CN-122020080-A - Main radar signal sorting method and system based on peak clustering and optimized pulse searching

CN122020080ACN 122020080 ACN122020080 ACN 122020080ACN-122020080-A

Abstract

The invention belongs to the technical field of radar signal processing, and relates to a radar signal main sorting method and system based on peak clustering and optimized pulse search. The method comprises multi-order differential feature accumulation, self-adaptive density and distance calculation, automatic cluster center identification, interference point rejection, PRI estimation, optimized pulse search and iterative update. The method is suitable for efficient sorting of the multi-source radar pulse signals under the complex electromagnetic environment, realizes unsupervised clustering with strong robustness, realizes pulse sequence extraction with high matching precision and controllable error, has good noise immunity and parameter separation performance, remarkably improves sorting accuracy and stability in complex environments with multi-radar signal overlapping, serious false alarm interference and the like, overcomes the dependence of the traditional method on a fixed threshold, realizes high-precision and automatic PRI cluster recognition under the multi-target scene, accurately eliminates low-density pseudo cluster points, remarkably reduces false division rate, improves search matching efficiency, and ensures the integrity and continuity of pulse sequence extraction.

Inventors

XIE LONGHAO
WANG BIN
LI HUIYONG
LI MING
CHENG ZIYANG

Assignees

电子科技大学长三角研究院(衢州)

Dates

Publication Date: 20260512
Application Date: 20251222
Priority Date: 20251021

Claims (10)

