Radar signal online sorting method based on D-S evidence theory and adaptive micro-cluster

Abstract

The application provides a radar signal online sorting method based on D-S evidence theory and adaptive micro-cluster, which comprises the following steps: intercepting initial pulse samples from a radar pulse data stream; performing static clustering on the initial pulse samples to obtain an initial micro-cluster set; performing micro-batch condensation on the radar pulse data stream continuously arriving in time windows to obtain micro-batch centroids; performing evidence fusion matching on the multi-dimensional feature evidence of the micro-batch centroids and the micro-clusters in the initial micro-cluster set based on D-S evidence theory to obtain a matching result; performing adaptive updating and drift compensation on the micro-clusters successfully matched according to the matching result to obtain an updated micro-cluster set, and putting the micro-batches failing to match into an abnormal buffer area; performing dynamic maintenance and macro-cluster merging on the abnormal buffer area and the updated micro-cluster set to obtain a radar emitter sorting result. The application can improve the accuracy, stability, real-time performance and memory controllability of radar pulse stream online sorting.

Description

Technical Field

[0001] This invention belongs to the field of radar signal processing technology, specifically relating to an online radar signal sorting method based on DS evidence theory and adaptive microclusters. Background Technology

[0002] Radar signal sorting is a crucial processing step in electronic reconnaissance and electromagnetic situational awareness. Its task is to separate the independent pulse sequences corresponding to each radiation source from a massive, aliased, and interleaved radar pulse data stream. With the continuous evolution of modern radar systems, the battlefield electromagnetic environment exhibits characteristics such as high signal density, complex overlapping parameters, high-speed continuous pulse arrival, and dynamic changes in signal parameters. Radar signal sorting methods not only need to achieve accurate differentiation across multiple dimensions such as radio frequency, pulse width, and angle of arrival, but also need to maintain low latency and stable memory consumption under continuous data stream conditions.

[0003] In related technologies, radar signal sorting and clustering methods mainly include traditional static clustering methods and data stream clustering methods. Traditional static clustering methods typically require pre-acquiring and storing the complete pulse dataset before performing global clustering on the offline data. While this can achieve good results in some offline scenarios, it leads to high memory overhead and long processing latency in high-speed, continuously arriving reconnaissance scenarios. Data stream clustering methods generate micro-clusters in an online layer and perform macro-clustering in an offline layer, which can adapt to the needs of data stream processing to some extent. However, existing micro-cluster update and matching strategies mostly rely on a single distance metric, density statistics, or simple probability models, making it difficult to fully incorporate the physical properties of radar signal parameters.

[0004] Therefore, existing radar signal online sorting technologies still have the following technical problems: static clustering methods are difficult to meet the real-time sorting requirements under high-speed pulse streams; conventional flow clustering methods are easily affected by single-pulse measurement noise, leading to micro-cluster update oscillations or model drift; in scenarios with parameter jumps, frequency agility, or overlapping multi-dimensional parameters, a single distance metric is difficult to stably determine pulse affiliation, easily splitting the same radiation source into multiple clusters and causing batching phenomena; when the abnormal buffer and micro-cluster lifecycle lack physical constraints for maintenance, it is easy to cause fragment cluster accumulation and memory growth. Summary of the Invention

[0005] To address the aforementioned problems in existing technologies, this invention provides an online radar signal sorting method based on DS evidence theory and adaptive micro-clusters. The technical problem to be solved by this invention is achieved through the following technical solution: In a first aspect, the present invention provides an online radar signal sorting method based on DS evidence theory and adaptive micro-clusters, comprising: Acquire radar pulse data stream, and extract initial pulse samples from the radar pulse data stream; Static clustering is performed on the initial pulse samples to obtain an initial set of micro-clusters; The continuously arriving radar pulse data stream is condensed into micro-batches according to time windows to obtain the micro-batch centroids; Based on the DS evidence theory, evidence fusion matching is performed on the multidimensional feature evidence of the micro-batch centroid and the micro-clusters in the initial micro-cluster set to obtain the matching result; Based on the matching results, adaptive updates and drift compensation are performed on the successfully matched micro-clusters to obtain the updated set of micro-clusters, and the micro-batches that failed to match are placed in the exception buffer. Dynamic maintenance is performed on the abnormal buffer, and dynamic maintenance and macro-cluster merging are performed on the updated micro-cluster set to obtain the radar radiation source sorting results.

[0006] In one embodiment of the present invention, the step of performing static clustering on the initial pulse samples to obtain an initial micro-cluster set includes: The initial pulse samples are clustered offline to obtain initial clusters; Select target initial clusters from the initial clusters that meet the minimum sample number requirement; Calculate the centroid and variance of the target initial cluster in each feature dimension; The centroid of the target initial cluster is solidified as the anchor point of the corresponding micro-cluster; The initial micro-cluster set is generated based on the centroid, variance, and anchor point of the target initial cluster.

[0007] In one embodiment of the present invention, the step of performing micro-batch aggregation on the continuously arriving radar pulse data stream according to a time window to obtain the micro-batch centroid includes: According to a preset time window, the pulses to be processed are extracted from the continuously arriving radar pulse data stream; Based on the absolute deviation of the pulse to be processed in each feature dimension, the pulse to be processed is agglomerated in one pass to obtain a micro-batch set. The centroids of the micro-batch set are calculated on each feature dimension to obtain the micro-batch centroids.

