Radar pulse data preprocessing method and system based on sequential network identification scene

Abstract

The invention discloses a radar pulse data preprocessing method and system based on a sequential network identification scene, and the method comprises the steps: sorting radar pulse data based on the TOA (pulse arrival time) to form a radar pulse sequence, and removing the radar pulse data with the PRI (pulse repetition interval) smaller than a threshold value, the radar pulse data comprises a pulse repetition interval PRI, a carrier frequency RF and a pulse width PW; eliminating radar pulse data of which carrier frequency RF and pulse width PW simultaneously meet outliers in the radar pulse sequence based on Chebyshev inequality; cracking pulses in the radar pulse sequence are screened, repairing of the cracking pulses is completed in an auxiliary mode based on the DOA, and pulse characteristics are reconstructed; and performing segmentation and normalization processing on the radar pulse sequence so as to complete radar pulse data preprocessing. According to the method, the robustness of the algorithm is effectively enhanced while the sequence recognition accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of target recognition technology, and in particular to a radar pulse data preprocessing method and system based on a time series network recognition scene. Background Art

[0002] With the increasing complexity of electromagnetic environment and the diversification of radar systems, it is difficult to extract radar pulse sequence features, and identification and matching are prone to failure. In engineering, time series networks are gradually introduced to identify radar pulse sequences. At present, the patent literature related to radar pulse sequence feature extraction is as follows:

[0003] CN116451131A discloses a radar pulse sequence recognition method based on a self-supervised temporal convolutional network. The method utilizes continuous pulse parameter features to identify radar radiation sources, and mines and models a large number of pulse parameter sequences through a data-driven deep learning method. First, a temporal convolutional deep neural network is designed according to the characteristics of the pulse sequence, and then the self-supervised temporal convolutional network is used to extract features and train the pulse sequence. Finally, a trained pulse sequence classifier is obtained for identifying the radiation source type.

[0004] CN118707457A discloses a radar pulse stream sorting method based on wide range embedding and mask interleaving, comprising the following steps: step 1, preprocessing the target template time series signal; segmenting the time series data by sliding window; step 2, processing the data to generate simulated data, and upgrading the data from the original multivariate time series to the wide range embedding space as the input of the network; step 3, masking the wide range embedding space generated after conversion using the wide range embedding method with a probability; step 4, determining the loss function, optimizer and learning strategy to obtain a network model; step 5, reasoning the model, and adding a correction algorithm to the output result, and performing sliding window sequence segmentation on the multivariate time series signal data composed of the interleaved radar pulse description words in the reasoning stage as the input of the model reasoning stage.

[0005] CN115792848A discloses a detection method and system for state switching points of a multifunctional radar based on wavelet transform. The detection method comprises the following steps: collecting a pulse signal sequence of a multifunctional radar with multiple working modes, and obtaining a pulse signal sequence of a single radar radiation source after characteristic parameter extraction and pulse sorting; detecting and processing leaking pulses and false pulses of the pulse signal sequence of the single radar radiation source and normalizing the same to obtain a pre-processed pulse signal sequence of the single radar; performing state switching point detection based on multidimensional parameters of the pre-processed pulse signal sequence of the single radar to obtain a multidimensional parameter mutation point matrix; determining the final mutation point of the pulse signal sequence of the single radar radiation source based on the multidimensional parameter mutation point matrix to realize state switching point detection of the pulse signal sequence of the single radar radiation source and perform sequence segmentation based on the ...

[0006] However, radar pulse sequences have strong professional barriers, and conventional sequence preprocessing methods are difficult to match the timing network model well, which affects the final recognition results. Summary of the invention

[0007] In order to solve the above problems, the present invention proposes a radar pulse data preprocessing method and system based on time series network recognition scene, which improves the sequence recognition accuracy and effectively enhances the robustness of the algorithm.

[0008] The technical solution adopted by the present invention is as follows:

[0009] A radar pulse data preprocessing method based on a time series network recognition scene comprises:

[0010] Based on the pulse arrival time TOA, radar pulse data are sorted to form a radar pulse sequence, and radar pulse data with a pulse repetition interval PRI less than a threshold are eliminated, wherein the radar pulse data includes a pulse repetition interval PRI, a carrier frequency RF, and a pulse width PW;

[0011] Based on Chebyshev inequality, radar pulse data with carrier frequency RF and pulse width PW that meet outliers in radar pulse sequence are eliminated;

[0012] Screen the cracked pulses in the radar pulse sequence, and assist in repairing the cracked pulses based on the pulse arrival angle DOA to reconstruct the pulse characteristics;

[0013] The radar pulse sequence is segmented and normalized to complete the radar pulse data preprocessing.