1. The radar signal main sorting method based on peak clustering and optimized pulse searching is characterized by comprising multi-order differential feature accumulation, self-adaptive density and distance calculation, cluster center automatic identification, interference point rejection, PRI estimation, optimized pulse searching and iterative updating; the multi-order differential feature accumulation comprises the steps of sequentially reading an interleaved pulse arrival time sequence, sequentially calculating a plurality of orders of differential, and accumulating differential results of all the orders into a mixed feature vector to obtain a PRI sample; the self-adaptive density and distance calculation comprises the steps of determining a neighborhood radius by combining a linear weighted neighborhood density estimation method and the distribution characteristics of PRI samples, respectively calculating a local density value of each PRI sample and a distance between each PRI sample and a neighbor with higher density, and representing the aggregation and separation degree of the PRI samples in a density and distance two-dimensional space; The cluster center is automatically identified, wherein the cluster center is obtained by evaluating the clustering results of different cluster numbers by utilizing a Karnsiki-Harabase index, determining the optimal cluster number, and identifying PRI samples meeting the set density and the set distance in a two-dimensional space of the density and the distance as the cluster center; The method comprises the steps of carrying out normalization on density values of PRI samples, determining an optimal density threshold value by using a maximum inter-class variance method, and eliminating PRI samples with density lower than the density threshold value to obtain PRI sample sets corresponding to PRI clusters; in the current iteration round, calculating an arithmetic mean value of PRI samples in the PRI cluster according to the PRI cluster subjected to density clustering and interference point eliminating processing, and taking the arithmetic mean value as a PRI estimated value of the round; The optimized pulse searching comprises the steps of constructing a dynamic searching window by taking a PRI estimated value as a center, executing sliding matching operation in a pulse arrival time sequence, sequencing candidate pulses according to time errors of theoretical matching positions, selecting pulses with the smallest errors for matching, and removing PRI samples with the errors exceeding a set threshold value to obtain a target pulse sequence; and iteratively updating, namely removing the target pulse sequence from the pulse arrival time sequence, re-extracting the target pulse sequence according to the residual pulse until the quantity of the residual pulse arrival time sequence meets the set condition, and finishing sorting.
2. The method for radar signal main sorting based on peak clustering and optimized pulse searching according to claim 1, wherein row vectors with pulse arrival time sequences arranged in ascending order are set as , , The number of radar pulses is indicated, Indicating receipt in the presence of noise interference or false pulses The time of arrival of the individual radar pulses, Represent the first The pulse of the order of the pulse, Indicating the set maximum differential order number, Representing pulse sequences The pulse sequence number of the pulse sequence number, Represent the first The time of arrival of the individual pulses, Represent the first Arrival time of each pulse, the first Differential of the time series of arrival of the order pulses Then: 。
3. The method for radar signal main sorting based on peak clustering and optimized pulse search according to claim 1, wherein accumulating the differential results of each order into a mixed feature vector comprises: Is provided with Represent the first The pulse of the order of the pulse, Represents the set maximum differential order, the th Differential of the time series of arrival of the order pulses , Represent the first The difference in the arrival time series of the order pulses, The number of data points for the difference of the pulse arrival time series, The hybrid feature vector is represented by a vector of features, Representing a real number, then: 。
4. The radar signal main sorting method based on peak clustering and optimized pulse search according to claim 1, wherein The number of data points for the difference of the pulse arrival time series, Represent the first A PRI sample accumulated by the multi-order differential features, Represent the first A PRI sample accumulated by the multi-order differential features, Represent the first The local density of the individual PRI samples, Representing the minimum distance from each PRI point to all higher density points, As the number of neighbors to be counted, Is a positive integer which is used for the preparation of the high-voltage power supply, The PRI sample accumulated by the multi-order differential features is , In order to truncate the distance threshold value, Is that Corresponding first after ascending order The data of the columns are stored in a memory, Then: ; ; 。
5. The method for main sorting of radar signals based on peak clustering and optimized pulse search according to claim 1, wherein the method for main sorting of radar signals based on peak clustering and optimized pulse search comprises evaluating clustering results of different cluster numbers by using a Karnsiki-Harabase index, determining an optimal cluster number, and identifying PRI samples satisfying a set density and a set distance in a two-dimensional space of density and distance as cluster centers, comprising setting a total pulse sample number as a cluster center The candidate cluster number is , Index of , Is a trace of the inter-cluster covariance matrix, used to characterize the degree of separation between different clusters, Is a trace of the intra-cluster covariance matrix, used to characterize the compactness of intra-cluster samples, Is the first The average value of the individual clusters is calculated, As a global average value of the values, Is the first The number of samples within a single cluster, Represent the first A collection of samples of the individual clusters, Represent the first PRI samples accumulated by the multi-order differential features are: ; ; ; for each of a range of preset candidate cluster numbers Calculating corresponding Kaolin-Ha Laba index value, selecting the cluster number corresponding to maximum value as cluster center number, setting the cluster center number as Then: ; Set the first The joint index of the two-dimensional space of the density and the distance of each sample is , Represent the first The local density of the individual PRI samples, Representing the minimum distance of each PRI sample to all higher density points, then: ; According to joint index The values of (2) are ordered from big to small and the previous one is selected The samples are taken as the final cluster center points.
6. The method for radar signal primary sorting based on peak clustering and optimized pulse searching of claim 1, wherein normalizing the density values of PRI samples comprises: Set the first Local density of individual PRI samples is , For a local density minimum for all PRI samples, For the local density maximum of all PRI samples, Is the first The local density of the individual PRI samples, For normalized local density, the local density is normalized to interval [0,1] by polar normalization, and the normalized local density is expressed as: 。