[0008] In one embodiment of the present invention, the step of performing evidence fusion matching on the multidimensional feature evidence of the micro-batch centroid and the micro-clusters in the initial micro-cluster set based on DS evidence theory to obtain a matching result includes: Based on the physical difference between the centroid of the micro-batch and the micro-cluster in the dimension of arrival angle, the micro-cluster is filtered by angle threshold to obtain a set of candidate micro-clusters; Based on the characteristic distances between the centroid of the micro-batch and each micro-cluster in the candidate micro-cluster set in the dimensions of radio frequency, pulse width, and angle of arrival, a basic probability assignment function is constructed; based on the basic probability assignment function, the support of radio frequency evidence, the support of pulse width evidence, and the support of angle of arrival evidence are determined. Based on the Dempster combination rule, a weighted evidence fusion is performed on the radio frequency evidence support, the pulse width evidence support, and the angle of arrival evidence support to obtain the total confidence of each micro-cluster in the candidate micro-cluster set corresponding to the centroid of the micro-batch. The matching result is obtained by comparing the maximum total confidence with the preset decision threshold.

[0009] In one embodiment of the present invention, the step of performing adaptive updates and drift compensation on the successfully matched micro-clusters according to the matching results to obtain an updated set of micro-clusters, and placing the micro-batches that failed to match into an exception buffer, includes: When the matching result indicates that the micro-batch matching is successful, the adaptive learning rate is determined based on the cumulative number of pulses of the successfully matched micro-clusters. Based on the adaptive learning rate and the micro-batch centroid, update the centroid and variance of the successfully matched micro-clusters; The drift distance is calculated based on the updated centroid and the anchor points of the successfully matched microclusters; If the drift distance exceeds a preset drift range, drift compensation is performed on the updated centroid. The updated microcluster set is obtained based on the centroid after drift compensation and the updated variance; If the matching result indicates that the micro-batch matching has failed, the micro-batch that failed to match will be placed in the exception buffer.

[0010] In one embodiment of the present invention, the dynamic maintenance of the exception buffer includes: The pulses in the abnormal buffer are sorted according to their angle of arrival, and a first-level slice is performed based on the difference in the angle of arrival of adjacent pulses to obtain a subset of the angles of arrival. The pulses arriving at the corner subset are sliced ​​in a secondary manner according to their radio frequency characteristics to obtain candidate slices; Based on the number of samples and feature dispersion of the candidate slices, a target candidate slice is determined from the candidate slices; Based on the multidimensional physical distance between the target candidate slice and each microcluster in the updated microcluster set, the slice processing result corresponding to the target candidate slice is determined; Based on the slice processing results, the target candidate slice is merged into an existing microcluster or initialized as a new microcluster, and the anomaly buffer is updated.

[0011] In one embodiment of the present invention, the step of performing dynamic maintenance and macro-cluster merging on the updated micro-cluster set to obtain radar radiation source sorting results includes: The weights of each microcluster in the updated microcluster set are decayed over time to obtain the decayed microcluster set. Remove microclusters with weights lower than a preset weight threshold from the decayed microcluster set to obtain a surviving microcluster set; Based on the physical gap between any two microclusters in the radio frequency dimension in the surviving microcluster set, a merging threshold screening is performed on any two microclusters. For microclusters that meet the merging threshold screening, the microcluster connectivity relationship is constructed based on the distribution overlap rate in the pulse width dimension and the angle of arrival dimension; Based on the micro-cluster connectivity relationship, the surviving micro-cluster set is merged by connectivity component merging to obtain a macro-cluster set; The radar radiation source sorting results are generated based on the macro-cluster set.

[0012] Secondly, the present invention provides an online radar signal sorting device based on DS evidence theory and adaptive micro-clusters, comprising: The pulse acquisition module is used to acquire radar pulse data stream and extract initial pulse samples from the radar pulse data stream. The micro-cluster construction module is used to perform static clustering on the initial pulse samples to obtain an initial micro-cluster set; The pulse aggregation module is used to perform micro-batch aggregation on the continuously arriving radar pulse data stream according to time windows to obtain the micro-batch centroid; The fusion matching module is used to perform evidence fusion matching between the centroid of the micro-batch and the multidimensional feature evidence of the micro-clusters in the initial micro-cluster set based on DS evidence theory, and to obtain the matching result. The micro-cluster update module is used to perform adaptive update and drift compensation on the successfully matched micro-clusters according to the matching results, to obtain the updated micro-cluster set, and to put the micro-batches that failed to match into the exception buffer. The result determination module is used to perform dynamic maintenance on the abnormal buffer, and to perform dynamic maintenance and macro-cluster merging on the updated micro-cluster set to obtain the radar radiation source sorting result.

[0013] Thirdly, the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; The memory is used to store computer programs; When the processor executes the computer program stored in the memory, it implements the steps of the radar signal online sorting method based on DS evidence theory and adaptive microclusters provided in the first aspect.