[0014] Furthermore, in the radar pulse data, each pulse is represented by a three-dimensional feature vector X n = {PRI n ,RF n ,PW n} to characterize, where PRI nis the pulse repetition interval characteristic of the nth pulse, RF n is the carrier frequency characteristic of the nth pulse, PW n is the pulse width characteristic of the nth pulse.

[0015] Furthermore, the method of eliminating radar pulse data whose carrier frequency RF and pulse width PW in the radar pulse sequence satisfy outliers based on the Chebyshev inequality includes:

[0016] For any ε>0, we have:

[0017]

[0018] Among them, the standard deviation D(X) = δ reflects the degree of dispersion of the data from the mean value. When ε = 5δ:

[0019]

[0020] That is, for the current radar pulse sequence, 96% of the radar pulse data are within 5 standard deviations of the mean, and 4% of the radar pulse data are outside 5 standard deviations of the mean, i.e., outliers;

[0021] Eliminate the radar pulse data whose carrier frequency RF and pulse width PW satisfy the outliers in the radar pulse sequence:

[0022] X delete =X dn {RF n -E(RF n )|≥5δ}∩X dn {RF n -E(RF n )|≥5δ}

[0023] X pn =X dn {X dn -X delete}

[0024] Among them, X delete represents the outlier pulse sequence that needs to be removed, X dn Indicates the elimination of pulse sequences that do not meet the pulse repetition interval PRI threshold condition, X pn Represents the pulse sequence after removing characteristic outliers.

[0025] Furthermore, the method of screening the cracked pulses in the radar pulse sequence and assisting in repairing the cracked pulses based on the pulse arrival angle DOA to reconstruct the pulse characteristics includes:

[0026] Traverse each type of radar pulse data and select the pulses whose adjacent pulse repetition interval PRI is smaller than the adjacent pulse width PW, that is, the pulses that meet TOA n-TOA n-1 <PW n-1 or TOA n -TOA n-1 <PW n The n-1th and nth pulses of the condition;

[0027] If |DOA n -DOA n-1 |<σ, where the parameter σ is determined by the direction finding error of the data environment, it is considered that the n-1th and nth pulses are split from the same pulse, so they are merged to obtain the reconstructed pulse. The characteristics of the reconstructed pulse are as follows:

[0028] TOA=TOA n-1

[0029] PW=TOA n +PW n -TOA n-1

[0030] Keep iterating until no pulses with adjacent pulse repetition intervals PRI smaller than adjacent pulse widths PW are found.

[0031] Furthermore, the segmenting and normalizing of the radar pulse sequence includes:

[0032] Count the histogram of the pulse repetition interval PRI in each type of radar pulse data, and select the value PRI corresponding to the maximum histogram value center , set the segment time interval threshold T thre = k PRI center , where k is a positive integer;

[0033] The pulse arrival time TOA of the traversal radar pulse sequence is calculated when the adjacent pulse repetition interval PRI is greater than the segment time interval threshold T thre Segmentation is performed to obtain sequence libraries L of varying lengths. n ;

[0034] If the length of the sample sequence is L seq , when L n <0.5·L seq When the sequence library L n Delete when L seq >L n >0.5·L seq When the sequence is filled with 0 to the same L seq equal length; when L n >L seq When the step size is 1 and the window length is L seq Perform sliding window segmentation;

[0035] For the segmented radar pulse sequence, the pulse repetition interval PRI feature of the first pulse of each segment is set to 0, and the pulse repetition interval PRI, pulse width PW and carrier frequency RF are normalized by mean variance to obtain sequence samples of three-dimensional feature vectors. The time series neural network is trained based on the processed sequence samples.

[0036] When testing data, the step length is 0.5·L seq Slide the window for the test data; there will be multiple recognition results within the reported beat time range, and the recognition result with the highest frequency will be selected as the final recognition result of the current beat.

