence, or for pulses from the same radiation source to be incorrectly assigned to different sequences. This results in insufficient consistency in the structure of the pulse sequence output by pre-sorting, directly affecting the effective sorting of the pulse sequence in the subsequent main sorting stage, and ultimately leading to mis-batch problems in the entire radiation source pulse sorting process.

[0004] Wei Yi et al. proposed a radar pulse sorting method based on multi-density features in their patent "A Parameter Adaptive Multi-Density Clustering Radar Signal Sorting Method" (patent number: CN202411176403.1). This method achieves adaptive selection of clustering parameters by analyzing the distance distribution between pulse descriptors, thus improving the adaptability of radiation source pulse pre-sorting to some extent. However, this method mainly relies on distance distribution statistical features and hierarchical traversal strategies for parameter adjustment, without jointly constraining the multiple structural performance aspects of the pre-sorting results under a unified evaluation framework, making it difficult to guarantee the basic consistency of the pulse sequence structure in the pre-sorting results. Wang Fei constructed a unified multi-objective optimization framework in his paper "UAV Delivery Route Planning Based on Multi-Objective Optimization Model and NSGA-II," but the NSGA-II search process still mainly relies on the genetic evolution mechanism of a single population. When the number of targets is large or target conflicts are significant, the solution set tends to cluster in local areas, limiting its global optimization capability. Therefore, if the key parameters of density clustering can be optimized under a unified multi-objective evaluation framework, enabling pre-sorting to perform preliminary dilution and effective structural separation of the received pulse sequences, it will help the subsequent main sorting to effectively sort the pulse sequences of each radiation source and reduce the problem of mis-sorting of radiation source pulses. Summary of the Invention

[0005] The purpose of this invention is to solve the problem that improper setting of key clustering parameters during radiation source pulse sorting leads to insufficient consistency in the structure of the pre-sorted output pulse sequence, making it difficult for subsequent main sorting to effectively sort the pulse sequence, resulting in mis-batch sorting of radiation source pulses. This invention provides a radiation source pulse sorting method, program, device, and storage medium.

[0006] A method for sorting radiation source pulses includes the following steps:

[0007] Obtain a pulse signal dataset in which the pulse descriptor and the corresponding real radiation source classification of each pulse signal are known; construct the feature vector of each pulse signal based on the pulse descriptor of each pulse signal.

[0008] The neighborhood radius and the minimum number of neighborhood points are used as optimization objectives to construct an objective vector;

[0009] Based on the neighborhood radius and minimum number of neighborhood points corresponding to the target vector, DBSCAN clustering is performed on the feature vectors of all pulse signals in the pulse signal dataset. The clustering accuracy, clustering purity, Rand index, and V-measure objective function of the clustering results are used as multi-objective clustering evaluation indicators.

[0010] Optimize the target vector to obtain the Pareto front approximate solution set of the target vector; from the Pareto front approximate solution set of the target vector, select the solution with the best comprehensive performance based on the multi-objective clustering evaluation index;

[0011] Obtain the pulse signal sequence to be sorted and construct the pulse feature vector to be sorted; perform DBSCAN clustering on the pulse feature vector to be sorted based on the neighborhood radius corresponding to the solution with the best comprehensive performance and the minimum number of neighborhood points to obtain the pre-sorting result of the radiation source pulse;

[0012] Based on the pre-sorting results of the radiation source pulses, pulse main sorting is performed to sort out the pulse sequences of each radiation source.

[0013] Furthermore, a multi-objective hoarfrost ice crystal optimization algorithm is used to optimize the target vector and construct a main population, where each individual in the main population represents a solution to a set of target vectors;

[0014] Non-dominated sorting is performed based on the multi-objective clustering evaluation index of all individuals in the main population. A specified number of individuals with the largest crowding distance in the first frontier level are selected as ice crystal nuclei. The remaining individuals in the main population are randomly assigned to each ice crystal nucleus as water molecules. Each ice crystal nucleus forms an ice crystal subpopulation.

