A two-stage fusion radar signal sorting method

Abstract

The application discloses a two-stage fused radar signal sorting method, which firstly utilizes the first stage processing of DBSCAN clustering to determine two important hyperparameters, i.e., a proper neighborhood radius and a minimum point number range, through a kernel function density estimation algorithm, preliminarily identifies potential clusters and noise points in radar signals, then utilizes the low complexity of a K-Means algorithm to determine the number of clusters and initial cluster centers, and carries out noise reduction processing on the noise points, refines the clustering results of DBSCAN by using the K-Means algorithm, further improves the consistency of signals in the clusters and the accuracy of clustering by optimizing the positions of the cluster centers, and completes the second stage processing of the entire sorting data. The advantages of the two kinds of clustering algorithms are fused, the sorting strategy is optimized, the adaptive clustering and sorting process can be completed in a complex scene with serious overlapping of feature parameters and a high pulse loss rate, the sorting accuracy is effectively improved, and the sorting complexity is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of electronic reconnaissance and situation awareness, and particularly relates to a two-stage fused radar signal sorting method. BACKGROUND

[0002] Radar signal sorting is one of the key technologies of electronic reconnaissance and battlefield situation awareness, and its accuracy directly affects subsequent combat decision-making and planning. In a complex electromagnetic environment, an interleaved radar pulse stream composed of different radar emitter sources is intercepted. Accurate sorting of different radar signals is a prerequisite for subsequent processing such as emitter identification and situation awareness analysis. Since the 1970s, radar signal sorting has been one of the research hotspots in the field of radar signal processing, and rich research results have been achieved. The main work is divided into three categories based on inter-pulse characteristics, intra-pulse characteristics, and machine learning.

[0003] Early radars have low distribution density, single modulation pattern, and low spatial electromagnetic signal complexity. Sorting research mainly focuses on mining inter-pulse characteristics and sorting through matching ideas. Typical methods include template matching, cumulative difference histogram (CDIF), sequence difference histogram (SDIF), PRI variation method, and plane variation method, and subsequent researchers have continuously optimized typical methods based on inter-pulse parameters. With the diversification of radar functions and the complication of signal modulation characteristics, sorting algorithms based on inter-pulse rules have difficulty in effectively representing different radars, and the discrimination ability is very limited. Researchers further introduced sorting methods based on intra-pulse characteristics to improve the dimensionality of signal features. Typical methods include time-frequency analysis, ambiguity function, and high-order statistics. Although the introduction of intra-pulse characteristics improves the information dimension and amplifies the signal difference, there are problems such as poor noise resistance and insufficient generalization ability.

[0004] In view of the problems of rapid increase in radar density, serious overlap of signal characteristics, and serious pulse loss, researchers introduced machine learning methods, especially neural networks and clustering methods, which have achieved excellent sorting results in specific complex electromagnetic spaces. Neural networks have strong classification ability and are suitable for any data distribution, and are widely used in radar signal sorting tasks. However, training the network requires a large amount of labeled data, which is unrealistic to obtain a large number of non-cooperative emitter samples in actual combat. Therefore, the actual sorting task is closer to the clustering model.

[0005] The clustering algorithm is widely used due to its unsupervised learning characteristics and small requirement for prior knowledge. Various clustering algorithms based on division, density, hierarchy and grid have been proposed, and the achievements are rich. The clustering algorithm divides the pulses into different clusters, so that the similarity of the pulses in the same cluster is as high as possible. Typical methods include K-Means, DBSCAN, spectral clustering, hierarchical clustering, etc. A large number of sorting research results based on clustering have achieved good sorting effect by optimizing typical clustering algorithms. However, the existing clustering algorithms still have serious problems in the sorting task. First, most clustering algorithms require a large amount of computing resources, and the real-time performance and accuracy of the sorting process are often difficult to balance. Second, it is often limited by fixed parameter settings and assumptions about specific signal patterns, making it difficult to achieve adaptive processing of the scene. Third, high-performance clustering algorithms have high complexity and are difficult to optimize, and are not practically applicable. SUMMARY

[0006] The purpose of the present application is to provide a two-stage fusion radar signal sorting method to solve the problem of low accuracy and efficiency of radar signal sorting in complex electromagnetic environment.