[0014] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the radar signal online sorting method based on DS evidence theory and adaptive microclusters provided in the first aspect.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention provides an online radar signal sorting method based on DS evidence theory and adaptive micro-clusters. It extracts initial pulse samples from the radar pulse data stream and performs static clustering to obtain an initial micro-cluster set with centroids, variances, and anchor points, allowing the online sorting process to start from a stable initial structure. Micro-batch agglomeration is performed according to time windows to reduce the sensitivity of point-by-point updates to single-pulse noise. Evidence fusion matching is performed based on the multi-dimensional feature evidence of micro-batch centroids and micro-clusters in the initial micro-cluster set, allowing multi-dimensional features such as radio frequency, pulse width, and angle of arrival to participate in the attribution judgment at the decision level, improving the matching robustness in complex electromagnetic environments. Adaptive updates and drift compensation are performed on successfully matched micro-clusters, enabling them to track dynamic changes in radar parameters, while anchor point constraints prevent abnormal model divergence. Dynamic maintenance and macro-cluster merging are performed on the abnormal buffer and the updated micro-cluster set, allowing unmatched pulses to be used for new radiation source detection, and fragmented micro-clusters are merged through physical gaps and multi-dimensional overlap relationships. Therefore, this invention can improve the accuracy, stability, real-time performance, and memory controllability of online radar pulse stream sorting.

[0016] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] Figure 1 This is a schematic diagram illustrating the implementation process of an online radar signal sorting method based on DS evidence theory and adaptive microclusters provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of real-time accuracy changes provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an online radar signal sorting device based on DS evidence theory and adaptive microclusters provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the hardware entity of an electronic device to which embodiments of the present invention are applied. Detailed Implementation

[0018] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0019] Please see Figure 1 , Figure 1 This is a schematic diagram illustrating the implementation process of an online radar signal sorting method based on DS evidence theory and adaptive micro-clusters, provided by an embodiment of the present invention. Figure 1 As shown, the online radar signal sorting method based on DS evidence theory and adaptive microclusters provided in this embodiment of the invention may include the following steps: Step S101: Obtain the radar pulse data stream and extract the initial pulse sample from the radar pulse data stream.

[0020] The radar pulse data stream can include multi-dimensional pulse descriptors such as radio frequency, pulse width, angle of arrival, time of arrival, and pulse amplitude.

[0021] For example, a small number of initial pulse samples can be extracted from the acquired radar pulse data stream as static data for subsequent construction of the initial microcluster set.

[0022] Step S102: Perform static clustering on the initial pulse samples to obtain the initial micro-cluster set.

[0023] In one possible implementation, initial pulse samples can be clustered offline to obtain initial clusters; target initial clusters that meet the minimum sample number requirement can be selected from the initial clusters; the centroids and variances of the target initial clusters in each feature dimension can be calculated; the centroids of the target initial clusters can be fixed as anchor points of the corresponding micro-clusters; and an initial set of micro-clusters can be generated based on the centroids, variances, and anchor points of the target initial clusters. .

[0024] Offline clustering can employ ensemble clustering, density clustering, hierarchical clustering, or other offline clustering algorithms capable of forming initial clusters. The minimum number of samples can be 5 to 10 pulses, or other numbers set according to actual needs.

[0025] Step S103: Perform micro-batch aggregation on the continuously arriving radar pulse data stream according to the time window to obtain the micro-batch centroid.

[0026] In one possible implementation, pulses to be processed can be extracted from the continuously arriving radar pulse data stream according to a preset time window; based on the absolute deviation of the pulses to be processed in each feature dimension, the pulses to be processed are condensed in a single pass to obtain a micro-batch set; the centroid of the micro-batch set in each feature dimension is calculated to obtain the micro-batch centroid.

[0027] The preset time window can be an extremely short time window with a length of 1ms to 10ms, or a time window of other lengths set according to actual needs.

[0028] For example, the pulse to be processed can be segmented in extremely short time windows of 1ms to 10ms, and a single-pass greedy strategy based on the smallest neighborhood can be used to aggregate the pulse in a single pass. Let the newly arrived radar pulse data stream be... and Only when its absolute deviation in all feature dimensions is less than the minimum tolerance threshold. When, that is, satisfied At that time, it is condensed into a micro-batch set. And extract the centroid of the micro-batch set. As the centroid of the micro-batch. The minimum tolerance threshold can include thresholds for three characteristic dimensions: Radio Frequency (RF), Pulse Width (PW), and Direction of Arrival (DOA). For example, the RF threshold can be 1 MHz, the Pulse Width threshold can be 0.1 μs, and the DOA threshold can be 1°. The thresholds for RF, Pulse Width, and DOA can also be other values ​​set according to actual needs. When the radar pulse data stream... and When the absolute deviations simultaneously satisfy the corresponding thresholds of the three characteristic dimensions of radio frequency, pulse width, and angle of arrival, they are aggregated into a micro-batch set.

[0029] In the above implementation, by using micro-batch aggregation, pulses with similar physical properties within a short time window can be compressed into micro-batch centroids to participate in subsequent matching, thereby reducing noise sensitivity and computational pressure caused by point-by-point updates.

[0030] Step S104: Based on the DS (Dempster-Shafer) evidence theory, perform evidence fusion matching between the multidimensional feature evidence of the micro-batch centroid and the micro-clusters in the initial micro-cluster set to obtain the matching result.