[0015] For each ice crystal population, select the individual with the closest Euclidean distance to the ice crystal nucleus from the global archive, and update the ice crystal nucleus through hard frost puncture; introduce a dual search mechanism of water molecule adsorption growth and soft frost search to update the water molecules in the ice crystal population.

[0016] Merge all updated ice crystal populations with the main population to obtain an extended population; perform non-dominated sorting on the individuals in the extended population, and add all individuals of the first frontier level in the extended population to the global archive to obtain the updated global archive;

[0017] Perform a non-dominated sort on all individuals in the updated global archive and delete dominated individuals; if the number of individuals still exceeds the global archive capacity after deletion, retain individuals with larger crowding distances.

[0018] With a specified probability, perform a hoarse attack on the individual with the smallest crowding distance in the updated global archive to escape the local Pareto front; perform screening on the individuals in the expanded population, and after reaching the preset population size, use them as the main population for the next iteration;

[0019] If the number of iterations meets the recombination interval, then disband all ice crystal population structures; otherwise, continue updating the ice crystal nuclei and water molecules.

[0020] After reaching the maximum number of iterations, the iteration stops, and the global archive is output as the Pareto front approximate solution set of the target vector.

[0021] Furthermore, the main population Each body in This represents a solution to a set of objective vectors. ;

[0022] According to the individual Corresponding neighborhood radius with minimum neighborhood number DBSCAN clustering is performed on the feature vectors of all pulse signals in the pulse signal dataset. The clustering accuracy of the clustering results is then evaluated. Cluster purity RAND Corporation Index and V-measure objective function As a multi-objective clustering evaluation index;

[0023] If individual The values ​​of each target clustering evaluation index are not less than ,Right now , And individuals There exists at least one target clustering evaluation index value greater than Then the individual Dominant Individual ;

[0024] main population Individuals not dominated by any other individual are classified as the first frontier level. ;

[0025] Excluding the first frontier level After the individual, the main population The other individuals continue to execute the judgment until the non-dominated sorting is completed.

[0026] Furthermore, the method for calculating the congestion distance is as follows:

[0027] For the first Frontier Level Individuals in Crowded distance ;

[0028]

[0029] in, Indicates the first Frontier Level Except for individuals Other individuals besides, ; Indicates the first Frontier Level The corresponding number in the middle The maximum value of the target clustering evaluation index; Indicates the first Frontier Level The corresponding number in the middle The minimum value of the target clustering evaluation index.

[0030] Furthermore, for each ice crystal population, the individual with the closest Euclidean distance to the ice crystal nucleus is selected from the global archive, and the ice crystal nucleus is updated through hard frost puncture, specifically as follows:

[0031]

[0032] in, For the first In the nth iteration The ice crystal nuclei of the ice crystal population; Save globally Zhongyu The closest individual in the European distance; Puncture strength; The attenuation coefficient;

[0033] The dual search mechanism of water molecule adsorption growth and soft frost search is introduced to update the water molecules in this ice crystal population, specifically as follows:

[0034] For the The first in the ice crystal population water molecules Generate random numbers ;

[0035] Set adsorption probability ,like Then, water molecules are renewed through adsorption and growth. ;

[0036]

[0037] in, The adsorption coefficient of water molecules;

[0038] like Then, the soft cream searches for and updates water molecules. ;

[0039]

[0040] in, The disturbance intensity; Random numbers that satisfy a standard normal distribution; target vector The upper realm, target vector The lower bound.

[0041] Furthermore, the step of performing a haze attack on the individual with the smallest crowding distance in the updated global archive with a specified probability to escape the local Pareto front specifically involves:

[0042] For the updated global archive The individual with the smallest distance in a crowded area With a specified probability Conduct a rime ice attack;

[0043] If the individual If attacked by frost, the update will be: ;

[0044] in, The range of frost attack; Random numbers that satisfy a standard normal distribution.