[0007] In order to achieve the above task, the present application adopts the following technical scheme:

[0008] A two-stage fusion radar signal sorting method, comprising:

[0009] Importing a pulse description word sample set to be sorted and processed, and determining the inter-pulse parameters in the pulse description word sample for radar signal sorting;

[0010] Normalizing the inter-pulse parameters contained in each pulse description sub-sample in the pulse description word sample set;

[0011] Dividing the normalized pulse description word sample set, and randomly dividing the pulse description word sample set into multiple sample subsets;

[0012] Selecting any one of the sampled sample subsets to further determine two hyperparameters of the DBSCAN algorithm, namely the neighborhood radius and the minimum number of neighborhood points; wherein the distribution of the distance between the pulse description word samples in the sample subset is estimated by using the method of estimating the probability density of the kernel function, so as to determine the value of the hyperparameters: first, determine the Euclidean distance between the pulse description word samples in the selected sample subset, construct a Euclidean distance sample set using the Euclidean distance, determine the maximum value in the Euclidean distance sample set, and then determine the value range of the independent variable for estimating the kernel function density, and then perform kernel function density estimation on the Euclidean distance sample set; according to the kernel function probability density estimation result, the value at the first peak position is selected as the neighborhood radius, and the mean value of the number of pulse description word samples within the neighborhood radius is selected as the minimum number of neighborhood points;

[0013] The first stage DBSCAN clustering process is performed on the selected sample subset to determine the core samples in the sample subset; for the pulse description word samples, if the neighborhood radius thereof contains at least the neighborhood minimum number of pulse description word samples, the pulse description word sample is taken as a core sample; after the DBSCAN clustering is completed, the core samples are gathered together to form independent clusters one by one;

[0014] Based on the DBSCAN clustering result, the cluster number and the initial clustering center of each cluster of the K-Means algorithm in the second stage sorting are determined; wherein the number of the core sample gathering areas is taken as the cluster number, and the mean value of the core sample gathering area is taken as the initial clustering center of each cluster;

[0015] The Euclidean distance of all pulse description word samples except the core samples in the pulse description word sample set to each initial clustering center is calculated, and each sample is assigned to the cluster corresponding to the nearest clustering center based on the K-Means algorithm;

[0016] The mean value of all data samples in each cluster is calculated, and the mean value is updated as a new clustering center; the clustering process is repeated until the clustering center no longer changes or the preset iteration number is reached, and then the clustering process is ended;

[0017] The final cluster clustering result is output, the pulse description word samples in each cluster are a class, and the radar signal sorting process is completed.

[0018] Further, in the pulse description word sample set, the frequency RF, the pulse width PW and the angle DOA three inter-pulse parameters in each pulse description word sample are selected for multi-stage radar signal sorting processing.

[0019] Further, when the sample subset is divided, each pulse description word sample is equally divided into each sample subset.

[0020] Further, the maximum value d ijmax of the Euclidean distance sample set is set as the bandwidth of the kernel function; wherein the value range of the independent variable for estimating the kernel function density is (, d ijmax ); wherein represents the left end point of the range, and the value thereof is [-0.3, 0.3].

[0021] Further, the kernel function density estimation on the Euclidean distance sample set is represented as:

[0022]

[0023] wherein, represents the kernel function density estimation result, n1 is the number of Euclidean distance samples, h is the bandwidth of the kernel function, d k is the kth Euclidean distance in the Euclidean distance sample set, and d is the independent variable for estimating the kernel function density.

[0024] K(·) is a non-negative weighted function, adopting a Gaussian kernel function;

[0025] The calculation formula of the kernel function bandwidth h is:

[0026]

[0027] Wherein R(K) is the variance of the kernel function K(·), m2(K) is the second moment of the kernel function, and R(f”) is the variance of the second derivative f” of the normal distribution probability density function.

[0028] Further, the initial clustering center of each cluster is represented as:

[0029]

[0030] Wherein C i represents the i-th core sample aggregation area, is the corresponding clustering center, PDW k is the pulse description word sample in the core sample aggregation area.

[0031] A radar signal sorting device, comprising a processor, a memory and a computer program stored in the memory; when the processor executes the computer program, the two-stage fused radar signal sorting method is realized.

