A radar individual sorting method, device, storage medium and equipment

Abstract

The application discloses a radar individual sorting method and device, a storage medium and equipment. The method comprises the following steps: first, dividing a radar pulse sequence into P beats; obtaining a feature parameter vector of a preset dimension of each pulse description word in the nth beat; then, performing clustering processing to obtain an abnormal clustering cluster and M effective clustering clusters contained in the nth beat; then, constructing a mixed Gaussian model corresponding to each of the M effective clustering clusters; selecting a target abnormal pulse description word from the abnormal clustering cluster in sequence to calculate a maximum likelihood score of the target abnormal pulse description word in the M mixed Gaussian models; then, selecting a maximum score and judging whether the maximum score is higher than a preset first threshold; if yes, sorting the target abnormal pulse description word into an effective clustering cluster corresponding to the maximum score; if no, sorting the target abnormal pulse description word into the abnormal clustering cluster; and the same is repeated until sorting results of all pulse description words in the P beats are obtained, so that the sorting accuracy is effectively improved.

Description

Technical Field

[0001] This application relates to the field of radar technology, and in particular to a radar individual sorting method, apparatus, storage medium and device. Background Technology

[0002] With the increasing sophistication of radar systems, accurately classifying signals belonging to the same radiation source is crucial for further processing of radar pulses from different sources. Individual radar sorting is a key step in radar signal processing, referring to the method of separating individual radar source pulse sequences from a randomly interleaved stream of radar pulse description words.

[0003] Typical radar individual sorting methods include template matching and model-driven methods based on intra-pulse modulation characteristics. However, these traditional radar individual sorting methods have two problems: First, because radars have several relatively fixed operating modes or states, the conventional parameters used for sorting, such as pulse arrival time (TOA), carrier frequency (RF), pulse width (PW), pulse amplitude (PA), and pulse direction of arrival (DOA), usually contain redundant information. Second, anomalous radar pulse descriptors sorted using traditional methods are usually discarded, but these discarded pulse descriptors may contain valuable radar signals. This results in inaccurate final radar individual sorting results. Summary of the Invention

[0004] The main objective of this application is to provide a radar individual sorting method, apparatus, storage medium, and device that can effectively improve the accuracy of sorting results when performing radar individual sorting.

[0005] This application provides a radar individual sorting method, including:

[0006] The radar pulse sequence to be sorted is divided into P beats; and in the nth beat of the P beats, the feature parameter vector of a preset dimension corresponding to each pulse descriptor is obtained; where P is a positive integer greater than 0; and n is a positive integer greater than 0 and not less than P.

[0007] Clustering is performed on the feature parameter vector of a preset dimension corresponding to each pulse descriptor in the nth beat to obtain the abnormal cluster and M valid clusters contained in the nth beat; where M is a positive integer greater than 0.

[0008] Construct Gaussian mixture models corresponding to the M effective clusters respectively; and sequentially select target abnormal pulse descriptors from the abnormal clusters to calculate the maximum likelihood scores of the feature parameter vectors of the preset dimensions corresponding to the target abnormal pulse descriptors in the Gaussian mixture models corresponding to the M effective clusters respectively.

[0009] The maximum score is selected from all the maximum likelihood scores, and it is determined whether the maximum score is higher than a first preset threshold. If so, the target abnormal pulse descriptor is sorted into the effective cluster corresponding to the maximum score. If not, the target abnormal pulse descriptor is still sorted into the abnormal cluster, and so on, until the individual sorting result of each pulse descriptor in the P beats is obtained, which is taken as the final sorting result corresponding to the radar pulse sequence.

[0010] In one possible implementation, after dividing the radar pulse sequence to be sorted into P beats and obtaining the feature parameter vector of a preset dimension corresponding to each pulse descriptor in the nth beat of the P beats, the method further includes:

[0011] The mean and variance of the feature parameters corresponding to each pulse descriptor in the nth beat are normalized to obtain the normalized feature parameter vector corresponding to each pulse descriptor in the nth beat.

[0012] The clustering process performed on the feature parameter vector of a preset dimension corresponding to each pulse descriptor within the nth beat yields the abnormal clusters and M valid clusters contained in the nth beat, including:

[0013] Clustering is performed on the normalized feature parameter vector corresponding to each pulse descriptor in the nth beat to obtain the abnormal cluster and M valid clusters contained in the nth beat.

[0014] In one possible implementation, after performing clustering processing on the feature parameter vector of a preset dimension corresponding to each pulse descriptor within the nth beat to obtain the abnormal clusters and M valid clusters contained in the nth beat, the method further includes:

[0015] Select target pulse descriptors sequentially from each valid cluster and calculate the Euclidean distance between the target pulse descriptor and other pulse descriptors in its own valid cluster.

[0016] When it is determined that the Euclidean distance is less than the second preset threshold, the target pulse descriptor is taken as the representative pulse descriptor of its effective cluster; and so on, until the K representative pulse descriptors contained in each of the M effective clusters are obtained; where K is a positive integer greater than 0;

[0017] Calculate the intra-class scatter matrix and inter-class scatter matrix for each of the K representative pulse descriptors in the M effective clusters;

[0018] Calculate the product of the inverse of the inter-class scatter matrix and the intra-class scatter matrix, and perform eigenvalue decomposition on the product result to determine the projection matrix based on the decomposition result;

[0019] Each pulse descriptor in the nth beat is feature-projected according to the projection matrix to obtain the projected parameter vector.

[0020] In one possible implementation, the method further includes:

[0021] Clustering is performed on the projected parameter vector corresponding to each pulse descriptor in the nth beat to obtain the target anomaly cluster and N valid clusters contained in the nth beat; where N is a positive integer greater than 0.

[0022] In one possible implementation, the step of constructing Gaussian mixture models corresponding to the M effective clusters and sequentially selecting target anomalous impulse descriptors from the anomalous clusters to calculate the maximum likelihood scores of the feature parameter vectors of a preset dimension corresponding to the target anomalous impulse descriptors in the Gaussian mixture models corresponding to the M effective clusters includes:

[0023] Construct Gaussian mixture models corresponding to the N effective clusters respectively; and sequentially select the abnormal pulse descriptors to be sorted from the target abnormal clusters, so as to calculate the maximum likelihood scores of the projected parameter vectors corresponding to the abnormal pulse descriptors to be sorted in the Gaussian mixture models corresponding to the N effective clusters respectively.

[0024] In one possible implementation, the step of selecting the maximum score from all the maximum likelihood scores and determining whether the maximum score is higher than a preset first threshold; if so, the target anomalous pulse descriptor is sorted into the effective cluster corresponding to the maximum score; if not, the target anomalous pulse descriptor is still sorted into the anomalous cluster, and so on, until the individual sorting result of each pulse descriptor in the P beats is obtained, which is taken as the final sorting result corresponding to the radar pulse sequence, includes:

[0025] The maximum score to be sorted is selected from all the maximum likelihood scores to be sorted, and it is determined whether the maximum score to be sorted is higher than a first preset threshold. If so, the abnormal pulse descriptor to be sorted is sorted into the effective cluster corresponding to the maximum score to be sorted. If not, the abnormal pulse descriptor to be sorted is still sorted into the abnormal cluster, and so on, until the individual sorting result of each pulse descriptor in the P beats is obtained, which is taken as the final sorting result corresponding to the radar pulse sequence.