[0045] Furthermore, the step of selecting the solution with the best overall performance from the Pareto front approximation solution set of the target vector based on a multi-objective clustering evaluation index specifically involves:

[0046] For the target vector The clustering evaluation index values ​​of each objective in the Pareto front approximate solution set are normalized. After normalization, the range of each objective clustering evaluation index value is [missing value]. ;

[0047] Constructing the ideal evaluation vector For the target vector For each candidate solution in the Pareto front approximation solution set, calculate its normalized Euclidean distance to the ideal evaluation vector, and select the candidate solution with the smallest corresponding Euclidean distance as the solution with the best overall performance. .

[0048] A computer device includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the radiation source pulse sorting method described above.

[0049] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described radiation source pulse sorting method.

[0050] A computer program product includes computer instructions that, when executed by a processor, implement the steps of the radiation source pulse sorting method described above.

[0051] The beneficial effects of this invention are as follows:

[0052] This invention models the process of determining the neighborhood radius and minimum number of neighborhood points for DBSCAN clustering in the pre-sorting of radiation source pulses as a multi-objective optimization problem. A multi-objective hoarfrost ice crystal optimization algorithm is designed, introducing multiple clustering performance indicators as optimization objectives within a unified evaluation framework to jointly constrain the multiple structural performance aspects of the pre-sorting results. Simultaneously, an ice crystal subpopulation structure is introduced into the hoarfrost algorithm, integrating water molecule adsorption and soft frost search mechanisms. This ensures that the parameter search process maintains both global search capability and good local search capability, avoiding the situation where pulses from many different radiation sources are incorrectly assigned to the same pulse sequence, or pulses from the same radiation source are incorrectly assigned to different sequences due to improper parameter selection. This guarantees the basic consistency of the pulse sequence structure in the pre-sorting results, providing reliable input for the subsequent main sorting of each radiation source pulse sequence and reducing the problem of mis-batch sorting of radiation source pulses. Attached Figure Description

[0053] Figure 1 This is the overall flowchart of the present invention.

[0054] Figure 2 This is a flowchart of the multi-objective rime ice crystal optimization algorithm in this invention. Detailed Implementation

[0055] The present invention will now be further described with reference to the accompanying drawings.

[0056] The purpose of this invention is to design a novel method for sorting radiation source pulses. The main aim is to address the problem that improper setting of key clustering parameters during radiation source pulse sorting leads to insufficient consistency in the structure of the pre-sorted pulse sequences, making it difficult for subsequent main sorting to effectively sort the pulse sequences from different radiation sources, thus causing mis-sorting. This invention models the process of determining the neighborhood radius and minimum number of neighborhood points for DBSCAN clustering in the pre-sorting of radiation source pulses as a multi-objective optimization problem. A multi-objective hoarfrost ice crystal optimization algorithm is designed, introducing multiple clustering performance indicators as optimization objectives. Simultaneously, an ice crystal subpopulation structure is incorporated into the hoarfrost algorithm, integrating water molecule adsorption and soft frost search mechanisms. This gives the parameter search process both good global and local search capabilities, avoiding the situation where pulses from many different radiation sources are incorrectly assigned to the same pulse sequence, or pulses from the same radiation source are incorrectly assigned to different sequences due to improper parameter selection. This provides reliable input for the main sorting to effectively sort the pulse sequences from different radiation sources, reducing the problem of mis-sorting.

[0057] As attached Figure 1 As shown, a method for sorting radiation source pulses includes the following steps:

[0058] Step 1: Obtain known information One radiation source The pulse description word of the pulse signal, including the first pulse signal. Pulse width of pulse signal carrier frequency ,bandwidth , Construct pulse feature vector And determine the characteristic vector of each pulse. Corresponding classification of real radiation sources;

[0059] Step 2: Calculate the neighborhood radius in the DBSCAN clustering method. with minimum neighborhood number As the optimization objective, construct the objective vector. ;

[0060] Determine the neighborhood radius The upper realm and the lower world Minimum number of neighborhood points The upper realm and the lower world Thus, the target vector is determined. The upper realm and the lower world That is, the feasible region of the target vector;

[0061] In one embodiment of the present invention, the neighborhood radius Minimum number of neighborhood points Upper bounds of neighborhood radius and minimum number of neighboring points Set as The lower bound of the neighborhood radius and the minimum number of neighboring points. Set as ;

