A radar range fuzzy point classification method based on support vector machine

Abstract

The invention discloses a radar distance fuzzy point trace classification method based on a support vector machine, comprising: receiving radar multi-pulse echo complex data formed by scattering of all objects in a radar detection area, processing the radar echo complex data to obtain radar detection point trace data, obtaining logarithmic amplitude data of N pulses in a distance unit where the radar detection point trace is located according to the radar detection point trace data and the radar multi-pulse echo complex data; constructing and training a binary classification support vector machine; and applying the binary classification support vector machine to a radar detection point trace preprocessing link in radar signal processing to judge whether the radar detection point trace belongs to a distance fuzzy point trace, thereby improving the radar's suppression performance on the distance fuzzy point trace.

Description

Technical Field

[0001] The invention relates to the technical field of radar signal processing, and in particular to a radar range fuzzy point trace classification method based on a support vector machine. Background Art

[0002] Radar is an electronic device that can detect targets in space using electromagnetic waves. Pulse radar is a type of radar that detects targets by emitting pulse signals of a certain length. When a pulse radar continuously emits multiple pulse signals to the same detection area, it may receive echoes from targets that exceed the maximum unambiguous distance. The traces detected by these target echoes after radar signal processing are distance-ambiguous, and these traces are called distance-ambiguous traces. Distance ambiguity means that in a pulse radar, when the distance of a target from the radar is greater than the maximum distance corresponding to the pulse repetition period of the current transmitted pulse, the target echo will not be received during the receiving period of the current transmitted pulse, but will be received during the receiving period of the subsequent transmitted pulse. At this time, the measured target distance is not the real distance, which is called distance ambiguity. For example, the first pulse signal emitted by the radar is backscattered by the target and is received during the receiving period of the second transmitted pulse. This type of distance ambiguity trace is a first-time distance ambiguity trace; by analogy, if it is received during the receiving period of the Dth transmitted pulse, it is called a D-1th distance ambiguity trace, where D represents the number of pulses emitted by the radar.

[0003] The echo data received by the radar receiver contains target echoes with unambiguous range and target echoes with ambiguous range. In the radar detection trace data obtained after processing the radar echo data by the radar target detection algorithm, there will be target traces with ambiguous range, but these target traces with ambiguous range are often not of interest and will cause great interference to the radar target tracking link. Therefore, it is necessary to judge whether all traces in the radar detection trace data belong to the range ambiguous traces, so as to achieve the purpose of improving the radar target tracking performance by suppressing the traces with ambiguous range. The currently used range ambiguous trace classification method is realized by calculating the amplitude difference of adjacent pulses and hard threshold judgment. However, due to the large dynamic range of the scattering cross-sectional area of ​​the radar target, the accuracy of this hard threshold judgment range ambiguous trace classification method is not high, which leads to a large number of range ambiguous traces entering the radar target tracking link, which seriously affects the tracking performance of the radar target. Summary of the invention

[0004] The purpose of the present invention is to provide a radar distance fuzzy point trace classification method based on support vector machine to solve the problem that the accuracy of the existing distance fuzzy point trace classification method based on hard threshold judgment is not high, thereby improving the accuracy of distance fuzzy point trace classification.

[0005] The objective of the present invention is achieved by the following technical means: a radar range fuzzy point trace classification method based on support vector machine, comprising the following steps:

[0006] S1, receives radar multi-pulse echo complex data Z formed by scattering from all objects in the radar detection area N×K =[z 1 ,z 2 ,...,z K ] N×K , and the radar echo complex data is processed to obtain the radar detection point data d M×1 =[d 1 ,d 2 ,...,d M ] M×1 , where K represents the number of range cells contained in the radar data, represents the column vector composed of N pulse echoes of the kth range unit, M represents the number of radar detection points, d m ,m=1,2,...,M represents the distance unit number where the mth radar point is located;