[0037] A radar pulse data preprocessing system based on time series network scene recognition, comprising:

[0038] A primary processing module is configured to sort radar pulse data based on pulse arrival time TOA to form a radar pulse sequence, and remove radar pulse data with a pulse repetition interval PRI less than a threshold, wherein the radar pulse data includes a pulse repetition interval PRI, a carrier frequency RF and a pulse width PW;

[0039] The secondary processing module is configured to remove radar pulse data whose carrier frequency RF and pulse width PW in the radar pulse sequence satisfy outliers based on Chebyshev inequality;

[0040] A cracking pulse processing module is configured to screen cracking pulses in the radar pulse sequence, and assist in repairing the cracking pulses based on the pulse arrival angle DOA, and reconstruct the pulse characteristics;

[0041] The segmentation and normalization processing module is configured to perform segmentation and normalization processing on the radar pulse sequence, thereby completing radar pulse data preprocessing.

[0042] Furthermore, in the radar pulse data processed by the primary processing module, each pulse is represented by a three-dimensional feature vector X n = {PRI n ,RF n ,PW n} to characterize, where PRI n is the pulse repetition interval characteristic of the nth pulse, RF n is the carrier frequency characteristic of the nth pulse, PW n is the pulse width characteristic of the nth pulse.

[0043] Furthermore, in the secondary processing module, radar pulse data whose carrier frequency RF and pulse width PW in the radar pulse sequence satisfy outliers are eliminated based on Chebyshev inequality, including:

[0044] For any ε>0, we have:

[0045]

[0046] Among them, the standard deviation D(X) = δ reflects the degree of dispersion of the data from the mean value. When ε = 5δ:

[0047]

[0048] That is, for the current radar pulse sequence, 96% of the radar pulse data are within 5 standard deviations of the mean, and 4% of the radar pulse data are outside 5 standard deviations of the mean, i.e., outliers;

[0049] Eliminate the radar pulse data whose carrier frequency RF and pulse width PW satisfy the outliers in the radar pulse sequence:

[0050] X delete =X dn {RF n -E(RF n )|≥5δ}∩X dn {RF n -E(RF n )|≥5δ}

[0051] X pn =X dn {X dn -X delete}

[0052] Among them, X delete represents the outlier pulse sequence that needs to be removed, X dn Indicates the elimination of pulse sequences that do not meet the pulse repetition interval PRI threshold condition, X pn Represents the pulse sequence after removing characteristic outliers.

[0053] Furthermore, in the cracking pulse processing module, the cracking pulses in the radar pulse sequence are screened, and the cracking pulses are repaired based on the pulse arrival angle DOA to reconstruct the pulse characteristics, including:

[0054] Traverse each type of radar pulse data and select the pulses whose adjacent pulse repetition interval PRI is smaller than the adjacent pulse width PW, that is, the pulses that meet TOA n -TOA n-1 <PW n-1 or TOA n -TOA n-1 <PW n The n-1th and nth pulses of the condition;

[0055] If |DOA n -DOA n-1|<σ, where the parameter σ is determined by the direction finding error of the data environment, it is considered that the n-1th and nth pulses are split from the same pulse, so they are merged to obtain the reconstructed pulse. The characteristics of the reconstructed pulse are as follows:

[0056] TOA=TOA n-1

[0057] PW=TOA n +PW n -TOA n-1

[0058] Keep iterating until no pulses with adjacent pulse repetition intervals PRI smaller than adjacent pulse widths PW are found.

[0059] Furthermore, in the segmentation and normalization processing module, segmentation and normalization processing are performed on the radar pulse sequence, including:

[0060] Count the histogram of the pulse repetition interval PRI in each type of radar pulse data, and select the value PRI corresponding to the maximum histogram value center , set the segment time interval threshold T thre = k PRI center , where k is a positive integer;

[0061] The pulse arrival time TOA of the traversal radar pulse sequence is calculated when the adjacent pulse repetition interval PRI is greater than the segment time interval threshold T thre Segmentation is performed to obtain sequence libraries L of varying lengths. n ;

[0062] If the length of the sample sequence is L seq , when L n <0.5·L seq When the sequence library L n Delete when L seq >L n >0.5·L seq When the sequence is filled with 0 to the same L seq equal length; when L n >L seq When the step size is 1 and the window length is L seq Perform sliding window segmentation;

[0063] For the segmented radar pulse sequence, the pulse repetition interval PRI feature of the first pulse of each segment is set to 0, and the pulse repetition interval PRI, pulse width PW and carrier frequency RF are normalized by mean variance to obtain sequence samples of three-dimensional feature vectors. The time series neural network is trained based on the processed sequence samples.