[0062] Step 3: Employ a multi-objective hoarfrost ice crystal optimization algorithm. This algorithm incorporates an ice crystal subpopulation structure and integrates a dual search mechanism combining water molecule adsorption growth and soft frost search to optimize the target vector. Perform optimization to obtain the target vector. The approximate solution set of the Pareto front;

[0063] Step 3.1: Set the population size Number of ice crystal nuclei Maximum number of iterations Global archive capacity Adsorption probability Disturbance intensity Water molecule adsorption coefficient Puncture strength attenuation coefficient Recombination interval Rime attack probability , Frost attack range ;

[0064] In one embodiment of the present invention, It can be set to 50; number of ice crystal cores It can be set to 5; maximum number of iterations. Can be set to 100; External archive capacity It can be set to 100; adsorption probability It can be set to 0.7; Disturbance intensity It can be set to 0.3; water molecule adsorption coefficient It can be set to 0.5; puncture strength It can be set to 0.8; attenuation coefficient It can be set to 0.01; recombination interval period It can be set to 20; frost attack probability It can be set to 0.05; Rime Attack Range It can be set to 0.2;

[0065] Initialize the number of iterations Initialize random generation within the feasible region of the target vector. individual , To form the initial population ;

[0066] Initialization will individual Add to global save ;

[0067] Step 3.2: For the population Each individual Based on the neighborhood radius corresponding to this individual with minimum neighborhood number right pulse feature vectors Perform clustering;

[0068] Clustering accuracy of clustering results Cluster purity RAND Corporation Index and V-measure objective function As a multi-objective clustering evaluation index , ;

[0069] Among them, clustering accuracy objective function :

[0070]

[0071] in, The confusion matrix is ​​the result of clustering. These are elements of the optimal allocation matrix. The number of clusters, The number of true categories;

[0072] Cluster purity objective function :

[0073]

[0074] in, For the first A cluster, , For the first One real category, , This indicates that they belong to the same cluster. and real-world categories The number of pulses;

[0075] RAND Corporation objective function :

[0076]

[0077] in, Belonging to both clusters and real-world categories The number of pulses, From The number of combinations of selecting 2 from a total of pulses. To be from real class The number of combinations of selecting 2 from a sample. To get from clusters The number of combinations of selecting 2 from a sample. To belong to the same cluster and real-world categories The number of combinations of selecting 2 items from a sample;

[0078] V-measure objective function :

[0079]

[0080] in, The extent to which each cluster contains only a single true class. The extent to which samples of each true class are assigned to the same cluster:

[0081]

[0082]

[0083] in, Given cluster labels, the conditional entropy of the true class. ; Given the true class, the conditional entropy of the cluster labels. ; The entropy of the true category, ; The entropy of the clustering results, ; The sample belongs to a cluster And the real category is The joint probability, ; The sample belongs to a cluster The marginal probability, ; The sample belongs to the real class The marginal probability, ;

[0084] Step 3.3: Population Individuals in the process perform non-dominated sorting;

[0085] If individual The values ​​of each target clustering evaluation index are not less than ,Right now And individuals There exists at least one target clustering evaluation index value greater than Then the individual Dominant Individual ;

[0086] population Individuals not dominated by any other individual are classified as the frontier level. ;

[0087] Excluding the frontier level After determining the individual, the judgment is performed on the other individuals; the above steps are repeated until the frontier level classification is completed;

[0088] For the first Frontier Level Individuals in Crowded distance ;

[0089]

[0090] in, Indicates the first Frontier Level Except for individuals Other individuals besides, ; Indicates the first Frontier Level The corresponding number in the middle The maximum value of the target clustering evaluation index; Indicates the first Frontier Level The corresponding number in the middle The minimum value of the target clustering evaluation index;

[0091] Step 3.4: Select the leading edge level The middle corresponds to the largest congestion distance Each individual crystal acts as the core of an ice crystal, if the leading level... The number of individuals in it is less than Then the frontier level All individuals in the population serve as ice crystal nuclei, and the remaining individuals in the population are randomly distributed as water molecules to each ice crystal nucleus, forming... A population of ice crystals;