[0031] In one possible implementation, microclusters can be screened by angle threshold based on the physical difference between the micro-batch centroid and the microcluster in the angle of arrival dimension to obtain a set of candidate microclusters. A basic probability assignment function is constructed based on the feature distances between the micro-batch centroid and each microcluster in the candidate microcluster set in the radio frequency (RF), pulse width (PWM), and angle of arrival dimensions. The RF evidence support, PWM evidence support, and angle of arrival evidence support are determined based on the basic probability assignment function. Weighted evidence fusion is performed on the RF evidence support, PWM evidence support, and angle of arrival evidence support based on Dempster's combination rule to obtain the total confidence score of each microcluster in the candidate microcluster set corresponding to the micro-batch centroid. The maximum total confidence score is compared with a preset decision threshold to obtain the matching result.

[0032] The preset decision threshold can be a value between 0.6 and 0.7, or other values ​​set according to actual needs. The closer the micro-batch centroid is to the micro-cluster centroid in one of the dimensions of radio frequency, pulse width, and angle of arrival, the stronger the evidence support in that dimension.

[0033] For example, microclusters can be defined Physical angle threshold For micro-batch sets If its micro-batch center of gravity In the dimension of arrival direction, it is similar to micro-clusters. If the physical difference between the centroids exceeds the physical angle threshold, the matching result will be directly judged as a mismatch; if the centroids of the micro-batch... In the dimension of arrival direction, it is similar to micro-clusters. If the physical difference between the centroids does not exceed the physical angle threshold, then the microclusters will be... The candidate microclusters are included in the set. The process of angular threshold screening of microclusters can be expressed by the following formula: ; in, The inverse transform factor for the Z-score normalization coefficients is used to reduce the normalized distance to the physical differences. The eigenvalues ​​of the micro-batch centroids along the DOA dimension. microclusters The centroid on the DOA dimension , where is the confidence level.

[0034] To make micro clusters Taking the inclusion of candidate microclusters as an example, radio frequency, pulse width, and DOA can be regarded as three independent sources of evidence. A parameterized Basic Probability Assignment (BPA) function is used, defined as follows: ; in, microclusters In dimensions The confidence level (i.e., the degree of support of the evidence) on the surface. For the centroid of the micro-batch in dimension The eigenvalues ​​of this dimension It can be any of the following: radio frequency, pulse width, and DOA. microclusters In dimensions The center of mass on top, For dimension The matching tolerance hyperparameter is used to control the distance decay rate of evidence support. This basic probability assignment function can make the evidence support approach 1 as the sample features are closer to the centroid.

[0035] Based on the aforementioned basic probability assignment functions, the support levels for radio frequency (RF) evidence, pulse width (PWM) evidence, and angle of arrival (AHA) evidence can be determined. The total confidence level of each micro-cluster in the candidate micro-cluster set corresponding to the micro-batch centroid can be calculated using the BPA function. To avoid misleading the final decision with a single unreliable piece of evidence, a weighted evidence fusion can be performed on the RF evidence support, PWM evidence support, and AHA evidence support based on Dempster's combination rule to obtain the total confidence level. The calculation method is as follows: ; in, The centroid of a micro-batch corresponds to a micro-cluster. The overall confidence level For dimension Prior confidence weights, microclusters In dimensions The confidence level (i.e., the degree of evidence support) of the sample is determined. If the maximum total confidence level of the total confidence level of each micro-cluster corresponding to the centroid of the micro-batch is greater than the preset decision threshold, the sample is determined to be a successful match and enters the subsequent processing flow; otherwise, it is placed in the exception buffer to be processed.

[0036] Step S105: Perform adaptive update and drift compensation on the successfully matched micro-clusters according to the matching results to obtain the updated micro-cluster set, and put the micro-batches that failed to match into the exception buffer.

[0037] In one possible implementation, if the matching result indicates that the micro-batch matching is successful, an adaptive learning rate is determined based on the cumulative number of pulses of the successfully matched micro-clusters; based on the adaptive learning rate and the centroid of the micro-batch, the centroid and variance of the successfully matched micro-clusters are updated; based on the updated centroid and the anchor point of the successfully matched micro-clusters, the drift distance is calculated; if the drift distance exceeds a preset drift range, drift compensation is performed on the updated centroid; based on the drift-compensated centroid and the updated variance, an updated set of micro-clusters is obtained; if the matching result indicates that the micro-batch matching is unsuccessful, the unmatched micro-batches are placed in an exception buffer.

[0038] For example, it can be determined by the following formula Adaptive learning rate at time step: ; in, for Adaptive learning rate at any given time. This is the minimum learning rate lower bound to prevent the model from completely losing its update ability after long-term operation. For microclusters in The cumulative number of pulses at any given time. The warm-up threshold, Based on the learning rate.

[0039] The centroid of successfully matched microclusters can be updated using the following formula: ; in, for Adaptive learning rate at any given time. For micro-batch centroid, For the updated centroid of the microcluster, This is the centroid of the microcluster before the update.

[0040] The variance of successfully matched microclusters can be updated using the following formula: ; in, for Adaptive learning rate at any given time. For the updated variance of the microclusters, The variance of the microclusters before the update. The centroid of the microcluster before the update. For micro-batch sets, This represents the number of microclusters.

[0041] In one possible implementation, the radio frequency feature envelope range can be maintained when updating the centroid and variance of the microcluster. pulse width feature envelope range By setting an attenuation factor, the envelope is gradually shrunk to adapt to the dynamic changes in the frequency hopping range.