[0064] When testing, the step length is 0.5 L seqSlide the window for the test data; there will be multiple recognition results within the reported beat time range, and the recognition result with the highest frequency will be selected as the final recognition result of the current beat.

[0065] The beneficial effects of the present invention are:

[0066] The present invention sorts radar pulse data based on pulse arrival time TOA and eliminates pulses with too small pulse repetition interval PRI, then removes pulse data outliers based on Chebyshev inequality, and then assists in repairing cracked pulses based on pulse arrival angle DOA, and finally achieves segmentation and normalization of radar pulse sequences. In addition, for test data, normalization and sliding window processing are performed, and the results of multiple windows within the reporting time interval are summarized, and the one with the largest proportion is taken as the final recognition result of the current beat. While improving the accuracy of sequence recognition, the present invention effectively enhances the robustness of the algorithm. Under the same data scenario and the same timing model conditions, the data preprocessing method of the present invention is about 5% more accurate than the model trained by conventional methods, and has stronger robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a flow chart of a radar pulse data preprocessing method based on a timing network recognition scene according to Example 1 of the present invention.

[0068] Figure 2 This is a flow chart of another radar pulse data preprocessing method based on timing network scene recognition in Example 1 of the present invention.

[0069] Figure 3 It is a schematic diagram of radar pulse data representation according to Example 1 of the present invention. DETAILED DESCRIPTION

[0070] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, the specific implementation methods of the present invention are now described. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the embodiments described are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0071] Example 1

[0072] With the increasing complexity of electromagnetic environment and the diversification of radar systems, it is difficult to extract radar pulse sequence features and identification and matching are prone to failure. Time series networks are gradually introduced in engineering to identify radar pulse sequences. However, radar pulse sequences have strong professional barriers, and conventional sequence preprocessing methods are difficult to match the time series network model well, which affects the final recognition results.

[0073] Based on this, this embodiment provides a radar pulse data preprocessing method based on a time series network recognition scene, such as Figure 1 As shown, including:

[0074] Based on the pulse arrival time TOA, radar pulse data are sorted to form a radar pulse sequence, and radar pulse data with a pulse repetition interval PRI less than a threshold are eliminated. The radar pulse data includes the pulse repetition interval PRI, carrier frequency RF and pulse width PW;

[0075] Based on Chebyshev inequality, radar pulse data with carrier frequency RF and pulse width PW that meet outliers in radar pulse sequence are eliminated;

[0076] Screen the cracked pulses in the radar pulse sequence, and assist in repairing the cracked pulses based on the pulse arrival angle DOA to reconstruct the pulse characteristics;

[0077] The radar pulse sequence is segmented and normalized to complete the radar pulse data preprocessing.

[0078] Specifically, Figure 2 As shown, the radar pulse data preprocessing method of this embodiment can be implemented by the following steps:

[0079] Step 1: Each pulse is a three-dimensional vector X consisting of pulse repetition interval PRI characteristics, carrier frequency RF characteristics, and pulse width PW characteristics. n = {PRI n ,RF n ,PW n} characterization, each type of radar pulse data received by the receiver can be represented by a series of continuous three-dimensional feature vectors, such as Figure 3 shown.

[0080] For each type of radar pulse data, sort them according to the pulse arrival time TOA, and delete the pulses with a pulse repetition interval PRI that is too small. The specific expression is as follows:

[0081] X sn =X n (Sort(TOA n ))

[0082] X dn =X sn (Find(PRI n >0.05))

[0083] Step 2: Use the 5δ method to remove outliers from pulse data. This method is actually derived from the Chebyshev inequality. For any ε>0, we have:

[0084]

[0085] The standard deviation D(X) = δ reflects the degree of dispersion of the data from the mean value. When ε = 5δ:

[0086]

[0087] That is, for the current pulse data, 96% of the pulse data are within 5 standard deviations of the mean, and 4% of the data are outside 5 standard deviations of the mean. These data are outliers.

[0088] Eliminate the pulses in which the carrier frequency RF and pulse width PW characteristics meet the outlier point. The expression is as follows:

[0089] X delete =X dn {RF n -E(RF n )|≥5δ}∩X dn {RF n -E(RF n )|≥5δ}

[0090] X pn =X dn {X dn -X delete}

[0091] Among them, X delete represents the outlier pulse sequence that needs to be removed, X dn Indicates the elimination of pulse sequences that do not meet the pulse repetition interval PRI threshold condition, X pn Represents the pulse sequence after removing characteristic outliers.