[0007] S2, based on the radar detection point data d in S1 M×1 =[d 1 ,d 2 ,...,d M ] M×1 and radar multi-pulse echo complex data Z N×K =[z 1 ,z 2 ,...,z K ] N×K , get the logarithmic amplitude data R of N pulses in the distance unit where the radar detection point is located N×M =[r 1 ,r 2 ,...,r M ] N×M ;

[0008] S3 constructs a binary classification support vector machine whose input is the logarithmic amplitude data of N pulses in the distance unit where the radar detection point trace is located, and outputs the binary classification support vector machine of the radar detection point trace category, and uses the logarithmic amplitude data R of N pulses in the distance unit where the radar detection point trace is located in S2 N×M =[r 1 ,r 2 ,...,r M ] N×M Train a binary classification support vector machine;

[0009] S4, based on the output of S3, is a binary classification support vector machine for the radar detection point category, which is applied to the radar detection point preprocessing link in radar signal processing to determine whether the radar detection point belongs to the range ambiguity point, thereby improving the radar's suppression performance on the range ambiguity point.

[0010] In S2, the logarithmic amplitude data R of the N pulses of the distance unit where the radar detection point trace is located is obtained. N×M =[r 1 ,r 2 ,...,r M ] N×M The method is,

[0011] S01: According to the radar detection point data d in S1 M×1 =[d 1 ,d 2 ,...,d M ] M×1 , from radar multi-pulse echo complex data Z N×K =[z 1 ,z 2 ,...,z K ] N×K According to the index k=d m , m=1,2,...,M take out the echo complex data W of the distance unit where all radar detection points are located N×M =[w 1 ,w 2 ,...,w M ] N×M , where w m =z dm ,m=1,2,...,M,w m Represents a column vector consisting of N pulse complex data of the mth radar detection point;

[0012] S02: Based on the echo complex data W of the distance unit where the radar detection point is located N×M =[w 1 ,w 2 ,...,w M ] N×M , through r m =20×log 10 (|w m |), m = 1, 2, ..., M to obtain the logarithmic amplitude data R of the distance unit where the radar detection point is located N×M =[r 1 ,r 2 ,...,r M ] N×M , where r m =20×log 10 (|wm |),m=1,2,...,M,r m A column vector consisting of the logarithmic amplitudes of the N pulses representing the distance unit where the mth radar detection point is located.

[0013] In S3, the method for training a binary classification support vector machine is:

[0014] S001: The kernel function of the binary classification support vector machine is set to be the radial basis function, the input is the logarithmic amplitude data of N pulses in the distance unit where the radar detection point trace is located, and the output is the category label of the radar detection point trace;

[0015] S002: Logarithmic amplitude data R of N pulses in the distance unit where the radar detection point trace in S2 is located N×M =[r 1 ,r 2 ,...,r M ] N×M Randomly divide it into two parts: training data set and verification data set to obtain training data set and validation dataset H is the number of training data sets, G is the number of validation data sets, Indicates that the product of M and h is rounded down;

[0016] S003: Based on the logarithmic amplitude data R of the N pulses in the distance unit where the radar detection point trace is located in S2 N×M =[r 1 ,r 2 ,...,r M ] N×M , get the training data set The corresponding label vector and validation dataset The corresponding label vector in Represents the training data set The category label of the point trace of the mth distance unit in , if the point trace belongs to the distance fuzzy point trace, the category label value is 1, otherwise it is -1; Represents the validation dataset The category label of the point trace of the nth distance unit in ;

[0017] S004: Using the training data set in S002 and its corresponding training dataset label vector in S003 Train the binary classification support vector machine in S001;

[0018] S005: The validation data set in S002 Input into the support vector machine obtained in S004 to obtain the category label predicted by the support vector machine for the validation data set

[0019] S006: By comparing S005 and the validation dataset label vector in S003 Calculate the accuracy of the validation data set on the support vector machine obtained in S004 Representation calculation and The number of equal values ​​at the same distance unit position;

[0020] S007: Determine whether the training cycle flag Q is less than the maximum number of cycles P: If it is less than P, then Q = Q + 1, go to S002; otherwise select A Q ,Q=1,2,...,the support vector machine corresponding to the maximum value in P is used as the final support vector machine for radar range fuzzy point classification.