[0042] The drift vector of the updated centroid relative to the anchor point of the successfully matched microcluster can be determined using the following formula. Thus based on the drift vector Determine the drift distance : ; in, Let be the drift vector. For the updated centroid of the microcluster, It serves as the anchor point for the micro-cluster.

[0043] In drift distance Exceeding the preset drift range In this case, drift compensation is performed on the updated centroid, for example, by forcibly pulling the centroid of the micro-cluster back to the maximum allowable boundary through vector projection. The preset drift range is specified. It can include the range of three characteristic dimensions: radio frequency, pulse width, and angle of arrival. For example, the range of radio frequency can be 10 MHz, the range of pulse width can be 0.2 μs, and the range of angle of arrival can be 3°. The range of radio frequency, pulse width, and angle of arrival can also be other values ​​set according to actual needs.

[0044] Step S106: Perform dynamic maintenance on the abnormal buffer, and perform dynamic maintenance and macro-cluster merging on the updated micro-cluster set to obtain the radar radiation source sorting results.

[0045] In one possible implementation, dynamic maintenance of the anomaly buffer includes: sorting pulses in the anomaly buffer according to their angle of arrival, and performing a first-level slice based on the difference in the angle of arrival between adjacent pulses to obtain a subset of angles of arrival; performing a second-level slice based on the radio frequency characteristics of the pulses in the subset of angles of arrival to obtain candidate slices; determining a target candidate slice from the candidate slices based on the number of samples and the feature dispersion of the candidate slices; determining the slice processing result corresponding to the target candidate slice based on the multidimensional physical distance between the target candidate slice and each microcluster in the updated microcluster set; and merging the target candidate slice into an existing microcluster or initializing it as a new microcluster according to the slice processing result, and updating the anomaly buffer.

[0046] In one possible implementation, if the multidimensional physical distance between the target candidate slice and the existing microcluster is less than a preset overlap threshold, the target candidate slice is merged into the existing microcluster as a residual fragment of the existing microcluster; if the target candidate slice meets the new class formation conditions and does not overlap with the existing microcluster, the target candidate slice is initialized as a new microcluster; if the pulses in the abnormal buffer cannot meet the clustering conditions for a long time, they are cleared as isolated noise.

[0047] For example, all pulses in the exception buffer can be sorted according to their DOA, and the first-order difference between adjacent pulses can be calculated. Let there be a total of [number missing] pulses in the exception buffer. Each pulse is sorted in ascending order according to its DOA eigenvalue, resulting in an ordered set of pulses. The formula for calculating the first-order difference between adjacent pulses is: ; in, The first-order difference between adjacent pulses, For the first DOA characteristic value of each pulse, For the first DOA characteristic value of each pulse.

[0048] When the first-order difference between adjacent pulses exceeds the preset physical angle resolution threshold When the angle is between 0.5° and 2°, or other values ​​set according to actual needs, it is considered that a spatial break exists, thus dividing the abnormal buffer into several subsets of arrival angles in the DOA dimension. Within each arrival corner subset, secondary slicing can be performed using radio frequency (RF) features. By calculating differences along the RF dimension and identifying high-density, low-variance contiguous segments, the method can separate pulses belonging to the same new radiation source from the noise background. If some discrete pulses fail to meet the minimum clustering density requirement (e.g., 3 to 10 pulses, or other values ​​set according to actual needs) and remain in the abnormal buffer for longer than the set maximum timeout period (e.g., 10ms to 50ms, or other durations set according to actual needs), the system will determine them as isolated measurement noise and forcibly discard them.

[0049] The above implementation method can ensure that the exception buffer memory will not overflow, realizing the logical closed loop of high-speed streaming processing.

[0050] For each candidate slice, its sample size can be calculated. and standard deviation of each dimension A slice enters the candidate state and is designated as a target candidate slice only when the number of samples exceeds the minimum clustering threshold (e.g., 5 to 10 pulses, or other values ​​set according to actual needs) and the variances of the feature parameters are all below the physical consistency thresholds (including the RF variance threshold, pulse width variance threshold, and angle of arrival variance threshold; the RF variance threshold can be 1MHz to 5MHz; the pulse width variance threshold can be 0.05μs to 0.2μs; and the angle of arrival variance threshold can be 0.5° to 2°; the above RF variance threshold, pulse width variance threshold, and angle of arrival variance threshold can also be other values ​​set according to actual needs). The centroid of this target candidate slice is calculated to be the physical distance between it and all current microclusters in terms of DOA, RF, and PW. Let... The centroid of the target candidate slice, , , For different dimensions, preset overlap thresholds are set. If the three-dimensional distance is less than the preset overlap threshold, it is determined to be a residual fragment of an existing microcluster, and its pulse weight is directly added to the microcluster. Otherwise, the slice is officially promoted to a new microcluster, its centroid, variance, envelope range, and anchor point position are initialized, and the registration of the new radiation source is completed. The process of determining the residual fragment of an existing microcluster can be expressed by the following formula: ; in, The centroid of the target candidate slice in the DOA dimension. For micro-clusters in the DOA dimension, The centroid of the target candidate slice in the RF dimension. For micro-clusters in the RF dimension, The centroid of the target candidate slice in the PW dimension. The centroid of the target candidate slice in the PW dimension. The overlap threshold in the DOA dimension. For the overlap threshold in the RF dimension, This is the overlap threshold in the PW dimension.