[0092] Step 3: Confirm the crack pulse, and assist in repairing the crack pulse based on the pulse arrival angle DOA, and reconstruct the pulse characteristics. Preferably, the specific processing method is as follows:

[0093] Step 3.1: Traverse each type of radar pulse data and select the pulses whose adjacent pulse intervals are smaller than the adjacent pulse widths, that is, the pulses that meet the TOA n -TOA n-1 <PW n-1 or TOA n -TOA n-1 <PW n Condition for the n-1th and nth pulses.

[0094] Step 3.2: If abs(DOA n -DOA n-1 )<σ (σ is determined by the direction finding error of the data environment and is generally set to 1°), it can be considered that the n-1th and nth pulses are split from the same pulse, and the reconstructed pulse can be obtained by merging them. The characteristics of the reconstructed pulse are as follows:

[0095] TOA=TOA n-1

[0096] PW=TOA n +PW n -TOA n-1

[0097] Step 3.3: Repeat steps 3.1 and 3.2 until no pulses whose intervals between adjacent pulses are smaller than the widths of adjacent pulses are found.

[0098] Step 4: segment the radar pulse sequence. Preferably, the specific processing method is as follows:

[0099] Step 4.1: Count the histogram of the pulse repetition interval PRI sequence of each type of radar pulse, and select the value PRI corresponding to the maximum histogram value center , set the segment time interval threshold T thre = k PRI center , k is a positive integer, usually set to 10.

[0100] Step 4.2: Traverse the pulse arrival time TOA sequence of radar pulses, and the interval between adjacent pulses is greater than T thre Segmentation is performed to obtain sequence libraries L of varying lengths. n .

[0101] Step 4.3: Assume that the length of the sample sequence is L seq , when L n <0.5·L seq When L seq >L n >0.5·L seq When the sequence is filled with 0 to the same L seq When L n >L seq When the step size is 1 and the window length is L seq Sliding window segmentation.

[0102] Step 5: For segmented pulse data, the pulse repetition interval PRI feature of the first pulse in each segment is set to 0. At the same time, the pulse repetition interval PRI, pulse width PW, and carrier frequency RF are normalized using mean variance to obtain sequence samples of three-dimensional feature vectors. The time series neural network is trained based on the processed sequence samples.

[0103] Step 6: During the test, use a step length of 0.5 L seq For the test data sliding window, there will be multiple recognition results within the reported beat time range. The one with the highest frequency among the recognition results is selected as the final recognition result of the beat.

[0104] Example 2

[0105] This embodiment provides a radar pulse data preprocessing system based on a time series network recognition scene, including:

[0106] The primary processing module is configured to sort the radar pulse data based on the pulse arrival time TOA to form a radar pulse sequence, and remove the radar pulse data with a pulse repetition interval PRI less than a threshold, the radar pulse data including the pulse repetition interval PRI, the carrier frequency RF and the pulse width PW;

[0107] The secondary processing module is configured to remove radar pulse data whose carrier frequency RF and pulse width PW in the radar pulse sequence satisfy outliers based on Chebyshev inequality;

[0108] A cracking pulse processing module is configured to screen cracking pulses in the radar pulse sequence, and assist in repairing the cracking pulses based on the pulse arrival angle DOA, and reconstruct the pulse characteristics;

[0109] The segmentation and normalization processing module is configured to perform segmentation and normalization processing on the radar pulse sequence, thereby completing radar pulse data preprocessing.

[0110] Preferably, in the radar pulse data processed by the primary processing module, each pulse is represented by a three-dimensional feature vector X n = {PRI n ,RF n ,PW n} to characterize, where PRI n is the pulse repetition interval characteristic of the nth pulse, RF n is the carrier frequency characteristic of the nth pulse, PW n is the pulse width characteristic of the nth pulse.