[0026] In one possible implementation, after performing clustering processing on the feature parameter vector of a preset dimension corresponding to each pulse descriptor within the nth beat to obtain the abnormal clusters and M valid clusters contained in the nth beat, the method further includes:

[0027] The abnormal clusters and M valid clusters contained in the nth beat are labeled to obtain the respective identifiers of the abnormal clusters and M valid clusters contained in the nth beat.

[0028] In one possible implementation, after obtaining the individual sorting result of each pulse descriptor in the P beats as the final sorting result corresponding to the radar pulse sequence, the method further includes:

[0029] Starting from the first beat of the P beats, the Euclidean distance is calculated between the central feature parameter of any effective cluster in the i-th beat and the central feature parameter of any effective cluster in the (i+1)-th beat, and it is determined whether the minimum Euclidean distance obtained is less than the third preset threshold.

[0030] If yes, then change the identifier of the effective cluster corresponding to the minimum Euclidean distance in the (i+1)th beat to the identifier of the effective cluster corresponding to the minimum Euclidean distance in the ith beat; otherwise, do not change the identifier of the effective cluster.

[0031] Where i is a positive integer greater than 0 and less than P.

[0032] This application also provides a radar individual sorting device, including:

[0033] The acquisition unit is used to divide the radar pulse sequence to be sorted into P beats; and in the nth beat of the P beats, acquire the feature parameter vector of a preset dimension corresponding to each pulse descriptor; where P is a positive integer greater than 0; and n is a positive integer greater than 0 and not less than P.

[0034] The first clustering unit is used to perform clustering processing on the feature parameter vector of a preset dimension corresponding to each pulse descriptor in the nth beat, to obtain the abnormal clusters and M valid clusters contained in the nth beat; where M is a positive integer greater than 0.

[0035] The first calculation unit is used to construct Gaussian mixture models corresponding to the M effective clusters respectively; and to select target abnormal pulse descriptors from the abnormal clusters in turn, so as to calculate the maximum likelihood scores of the feature parameter vectors of the preset dimension corresponding to the target abnormal pulse descriptors in the Gaussian mixture models corresponding to the M effective clusters respectively.

[0036] The sorting unit is used to select the maximum score from all the maximum likelihood scores and determine whether the maximum score is higher than a preset first threshold. If so, the target abnormal pulse descriptor is sorted into the effective cluster corresponding to the maximum score. If not, the target abnormal pulse descriptor is still sorted into the abnormal cluster, and so on, until the individual sorting result of each pulse descriptor in the P beats is obtained, which is used as the final sorting result corresponding to the radar pulse sequence.

[0037] In one possible implementation, the device further includes:

[0038] The normalization unit is used to normalize the mean and variance of the feature parameters of a preset dimension corresponding to each pulse descriptor in the nth beat, so as to obtain the normalized feature parameter vector corresponding to each pulse descriptor in the nth beat.

[0039] The first clustering unit is specifically used for:

[0040] Clustering is performed on the normalized feature parameter vector corresponding to each pulse descriptor in the nth beat to obtain the abnormal cluster and M effective clusters contained in the nth beat.

[0041] In one possible implementation, the device further includes:

[0042] The second calculation unit is used to sequentially select target pulse descriptors from each effective cluster and calculate the Euclidean distance between the target pulse descriptor and other pulse descriptors in its effective cluster.

[0043] The obtaining unit is used to, when it is determined that the Euclidean distance is less than a second preset threshold, take the target pulse descriptor as a representative pulse descriptor within its effective cluster; and so on, until the K representative pulse descriptors contained in each of the M effective clusters are obtained; where K is a positive integer greater than 0;

[0044] The third calculation unit is used to calculate the intra-class scatter matrix and inter-class scatter matrix of K representative pulse descriptors in the M effective clusters, respectively.

[0045] The fourth calculation unit is used to calculate the product of the inverse of the inter-class scatter matrix and the intra-class scatter matrix, and to perform eigenvalue decomposition on the product result in order to determine the projection matrix based on the decomposition result.

[0046] The projection unit is used to project each pulse descriptor in the nth beat according to the projection matrix to obtain the projected parameter vector.

[0047] In one possible implementation, the device further includes:

[0048] The second clustering unit is used to perform clustering processing on the projected parameter vector corresponding to each pulse descriptor in the nth beat, to obtain the target abnormal cluster and N effective clusters contained in the nth beat; where N is a positive integer greater than 0.

[0049] In one possible implementation, the first computing unit is specifically used for:

[0050] Construct Gaussian mixture models corresponding to the N effective clusters respectively; and sequentially select the abnormal pulse descriptors to be sorted from the target abnormal clusters, so as to calculate the maximum likelihood scores of the projected parameter vectors corresponding to the abnormal pulse descriptors to be sorted in the Gaussian mixture models corresponding to the N effective clusters respectively.

[0051] In one possible implementation, the sorting unit is specifically used for:

[0052] The maximum score to be sorted is selected from all the maximum likelihood scores to be sorted, and it is determined whether the maximum score to be sorted is higher than a first preset threshold. If so, the abnormal pulse descriptor to be sorted is sorted into the effective cluster corresponding to the maximum score to be sorted. If not, the abnormal pulse descriptor to be sorted is still sorted into the abnormal cluster, and so on, until the individual sorting result of each pulse descriptor in the P beats is obtained, which is taken as the final sorting result corresponding to the radar pulse sequence.

[0053] In one possible implementation, the device further includes:

[0054] The annotation unit is used to annotate the abnormal clusters and M valid clusters contained in the nth beat, and obtain the respective identifiers of the abnormal clusters and M valid clusters contained in the nth beat.

[0055] In one possible implementation, the device further includes:

[0056] The fifth calculation unit is used to start from the first beat of the P beats, sequentially select the center feature parameter of any effective cluster in the i-th beat and the center feature parameter of any effective cluster in the (i+1)-th beat to calculate the Euclidean distance, and determine whether the obtained minimum Euclidean distance is less than the third preset threshold.

[0057] The modification unit is used to change the identifier of the effective cluster corresponding to the minimum Euclidean distance in the (i+1)th beat to the identifier of the effective cluster corresponding to the minimum Euclidean distance in the ith beat if it is determined that the obtained minimum Euclidean distance is less than a third preset threshold; otherwise, the identifier of the effective cluster is not changed.

[0058] Where i is a positive integer greater than 0 and less than P.

[0059] This application also provides a radar individual sorting device, including: a processor, a memory, and a system bus;

[0060] The processor and the memory are connected via the system bus;

[0061] The memory is used to store one or more programs, the one or more programs including instructions, which, when executed by the processor, cause the processor to perform any of the above-described implementations of the radar individual sorting method.

[0062] This application also provides a computer-readable storage medium storing instructions that, when executed on a terminal device, cause the terminal device to perform any of the above-described radar individual sorting methods.

[0063] This application also provides a computer program product, which, when run on a terminal device, causes the terminal device to execute any of the above-described radar individual sorting methods.