[0032] A computer readable storage medium, the medium stores a computer program; when the computer program is executed by the processor, the two-stage fused radar signal sorting method is realized.

[0033] Compared with the prior art, the present application has the following technical features:

[0034] The present application aims at the capability limitation of the traditional sorting algorithm under the condition of high pulse loss rate and serious parameter overlap, considers the sorting task characteristics, utilizes the characteristics of DBSCAN and K-Means algorithm, and proposes a multi-stage radar signal sorting strategy fusing DBSCAN and K-Means. The advantages of different algorithms are fused in different stages, aiming at improving the sorting accuracy under complex conditions. The specific advantages are as follows:

[0035] (1) According to the radar signal sorting demand under complex environment, the multi-stage processing strategy is utilized, the advantages are complementary, and the contradiction between the complexity of clustering algorithm, sorting real-time and sorting performance is effectively solved.

[0036] (2) Combined with the multi-stage processing strategy, the DBSCAN clustering preprocessing of small sample is introduced, the sorting process does not need hyperparameters, the manual parameter adjustment is effectively avoided, and the generalization ability of the algorithm is improved.

[0037] The simulation results show that the sorting accuracy of the invented method can reach over 90%. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a schematic diagram of radar signal sorting of the present invention;

[0039] Figure 2 1 is a schematic flow chart of the sorting method of the present invention;

[0040] Figure 3 This is a schematic diagram of the first stage DBSCAN algorithm processing of the present invention;

[0041] Figure 4 This is a schematic diagram of the second stage K-Means algorithm processing of the present invention;

[0042] Figure 5 3D parameter coordinate diagram of RF, PW and DOA of full pulse data in an embodiment of the present invention;

[0043] Figure 6 1 is a normalized coordinate diagram of the RF, PW, and DOA three-dimensional parameters of full pulse data in an embodiment of the present invention;

[0044] Figure 7 : is a density distribution result diagram of kernel function parameter estimation of sample point spacing in an embodiment of the present invention;

[0045] Figure 8 is the first stage clustering result of the full pulse data in the embodiment of the present invention;

[0046] Figure 9 is a schematic diagram of the initial cluster center of full pulse data in an embodiment of the present invention;

[0047] Figure 10 This is the second stage clustering result of the full pulse data in the embodiment of the present invention. DETAILED DESCRIPTION

[0048] In view of the problems of the existing algorithm, such as limited generalization ability, high complexity and difficulty in performing actual sorting tasks under complex conditions, the application selects typical DBSCAN and K-Means clustering algorithms based on the advantages of different clustering algorithms, and proposes a two-stage fusion radar signal sorting method. First, the DBSCAN clustering algorithm can process clusters of any shape and has strong noise resistance, so as to complete the first stage processing of the sorted data. The sample subset is used to determine two important DBSCAN clustering parameters, such as the appropriate neighborhood radius (Eps) and the minimum point range (MinPts), by using the kernel function density estimation algorithm, and the potential clusters and noise points in the radar signal are preliminarily identified. Then, the number of clusters and the initial cluster center are determined by using the low complexity of the K-Means algorithm. The K-Means algorithm is used to refine the clustering results of DBSCAN, the position of the cluster center is optimized, the consistency of the signals in the cluster and the accuracy of the clustering are further improved, and the second stage processing of the entire sorted data is completed. The advantages of the two kinds of clustering algorithms are fused, the sorting strategy is optimized, the adaptive clustering and sorting process can be completed in the complex scene with serious overlapping of feature parameters and high pulse loss rate, the sorting accuracy is effectively improved, and the sorting complexity is reduced.

[0049] Referring to the accompanying drawings Figure 1 and Figure 2 The two-stage fusion radar signal sorting method provided by the application comprises the following steps:

[0050] Step 1, import the pulse description word (PDW) sample set to be sorted and processed, the sample set contains N pulse description word samples; select three stable inter-pulse parameters, such as frequency (RF), pulse width (PW) and angle (DOA) in each pulse description word sample for multi-stage radar signal sorting processing.

[0051] RF is one of the most important parameters of radar, PW is related to the working characteristics of radar, and DOA is related to the position of radar and observation station, which is a reliable parameter for sorting.