[0111] Preferably, in the secondary processing module, radar pulse data whose carrier frequency RF and pulse width PW in the radar pulse sequence satisfy outliers are eliminated based on Chebyshev inequality, including:

[0112] For any ε>0, we have:

[0113]

[0114] Among them, the standard deviation D(X) = δ reflects the degree of dispersion of the data from the mean value. When ε = 5δ:

[0115]

[0116] That is, for the current radar pulse sequence, 96% of the radar pulse data are within 5 standard deviations of the mean, and 4% of the radar pulse data are outside 5 standard deviations of the mean, i.e., outliers;

[0117] Eliminate the radar pulse data whose carrier frequency RF and pulse width PW satisfy the outliers in the radar pulse sequence:

[0118] X delete =X dn {RF n -E(RF n )|≥5δ}∩X dn {RF n -E(RF n )|≥5δ}

[0119] X pn =X dn {X dn -X delete}

[0120] Among them, X delete represents the outlier pulse sequence that needs to be removed, X dn Indicates the elimination of pulse sequences that do not meet the pulse repetition interval PRI threshold condition, X pn Represents the pulse sequence after removing characteristic outliers.

[0121] Preferably, in the cracking pulse processing module, the cracking pulses in the radar pulse sequence are screened, and the repair of the cracking pulses is assisted based on the pulse arrival angle DOA to reconstruct the pulse characteristics, including:

[0122] Traverse each type of radar pulse data and select the pulses whose adjacent pulse repetition interval PRI is smaller than the adjacent pulse width PW, that is, the pulses that meet TOA n -TOA n-1 <PW n-1 or TOA n -TOA n-1 <PW n The n-1th and nth pulses of the condition;

[0123] If |DOA n -DOA n-1 |<σ, where the parameter σ is determined by the direction finding error of the data environment, it is considered that the n-1th and nth pulses are split from the same pulse, so they are merged to obtain the reconstructed pulse. The characteristics of the reconstructed pulse are as follows:

[0124] TOA=TOA n-1

[0125] PW=TOA n +PW n -TOA n-1

[0126] Keep iterating until no pulses with adjacent pulse repetition intervals PRI smaller than adjacent pulse widths PW are found.

[0127] Preferably, in the segmentation and normalization processing module, segmentation and normalization processing is performed on the radar pulse sequence, including:

[0128] Count the histogram of the pulse repetition interval PRI in each type of radar pulse data, and select the value PRI corresponding to the maximum histogram value center , set the segment time interval threshold T thre = k PRI center , where k is a positive integer;

[0129] The pulse arrival time TOA of the traversal radar pulse sequence is calculated when the adjacent pulse repetition interval PRI is greater than the segment time interval threshold T thre Segmentation is performed to obtain sequence libraries L of varying lengths. n ;

[0130] If the length of the sample sequence is L seq , when L n <0.5·L seq When the sequence library L n Delete when L seq >L n >0.5·L seq When the sequence is filled with 0 to the same L seq equal length; when L n >L seq When the step size is 1 and the window length is L seq Perform sliding window segmentation;

[0131] For the segmented radar pulse sequence, the pulse repetition interval PRI feature of the first pulse of each segment is set to 0, and the pulse repetition interval PRI, pulse width PW and carrier frequency RF are normalized by mean variance to obtain sequence samples of three-dimensional feature vectors. The time series neural network is trained based on the processed sequence samples.

[0132] When testing, the step length is 0.5 L seq Slide the window for the test data; there will be multiple recognition results within the reported beat time range, and the recognition result with the highest frequency will be selected as the final recognition result of the current beat.

[0133] Example 3

[0134] This embodiment is based on embodiment 1:

[0135] This embodiment provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the radar pulse data preprocessing method based on time series network scene recognition of embodiment 1 is implemented. The computer program may be in source code form, object code form, executable file, or some intermediate form.

[0136] Example 4

[0137] This embodiment is based on embodiment 1:

[0138] This embodiment provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the radar pulse data preprocessing method based on timing network recognition scene of embodiment 1. The computer program may be in source code form, object code form, executable file or some intermediate form, etc. The storage medium includes: any entity or device capable of carrying computer program code, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the storage medium does not include electric carrier signals and telecommunication signals.

[0139] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

Claims

1. A radar pulse data preprocessing method based on time series network scene recognition, characterized in that: include: Based on the pulse arrival time TOA, radar pulse data are sorted to form a radar pulse sequence, and radar pulse data with a pulse repetition interval PRI less than a threshold are eliminated, wherein the radar pulse data includes a pulse repetition interval PRI, a carrier frequency RF and a pulse width PW; Based on Chebyshev inequality, radar pulse data with carrier frequency RF and pulse width PW that meet outliers in radar pulse sequence are eliminated; Screen the cracked pulses in the radar pulse sequence, and assist in repairing the cracked pulses based on the pulse arrival angle DOA to reconstruct the pulse characteristics; The radar pulse sequence is segmented and normalized to complete the radar pulse data preprocessing.