[0064] This application provides a radar individual sorting method, apparatus, storage medium, and device. First, the radar pulse sequence to be sorted is divided into P beats. Then, within the nth beat of the P beats, a feature parameter vector of a preset dimension corresponding to each pulse descriptor is obtained; where P is a positive integer greater than 0; and n is a positive integer greater than 0 and not less than P. Next, the feature parameter vector of the preset dimension corresponding to each pulse descriptor within the nth beat is clustered to obtain anomaly clusters and M valid clusters contained in the nth beat; where M is a positive integer greater than 0. Then, Gaussian mixture models corresponding to the M valid clusters are constructed respectively. Target anomalous pulse descriptors are selected sequentially from the anomalous clusters. The maximum likelihood scores of the feature parameter vectors of the preset dimension corresponding to the target anomalous pulse descriptors in the Gaussian mixture models corresponding to the M effective clusters are calculated. Then, the maximum score is selected from all the maximum likelihood scores, and it is determined whether the maximum score is higher than a preset first threshold. If so, the target anomalous pulse descriptor is selected as the effective cluster corresponding to the maximum score. If not, the target anomalous pulse descriptor is still selected as an anomalous cluster, and so on, until the individual selection result of each pulse descriptor in these P beats is obtained, which is taken as the final selection result corresponding to the radar pulse sequence.

[0065] As can be seen, since this application uses the maximum likelihood scoring method based on the Gaussian mixture model to perform secondary sorting on the abnormal pulse descriptors sorted by the existing methods, it can further reduce the probability of missed alarms to a certain extent and promote the radar individual sorting of pulse descriptors with weak distinguishability, thereby effectively improving the accuracy of the sorting results. Attached Figure Description

[0066] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0067] Figure 1 A schematic flowchart illustrating a radar individual sorting method provided in an embodiment of this application;

[0068] Figure 2 This is a schematic diagram illustrating the process of compensating for pulse descriptor information provided in an embodiment of this application;

[0069] Figure 3 A schematic diagram illustrating the process of "information compensation for pulse descriptors" and the compensated "secondary sorting of abnormal pulse descriptors" provided in the embodiments of this application;

[0070] Figure 4 This is a schematic diagram of the composition of a radar individual sorting device provided in an embodiment of this application. Detailed Implementation

[0071] Radar individual sorting technology typically refers to the use of signal processing techniques to measure characteristic parameters, analyze electromagnetic features, and identify acquired radar radiation source signals. This allows for the differentiation of target radar radiation source systems and models, providing a basis for high-level information fusion, situational awareness, and threat assessment. For example, a radar signal receiver may simultaneously receive randomly interleaved radar pulse descriptors from regions A, B, and C within a given period. These radar pulse descriptors can usually be characterized using conventional feature parameters such as TOA, RF, PW, PA, and DOA. Therefore, accurately separating the pulse sequences from these randomly interleaved radar pulse descriptor streams from each radar radiation source (e.g., regions A, B, and C) is particularly important.

[0072] For target sorting, traditional radar individual sorting typically employs model-driven sorting methods. Typical sorting schemes include template matching and intra-pulse modulation feature-based sorting methods. The specific implementation processes of these two sorting methods are as follows:

[0073] I. Template matching method:

[0074] Template matching requires prior collection of key characteristic parameters of relevant radar sources, such as pulse arrival time (TOA), carrier frequency (RF), pulse width (PW), pulse amplitude (PA), and pulse direction of arrival (DOA), and storing this information in a database. By matching the five-dimensional conventional parameters (TOA, RF, PW, PA, DOA) in the pulse descriptor of the received signal with the parameters stored in the template database, pulse descriptors with the same or similar parameter values ​​are selected for sorting. This method is relatively simple to operate and suitable for scenarios with few radar sources. However, with the rapid development of radar signal transmission and modulation technologies, radar systems are extremely diverse and complex, posing a significant challenge to the completeness of the template database.

[0075] II. Sorting methods based on intra-pulse modulation characteristics:

[0076] With the increasing sophistication of radar systems, the proportion of conventional pulse radar signals in the electromagnetic environment is gradually decreasing, while various modulation signals such as linear frequency modulation (LFM), non-linear frequency modulation (NLFM), and phase coding are increasing. The increasingly complex electromagnetic environment continuously presents numerous severe challenges to radar radiation source signal sorting methods based on inter-pulse modulation characteristics.

[0077] In traditional radar individual sorting, the most crucial step is extracting effective feature parameters that reflect the individual fingerprint of the radar emitter. However, due to the complex generation mechanism of radar emitter fingerprints and the lack of a precise physical definition, they cannot be accurately modeled and expressed using mathematical tools. Furthermore, radar emitter fingerprints are unintentional modulations attached to radio signals, with weaker energy compared to carrier signals, making them particularly susceptible to interference from complex channel conditions, multipath effects, and environmental noise. These issues typically prevent the extraction of true radar emitter fingerprints. Therefore, it is necessary to extract effective feature parameters from multiple perspectives and then perform weighted fusion at the feature level to obtain a more accurate, unique, independent, and stable radar emitter fingerprint. However, this feature fusion method increases the complexity of radar individual sorting technology, requiring significant hardware and software support, which hinders widespread application.

[0078] It is evident that with the increasing complexity of the current electromagnetic environment and radar systems, relying solely on measuring the five-dimensional parameters in the radar pulse descriptor word is no longer sufficient to meet the sorting requirements. The overlapping numerical ranges and complex and variable parameter periods make it impossible for traditional model-driven sorting methods to fit similar feature patterns, resulting in low sorting accuracy.

[0079] Currently, radar individual sorting methods for pulse descriptors typically construct feature vectors from the five-dimensional conventional parameters of each radar pulse signal's pulse descriptor, and then perform cluster analysis. Density-Based Spatial Clustering of Applications with Noise (DBSCAN) is commonly used for clustering. This clustering method is a density-based spatial clustering approach that leverages the high-density connectivity of clusters to quickly discover clusters of arbitrary shapes, offering significant advantages for spatial data clustering. After this clustering analysis, most radar pulse descriptors are effectively sorted, and the sorted radar individuals can be labeled. However, some radar pulse descriptors are not sorted. These radar pulse descriptors are considered unsustainable for radar individual information or useless for sorting and are actively discarded.

[0080] However, existing radar individual sorting methods based on five-dimensional conventional parameters in pulse descriptors often contain redundant information. This is because radars have relatively fixed operating modes or states, and these parameters typically contain redundant information. For example, carrier frequency (RF), pulse width (PW), and pulse amplitude (PA) may have certain correlations on a given radar individual, potentially interfering with the subsequent sorting process and reducing the accuracy of the sorting results. Furthermore, anomalous radar pulse descriptors sorted using traditional methods are usually discarded. However, these discarded pulse descriptors may contain valuable radar signals, such as those from narrow-beam high-speed radars or low probability of intercept (LPI) radars. While these signals may not belong to typical radar systems, they may employ novel modulation or transmission methods. Therefore, it is crucial to pay attention to this type of data and extract valuable radar signals to provide a basis for subsequent high-level information fusion, situational awareness, and threat assessment. Thus, effectively improving the accuracy of radar individual sorting is a pressing technical problem that needs to be solved.