[0092] in, equal or the The leading-edge level in the next iteration The number of individuals in;

[0093] Step 3.5: Update the populations of each ice crystal;

[0094] For the Ice crystal nuclei in the ice crystal population From global archive Take from the middle and The closest individual in the European distance Updating by hard frost puncture of the ice crystal core ;

[0095]

[0096] For the The first in the ice crystal population water molecules Generate random numbers ;

[0097] If the random number is less than the adsorption probability ,Right now Then, water molecules are renewed through adsorption and growth. ;

[0098]

[0099] like Then, the soft cream searches for and updates water molecules. ;

[0100]

[0101] in, To obtain random numbers that satisfy the standard normal distribution, i.e. ;

[0102] Step 3.6: Update all ice crystal populations and populations. Merging yields an expanded population. ;

[0103] For expanding the population Individuals in the group are subjected to non-dominated ranking, and the frontier rank is calculated. The crowding distance between individuals;

[0104] Expand the population The frontier level is Add all individuals to the global archive In the middle, get the updated global save. ; For the updated global archive Perform a dominance relationship determination on all individuals, delete the dominated individual, and save the global archive if deletion occurs. The number of individuals exceeds the global archive capacity. Then, based on the crowding distance of each individual, individuals with smaller crowding distances are deleted first, until the global archive is completed. The number of individuals in the middle reaches the global archive capacity ;

[0105] With probability Global archive The individual with the smallest distance in a crowded area Perform a rime attack to escape the local Pareto front and avoid getting trapped in a local optimum;

[0106]

[0107] in, To obtain random numbers that satisfy the standard normal distribution, i.e. ;

[0108] Step 3.7: Expand the population Individuals within the population are selected by prioritizing those at the forefront level. Within the same forefront level, individuals with greater crowding distance are prioritized, until the population size reaches the preset target size. This yields the population for the next iteration. ;

[0109] Step 3.8: If the number of iterations... Recombination interval period reached Multiples of, i.e. If so, disband all ice crystal population structures and proceed to step 3.9; otherwise, let Return to step 3.5;

[0110] Step 3.9: If the number of iterations... Reaching the maximum number of iterations If the iteration stops, output the global archive. As the target vector The approximate solution set of the Pareto front; otherwise, let Return to step 3.3;

[0111] Step 4: From the target vector From the Pareto front approximate solution set, the solution with the best overall performance is selected based on a multi-objective clustering evaluation index. ;

[0112] For the target vector The clustering evaluation index values ​​of each objective in the Pareto front approximate solution set are normalized. After normalization, the range of each objective clustering evaluation index value is [missing value]. ;

[0113] Constructing the ideal evaluation vector For the target vector For each candidate solution in the Pareto front approximation solution set, calculate its normalized Euclidean distance to the ideal evaluation vector, and select the candidate solution with the smallest corresponding Euclidean distance as the solution with the best overall performance. ;

[0114] Since cluster purity, cluster accuracy, RAND index, and V-measure are all maximization metrics, the ideal evaluation vector can be set as the theoretical upper limit of each objective clustering evaluation metric, i.e. ;

[0115] Step 5: Obtain the pulse signal sequence to be sorted and construct the pulse feature vector to be sorted; based on the solution with the best overall performance... The corresponding neighborhood radius and minimum number of neighborhood points are used to perform DBSCAN clustering on the pulse feature vectors to be sorted, obtaining the pre-sorting results of the radiation source pulses, including... One cluster;

[0116] Step 6: Based on the pre-sorting results of radiation source pulses, perform pulse main sorting, calculate the PRI estimate value corresponding to each cluster, and perform sequence retrieval based on the PRI estimate value to sort out the pulse sequences of each radiation source;

[0117] Step 6.1: Minimum PRI value for each cluster PRI value range maximum value PRI value range Divide the PRI value range into equal parts. Each part is called a PRI box, and the box width is... Box width The formula is:

[0118]

[0119] No. The center coordinates of each PRI box for:

[0120]

[0121] in, ;

[0122] Step 6.2: Using the first One pulse is used as the current reference pulse, and the pulse index is adjusted accordingly. A sliding search is performed on the pre-sorted pulse sequence to determine the arrival time difference between pulses. Does the PRI constraint condition satisfy?