2. The radar pulse data preprocessing method based on time series network scene recognition according to claim 1 is characterized in that: In the radar pulse data, each pulse is represented by a three-dimensional feature vector X n = {PRI n ,RF n ,PW n } to characterize, where PRI n is the pulse repetition interval characteristic of the nth pulse, RF n is the carrier frequency characteristic of the nth pulse, PW n is the pulse width characteristic of the nth pulse.

3. The radar pulse data preprocessing method based on time series network scene recognition according to claim 2 is characterized in that: The method of eliminating radar pulse data in a radar pulse sequence where both carrier frequency RF and pulse width PW satisfy outliers based on Chebyshev inequality includes: For any ε>0, we have: Among them, the standard deviation D(X) = δ reflects the degree of dispersion of the data from the mean value. When ε = 5δ: That is, for the current radar pulse sequence, 96% of the radar pulse data are within 5 standard deviations of the mean, and 4% of the radar pulse data are outside 5 standard deviations of the mean, i.e., outliers; Eliminate the radar pulse data whose carrier frequency RF and pulse width PW satisfy the outliers in the radar pulse sequence: X delete =X dn {RF n -E(RF n )|≥5δ}∩X dn {RF n -E(RF n )|≥5δ} X pn =X dn {X dn -X delete } Among them, X delete represents the outlier pulse sequence that needs to be removed, X dn Indicates the elimination of pulse sequences that do not meet the pulse repetition interval PRI threshold condition, X pn Represents the pulse sequence after removing characteristic outliers.

4. The radar pulse data preprocessing method based on time series network scene recognition according to claim 3 is characterized in that: The method of screening the cracked pulses in the radar pulse sequence and assisting in repairing the cracked pulses based on the pulse arrival angle DOA to reconstruct the pulse characteristics includes: Traverse each type of radar pulse data and select the pulses whose adjacent pulse repetition interval PRI is smaller than the adjacent pulse width PW, that is, the pulses that meet TOA n -TOA n-1 <PW n-1 or TOA n -TOA n-1 <PW n The n-1th and nth pulses of the condition; If |DOA n -DOA n-1 |<σ, where the parameter σ is determined by the direction finding error of the data environment, it is considered that the n-1th and nth pulses are split from the same pulse, so they are merged to obtain the reconstructed pulse. The characteristics of the reconstructed pulse are as follows: TOA=TOA n-1 PW = TOA n +PW n -TOA n-1 Keep iterating until no pulses with adjacent pulse repetition intervals PRI smaller than adjacent pulse widths PW are found.

5. The radar pulse data preprocessing method based on time series network scene recognition according to claim 2 is characterized in that: The segmentation and normalization processing of the radar pulse sequence includes: Count the histogram of the pulse repetition interval PRI in each type of radar pulse data, and select the value PRI corresponding to the maximum histogram value center , set the segment time interval threshold T thre = k PRI center , where k is a positive integer; The pulse arrival time TOA of the traversal radar pulse sequence is calculated when the adjacent pulse repetition interval PRI is greater than the segment time interval threshold T thre Segmentation is performed to obtain sequence libraries L of varying lengths. n ; If the length of the sample sequence is L seq , when L n <0.5·L seq When the sequence library L n Delete when L seq >L n >0.5·L seq When the sequence is filled with 0 to the same L seq equal length; when L n >L seq When the step size is 1 and the window length is L seq Perform sliding window segmentation; For the segmented radar pulse sequence, the pulse repetition interval PRI feature of the first pulse of each segment is set to 0, and the pulse repetition interval PRI, pulse width PW and carrier frequency RF are normalized by mean variance to obtain sequence samples of three-dimensional feature vectors. The time series neural network is trained based on the processed sequence samples. When testing data, the step length is 0.5·L seq Slide the window for the test data; there will be multiple recognition results within the reported beat time range, and the recognition result with the highest frequency will be selected as the final recognition result of the current beat.

6. A radar pulse data preprocessing system based on time series network scene recognition, characterized in that: include: A primary processing module is configured to sort radar pulse data based on pulse arrival time TOA to form a radar pulse sequence, and remove radar pulse data with a pulse repetition interval PRI less than a threshold, wherein the radar pulse data includes a pulse repetition interval PRI, a carrier frequency RF and a pulse width PW; The secondary processing module is configured to remove radar pulse data whose carrier frequency RF and pulse width PW in the radar pulse sequence satisfy outliers based on Chebyshev inequality; A cracking pulse processing module is configured to screen cracking pulses in the radar pulse sequence, and assist in repairing the cracking pulses based on the pulse arrival angle DOA, and reconstruct the pulse characteristics; The segmentation and normalization processing module is configured to perform segmentation and normalization processing on the radar pulse sequence, thereby completing radar pulse data preprocessing.