[0081] To address the aforementioned shortcomings, this application provides a radar individual sorting method. First, the radar pulse sequence to be sorted is divided into P beats. Then, within the nth beat of the P beats, a feature parameter vector of a preset dimension corresponding to each pulse descriptor is obtained; where P is a positive integer greater than 0, and n is a positive integer greater than 0 and not less than P. Next, the feature parameter vector of the preset dimension corresponding to each pulse descriptor within the nth beat is clustered to obtain anomaly clusters and M valid clusters contained in the nth beat; where M is a positive integer greater than 0. Then, Gaussian mixture models corresponding to the M valid clusters are constructed respectively; and sequentially... Target anomalous pulse descriptors are selected from the anomalous clusters. The maximum likelihood scores of the feature parameter vectors of the preset dimension corresponding to the target anomalous pulse descriptors in the Gaussian mixture models corresponding to the M effective clusters are calculated. Then, the maximum score is selected from all the maximum likelihood scores, and it is determined whether the maximum score is higher than a preset first threshold. If so, the target anomalous pulse descriptor is selected as the effective cluster corresponding to the maximum score. If not, the target anomalous pulse descriptor is still selected as an anomalous cluster, and so on, until the individual selection result of each pulse descriptor in these P beats is obtained, which is taken as the final selection result corresponding to the radar pulse sequence.

[0082] As can be seen, since this application uses the maximum likelihood scoring method based on the Gaussian mixture model to perform secondary sorting on the abnormal pulse descriptors sorted by the existing methods, it can further reduce the probability of missed alarms to a certain extent and promote the radar individual sorting of pulse descriptors with weak distinguishability, thereby effectively improving the accuracy of the sorting results.

[0083] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0084] First Embodiment

[0085] See Figure 1 This is a flowchart illustrating a radar individual sorting method provided in this embodiment. The method includes the following steps:

[0086] S101: Divide the radar pulse sequence to be sorted into P beats; and in the nth beat of the P beats, obtain the feature parameter vector of the preset dimension corresponding to each pulse descriptor; where P is a positive integer greater than 0; and n is a positive integer greater than 0 and not less than P.

[0087] In this embodiment, any radar pulse sequence that needs to be individually sorted is defined as the radar pulse sequence to be sorted. It is understood that a radar pulse sequence is typically composed of a radar pulse descriptor word stream, and each pulse descriptor word in the radar pulse sequence is typically characterized by conventional parameters such as pulse time of arrival (TOA), carrier frequency (RF), pulse width (PW), pulse amplitude (PA), and pulse direction of arrival (DOA). It should be noted that this embodiment does not limit the length of the radar pulse sequence to be sorted, that is, it does not limit the number of pulse descriptor words contained in the pulse sequence, nor does it limit the method of obtaining the radar pulse sequence to be sorted.

[0088] In practical applications, after acquiring the radar pulse sequence to be sorted, since it usually contains a large number of pulse descriptors, in order to improve the efficiency and accuracy of the sorting results when sorting individual radars, the radar pulse sequence to be sorted can first be divided into P beats according to the pulse arrival time (TOA), where P is a positive integer greater than 0. For example, it can be divided into a beat every 2 seconds according to the pulse arrival time (TOA). Each beat contains several pulse descriptors characterized by conventional parameters such as pulse arrival time (TOA), carrier frequency (RF), pulse width (PW), pulse amplitude (PA), and pulse direction of arrival (DOA).

[0089] Furthermore, it should be noted that this application does not limit the representation method corresponding to the pulse descriptor. Each pulse descriptor can be represented by a feature parameter vector of a corresponding preset dimension according to the number of preset parameters. For example, if the preset setting is to represent a pulse descriptor by five conventional feature parameters: pulse arrival time (TOA), carrier frequency (RF), pulse width (PW), pulse amplitude (PA), and pulse arrival direction (DOA), then the preset dimension of the feature parameter vector corresponding to each pulse descriptor is 5-dimensional. In this way, after obtaining the 5-dimensional feature parameter vector corresponding to each pulse descriptor in the nth (n is a positive integer greater than 0 and not less than P) of P beats, that is, after obtaining the feature parameter vector of the preset dimension corresponding to all pulse descriptors in P beats, it can be defined as s, that is, s = [TOA, RF, PW, PA, DOA], to continue to execute the subsequent steps S102-S104 to achieve accurate sorting of all pulse descriptors in these P beats.

[0090] S102: Perform clustering processing on the feature parameter vector of the preset dimension corresponding to each pulse descriptor in the nth beat to obtain the abnormal cluster and M valid clusters contained in the nth beat; where M is a positive integer greater than 0.

[0091] In this embodiment, after obtaining the feature parameter vector of a preset dimension corresponding to each pulse descriptor in the nth beat of P beats through step S101, such as obtaining the 5-dimensional feature parameter vector corresponding to each pulse descriptor in the nth beat, in order to improve the accuracy of the sorting results, existing or future clustering algorithms can be used to cluster the feature parameter vector of the preset dimension (such as the 5-dimensional feature parameter vector) corresponding to each pulse descriptor in the nth beat. For example, the DBSCAN algorithm can be used to cluster the feature parameter vector of the preset dimension (such as the 5-dimensional feature parameter vector) corresponding to each pulse descriptor in the nth beat, obtaining the abnormal cluster and M valid clusters contained in the nth beat, which are then used to execute the subsequent step S103, where M is a positive integer greater than 0. The abnormal cluster contains pulse descriptors that have not been clustered into any valid cluster.

[0092] Specifically, one possible implementation is to obtain the feature parameter vector of a preset dimension corresponding to each pulse descriptor within the nth beat of P beats. To improve the accuracy of the sorting results, the mean and variance of the feature parameters of the preset dimension corresponding to each pulse descriptor within the nth beat can be normalized to ensure that the feature parameters of each dimension are on a uniform computational metric, thereby determining the normalized feature parameter vector corresponding to each pulse descriptor within the nth beat. Then, clustering algorithms such as DBSCAN can be used to cluster the normalized feature parameter vector corresponding to each pulse descriptor within the nth beat to obtain the abnormal clusters and M valid clusters contained in the nth beat, which are then used to execute the subsequent step S103.

[0093] In addition, an alternative implementation is to label the abnormal clusters and M valid clusters contained in the nth beat to obtain the respective identifiers of the abnormal clusters and M valid clusters contained in the nth beat. However, this application does not limit the specific content of the identifiers, as long as they can characterize the uniqueness of the abnormal clusters and the M valid clusters. For example, all pulse descriptors in the abnormal clusters can be uniformly labeled with the number "-1", while the pulse descriptors of the valid clusters can be labeled with "1", "2", "3", ..., "M" respectively according to their respective clusters.

[0094] S103: Construct Gaussian mixture models corresponding to M effective clusters respectively; and select target abnormal pulse descriptors from the abnormal clusters in turn to calculate the maximum likelihood scores of the feature parameter vectors of the preset dimensions corresponding to the target abnormal pulse descriptors in the Gaussian mixture models corresponding to the M effective clusters respectively.

[0095] In this embodiment, after obtaining the abnormal clusters and M effective clusters contained in the nth beat through step S102, in order to improve the accuracy of radar individual sorting results, it is further necessary to construct a Gaussian mixture model for each effective cluster. Then, an abnormal pulse descriptor is selected from the abnormal clusters as the target abnormal pulse descriptor, and the maximum likelihood score of the feature parameter vector of the preset dimension corresponding to the target abnormal pulse descriptor in the Gaussian mixture model corresponding to the M effective clusters is calculated. The specific calculation formula is as follows:

[0096]

[0097] Where c represents the Gaussian mixture number, which can be determined based on actual conditions and empirical values; α c μ c and ∑ cLet represent the weight, mean, and variance of the c-th Gaussian component, respectively; s represents the feature parameter vector of a preset dimension corresponding to the target abnormal pulse descriptor; D represents the dimension of s, which takes the value of 5 when s = [TOA, RF, PW, PA, DOA]; p(s|θ m ) represents the maximum likelihood score of the feature parameter vector s of a preset dimension corresponding to the target abnormal pulse descriptor in the Gaussian mixture model of the m-th effective cluster, where m∈{1,2,3,...,M}.