[0051] In one possible implementation, dynamic maintenance and macro-cluster merging are performed on the updated micro-cluster set to obtain radar radiation source sorting results. This includes: time-attenuating the weights of each micro-cluster in the updated micro-cluster set to obtain an attenuated micro-cluster set; deleting micro-clusters with weights lower than a preset weight threshold from the attenuated micro-cluster set to obtain a surviving micro-cluster set; performing merging threshold screening on any two micro-clusters in the surviving micro-cluster set based on the physical gap between any two micro-clusters in the radio frequency dimension; constructing micro-cluster connectivity relationships for micro-clusters that meet the merging threshold screening based on the distribution overlap rate in the pulse width and angle of arrival dimensions; merging connected components of the surviving micro-cluster set based on the micro-cluster connectivity relationships to obtain a macro-cluster set; and generating radar radiation source sorting results based on the macro-cluster set.

[0052] For example, a decay factor can be introduced. At the end of each time slice, the weights of each microcluster in the updated microcluster set are decayed over time. Let... Microclusters in the constantly updated microcluster set The weight is After a time interval Then, its weight update formula is: ; in, Time interval Post-microcluster The weight, for constantly updated microclusters The weight, This is the attenuation factor.

[0053] When the weight of a microcluster is lower than the preset weight threshold (such as 0.05 to 0.1, or other values ​​set according to actual needs), it is regarded as a noise cluster and deleted. The remaining microclusters after deletion are then included in the set of surviving microclusters.

[0054] The RF physical gaps between microclusters in the surviving microcluster set can be calculated. In the macrocluster graph merging strategy, let any two microclusters to be determined... and Their physical envelope ranges in the RF dimension are respectively and First, calculate the RF physical gap between the two. The formula is: ; in, microclusters and RF physical gap between microclusters Within the physical envelope of the RF dimension, microclusters The physical envelope range in the RF dimension.

[0055] If microclusters and If the physical RF gap between the transmitters exceeds the preset maximum physical isolation threshold, a veto will be applied, and merging is strictly prohibited.

[0056] Secondly, if the gap condition is met, the microclusters are further calculated. and Distribution overlap rate in the PW and DOA dimensions. Taking the PW dimension as an example, the distribution overlap rate... Defined as the ratio of the intersection length to the shortest interval length, the formula is: ; in, microclusters and Distribution overlap rate in the PW dimension microclusters Within the physical envelope of the PW dimension, microclusters The physical envelope range in the PW dimension.

[0057] Only when the overlap rates of both PW and DOA exceed a preset overlap threshold are they determined to be fragments of the same target that were accidentally severed. Then, a micro-cluster physical connectivity adjacency matrix is ​​constructed. Search the interconnected subgraph, weighted merge the micro-clusters within the connected components into a new macro-cluster, reset the anchor point, and assign the new macro-cluster to the macro-cluster set.

[0058] The above implementation method uses the radio frequency physical gap, pulse width overlap rate and angle of arrival overlap rate to jointly constrain macro-cluster merging, which can avoid erroneously merging micro-clusters that cannot physically belong to the same radiation source. At the same time, it can merge multiple micro-clusters formed by parameter jumps, frequency agility or local fragmentation into the same macro-cluster, thereby suppressing the batching phenomenon.

[0059] In one possible implementation, a continuous online sorting test lasting up to 10 seconds can be conducted using over 900,000 pulse data streams from seven radars of different systems.

[0060] Test results show that the method of this invention achieves an overall accuracy of 96.54%, a precision of 97.20%, a recall of 96.73%, and a comprehensive F1 score of 96.97%. This demonstrates that the multi-dimensional joint verification strategy of this invention constructs a strong anti-interference barrier, effectively preventing the target from being erroneously fragmented, and achieving an excellent balance between precision and recall. The method of this invention processes nearly 900,000 pulses in a total time of only 26.94 seconds, with an average throughput of 34,874 pulses / second and an average processing latency of only 0.02867 milliseconds per pulse. More importantly, because the algorithm uses micro-cluster probability maintenance, the peak memory usage is only 19.10KB and does not increase with the length of the data stream.

[0061] In one possible implementation, the Density Stream clustering algorithm DenStream and the Adaptive Density Peak Clustering-Gaussian Mixture Model (ADPC-GMM) are selected as control groups for comparison. Figure 2 As shown, the accuracy rates obtained using the method described in this chapter (i.e., the method of this invention), ADPC-GMM, and DenStream are illustrated. It is evident that the global accuracy of the complete data stream obtained by the method of this invention is superior to ADPC-GMM and DenStream. It is also clear that traditional stream clustering algorithms, limited by a single decision principle or probability density constraints, exhibit significant lag and vulnerability when dealing with complex dynamic radar data streams. In contrast, the method of this invention can more accurately and stably complete online sorting tasks in non-cooperative scenarios.