7. The radar pulse data preprocessing system based on time series network scene recognition according to claim 6 is characterized in that: In the radar pulse data processed by the primary processing module, each pulse is represented by a three-dimensional feature vector X n = {PRI n ,RF n ,PW n } to characterize, where PRI n is the pulse repetition interval characteristic of the nth pulse, RF n is the carrier frequency characteristic of the nth pulse, PW n is the pulse width characteristic of the nth pulse.

8. The radar pulse data preprocessing system based on time series network scene recognition according to claim 7 is characterized in that: In the secondary processing module, radar pulse data whose carrier frequency RF and pulse width PW in the radar pulse sequence satisfy outliers are eliminated based on Chebyshev inequality, including: For any ε>0, we have: Among them, the standard deviation D(X) = δ reflects the degree of dispersion of the data from the mean value. When ε = 5δ: That is, for the current radar pulse sequence, 96% of the radar pulse data are within 5 standard deviations of the mean, and 4% of the radar pulse data are outside 5 standard deviations of the mean, i.e., outliers; Eliminate the radar pulse data whose carrier frequency RF and pulse width PW satisfy the outliers in the radar pulse sequence: X delete =X dn {|RF n -E(RF n )|≥5δ}∩X dn {|RF n -E(RF n )|≥5δ} X pn =X dn {X dn -X delete } Among them, X delete represents the outlier pulse sequence that needs to be removed, X dn Indicates the elimination of pulse sequences that do not meet the pulse repetition interval PRI threshold condition, X pn Represents the pulse sequence after removing characteristic outliers.

9. The radar pulse data preprocessing system based on time series network scene recognition according to claim 8 is characterized in that: In the cracking pulse processing module, the cracking pulses in the radar pulse sequence are screened, and the cracking pulses are repaired based on the pulse arrival angle DOA to reconstruct the pulse characteristics, including: Traverse each type of radar pulse data and select the pulses whose adjacent pulse repetition interval PRI is smaller than the adjacent pulse width PW, that is, the pulses that meet TOA n -TOA n-1 <PW n-1 or TOA n -TOA n-1 <PW n The n-1th and nth pulses of the condition; If |DOA n -DOA n-1 |<σ, where the parameter σ is determined by the direction finding error of the data environment, it is considered that the n-1th and nth pulses are split from the same pulse, so they are merged to obtain the reconstructed pulse. The characteristics of the reconstructed pulse are as follows: TOA=TOA n-1 PW = TOA n +PW n -TOA n-1 Keep iterating until no pulses with adjacent pulse repetition intervals PRI smaller than adjacent pulse widths PW are found.

10. The radar pulse data preprocessing system based on time series network scene recognition according to claim 7, characterized in that: In the segmentation and normalization processing module, segmentation and normalization processing are performed on the radar pulse sequence, including: Count the histogram of the pulse repetition interval PRI in each type of radar pulse data, and select the value PRI corresponding to the maximum histogram value center , set the segment time interval threshold T thre = k PRI center , where k is a positive integer; The pulse arrival time TOA of the traversal radar pulse sequence is calculated when the adjacent pulse repetition interval PRI is greater than the segment time interval threshold T thre Segmentation is performed to obtain sequence libraries L of varying lengths. n ; If the length of the sample sequence is L seq , when L n <0.5·L seq When the sequence library L n Delete when L seq >L n >0.5·L seq When the sequence is filled with 0 to the same L seq equal length; when L n >L seq When the step size is 1 and the window length is L seq Perform sliding window segmentation; For the segmented radar pulse sequence, the pulse repetition interval PRI feature of the first pulse of each segment is set to 0, and the pulse repetition interval PRI, pulse width PW and carrier frequency RF are normalized by mean variance to obtain sequence samples of three-dimensional feature vectors. The time series neural network is trained based on the processed sequence samples. When testing, the step length is 0.5 L seq Slide the window for the test data; there will be multiple recognition results within the reported beat time range, and the recognition result with the highest frequency will be selected as the final recognition result of the current beat.

Images