[0123] Step 6.2.1: Using the first Using the first pulse as a reference pulse, calculate its relationship with the second pulse. Time difference of arrival between pulses :

[0124]

[0125] Step 6.2.2: If If the current pulse satisfies the PRI constraint, then step 6.3 is executed; if or This indicates the current candidate pulse. With reference pulse The interval is too large or the current reference pulse The search has ended, so let and reinitialize Return to step 6.2.1, where... The first in the pre-sorting results The number of pulses in a pulse sequence; if This indicates the current candidate pulse. With reference pulse If the interval is too small, then let Return to step 6.2.1; when When the current pulse sequence search ends, proceed to step 6.4;

[0126] Step 6.3: From The corresponding can be determined Value, update ,make Accumulate PRI boxes. The calculation formula is:

[0127]

[0128] in, For the current conditions, the first Spectral values ​​of each PRI box, For integration variables;

[0129] Step 6.4: For each PRI box, calculate the detection threshold. When the accumulated value of the PRI bin exceeds the corresponding threshold, the center coordinates of the PRI bin are changed. As an estimate of PRI, The formula and the estimated PRI value are:

[0130]

[0131]

[0132] in, This is an adjustable coefficient for observation time. To eliminate the subharmonic adjustable coefficient, To eliminate noise, the adjustable coefficient, For observation time, pulse density, For the first The width of each PRI box;

[0133] Step 6.5: Based on the PRI estimate, in the pre-sorting results The sequence is searched and the pulse sequences of each radiation source are finally sorted out.

[0134] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for sorting radiation source pulses, characterized in that: Obtain the pulse signal dataset, in which the pulse descriptor and the corresponding real radiation source classification of each pulse signal are known; Based on the pulse description words of each pulse signal, construct the feature vector of each pulse signal; The neighborhood radius and the minimum number of neighborhood points are used as optimization objectives to construct an objective vector; Based on the neighborhood radius and minimum number of neighborhood points corresponding to the target vector, DBSCAN clustering is performed on the feature vectors of all pulse signals in the pulse signal dataset. The clustering accuracy, clustering purity, Rand index, and V-measure objective function of the clustering results are used as multi-objective clustering evaluation indicators. Optimize the target vector to obtain the Pareto front approximate solution set of the target vector; from the Pareto front approximate solution set of the target vector, select the solution with the best comprehensive performance based on the multi-objective clustering evaluation index; Obtain the pulse signal sequence to be sorted and construct the pulse feature vector to be sorted; Based on the neighborhood radius and minimum number of neighborhood points corresponding to the solution with the best overall performance, DBSCAN clustering is performed on the pulse feature vector to be sorted to obtain the pre-sorting results of radiation source pulses; Based on the pre-sorting results of the radiation source pulses, pulse main sorting is performed to sort out the pulse sequences of each radiation source.

2. The radiation source pulse sorting method according to claim 1, characterized in that: A multi-objective hoarfrost and ice crystal optimization algorithm is used to optimize the target vector and construct a main population, where each individual in the main population represents a solution to a set of target vectors. Non-dominated sorting is performed based on the multi-objective clustering evaluation index of all individuals in the main population. A specified number of individuals with the largest crowding distance in the first frontier level are selected as ice crystal nuclei. The remaining individuals in the main population are randomly assigned to each ice crystal nucleus as water molecules. Each ice crystal nucleus forms an ice crystal subpopulation. For each ice crystal population, select the individual with the closest Euclidean distance to the ice crystal nucleus from the global archive, and update the ice crystal nucleus through hard frost puncture; introduce a dual search mechanism of water molecule adsorption growth and soft frost search to update the water molecules in the ice crystal population. Merge all updated ice crystal populations with the main population to obtain an extended population; perform non-dominated sorting on the individuals in the extended population, and add all individuals of the first frontier level in the extended population to the global archive to obtain the updated global archive; Perform a non-dominated sort on all individuals in the updated global archive and delete dominated individuals; if the number of individuals still exceeds the global archive capacity after deletion, retain individuals with larger crowding distances. With a specified probability, perform a hoarse attack on the individual with the smallest crowding distance in the updated global archive to escape the local Pareto front; perform screening on the individuals in the expanded population, and after reaching the preset population size, use them as the main population for the next iteration; If the number of iterations meets the recombination interval, then disband all ice crystal population structures; otherwise, continue updating the ice crystal nuclei and water molecules. After reaching the maximum number of iterations, the iteration stops, and the global archive is output as the Pareto front approximate solution set of the target vector.