7. The method for main sorting radar signals based on peak clustering and optimized pulse search according to claim 1, wherein the method for determining the optimal density threshold by using the maximum inter-class variance method, removing PRI samples with density lower than the density threshold to obtain a sample set corresponding to the PRI cluster comprises dividing the PRI samples into a first density cluster which is greater than or equal to a set threshold and a second density cluster which is smaller than the set threshold, and setting intra-class accumulation probability of the first density cluster as The intra-class cumulative probability of the second density cluster is The first density cluster has an average density of The second density cluster has an average density of Inter-class variance of , To normalize the first in the density histogram The probability values for the individual density classes, For the number of density steps, For the density level index, As a candidate density threshold, then: ; ; ; ; the inter-class variance is expressed as: ; Setting the optimal density threshold value as Then: ; let the original density threshold be First, the Local density of individual PRI samples is , For a local density minimum for all PRI samples, For the local density maximum of all PRI samples, Is the first Local density of individual PRI samples; Will be the optimal threshold Mapping back to the original density space to obtain a corresponding original density threshold, and then expressing the original density threshold as: ; classifying and judging all PRI samples according to the optimal density threshold value, if And otherwise, reserving the PRI sample to obtain a PRI sample set corresponding to the PRI cluster.
8. The method for radar signal main sorting based on peak clustering and optimized pulse search according to claim 1, wherein in the current iteration round, according to the PRI cluster having completed density clustering and interference point rejection processing, calculating an arithmetic mean value of PRI samples in the PRI cluster as a PRI estimation value of the round, comprising: Let the number of data points be , Is the first in the current PRI cluster The value of the individual data points is calculated, Is the first PRI cluster The PRI cluster sequence after clustering and interference elimination is the value of each data point , A PRI estimate representing the current PRI cluster, Indicating the number of data points in the current PRI cluster, the PRI estimated value of the current PRI cluster is: 。
9. The method for main sorting radar signals based on peak clustering and optimized pulse search according to claim 1, wherein a dynamic search window is constructed with PRI estimation value as the center, sliding matching operation is performed in pulse arrival time sequence, candidate pulses are ordered according to time errors with theoretical matching positions, pulses with minimum errors are selected for matching, PRI samples with errors exceeding a set threshold are removed, and a target pulse sequence is obtained, including setting current reference pulse time as PRI estimate is The theoretical arrival time of the next real pulse is The upper search limit is The candidate set is , As a candidate set Middle (f) The time of arrival of the individual pulses, For the pulse arrival time set which is not matched currently, screening all pulses meeting the set condition from the rest arrival time sequences to form a candidate set: ; Setting the matching error of each candidate pulse in the candidate set and the theoretical value as Then: ; set the matching target as , For selecting an error threshold for screening valid matching pulses, selecting a matching error that satisfies And match error The smallest pulse serves as the matching target, then: ; Is provided with Is the first in the pulse sequence The time of arrival of the individual pulses, In the form of a sequence of adjacent pulse intervals, In order to be the pulse sequence number, For the total number of pulse sequences, the adjacent pulse interval sequences are: ; Is provided with Represents the median, the absolute deviation of the median is , Is based on The standard deviation 1.4826 of the generated pulse data is a normalized constant under normal distribution, , The abnormal interval is identified according to the following criteria: 。
10. The radar signal main sorting system based on peak clustering and optimized pulse searching is characterized by comprising a multi-order differential feature accumulation unit, a self-adaptive density and distance calculation unit, a cluster center automatic identification unit, an interference point rejection unit, a PRI estimation unit, an optimized pulse searching unit and an iteration updating unit; The multi-order differential feature accumulation unit is used for sequentially reading the staggered pulse arrival time sequence, sequentially calculating a plurality of orders of differential, and accumulating differential results of all the orders into a mixed feature vector to obtain a PRI sample; The self-adaptive density and distance calculation unit is used for determining a neighborhood radius by combining the distribution characteristics of PRI samples based on a linear weighted neighborhood density estimation method, respectively calculating the local density value of each PRI sample and the distance between each PRI sample and a neighbor with higher density, and representing the aggregation and separation degree of the PRI samples in a density and distance two-dimensional space; The cluster center automatic identification unit is used for evaluating the clustering results of different cluster numbers by utilizing the Karnsiki-Harabase index, determining the optimal cluster number, and identifying PRI samples meeting the set density and the set distance in the two-dimensional space of the density and the distance as cluster centers; The interference point removing unit is used for normalizing the density value of the PRI samples, determining an optimal density threshold value by using a maximum inter-class variance method, removing PRI samples with the density lower than the density threshold value, and obtaining a PRI sample set corresponding to the PRI cluster; the PRI estimation unit is used for calculating an arithmetic mean value of PRI samples in the PRI cluster according to the PRI cluster subjected to density clustering and interference point eliminating processing in the current iteration round, and taking the arithmetic mean value as a PRI estimation value of the round; The optimized pulse searching unit is used for constructing a dynamic searching window by taking the PRI estimated value as the center, executing sliding matching operation in the pulse arrival time sequence, sequencing candidate pulses according to the time error of the candidate pulses at the position matched with the theory, selecting the pulse with the smallest error for matching, and eliminating PRI samples with the error exceeding a set threshold value to obtain a target pulse sequence; and the iteration updating unit is used for removing the target pulse sequence from the pulse arrival time sequence, re-extracting the target pulse sequence according to the residual pulse until the quantity of the residual pulse arrival time sequence meets the set condition, and finishing sorting.