[0021] The category label of the distance blurred point trace is 1, the category label of the distance unblurred point trace is -1, the training cycle flag Q=1 is set, and the maximum number of cycles P=10.

[0022] In S004, the training algorithm of the binary support vector machine is a sequential minimum optimization algorithm.

[0023] In S002, the logarithmic amplitude data R of the N pulses of the distance unit where the radar detection point is located is N×M =[r 1 ,r 2 ,...,r M ] N×M The dataset is randomly divided into a training data set and a validation data set according to the ratio of h:g along the distance dimension, where h+g=1,h>g.

[0024] In S1, radar detection point data d is obtained by processing radar echo complex data using a radar constant false alarm rate detection method. M×1 =[d 1 ,d 2 ,...,d M ] M×1 .

[0025] The beneficial effects of the present invention are as follows: a support vector machine is used to classify radar distance fuzzy points, which improves the accuracy of radar distance fuzzy point classification compared with the existing classification method based on hard threshold judgment, and is applied to the detection point preprocessing link in radar signal processing, effectively suppressing radar distance fuzzy points and improving the radar's tracking performance for targets of interest. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a schematic diagram of the process of the present invention;

[0027] Figure 2 This is a comparison chart of simulation data of the present invention.

[0028] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. DETAILED DESCRIPTION

[0029] [Example 1]

[0030] like Figure 1 As shown, a radar distance fuzzy point classification method based on support vector machine includes the following steps:

[0031] S1, receives radar multi-pulse echo complex data Z formed by scattering from all objects in the radar detection area N×K =[z 1 ,z 2 ,...,z K ] N×K , and the radar echo complex data is processed to obtain the radar detection point data d M×1 =[d 1 ,d 2 ,...,d M ] M×1 , where K represents the number of range cells contained in the radar data, represents the column vector composed of N pulse echoes of the kth range unit, M represents the number of radar detection points, d m ,m=1,2,...,M represents the distance unit number where the mth radar point is located;

[0032] S2, based on the radar detection point data d in S1 M×1 =[d 1 ,d 2 ,...,d M ] M×1 and radar multi-pulse echo complex data Z N×K =[z 1 ,z 2 ,...,z K ] N×K , get the logarithmic amplitude data R of N pulses in the distance unit where the radar detection point is located N×M =[r 1 ,r 2 ,...,r M ] N×M ;

[0033] S3 constructs a binary classification support vector machine whose input is the logarithmic amplitude data of N pulses in the distance unit where the radar detection point trace is located, and outputs the binary classification support vector machine of the radar detection point trace category, and uses the logarithmic amplitude data R of N pulses in the distance unit where the radar detection point trace is located in S2 N×M =[r 1 ,r 2 ,...,r M ] N×M Train a binary classification support vector machine;

[0034] S4, based on the output of S3, is a binary classification support vector machine for the radar detection point category, which is applied to the radar detection point preprocessing link in radar signal processing to determine whether the radar detection point belongs to the range ambiguity point, thereby improving the radar's suppression performance on the range ambiguity point.

[0035] S1, the radar transmitter continuously transmits N pulse signals, and the radar receiver receives radar multi-pulse echo complex data Z formed by all target scattering in the radar detection area. N×K =[z 1 ,z 2 ,...,z K ] N×K , through the target detection algorithm, the radar echo data is processed to obtain the radar detection point data d M×1 =[d 1 ,d 2 ,...,d M ] M×1 The radar detection point data contains multiple distance fuzzy point data and distance unfuzzy point data, and finally the fuzzy point data needs to be identified.