[0062] This invention provides an online radar signal sorting method based on DS evidence theory and adaptive micro-clusters. It extracts initial pulse samples from the radar pulse data stream and performs static clustering to obtain an initial micro-cluster set with centroids, variances, and anchor points, allowing the online sorting process to start from a stable initial structure. Micro-batch agglomeration is performed according to time windows to reduce the sensitivity of point-by-point updates to single-pulse noise. Evidence fusion matching is performed based on the multi-dimensional feature evidence of micro-batch centroids and micro-clusters in the initial micro-cluster set, allowing multi-dimensional features such as radio frequency, pulse width, and angle of arrival to participate in the attribution judgment at the decision level, improving the matching robustness in complex electromagnetic environments. Adaptive updates and drift compensation are performed on successfully matched micro-clusters, enabling them to track dynamic changes in radar parameters, while anchor point constraints prevent abnormal model divergence. Dynamic maintenance and macro-cluster merging are performed on the abnormal buffer and the updated micro-cluster set, allowing unmatched pulses to be used for new radiation source detection, and fragmented micro-clusters are merged through physical gaps and multi-dimensional overlap relationships. Therefore, this invention can improve the accuracy, stability, real-time performance, and memory controllability of online radar pulse stream sorting.

[0063] Please see Figure 3 , Figure 3 This is a schematic diagram of the structure of an online radar signal sorting device based on DS evidence theory and adaptive microclusters provided by an embodiment of the present invention. The online radar signal sorting device based on DS evidence theory and adaptive microclusters provided by the present invention includes a pulse acquisition module 301, a microcluster construction module 302, a pulse agglomeration module 303, a fusion matching module 304, a microcluster update module 305, and a result determination module 306.

[0064] Among them, the pulse acquisition module 301 is used to acquire radar pulse data stream and extract initial pulse samples from radar pulse data stream; The micro-cluster construction module 302 is used to perform static clustering on the initial pulse samples to obtain an initial micro-cluster set; The pulse aggregation module 303 is used to perform micro-batch aggregation on the continuously arriving radar pulse data stream according to the time window to obtain the micro-batch centroid. The fusion matching module 304 is used to perform evidence fusion matching between the centroid of the micro-batch and the multidimensional feature evidence of the micro-clusters in the initial micro-cluster set based on the DS evidence theory, and obtain the matching result. The micro-cluster update module 305 is used to perform adaptive update and drift compensation on the successfully matched micro-clusters according to the matching results, to obtain the updated micro-cluster set, and to put the micro-batches that failed to match into the exception buffer. The result determination module 306 is used to perform dynamic maintenance and macro-cluster merging on the abnormal buffer and the updated micro-cluster set to obtain the radar radiation source sorting results.

[0065] Please see Figure 4 , Figure 4 This is a schematic diagram of the hardware entity of an electronic device to which an embodiment of the present invention applies. The electronic device provided in this embodiment includes a processor 401, a communication interface 402, a memory 403, and a communication bus 404, wherein the processor 401, the communication interface 402, and the memory 403 communicate with each other via the communication bus 404. Memory 403 is used to store computer programs; When the processor 401 executes the program stored in the memory 403, it implements the steps in the above method embodiments.

[0066] The communication bus 404 mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus 404 can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, it is represented by only one thick line in the figure, but this does not indicate that there is only one bus or one type of bus.

[0067] Communication interface 402 is used for communication between the above-mentioned electronic device and other devices.

[0068] The memory 403 may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory 403 may also be at least one storage device located remotely from the aforementioned processor.

[0069] The processor 401 mentioned above can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0070] The method provided in this invention can be applied to electronic devices. Specifically, the electronic device can be a desktop computer, a portable computer, a smart mobile terminal, a server, etc. No limitation is made herein; any electronic device that can implement this invention falls within the protection scope of this invention.

[0071] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps provided in the above-described method embodiments.

[0072] For the embodiments of the device / electronic device / storage medium, since they are basically similar to the method embodiments, the description is relatively simple. For specific details and beneficial effects, please refer to the description of the method embodiments.

[0073] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0074] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0075] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus (devices), or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects, all of which are collectively referred to herein as "modules" or "systems." Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The computer program may be stored / distributed in a suitable medium, provided with or as part of other hardware, or may take other distribution forms, such as via the Internet or other wired or wireless telecommunications systems.

[0076] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0077] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0078] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0079] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A radar signal online sorting method based on DS evidence theory and adaptive micro-clusters, characterized in that, include: Acquire radar pulse data stream, and extract initial pulse samples from the radar pulse data stream; Static clustering is performed on the initial pulse samples to obtain an initial set of micro-clusters; The continuously arriving radar pulse data stream is condensed into micro-batches according to time windows to obtain the micro-batch centroids; Based on the DS evidence theory, evidence fusion matching is performed on the multidimensional feature evidence of the micro-batch centroid and the micro-clusters in the initial micro-cluster set to obtain the matching result; Based on the matching results, adaptive updates and drift compensation are performed on the successfully matched micro-clusters to obtain the updated set of micro-clusters, and the micro-batches that failed to match are placed in the exception buffer. Dynamic maintenance is performed on the abnormal buffer, and dynamic maintenance and macro-cluster merging are performed on the updated micro-cluster set to obtain the radar radiation source sorting results.

2. The method according to claim 1, characterized in that, The initial pulse samples are statically clustered to obtain an initial micro-cluster set, including: The initial pulse samples are clustered offline to obtain initial clusters; Select target initial clusters from the initial clusters that meet the minimum sample number requirement; Calculate the centroid and variance of the target initial cluster in each feature dimension; The centroid of the target initial cluster is solidified as the anchor point of the corresponding micro-cluster; The initial micro-cluster set is generated based on the centroid, variance, and anchor point of the target initial cluster.