Description

Main radar signal sorting method and system based on peak clustering and optimized pulse searching Technical Field The invention belongs to the technical field of radar signal processing, and particularly relates to a radar signal main sorting method and system based on peak clustering and optimized pulse searching. Background An electronic support system (Electronic Support Measures, ESM for short) is faced with a large number of closely spaced and complex overlapping pulse signal streams from multiple source, multiple type radar systems, requiring high precision sorting of the radar pulse signals. Pulse repetition interval (Pulse Repetition Interval, PRI) is a primary sorting parameter characterizing the periodic nature of radar pulses, and is typically calculated from a sequence of pulse Arrival Times (TOAs). Existing PRI-based main sorting methods generally include two core steps, PRI estimation and pulse search. In practical application, however, an ESM system is often in a non-ideal electromagnetic environment such as high noise, pulse loss, false pulse interference and the like, and the robustness and adaptability of the method are poor, so that false PRI peaks or missing real PRIs are easy to generate, the sorting precision is reduced, and the stability is insufficient. With the development of an unsupervised learning algorithm, a cluster analysis method is introduced into the PRI sorting field, for example, a PRI sorting algorithm combining DBSCAN (Density-Based Spatial Clustering of Applications with Noise, density-based noisy application spatial clustering algorithm) Density clustering and curve fitting search strategies shows better sorting performance in a high-noise background than a traditional method, but the method relies on sensitivity setting of super-parameters such as neighborhood radius, minimum point number and the like, has limited adaptability to pulse distribution Density change, and is difficult to meet the steady sorting requirement in a complex electromagnetic environment. The density peak clustering (DENSITY PEAK Clustering, DPC for short) is based on a cluster center discrimination criterion with high density and long distance, has the capability of processing non-uniform density distribution, but the application of the method in an actual PRI sorting task still faces the following challenges that firstly, the density estimation is not robust enough, the traditional DPC relies on hard threshold counting with fixed cut-off distance to carry out local density estimation, the local density estimation is easy to deviate from a real aggregation structure under the non-uniform density distribution, so that cluster division errors are caused, pseudo PRI is introduced, sorting accuracy is influenced, secondly, the cluster center selection relies on manual judgment, the DPC algorithm needs to carry out manual cluster center selection through observing density and a distance decision diagram, subjectivity is strong, efficiency is low, the method is difficult to adapt to a scene with unknown cluster number or real-time response, thirdly, the pseudo cluster interference is serious, and if the pseudo cluster interference is not restrained, PRI estimation and pulse searching process are easy to be led, and sorting accuracy is influenced. In addition, the traditional pulse search based on PRI estimation value adopts a global traversal strategy with fixed step length, and under the background of high-density pulse flow, the strategy has certain limitation that on one hand, pulse loss or noise interference is easy to cause path offset and accumulate errors, and the sorting accuracy is reduced, on the other hand, the calculation resources required by global traversal are larger, the processing efficiency is lower, and the real-time processing requirement of large-scale pulse data in a complex environment is difficult to meet. In a word, the existing PRI-based radar signal sorting method generally depends on a fixed threshold or preset parameters, when facing complex electromagnetic environments such as high noise, pulse loss and the like, false PRI peaks or missing real PRIs are easy to generate, so that the accuracy and stability of sorting results are greatly reduced, the existing method lacks self-adaptive recognition capability on pulse data distribution characteristics (such as local density and cluster center position), the real PRIs are difficult to accurately extract from the internal structure of data, particularly when pulse streams show non-uniform density distribution, the sorting effect is further limited, in addition, the related pulse searching strategy mostly adopts a global fixed step traversing mode, the calculation efficiency is low, errors are easy to accumulate when the interference is serious or the data scale is large, and stable and reliable sorting performance is difficult to maintain in the complex electromagnetic environments. Disclosure of Invention In order to solve