3. The radiation source pulse sorting method according to claim 2, characterized in that: The main population Each body in This represents a solution to a set of objective vectors. ; According to the individual Corresponding neighborhood radius with minimum neighborhood number DBSCAN clustering is performed on the feature vectors of all pulse signals in the pulse signal dataset. The clustering accuracy of the clustering results is then evaluated. Cluster purity RAND Corporation Index and V-measure objective function As a multi-objective clustering evaluation index; If individual The values ​​of each target clustering evaluation index are not less than ,Right now , And individuals There exists at least one target clustering evaluation index value greater than Then the individual Dominant Individual ; main population Individuals not dominated by any other individual are classified as the first frontier level. ; Excluding the first frontier level After the individual, the main population The other individuals continue to execute the judgment until the non-dominated sorting is completed.

4. The radiation source pulse sorting method according to claim 3, characterized in that: The method for calculating the congestion distance is as follows: For the first Frontier Level Individuals in Crowded distance ; in, Indicates the first Frontier Level Except for individuals Other individuals besides, ; Indicates the first Frontier Level The corresponding number in the middle The maximum value of the target clustering evaluation index; Indicates the first Frontier Level The corresponding number in the middle The minimum value of the target clustering evaluation index.

5. The radiation source pulse sorting method according to claim 3, characterized in that: For each ice crystal population, the individual with the closest Euclidean distance to the ice crystal core is selected from the global archive, and the ice crystal core is updated through hard frost puncture. Specifically: in, For the first In the nth iteration The ice crystal nuclei of the ice crystal population; Save globally Zhongyu The closest individual in the European distance; Puncture strength; The attenuation coefficient; The dual search mechanism of water molecule adsorption growth and soft frost search is introduced to update the water molecules in this ice crystal population, specifically as follows: For the The first in the ice crystal population water molecules Generate random numbers ; Set adsorption probability ,like Then, water molecules are renewed through adsorption and growth. ; in, The adsorption coefficient of water molecules; like Then, the soft cream searches for and updates water molecules. ; in, The disturbance intensity; Random numbers that satisfy a standard normal distribution; target vector The upper realm, target vector The lower bound.

6. The radiation source pulse sorting method according to claim 5, characterized in that: The step of performing a frost attack on the individual with the smallest crowding distance in the updated global archive with a specified probability to escape the local Pareto front specifically involves: For the updated global archive The individual with the smallest distance in a crowded area With a specified probability Conduct a rime ice attack; If the individual If attacked by frost, the update will be: ; in, The range of frost attack; Random numbers that satisfy a standard normal distribution.

7. The radiation source pulse sorting method according to claim 3, characterized in that: The step of selecting the solution with the best overall performance from the Pareto front approximation solution set of the target vector based on a multi-objective clustering evaluation index is as follows: For the target vector The clustering evaluation index values ​​of each objective in the Pareto front approximate solution set are normalized. After normalization, the range of each objective clustering evaluation index value is [missing value]. ; Constructing the ideal evaluation vector For the target vector For each candidate solution in the Pareto front approximation solution set, calculate its normalized Euclidean distance to the ideal evaluation vector, and select the candidate solution with the smallest corresponding Euclidean distance as the solution with the best overall performance. .

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 7.

10. A computer program product comprising computer instructions, characterized in that: When executed by a processor, the computer instructions implement the steps of the method according to any one of claims 1 to 7.

Images