[0098] S104: Select the maximum score from all maximum likelihood scores and determine whether the maximum score is higher than a preset first threshold; if so, the target abnormal pulse descriptor is sorted into the effective cluster corresponding to the maximum score; if not, the target abnormal pulse descriptor is still sorted into the abnormal cluster, and so on, until the individual sorting result of each pulse descriptor in P beats is obtained, which is used as the final sorting result corresponding to the radar pulse sequence.

[0099] In this embodiment, after calculating the maximum likelihood scores of the feature parameter vectors of the preset dimension corresponding to the target abnormal pulse descriptor in the Gaussian mixture model corresponding to the M effective clusters in step S103, in order to effectively improve the accuracy of the target abnormal pulse descriptor sorting results, the maximum score can be selected from all the maximum likelihood scores and defined as max{p(s|θ)} m Then determine the maximum score max{p(s|θ)}. m If the target abnormal pulse descriptor is higher than the first preset threshold, then the target abnormal pulse descriptor can be selected as the maximum score max{p(s|θ)}. m If the target anomalous pulse descriptor is not identified, the corresponding effective cluster can be determined. If not, the target anomalous pulse descriptor can still be sorted into an anomalous cluster, and so on. By repeating the above steps, the individual sorting result of each pulse descriptor in the P beats can be determined, and it can be used as the final sorting result corresponding to the radar pulse sequence. In this way, by performing secondary sorting on the anomalous noise data (i.e., all anomalous pulse descriptors contained in the anomalous clusters) in the initial radar individual pre-sorting, the effective sorting of the radar signal that may contain high value can be achieved. That is, the easily confused and high-value effective radar pulse descriptors are extracted from the anomalous radar pulse descriptors, and finally the sorting of all radar individuals in each beat of the entire radar pulse sequence is completed, which effectively improves the accuracy of the sorting results.

[0100] The specific value of the first preset threshold can be determined based on actual conditions and experience. This application does not limit this value. For example, the first preset threshold can be set to 0.5.

[0101] Furthermore, to ensure the neatness and regularity of data identification, an optional implementation method is to, after obtaining the sorting results of all radar individuals within each beat in the radar pulse sequence, start from the first beat of these P beats, sequentially select the central feature parameter of any effective cluster within the i-th beat (i is a positive integer greater than 0 and less than P) and calculate the Euclidean distance between it and the central feature parameter of any effective cluster within the (i+1)-th beat. Then, determine whether the obtained minimum Euclidean distance is less than a third preset threshold. If so, change the identifier of the effective cluster corresponding to the minimum Euclidean distance within the (i+1)-th beat to the identifier of the effective cluster corresponding to the minimum Euclidean distance within the i-th beat; otherwise, do not change the identifier of the effective cluster, such as still marking it as "1". It should be noted that the abnormal pulse descriptor marked as an abnormal cluster ("-1") within each beat does not participate in the radar individual identifier alignment process between beats mentioned above, and is still marked as "-1".

[0102] The specific value of the third preset threshold can be determined based on actual conditions and experience. This application does not limit this value. For example, the third preset threshold can be set to 0.3.

[0103] In summary, the radar individual sorting method provided in this embodiment first divides the radar pulse sequence to be sorted into P beats; and within the nth beat of the P beats, obtains the feature parameter vector of a preset dimension corresponding to each pulse descriptor; where P is a positive integer greater than 0; n is a positive integer greater than 0 and not less than P. Then, clustering is performed on the feature parameter vector of the preset dimension corresponding to each pulse descriptor within the nth beat to obtain the abnormal cluster and M effective clusters contained in the nth beat; where M is a positive integer greater than 0. Next, Gaussian mixture models corresponding to the M effective clusters are constructed respectively; and then, from the abnormal clusters... Target anomalous pulse descriptors are selected from clusters to calculate the maximum likelihood scores of the feature parameter vectors of the preset dimension corresponding to the target anomalous pulse descriptors in the Gaussian mixture models corresponding to the M effective clusters. Then, the maximum score can be selected from all the maximum likelihood scores, and it is determined whether the maximum score is higher than a preset first threshold. If so, the target anomalous pulse descriptor is selected as the effective cluster corresponding to the maximum score. If not, the target anomalous pulse descriptor is still selected as an anomalous cluster, and so on, until the individual selection result of each pulse descriptor in these P beats is obtained, which is taken as the final selection result corresponding to the radar pulse sequence.

[0104] As can be seen, since this application uses the maximum likelihood scoring method based on the Gaussian mixture model to perform secondary sorting on the abnormal pulse descriptors sorted by the existing methods, it can further reduce the probability of missed alarms to a certain extent and promote the radar individual sorting of pulse descriptors with weak distinguishability, thereby effectively improving the accuracy of the sorting results.

[0105] Second Embodiment

[0106] It should be noted that, in order to further improve the accuracy of radar individual sorting results, this embodiment proposes a more accurate sorting method based on the first embodiment described above. Specifically, this embodiment proposes to compensate for the information of the pulse descriptors before performing secondary sorting on the abnormal pulse descriptors, so as to fully eliminate redundant information such as correlation in conventional feature parameters (such as TOA, RF, PW, PA, DOA), enhance the distinguishability of feature parameters, and obtain a more accurate feature parameter vector of preset dimensions corresponding to each pulse descriptor. Then, based on the more accurate parameter vector, each pulse descriptor is sorted to effectively improve the accuracy of the sorting results. Next, this embodiment will describe in detail the "information compensation for pulse descriptors" and the compensated "secondary sorting of abnormal pulse descriptors", which may include the following steps:

[0107] S201: Select the target pulse descriptor from each effective cluster in turn, and calculate the Euclidean distance between the target pulse descriptor and other pulse descriptors in its effective cluster.

[0108] In this embodiment, after obtaining the abnormal cluster and M valid clusters contained in the nth beat out of P beats using the above steps S101-S102, further, for each valid pulse descriptor's cluster, any pulse descriptor can be selected sequentially from each valid cluster as the target pulse descriptor, and the Euclidean distance between the target pulse descriptor and other pulse descriptors in its valid cluster can be calculated. The specific calculation formula is as follows:

[0109]

[0110] Among them, s i,n s represents a feature parameter vector of a preset dimension for the i-th pulse descriptor in the n-th valid cluster; j,n d represents the feature parameter vector of a preset dimension for the j-th pulse descriptor in the n-th valid cluster; J represents the total number of pulse descriptors contained in the n-th valid cluster; i,n This represents the Euclidean distance between the i-th impulse descriptor and the j-th impulse descriptor in the n-th valid cluster.

[0111] S202: When it is determined that the Euclidean distance is less than the second preset threshold, the target pulse descriptor is taken as the representative pulse descriptor in its effective cluster; and so on, until the K representative pulse descriptors contained in each of the M effective clusters are obtained; where K is a positive integer greater than 0.