[0052] D1={PDW1,PDW2,...,PDW N} (1)

[0053] Wherein, D1 represents the pulse description word sample set, PDW i represents the pulse description word sample of the i-th pulse, which contains three inter-pulse parameters, such as RF, PW and DOA.

[0054] Step 2, for each pulse description word sample PDW iNormalization processing is performed to normalize the three pulse-to-pulse parameters of RF, PW, and DOA to between 0 and 1 to eliminate the dimension effect and ensure that all parameters have the same weight in the clustering process.

[0055] Step 3: Divide the normalized pulse description word sample set into m sample subsets at random; when dividing the sample subsets, each pulse description word sample is equally likely to be assigned to each sample subset.

[0056] The purpose of designing this step is to take into account that the sample set to be sorted is large, resulting in high time complexity of sorting, and not all samples need to participate in the first stage, so sampling and division of sample subsets are performed.

[0057] Step 4: Select any sample subset after sampling to further determine the neighborhood radius (Eps) and the minimum number of neighborhood points (MinPts) of the DBSCAN algorithm. The processing effect of the DBSCAN algorithm greatly depends on the reasonable selection of the two parameters. The present invention uses the kernel function to estimate the probability density method to estimate the distribution of the sample spacing of each pulse description word in the sample subset to obtain a reasonable parameter range.

[0058] Neighborhood radius Eps: for pulse description word sample PDW i , whose neighborhood contains other pulse description word samples within the radius of Eps.

[0059] Minimum number of neighborhood points MinPts: defined as PDW i Density threshold within the Eps radius.

[0060] Step 4.1: First, calculate the Euclidean distance between the pulse description word samples in the selected sample subset, and use the Euclidean distance to construct a Euclidean distance sample set.

[0061] d ij =|PDW i -PDW j | 2 i,j∈{1,2,…,n} (2)

[0062] Among them, PDW i and PDW j represents the i,j pulse description word sample in the sample subset, n represents the number of pulse description word samples in the sample subset; the number of Euclidean distance samples after calculating the Euclidean distance is n 2 / 2.

[0063] Step 4.2, find the maximum value d in the Euclidean distance sample set ijmax , set the value range of the independent variable of the estimated kernel function density to (~,d ijmax); where ~ represents the left endpoint of the range, and its value is [-0.3, 0.3]; kernel function density estimation is performed on the Euclidean distance sample set:

[0064]

[0065] in Represents the kernel function density estimation result, n1=n 2 / 2 is the number of Euclidean distance samples, h is the kernel function bandwidth, d k is the kth Euclidean distance in the Euclidean distance sample set, and d is the independent variable for estimating the kernel function density.

[0066] K(·) is a non-negative weighting function, and the present invention uses the Gaussian kernel function:

[0067]

[0068] Where x represents the variable of the Gaussian kernel function, and e is a natural constant.

[0069] The kernel function bandwidth h determines the smoothness of the kernel function. The selection of the kernel function bandwidth is crucial to the performance of the distribution estimation. It is determined by the method of minimizing the average integral error. The calculation process is as follows:

[0070]

[0071] where R(K) is the variance of the kernel function K(·), m2(K) is the second moment of the kernel function, and R(f”) is the variance of the second derivative of the probability density function of the normal distribution, f”.

[0072] In step 4.3, for the kernel function probability density estimation result, select the value at the first peak position as the neighborhood radius Eps, and take the mean number of pulse description word samples within the neighborhood radius Eps as the minimum number of neighborhood points MinPts.

[0073] Step 5: After determining the neighborhood radius and the minimum number of neighborhood points, perform the first stage of DBSCAN clustering on the selected sample subset. The clustering results of the sample subset are divided into three categories: core samples, boundary samples, and noise samples:

[0074] For the pulse description word sample PDW i If its neighborhood radius Eps contains at least the minimum number of pulse description word samples in the neighborhood MinPts, it is regarded as a core sample; if it does not meet the conditions of the core sample, but it is within the neighborhood radius Eps of a core sample, it is regarded as a boundary sample; otherwise, it is regarded as a noise sample.

[0075] After DBSCAN clustering is completed, the core samples are clustered together to form independent clusters, and each cluster is a core sample clustering area.