[0036] S2, detects the point data d through the radar in S1 M×1 =[d 1 ,d 2 ,...,d M ] M×1 and radar multi-pulse echo complex data Z N×K =[z 1 ,z 2 ,...,z K ] N×K , get the logarithmic amplitude data R of N pulses in the distance unit where the radar detection point is located N×M =[r 1 ,r 2 ,...,r M ] N×M ;

[0037] S3, construct a binary classification support vector machine and use the logarithmic amplitude data R of the N pulses in the distance unit where the radar detection point trace in S2 is locatedN×M =[r 1 ,r 2 ,...,r M ] N×M Train a binary classification support vector machine;

[0038] S4, using the binary classification support vector machine trained in S3, performs radar detection point trace preprocessing on the received radar signal, so as to determine which radar detection point trace data belong to range fuzzy point traces and which belong to range unfuzzy point trace data, so as to achieve the purpose of improving the radar's suppression performance on range fuzzy point traces.

[0039] [Example 2]

[0040] On the basis of Example 1, in S2, the logarithmic amplitude data R of N pulses of the distance unit where the radar detection point trace is located is obtained. N×M =[r 1 ,r 2 ,...,r M ] N×M The method is,

[0041] S01: According to the radar detection point data d in S1 M×1 =[d 1 ,d 2 ,...,d M ] M×1 , from radar multi-pulse echo complex data Z N×K =[z 1 ,z 2 ,...,z K ] N×K According to the index k=d m , m=1,2,...,M take out the echo complex data W of the distance unit where all radar detection points are located N×M =[w 1 ,w 2 ,...,w M ] N×M ,in m=1,2,...,M,w m Represents a column vector consisting of N pulse complex data of the mth radar detection point;

[0042] S02: Based on the echo complex data W of the distance unit where the radar detection point is located N×M =[w 1 ,w 2 ,...,w M ] N×M , through r m =20×log 10 (|w m|), m = 1, 2, ..., M to obtain the logarithmic amplitude data R of the distance unit where the radar detection point is located N×M =[r 1 ,r 2 ,...,r M ] N×M , where r m =20×log 10 (|w m |),m=1,2,...,M,r m A column vector consisting of the logarithmic amplitudes of the N pulses representing the distance unit where the mth radar detection point is located.

[0043] In S3, the method for training a binary classification support vector machine is:

[0044] S001: The kernel function of the binary classification support vector machine is set to be the radial basis function, the input is the logarithmic amplitude data of N pulses in the distance unit where the radar detection point trace is located, and the output is the category label of the radar detection point trace;

[0045] S002: Logarithmic amplitude data R of N pulses in the distance unit where the radar detection point trace in S2 is located N×M =[r 1 ,r 2 ,...,r M ] N×M Randomly divide it into two parts: training data set and verification data set to obtain training data set and validation dataset H is the number of training data sets, G is the number of validation data sets, Indicates that the product of M and h is rounded down;

[0046] S003: Based on the logarithmic amplitude data R of the N pulses in the distance unit where the radar detection point trace is located in S2 N×M =[r 1 ,r 2 ,...,r M ] N×M , get the training data set The corresponding label vector and validation dataset The corresponding label vector in Represents the training data set The category label of the point trace of the mth distance unit in , if the point trace belongs to the distance fuzzy point trace, the category label value is 1, otherwise it is -1; Represents the validation dataset The category label of the point trace of the nth distance unit in ;

[0047] S004: Using the training data set in S002 and its corresponding training dataset label vector in S003 Train the binary classification support vector machine in S001;

[0048] S005: The validation data set in S002 Input into the support vector machine obtained in S004 to obtain the category label predicted by the support vector machine for the validation data set

[0049] S006: By comparing S005 and the validation dataset label vector in S003 Calculate the accuracy of the validation data set on the support vector machine obtained in S004 Representation calculation and The number of equal values ​​at the same distance unit position;

[0050] S007: Determine whether the training cycle flag Q is less than the maximum number of cycles P: If it is less than P, then Q = Q + 1, go to S002; otherwise select A Q ,Q=1,2,...,the support vector machine corresponding to the maximum value in P is used as the final support vector machine for radar range fuzzy point classification.