[0112] In this embodiment, after calculating the Euclidean distance between the target pulse descriptor and other pulse descriptors within its effective cluster in step S201, it can be further determined whether the Euclidean distance is less than a second preset threshold. If so, the target pulse descriptor is used as a representative pulse descriptor within its effective cluster to more accurately describe the distribution of the cluster's characteristic parameters. Otherwise, it is not used as a representative pulse descriptor within its effective cluster. This process is repeated to sequentially determine the K (K is a positive integer greater than 0) representative pulse descriptors contained in each of the M effective clusters. In this way, the number of pulse descriptors contained in each effective cluster can be compressed to K for subsequent step 203, which helps to reduce computational complexity and improve the accuracy of the final sorting result.

[0113] The specific value of the second preset threshold can be determined based on actual conditions and experience. This application does not limit this value. For example, the second preset threshold can be set to 0.5.

[0114] S203: Calculate the intra-class scatter matrix and inter-class scatter matrix for the K representative impulse descriptors in the M effective clusters.

[0115] In this embodiment, after calculating the K representative pulse descriptors contained in each of the M effective clusters in step S202, in order to fully eliminate redundant information in the conventional feature parameters, the intra-cluster scatter matrix and inter-cluster scatter matrix of the K representative pulse descriptors in the M effective clusters can be further calculated, and defined as S respectively. b and S w The specific calculation formula is as follows:

[0116]

[0117]

[0118] Among them, s n This represents the central characteristic parameter of the nth valid cluster. This represents a feature parameter vector of a preset dimension, such as a 5-dimensional feature parameter vector [TOA, RF, PW, PA, DOA], representing the i-th representative pulse descriptor in the n-th effective cluster. This represents the mean of the central characteristic parameters of all M valid clusters.

[0119] S204: Calculate the product of the inverse of the inter-class scatter matrix and the intra-class scatter matrix, and perform eigenvalue decomposition on the product result to determine the projection matrix based on the decomposition result.

[0120] In this embodiment, step S203 calculates the intra-class scatter matrix S of K representative pulse descriptors in M ​​effective clusters. b and the inter-class scatter matrix S w Then, we can further calculate the product of the inverse of the between-class scatter matrix and the within-class scatter matrix, i.e. The product result is obtained, and the product result is decomposed along the preset conventional feature parameter directions (such as the five directions of TOA, RF, PW, PA, and DOA). The feature vectors corresponding to the largest feature values ​​in the preset directions (such as the four directions) are selected to form a projection matrix (such as a 5x4 matrix), which is defined as A, and used to execute the subsequent step S205.

[0121] S205: Project the feature of each pulse descriptor in the nth beat according to the projection matrix to obtain the projected parameter vector.

[0122] In this embodiment, after determining the projection matrix A in step S204, the feature parameter vector (e.g., s = [TOA, RF, PW, PA, DOA]) of each pulse descriptor word in the nth beat can be projected onto the projection matrix A (e.g., a 5x4 matrix) to obtain the projected parameter vector, which is defined as y, i.e., y = A. T For example, when s = [TOA, RF, PW, PA, DOA], and A is a 5x4 matrix, then y is a 4-dimensional vector matrix.

[0123] By repeating steps S201-S205, feature projection can be performed on the feature parameters of all pulse descriptors within P beats, completing information compensation for all pulse descriptors. This effectively removes redundant information such as correlations mixed in the feature parameters, enhancing their discriminative power. Furthermore, after information compensation, more accurate "secondary sorting of abnormal pulse descriptors" can be achieved, such as... Figure 3 As shown, this is to obtain more accurate sorting results.

[0124] Specifically, one possible implementation is that after obtaining the projected parameter vector y corresponding to each pulse descriptor in the nth beat out of P beats, existing or future clustering algorithms can be used to cluster the projected parameter vector y corresponding to each pulse descriptor in the nth beat. For example, the DBSCAN algorithm can be used to cluster the projected parameter vector y corresponding to each pulse descriptor in the nth beat to obtain the abnormal cluster (defined here as the target abnormal cluster) and N valid clusters contained in the nth beat, which are labeled as "1", "2", "3", ..., "N", where N is a positive integer greater than 0.

[0125] Building upon this, an alternative implementation involves constructing a Gaussian mixture model for each of the N effective clusters. Then, an anomaly pulse descriptor is selected sequentially from the target anomaly cluster as the anomaly pulse descriptor to be sorted. The maximum likelihood score of the projected parameter vector y corresponding to this anomaly pulse descriptor in the Gaussian mixture models corresponding to the N effective clusters is calculated. The specific calculation formula is as follows:

[0126]

[0127] Where c represents the Gaussian mixture number, which can be determined based on actual conditions and empirical values; α c μ c and ∑ c Let represent the weight, mean, and variance of the c-th Gaussian component, respectively; y represents the projected parameter vector corresponding to the descriptor of the abnormal pulse to be sorted; D' represents the dimension of y, and when y is a 4-dimensional vector matrix, the value of D' is 4; p(y|θ n ) represents the maximum likelihood score of the projected parameter vector y corresponding to the abnormal pulse descriptor to be sorted in the Gaussian mixture model of the nth effective cluster, where n∈{1,2,3,...N}.

[0128] Building upon this, another alternative implementation is to first select the maximum likelihood score from all the maximum likelihood scores to be sorted, and define it as max{p(y|θ)}. n Then determine the maximum score to be sorted, max{p(y|θ)}. n If the value of the abnormal pulse descriptor to be sorted is higher than the first preset threshold, then the abnormal pulse descriptor to be sorted can be sorted as the maximum sorting score max{p(y|θ)}. nIf the corresponding valid cluster is not found, the abnormal pulse descriptor to be sorted can still be sorted into an abnormal cluster. This process is repeated to determine the individual sorting result of each pulse descriptor in the P beats, which is then used as the final sorting result corresponding to the radar pulse sequence. The specific value of the first preset threshold can still be determined based on actual conditions and experience; this application does not limit this, for example, the first preset threshold can still be set to 0.5.

[0129] Furthermore, to ensure the neatness and regularity of data identification, an optional implementation method is to perform radar individual identification alignment processing on the pulse descriptors within each beat of the radar pulse sequence after obtaining the sorting results of all radar individuals within each beat. The specific processing procedure can be found in the similar description in the corresponding section of the first embodiment above, and will not be repeated here. Additionally, it should be noted that abnormal pulse descriptors marked as abnormal clusters ("-1") within each beat are still not involved in the radar individual identification alignment process between beats, and are still marked with "-1" or similar identifiers.

[0130] In summary, this embodiment first compensates for the information of the pulse descriptor to fully eliminate redundant information such as correlation in conventional feature parameters (such as TOA, RF, PW, PA, DOA), thereby enhancing the discriminativeness of the feature parameters and obtaining a more accurate feature parameter vector of preset dimensions for each pulse descriptor. Then, based on this more accurate parameter vector, the pulse descriptors are further sorted using the maximum likelihood scoring method based on the Gaussian mixture model, thereby significantly improving the accuracy of the sorting results.

[0131] Third Embodiment

[0132] This embodiment will introduce a radar individual sorting device; please refer to the above method embodiment for related content.

[0133] See Figure 4 This is a schematic diagram of the composition of a radar individual sorting device provided in this embodiment. The device 400 includes:

[0134] The acquisition unit 401 is used to divide the radar pulse sequence to be sorted into P beats; and in the nth beat of the P beats, acquire the feature parameter vector of a preset dimension corresponding to each pulse descriptor; where P is a positive integer greater than 0; and n is a positive integer greater than 0 and not less than P.