[0076] Step 6: Determine the number of clusters k and the initial cluster centers of each cluster in the K-Means algorithm during the second stage sorting

[0077] In the first stage, the data samples have been divided into three categories: core samples, boundary samples, and noise samples. Core samples are the core part of the cluster. They have enough neighboring points around them to form a dense area. These core samples are clustered together to form independent clusters.

[0078] The number of core sample clusters is taken as the number of clusters k, and the mean of the core sample clusters is taken as the initial cluster center of each cluster.

[0079]

[0080] Among them C i represents the i-th cluster (core sample cluster area), is the corresponding cluster center, PDW k It is the pulse description word sample in the core sample cluster area.

[0081] Step 7: Calculate the Euclidean distances between all pulse description word samples except the core samples in the pulse description word sample set D1 and the initial cluster centers, and assign each sample to the cluster corresponding to the nearest cluster center based on the K-Means algorithm.

[0082] Step 8: Calculate the mean of all data samples in each cluster and update the mean as the new cluster center.

[0083] Step 9: Repeat steps 7 and 8 until the cluster center no longer changes or the preset number of iterations is reached, and then the clustering process ends;

[0084] Step 10: Output the final clustering result. The pulse description word samples in each cluster are one category, completing the sorting process.

[0085] Example:

[0086] Table 1 Sorting data set parameter table

[0087]

[0088] Step 1: Import the pulse description word (PDW) sample set D1 shown in Table 1. The sample set contains 26963 pulse samples. Select the three inter-pulse parameters with high stability, namely frequency (RF), pulse width (PW) and angle (DOA), to perform multi-stage three-dimensional clustering processing. The three-dimensional parameters are used as spatial coordinates as shown in Figure 5 shown.

[0089] Step 2, normalize the 26964 data samples given in this example, normalize the selected RF, PW and DOA parameters of the imported data sample set to 0-1, eliminate the dimension effect, and ensure that all parameters have the same weight in the clustering process, as shown in Figure 6 .

[0090] Step 3, randomly divide the sample set in this example into 20 subsets, where each sample point is equally divided into each subset, and one subset contains 1348 samples.

[0091] Step 4, select a subset to calculate the Euclidean distance of the samples, obtain a set composed of Euclidean distance values, and perform kernel function density estimation to obtain the distance set density distribution.

[0092] Step 4.1, calculate the Euclidean distance between samples in the subset, obtain a set composed of 908552 Euclidean distances.

[0093] Step 4.2, find the maximum value of the Euclidean distance subset 1.03, and determine the independent variable value range as (-0.2, 1.03). Perform kernel function density calculation to obtain the density distribution of the sample distance, as shown in Figure 7 .

[0094] Step 4.3, the reasonable range of Eps parameter value is 0-0.124, and the value at the first peak position is selected as the reasonable Eps value, which is 0.124 in this example.

[0095] Step 4.4, take the average number of sample points participating in the Eps neighborhood as the reasonable parameter value of MinPts, and calculate to obtain the MinPts value as 125.

[0096] Step 5, after determining the hyperparameters, perform the first stage of DBSCAN clustering process on the selected subset, and the clustered results of the subset can be divided into three categories: core samples, boundary samples and noise samples, as shown in Figure 8 , which shows the core samples and noise samples.

[0097] Step 6, determine the number of cluster classes k and the initial cluster centers It can be seen that the samples are clustered into 5 regions, so the number of cluster classes is 5, and the initial cluster centers are calculated by further calculating the mean value of the core samples, as shown in Figure 9 .

[0098] Step 7, calculate the Euclidean distance from all samples in the data sample set D1 except the core samples to the cluster centers, and assign each sample to the cluster corresponding to the nearest cluster center.

[0099] Step 8, update the cluster center, calculate the mean of all data samples in the cluster, and take the mean as the new cluster center.

[0100] Step 9, repeat steps 7 and 8 until the cluster center no longer changes, or the preset iteration number is reached, and end the clustering process.

[0101] Step 10, output the final clustering result to form a sub-pulse data, and complete the sorting process, as shown in Figure 10 The calculation of the five radiation source pulse sorting accuracy rates is 94.53%, 93.31%, 93.36%, 92.20%, and 93.60%.