3. The method according to claim 1, characterized in that, The step of performing micro-batch aggregation on the continuously arriving radar pulse data stream according to time windows to obtain the micro-batch centroid includes: According to a preset time window, the pulses to be processed are extracted from the continuously arriving radar pulse data stream; Based on the absolute deviation of the pulse to be processed in each feature dimension, the pulse to be processed is agglomerated in one pass to obtain a micro-batch set. The centroids of the micro-batch set are calculated on each feature dimension to obtain the micro-batch centroids.

4. The method according to claim 1, characterized in that, Based on the DS evidence theory, evidence fusion matching is performed on the multidimensional feature evidence of the micro-batch centroid and the micro-clusters in the initial micro-cluster set to obtain the matching results, including: Based on the physical difference between the centroid of the micro-batch and the micro-cluster in the dimension of arrival angle, the micro-cluster is filtered by angle threshold to obtain a set of candidate micro-clusters; Based on the characteristic distances between the centroid of the micro-batch and each micro-cluster in the candidate micro-cluster set in the dimensions of radio frequency, pulse width, and angle of arrival, a basic probability assignment function is constructed; based on the basic probability assignment function, the support of radio frequency evidence, the support of pulse width evidence, and the support of angle of arrival evidence are determined. Based on the Dempster combination rule, the radio frequency evidence support, the pulse width evidence support, and the angle of arrival evidence support are weighted and fused to obtain the total confidence of each micro-cluster in the candidate micro-cluster set corresponding to the centroid of the micro-batch. The matching result is obtained by comparing the maximum total confidence with the preset decision threshold.

5. The method according to claim 1, characterized in that, The process involves adaptively updating and drift compensation for successfully matched microclusters based on the matching results, obtaining an updated set of microclusters, and placing unmatched micro-batches into an exception buffer, including: When the matching result indicates that the micro-batch matching is successful, the adaptive learning rate is determined based on the cumulative number of pulses of the successfully matched micro-clusters. Based on the adaptive learning rate and the micro-batch centroid, update the centroid and variance of the successfully matched micro-clusters; The drift distance is calculated based on the updated centroid and the anchor points of the successfully matched microclusters; If the drift distance exceeds a preset drift range, drift compensation is performed on the updated centroid. The updated microcluster set is obtained based on the centroid after drift compensation and the updated variance; If the matching result indicates that the micro-batch matching has failed, the micro-batch that failed to match will be placed in the exception buffer.

6. The method according to claim 1, characterized in that, The dynamic maintenance of the exception buffer includes: The pulses in the abnormal buffer are sorted according to their angle of arrival, and a first-level slice is performed based on the difference in the angle of arrival of adjacent pulses to obtain a subset of the angles of arrival. The pulses arriving at the corner subset are sliced ​​in a secondary manner according to their radio frequency characteristics to obtain candidate slices; Based on the number of samples and feature dispersion of the candidate slices, a target candidate slice is determined from the candidate slices; Based on the multidimensional physical distance between the target candidate slice and each microcluster in the updated microcluster set, the slice processing result corresponding to the target candidate slice is determined; Based on the slice processing results, the target candidate slice is merged into an existing microcluster or initialized as a new microcluster, and the anomaly buffer is updated.

7. The method according to claim 1, characterized in that, The process of dynamically maintaining and merging macroclusters on the updated microcluster set to obtain radar radiation source sorting results includes: The weights of each microcluster in the updated microcluster set are decayed over time to obtain the decayed microcluster set. Remove microclusters with weights lower than a preset weight threshold from the decayed microcluster set to obtain a surviving microcluster set; Based on the physical gap between any two microclusters in the radio frequency dimension in the surviving microcluster set, a merging threshold screening is performed on any two microclusters. For microclusters that meet the merging threshold screening, the microcluster connectivity relationship is constructed based on the distribution overlap rate in the pulse width dimension and the angle of arrival dimension; Based on the micro-cluster connectivity relationship, the surviving micro-cluster set is merged by connectivity component merging to obtain a macro-cluster set; The radar radiation source sorting results are generated based on the macro-cluster set.

8. An online radar signal sorting device based on DS evidence theory and adaptive micro-clusters, characterized in that, include: The pulse acquisition module is used to acquire radar pulse data stream and extract initial pulse samples from the radar pulse data stream. The micro-cluster construction module is used to perform static clustering on the initial pulse samples to obtain an initial micro-cluster set; The pulse aggregation module is used to perform micro-batch aggregation on the continuously arriving radar pulse data stream according to time windows to obtain the micro-batch centroid; The fusion matching module is used to perform evidence fusion matching between the centroid of the micro-batch and the multidimensional feature evidence of the micro-clusters in the initial micro-cluster set based on DS evidence theory, and to obtain the matching result. The micro-cluster update module is used to perform adaptive update and drift compensation on the successfully matched micro-clusters according to the matching results, to obtain the updated micro-cluster set, and to put the micro-batches that failed to match into the exception buffer. The result determination module is used to perform dynamic maintenance on the abnormal buffer, and to perform dynamic maintenance and macro-cluster merging on the updated micro-cluster set to obtain the radar radiation source sorting result.

9. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; The memory is used to store computer programs; When the processor executes the computer program stored in the memory, it implements the steps of the radar signal online sorting method based on DS evidence theory and adaptive microclusters as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the radar signal online sorting method based on DS evidence theory and adaptive microclusters as described in any one of claims 1-7.

Images