[0051] The category label of the distance blurred point trace is 1, the category label of the distance unblurred point trace is -1, the training cycle flag Q=1 is set, and the maximum number of cycles P=10.

[0052] In S004, the training algorithm of the binary support vector machine is a sequential minimum optimization algorithm.

[0053] In S002, the logarithmic amplitude data R of the N pulses of the distance unit where the radar detection point is located is N×M =[r 1 ,r 2 ,...,r M ] N×M The dataset is randomly divided into a training data set and a validation data set according to the ratio of h:g along the distance dimension, where h+g=1,h>g.

[0054] In S1, radar detection point data d is obtained by processing radar echo complex data using a radar constant false alarm rate detection method. M×1 =[d 1 ,d 2 ,...,d M ] M×1 .

[0055] Simulation parameters

[0056] Matlab software is used to simulate the echo complex data of radar targets and radar detection point trace data. The number of category labels of radar detection point traces is two categories. The experimental simulation parameters are set as the number of pulses N=6, the number of unambiguous distance points is 10000, the number of fuzzy distance points is 10000, and the fuzzy distance points contain 2000 one-time fuzzy distance points, 2000 two-time fuzzy distance points, 2000 three-time fuzzy distance points, 2000 four-time fuzzy distance points, and 2000 five-time fuzzy distance points. A total of 5 data sets are randomly simulated to evaluate the method of the present invention, and the classification accuracy is used to evaluate the performance of the method of the present invention.

[0057] 2. Simulation experiment content

[0058] In the simulation experiment, the existing hard threshold method and the estimation method of the present invention are used to classify the radar detection point data respectively, and bar graphs corresponding to the accurate number of classifications and the classification accuracy of the radar detection points obtained by the present invention and the existing method on different data sets are drawn.

[0059] The experimental results are as follows Figure 2 As shown, the horizontal axis represents the data set number and the vertical axis represents the classification accuracy. Figure 2 The black-marked bar graph represents the existing hard threshold method, and the white-marked bar graph represents the method of the present invention.

[0060] Figure 2 In the first data set, the number of distance unblurred points and distance blurred points accurately classified by the hard threshold method is 14423, and the number of distance unblurred points and distance blurred points accurately classified by the method of the present invention is 18686; in the second data set, the number of distance unblurred points and distance blurred points accurately classified by the hard threshold method is 15923, and the number of distance unblurred points and distance blurred points accurately classified by the method of the present invention is 17921; in the third data set, the number of distance unblurred points and distance blurred points accurately classified by the hard threshold method is 14 753, the number of distance unblurred points and distance blurred points accurately classified by the method of the present invention is 18442; in the fourth group of data sets, the number of distance unblurred points and distance blurred points accurately classified by the hard threshold method is 16440, and the number of distance unblurred points and distance blurred points accurately classified by the method of the present invention is 17994; in the fifth group of data sets, the number of distance unblurred points and distance blurred points accurately classified by the hard threshold method is 13007, and the number of distance unblurred points and distance blurred points accurately classified by the method of the present invention is 17400. Figure 2It can be seen that for radar range fuzzy point trace data, the classification accuracy of the present invention is significantly higher than that of the existing method.