[0102] The above examples are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacements for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A two-stage fusion radar signal sorting method, characterized in that: include: Importing a pulse description word sample set to be sorted and determining the pulse-to-pulse parameters used for radar signal sorting in the pulse description word samples; Normalizing the inter-pulse parameters contained in each pulse description sub-sample in the pulse description word sample set; Dividing the normalized pulse description word sample set into a plurality of sample subsets at random; Select any sample subset after sampling to further determine the neighborhood radius and the minimum number of neighborhood points of the DBSCAN algorithm; wherein, the distribution of the distance between the pulse description word samples in the sample subset is estimated by using the kernel function probability density estimation method to determine the value of the hyperparameter: first, the Euclidean distance between the pulse description word samples in the selected sample subset is determined, and a Euclidean distance sample set is constructed using the Euclidean distance. The maximum value in the Euclidean distance sample set is determined to determine the value range of the independent variable for estimating the kernel function density, and then the kernel function density is estimated for the Euclidean distance sample set; for the kernel function probability density estimation result, the value at the first peak position is selected as the neighborhood radius, and the mean of the number of pulse description word samples within the neighborhood radius is taken as the minimum number of neighborhood points; The first stage of DBSCAN clustering process is performed on the selected sample subset to determine the core samples in the sample subset; for pulse description word samples, if their neighborhood radius contains at least the minimum number of pulse description word samples in the neighborhood, they are regarded as core samples. After the DBSCAN clustering is completed, the core samples are clustered together to form independent clusters; Based on the DBSCAN clustering results, the number of clusters and the initial cluster centers of each cluster in the K-Means algorithm during the two-stage sorting were determined; the number of core sample clusters was used as the number of clusters, and the mean of the core sample clusters was used as the initial cluster center of each cluster; Calculate the Euclidean distance between all pulse description word samples except the core samples in the pulse description word sample set and each initial cluster center, and assign each sample to the cluster corresponding to the nearest cluster center based on the K-Means algorithm; Calculate the mean of all data samples in each cluster and update the mean as the new cluster center; repeat the clustering process until the cluster center no longer changes or the preset number of iterations is reached; The final clustering results are output, and the pulse description word samples in each cluster are a category, completing the radar signal sorting process.

2. The radar signal sorting method of the two-stage fusion according to claim 1 is characterized in that: In the pulse description word sample set, the three inter-pulse parameters of frequency RF, pulse width PW and angle DOA in each pulse description word sample are selected for multi-stage radar signal sorting processing.

3. The radar signal sorting method of the two-stage fusion according to claim 1 is characterized in that: When dividing the sample subsets, each pulse description word sample is equally likely to be divided into each sample subset.

4. The radar signal sorting method of the two-stage fusion according to claim 1 is characterized in that: The maximum value d in the Euclidean distance sample set ijmax , set the value range of the independent variable of the estimated kernel function density to (~,d ijmax ); ~ represents the left endpoint of the range, and its value is [-0.3,0.3].

5. The radar signal sorting method of two-stage fusion according to claim 1 is characterized in that: The kernel function density estimation of the Euclidean distance sample set is expressed as: in, Represents the kernel function density estimation result, n1 is the number of Euclidean distance samples, h is the kernel function bandwidth, d k is the kth Euclidean distance in the Euclidean distance sample set, and d is the independent variable for estimating the kernel function density; K(·) is a non-negative weighting function using a Gaussian kernel function; The calculation formula of kernel function bandwidth h is: where R(K) is the variance of the kernel function K(·), m2(K) is the second moment of the kernel function, and R(f”) is the variance of the second derivative of the probability density function of the normal distribution, f”.

6. The radar signal sorting method of the two-stage fusion according to claim 1 is characterized in that: The initial cluster center of each cluster is expressed as: Among them C i represents the i-th core sample cluster, is the corresponding cluster center, PDW k It is the pulse description word sample in the core sample cluster area.

7. A radar signal sorting device comprising a processor, a memory, and a computer program stored in the memory; characterized in that: When the processor executes the computer program, the radar signal sorting method of the two-stage fusion according to any one of claims 1 to 6 is implemented.

8. A computer-readable storage medium storing a computer program; wherein: When the computer program is executed by a processor, the two-stage fusion radar signal sorting method according to any one of claims 1 to 6 is implemented.

Images