[0135] The first clustering unit 402 is used to perform clustering processing on the feature parameter vector of a preset dimension corresponding to each pulse descriptor in the nth beat, to obtain the abnormal clusters and M valid clusters contained in the nth beat; where M is a positive integer greater than 0.

[0136] The first calculation unit 403 is used to construct the Gaussian mixture model corresponding to the M effective clusters respectively; and to select the target abnormal pulse descriptor from the abnormal clusters in turn, so as to calculate the maximum likelihood score of the feature parameter vector of the preset dimension corresponding to the target abnormal pulse descriptor in the Gaussian mixture model corresponding to the M effective clusters respectively.

[0137] The sorting unit 404 is used to select the maximum score from all the maximum likelihood scores and determine whether the maximum score is higher than a preset first threshold. If so, the target abnormal pulse descriptor is sorted into the effective cluster corresponding to the maximum score. If not, the target abnormal pulse descriptor is still sorted into the abnormal cluster, and so on, until the individual sorting result of each pulse descriptor in the P beats is obtained, which is used as the final sorting result corresponding to the radar pulse sequence.

[0138] In one implementation of this embodiment, the apparatus further includes:

[0139] The normalization unit is used to normalize the mean and variance of the feature parameters of a preset dimension corresponding to each pulse descriptor in the nth beat, so as to obtain the normalized feature parameter vector corresponding to each pulse descriptor in the nth beat.

[0140] The first clustering unit 402 is specifically used for:

[0141] Clustering is performed on the normalized feature parameter vector corresponding to each pulse descriptor in the nth beat to obtain the abnormal cluster and M effective clusters contained in the nth beat.

[0142] In one implementation of this embodiment, the apparatus further includes:

[0143] The second calculation unit is used to sequentially select target pulse descriptors from each effective cluster and calculate the Euclidean distance between the target pulse descriptor and other pulse descriptors in its effective cluster.

[0144] The obtaining unit is used to, when it is determined that the Euclidean distance is less than a second preset threshold, take the target pulse descriptor as a representative pulse descriptor within its effective cluster; and so on, until the K representative pulse descriptors contained in each of the M effective clusters are obtained; where K is a positive integer greater than 0;

[0145] The third calculation unit is used to calculate the intra-class scatter matrix and inter-class scatter matrix of K representative pulse descriptors in the M effective clusters, respectively.

[0146] The fourth calculation unit is used to calculate the product of the inverse of the inter-class scatter matrix and the intra-class scatter matrix, and to perform eigenvalue decomposition on the product result in order to determine the projection matrix based on the decomposition result.

[0147] The projection unit is used to project each pulse descriptor in the nth beat according to the projection matrix to obtain the projected parameter vector.

[0148] In one implementation of this embodiment, the apparatus further includes:

[0149] The second clustering unit is used to perform clustering processing on the projected parameter vector corresponding to each pulse descriptor in the nth beat, to obtain the target abnormal cluster and N effective clusters contained in the nth beat; where N is a positive integer greater than 0.

[0150] In one implementation of this embodiment, the first computing unit 403 is specifically used for:

[0151] Construct Gaussian mixture models corresponding to the N effective clusters respectively; and sequentially select the abnormal pulse descriptors to be sorted from the target abnormal clusters, so as to calculate the maximum likelihood scores of the projected parameter vectors corresponding to the abnormal pulse descriptors to be sorted in the Gaussian mixture models corresponding to the N effective clusters respectively.

[0152] In one implementation of this embodiment, the sorting unit 404 is specifically used for:

[0153] The maximum score to be sorted is selected from all the maximum likelihood scores to be sorted, and it is determined whether the maximum score to be sorted is higher than a first preset threshold. If so, the abnormal pulse descriptor to be sorted is sorted into the effective cluster corresponding to the maximum score to be sorted. If not, the abnormal pulse descriptor to be sorted is still sorted into the abnormal cluster, and so on, until the individual sorting result of each pulse descriptor in the P beats is obtained, which is taken as the final sorting result corresponding to the radar pulse sequence.

[0154] In one implementation of this embodiment, the apparatus further includes:

[0155] The annotation unit is used to annotate the abnormal clusters and M valid clusters contained in the nth beat, and obtain the respective identifiers of the abnormal clusters and M valid clusters contained in the nth beat.

[0156] In one implementation of this embodiment, the apparatus further includes:

[0157] The fifth calculation unit is used to start from the first beat of the P beats, sequentially select the center feature parameter of any effective cluster in the i-th beat and the center feature parameter of any effective cluster in the (i+1)-th beat to calculate the Euclidean distance, and determine whether the obtained minimum Euclidean distance is less than the third preset threshold.

[0158] The modification unit is used to change the identifier of the effective cluster corresponding to the minimum Euclidean distance in the (i+1)th beat to the identifier of the effective cluster corresponding to the minimum Euclidean distance in the ith beat if it is determined that the obtained minimum Euclidean distance is less than a third preset threshold; otherwise, the identifier of the effective cluster is not changed.

[0159] Where i is a positive integer greater than 0 and less than P.

[0160] Furthermore, embodiments of this application also provide a radar individual sorting device, including: a processor, a memory, and a system bus;

[0161] The processor and the memory are connected via the system bus;

[0162] The memory is used to store one or more programs, the one or more programs including instructions, which, when executed by the processor, cause the processor to perform any of the above-described implementations of the radar individual sorting method.

[0163] Furthermore, embodiments of this application also provide a computer-readable storage medium storing instructions that, when executed on a terminal device, cause the terminal device to perform any of the above-described implementations of the radar individual sorting method.

[0164] Furthermore, this application embodiment also provides a computer program product, which, when run on a terminal device, causes the terminal device to execute any of the above-described implementation methods of the radar individual sorting method.

[0165] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that all or part of the steps in the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network communication device such as a media gateway, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0166] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0167] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0168] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A radar individual sorting method, characterized in that, include: The radar pulse sequence to be sorted is divided into P beats; And within the nth beat of the P beats, obtain the 5-dimensional feature parameter vector corresponding to each pulse descriptor; where P is a positive integer greater than 0; and n is a positive integer greater than 0 and not less than P; Clustering is performed on the 5-dimensional feature parameter vector corresponding to each pulse descriptor in the nth beat to obtain the abnormal cluster and M valid clusters contained in the nth beat; where M is a positive integer greater than 0. Construct Gaussian mixture models corresponding to the M effective clusters respectively; and select target abnormal pulse descriptors from the abnormal clusters in turn, and calculate the maximum likelihood scores of the 5-dimensional feature parameter vectors corresponding to the target abnormal pulse descriptors in the Gaussian mixture models corresponding to the M effective clusters respectively. Select the maximum score from all the maximum likelihood scores, and determine whether the maximum score is higher than a first preset threshold; If yes, the target anomalous pulse descriptor is sorted into the effective cluster corresponding to the maximum score; if no, the target anomalous pulse descriptor is still sorted into the anomalous cluster, and so on, until the individual sorting result of each pulse descriptor in the P beats is obtained, which is taken as the final sorting result corresponding to the radar pulse sequence. After performing clustering processing on the 5-dimensional feature parameter vector corresponding to each pulse descriptor within the nth beat to obtain the abnormal clusters and M valid clusters contained in the nth beat, the method further includes: Target pulse descriptors are selected sequentially from each valid cluster, and the Euclidean distance between the target pulse descriptor and other pulse descriptors in its own valid cluster is calculated. When it is determined that the Euclidean distance is less than the second preset threshold, the target pulse descriptor is taken as the representative pulse descriptor of its effective cluster; and so on, until the K representative pulse descriptors contained in each of the M effective clusters are obtained; where K is a positive integer greater than 0; Calculate the intra-class scatter matrix and inter-class scatter matrix for each of the K representative pulse descriptors in the M effective clusters; Calculate the product of the inverse of the inter-class scatter matrix and the intra-class scatter matrix, and perform eigenvalue decomposition on the product result to determine the projection matrix based on the decomposition result; Each pulse descriptor in the nth beat is feature-projected according to the projection matrix to obtain the projected parameter vector; Clustering is performed on the projected parameter vector corresponding to each pulse descriptor in the nth beat to obtain the target anomaly cluster and N valid clusters contained in the nth beat; where N is a positive integer greater than 0.