Claims

1. A radar range fuzzy point classification method based on support vector machine, characterized in that The following steps are involved: S1, receives radar multi-pulse echo complex data scattered by all objects in the radar detection area , and process the radar echo complex data to obtain radar detection point data ,in K Indicates the number of range cells contained in the radar data. Indicates k Distance unit N The column vector of pulse echoes is M Indicates the number of radar detection points, Indicates m The range unit number where each radar point trace is located; S2, based on the radar detection point data in S1 and radar multi-pulse echo complex data , get the logarithmic amplitude data of N pulses in the distance unit where the radar detection point is located ; S3, constructs the input as the distance unit where the radar detection point is located N The logarithmic amplitude data of the pulses is output as a binary classification support vector machine of the radar detection point category, and the logarithmic amplitude data of the N pulses in the distance unit where the radar detection point in S2 is located is used Train a binary classification support vector machine; S4, based on the output of S3, is a binary classification support vector machine for the radar detection point category, which is applied to the radar detection point preprocessing link in radar signal processing to determine whether the radar detection point belongs to the range ambiguity point, thereby improving the radar's suppression performance on the range ambiguity point.

2. The radar distance fuzzy point classification method based on support vector machine according to claim 1 is characterized in that: In S2, the logarithmic amplitude data of N pulses of the distance unit where the radar detection point trace is located is obtained. The method is, S01: According to the radar detection point data in S1 , from radar multi-pulse echo complex data By index Take out the echo complex data of all the distance units where the radar detection points are located ,in , Indicates m Radar detection points N A column vector consisting of complex data of pulses; S02: echo complex data based on the distance unit where the radar detection point is located ,pass Get the logarithmic amplitude data of the distance unit where the radar detection point is located ,in , Indicates m The distance unit where the radar detection point is located N A column vector of the logarithmic amplitudes of the pulses.

3. The radar distance fuzzy point classification method based on support vector machine according to claim 1 is characterized in that: In S3, the method for training a binary classification support vector machine is: S001: Set the kernel function of the binary support vector machine to the radial basis function, and the input is the distance unit where the radar detection point is located. N The logarithmic amplitude data of the pulses is output as the category label of the radar detection point trace; S002: The distance unit where the radar detection point trace in S2 is located N Logarithmic amplitude data of pulses Randomly divide it into two parts: training data set and verification data set to obtain training data set and validation dataset ; , , H is the number of training data sets, G is the number of validation datasets, Express M and h The product of is rounded down; S003: Logarithmic amplitude data of N pulses in the distance unit where the radar detection point trace in S2 is located , get the training data set The corresponding label vector and validation dataset The corresponding label vector ;in Represents the training data set Middle m The category label of the point trace of the distance unit. If the point trace belongs to the distance fuzzy point trace, the category label value is 1, otherwise it is -1; Represents the validation dataset Middle n The category label of the point trace of each distance unit; S004: Using the training data set in S002 and its corresponding training dataset label vector in S003 Train the binary classification support vector machine in S001; S005: The validation data set in S002 Input into the support vector machine obtained in S004 to obtain the category label predicted by the support vector machine for the validation data set ; S006: By comparing S005 and the validation dataset label vector in S003 , calculate the accuracy of the validation data set on the support vector machine obtained in S004 , Representation calculation and The number of equal values ​​at the same distance unit position; S007: Determine the training cycle flag Q Is it less than the maximum number of cycles? P : If less than P ,but Q = Q +1, go to S002; otherwise select The support vector machine corresponding to the maximum value is used as the final support vector machine for radar range fuzzy point classification.

4. The radar distance fuzzy point classification method based on support vector machine according to claim 3 is characterized in that: The class label of the distance blurred point is 1, and the class label of the distance unblurred point is -1, setting the training loop flag Q =1, maximum number of cycles P =10.

5. The radar range fuzzy point classification method based on support vector machine according to claim 3 is characterized by: In S004, the training algorithm of the binary support vector machine is a sequential minimum optimization algorithm.

6. The radar range fuzzy point classification method based on support vector machine according to claim 3 is characterized by: In the step S002, the distance unit where the radar detection point is located is N Logarithmic amplitude data of pulses Along the distance dimension h : g The ratio of is randomly divided into two parts: training data set and validation data set. .

7. The radar range fuzzy point classification method based on support vector machine according to claim 1 is characterized by: In S1, radar detection point data is obtained by processing radar echo complex data using a radar constant false alarm rate detection method. .

Images