2. The method according to claim 1, characterized in that, After dividing the radar pulse sequence to be sorted into P beats, and obtaining the 5-dimensional feature parameter vector corresponding to each pulse descriptor in the nth beat of the P beats, the method further includes: The mean and variance of the 5-dimensional feature parameters corresponding to each pulse descriptor in the nth beat are normalized to obtain the normalized feature parameter vector corresponding to each pulse descriptor in the nth beat. The clustering process performed on the 5-dimensional feature parameter vector corresponding to each pulse descriptor in the nth beat yields the abnormal clusters and M valid clusters contained in the nth beat, including: Clustering is performed on the normalized feature parameter vector corresponding to each pulse descriptor in the nth beat to obtain the abnormal cluster and M valid clusters contained in the nth beat.

3. The method according to claim 1, characterized in that, The process involves constructing Gaussian mixture models corresponding to the M effective clusters, and sequentially selecting target anomalous pulse descriptors from the anomalous clusters to calculate the maximum likelihood scores of the 5-dimensional feature parameter vectors corresponding to the target anomalous pulse descriptors in the Gaussian mixture models corresponding to the M effective clusters, including: Construct Gaussian mixture models corresponding to the N effective clusters respectively; and sequentially select the abnormal pulse descriptors to be sorted from the target abnormal clusters, so as to calculate the maximum likelihood scores of the projected parameter vectors corresponding to the abnormal pulse descriptors to be sorted in the Gaussian mixture models corresponding to the N effective clusters respectively.

4. The method according to claim 3, characterized in that, The step involves selecting the maximum score from all the maximum likelihood scores and determining whether the maximum score is higher than a preset first threshold. If so, the target anomalous pulse descriptor is sorted into the effective cluster corresponding to the maximum score. If not, the target anomalous pulse descriptor is still sorted into the anomalous cluster, and so on, until the individual sorting result of each pulse descriptor in the P beats is obtained, which serves as the final sorting result corresponding to the radar pulse sequence. This includes: The maximum score to be sorted is selected from all the maximum likelihood scores to be sorted, and it is determined whether the maximum score to be sorted is higher than a first preset threshold. If so, the abnormal pulse descriptor to be sorted is sorted into the effective cluster corresponding to the maximum score to be sorted. If not, the abnormal pulse descriptor to be sorted is still sorted into the abnormal cluster, and so on, until the individual sorting result of each pulse descriptor in the P beats is obtained, which is taken as the final sorting result corresponding to the radar pulse sequence.

5. The method according to claim 1, characterized in that, After performing clustering processing on the 5-dimensional feature parameter vector corresponding to each pulse descriptor within the nth beat to obtain the abnormal clusters and M valid clusters contained in the nth beat, the method further includes: The abnormal clusters and M valid clusters contained in the nth beat are labeled to obtain the respective identifiers of the abnormal clusters and M valid clusters contained in the nth beat.

6. The method according to claim 5, characterized in that, After obtaining the individual sorting result of each pulse descriptor in the P beats as the final sorting result corresponding to the radar pulse sequence, the method further includes: Starting from the first beat of the P beats, the Euclidean distance is calculated between the central feature parameter of any effective cluster in the i-th beat and the central feature parameter of any effective cluster in the (i+1)-th beat, and it is determined whether the minimum Euclidean distance obtained is less than the third preset threshold. If yes, then change the identifier of the effective cluster corresponding to the minimum Euclidean distance in the (i+1)th beat to the identifier of the effective cluster corresponding to the minimum Euclidean distance in the ith beat; otherwise, do not change the identifier of the effective cluster. Where i is a positive integer greater than 0 and less than P.

7. A radar-based individual sorting device, characterized in that, include: The acquisition unit is used to divide the radar pulse sequence to be sorted into P beats; And within the nth beat of the P beats, obtain the 5-dimensional feature parameter vector corresponding to each pulse descriptor; where P is a positive integer greater than 0; and n is a positive integer greater than 0 and not less than P; The first clustering unit is used to perform clustering processing on the 5-dimensional feature parameter vector corresponding to each pulse descriptor in the nth beat, to obtain the abnormal clusters and M valid clusters contained in the nth beat; where M is a positive integer greater than 0. The first calculation unit is used to construct Gaussian mixture models corresponding to the M effective clusters respectively; and to select target abnormal pulse descriptors from the abnormal clusters in turn, so as to calculate the maximum likelihood scores of the 5-dimensional feature parameter vectors corresponding to the target abnormal pulse descriptors in the Gaussian mixture models corresponding to the M effective clusters respectively. The sorting unit is used to select the maximum score from all the maximum likelihood scores and determine whether the maximum score is higher than a preset first threshold. If yes, the target anomalous pulse descriptor is sorted into the effective cluster corresponding to the maximum score; if no, the target anomalous pulse descriptor is still sorted into the anomalous cluster, and so on, until the individual sorting result of each pulse descriptor in the P beats is obtained, which is taken as the final sorting result corresponding to the radar pulse sequence. The device further includes: The second calculation unit is used to sequentially select target pulse descriptors from each effective cluster and calculate the Euclidean distance between the target pulse descriptor and other pulse descriptors in its effective cluster. The obtaining unit is used to, when it is determined that the Euclidean distance is less than a second preset threshold, take the target pulse descriptor as a representative pulse descriptor within its effective cluster; and so on, until the K representative pulse descriptors contained in each of the M effective clusters are obtained; where K is a positive integer greater than 0; The third calculation unit is used to calculate the intra-class scatter matrix and inter-class scatter matrix of K representative pulse descriptors in the M effective clusters, respectively. The fourth calculation unit is used to calculate the product of the inverse of the inter-class scatter matrix and the intra-class scatter matrix, and to perform eigenvalue decomposition on the product result in order to determine the projection matrix based on the decomposition result. The projection unit is used to perform feature projection on each pulse descriptor word in the nth beat according to the projection matrix to obtain the projected parameter vector. The second clustering unit is used to perform clustering processing on the projected parameter vector corresponding to each pulse descriptor in the nth beat, to obtain the target abnormal cluster and N effective clusters contained in the nth beat; where N is a positive integer greater than 0.

8. A radar-based individual sorting device, characterized in that, include: Processor, memory, system bus; The processor and the memory are connected via the system bus; The memory is used to store one or more programs, the one or more programs including instructions that, when executed by the processor, cause the processor to perform the method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a terminal device, cause the terminal device to perform the method described in any one of claims 1-6.

Images