Radar working mode identification method based on pulse description graph and comparative learning

Abstract

The invention discloses a radar working mode recognition method based on a pulse description graph and comparative learning, and the method comprises the steps: obtaining a radar pulse sequence which comprises a plurality of pulse description words arranged according to a time sequence; performing pulse waveform parameter feature transformation on the pulse description words in the radar pulse sequence to obtain a pulse waveform parameter feature sequence; performing pulse repetition frequency mode feature transformation on the pulse description words in the radar pulse sequence to obtain a pulse repetition frequency mode feature sequence; adding the pulse waveform parameter feature sequence and the pulse repetition frequency mode feature sequence to obtain a label-free pulse description graph; performing bilinear interpolation and normalization processing on the unlabeled pulse description graph to obtain a processed unlabeled pulse description graph; and inputting the processed label-free pulse description graph into the trained working mode recognition model for processing to obtain the predicted current working mode of the radar. According to the invention, many radar working modes can be identified and the identification accuracy is high.

Description

Technical Field

[0001] The invention belongs to the technical field of radar electronic reconnaissance, and in particular relates to a radar working mode recognition method based on a pulse description diagram and contrast learning. Background Art

[0002] Radar working mode recognition is an important part of the field of electronic reconnaissance technology. The current radar working modes are developing towards diversification, and the signal waveforms are complex and changeable. Among them, multi-function radars have a large and stable signal library, which can flexibly adjust signal parameters for different tasks and switch between different working modes. The transmission signal parameters have the characteristics of sensitivity, adaptability and overlap. In addition, the traditional clustering algorithm has failure problems in complex scenarios such as lost pulse scenarios and random noise scenarios, making radar working mode recognition more and more difficult. Therefore, radar working mode recognition technology has received more and more attention.

[0003] In the prior art, Wang Yubing et al. proposed a judgment system based on DS evidence theory in their paper “Determination of Airborne Fire Control Radar Air-to-Air Working Mode Based on DS Evidence Theory” (Modern Radar, 2017, 39(05): 79-84). First, feature parameters are extracted from the detected enemy radar signals and a radar signal feature library is constructed. The mode grouping under a single platform is quickly realized through a pre-judgment process. Then, the DS evidence theory is applied to the grouped multi-dimensional space pattern recognition results, and a distributed structure is used to perform single-platform multi-period time domain fusion and multi-platform airspace information fusion to realize radar working mode determination. The simulation conditions of this method are too ideal. The sample data only contains a single working mode during the judgment, and fails to restore the scene of multiple working modes mixed during the actual interception and reception, which makes its recognition accuracy low in complex scenes such as lost pulse scenes and random noise scenes.

[0004] Therefore, there is an urgent need to provide a radar working mode recognition method to improve the above defects. Summary of the invention

[0005] In order to solve the above problems existing in the prior art, the present invention provides a radar working mode recognition method based on pulse description diagram and contrast learning. The technical problem to be solved by the present invention is achieved by the following technical solutions:

[0006] In a first aspect, the present invention provides a radar working mode recognition method based on a pulse description diagram and contrast learning, comprising:

[0007] Acquire a radar pulse sequence, where the radar pulse sequence includes a plurality of pulse description words arranged in time sequence;

[0008] Perform pulse waveform parameter feature transformation on the pulse description words in the radar pulse sequence to obtain a pulse waveform parameter feature sequence;

[0009] The pulse description words in the radar pulse sequence are transformed into pulse repetition frequency mode characteristics to obtain a pulse repetition frequency mode characteristic sequence;

[0010] The pulse waveform parameter characteristic sequence and the pulse repetition rate pattern characteristic sequence are added together to obtain an unlabeled pulse description graph;

[0011] Performing bilinear interpolation and normalization processing on the unlabeled pulse description graph to obtain a processed unlabeled pulse description graph;

[0012] The processed unlabeled pulse description graph is input into the trained working mode recognition model for processing to obtain the predicted current working mode of the radar.

[0013] Beneficial effects of the present invention:

[0014] The present invention provides a radar working mode recognition method based on pulse description diagram and contrastive learning. The method performs pulse waveform parameter feature transformation and pulse repetition frequency mode feature transformation on an acquired radar pulse sequence to obtain an unlabeled pulse description diagram. The unlabeled pulse description diagram is then subjected to bilinear interpolation and normalization processing in sequence. The working mode is then recognized using a trained working mode recognition model. In this way, the method is not limited to a single working mode of the radar, more radar working modes can be recognized with a high recognition accuracy, and a scene in which a plurality of radar working modes are mixed can be restored. In the case of poor interception conditions, pulse loss and noise pulses, a high recognition accuracy can still be achieved.

[0015] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flow chart of a radar working mode recognition method based on pulse description diagram and contrast learning provided by an embodiment of the present invention;

[0017] Figure 2 It is a schematic diagram of a radar working mode recognition method based on pulse description diagram and contrast learning provided by an embodiment of the present invention;

[0018] Figure 3 It is a schematic diagram of a training working mode recognition model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The present invention is further described in detail below with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.

[0020] See also Figure 1 and Figure 2 , Figure 1is a flow chart of a radar working mode recognition method based on pulse description diagram and contrast learning provided by an embodiment of the present invention, Figure 2 1 is a schematic diagram of a radar working mode recognition method based on a pulse description diagram and contrastive learning provided by an embodiment of the present invention. The radar working mode recognition method based on a pulse description diagram and contrastive learning provided by the present invention includes:

[0021] S101. Acquire a radar pulse sequence, where the radar pulse sequence includes a plurality of pulse description words arranged in time sequence.

[0022] Specifically, in this embodiment, a radar pulse sequence is obtained, and parameter information such as pulse arrival time (TOA), pulse width (PW), pulse amplitude (PA), pulse bandwidth (BW), pulse repetition period (PRI), intra-pulse modulation characteristics (PM), pulse compression ratio (PCR), and duty cycle (D) in the radar pulse sequence is determined by parameter measurement to form a pulse description word (PDW), which is expressed as follows:

[0023] PDW=[TOA,PW,PA,BW,PM,PCR,DR];

[0024] The pulse description words arranged in time sequence constitute the radar pulse sequence, which is expressed as:

[0025] PDW={PDW1,PDW2,…,PDW n}.

[0026] S102, performing pulse waveform parameter feature transformation on the pulse description words in the radar pulse sequence to obtain a pulse waveform parameter feature sequence.

[0027] Specifically, in this embodiment, the pulse waveform parameter characteristic sequence is obtained through the following process.

[0028] The parameters corresponding to each pulse description word in the radar pulse sequence are converted into an L-dimensional vector, and each dimensional vector is linearly normalized and the value is limited to [-1,1] to obtain the pulse waveform parameter characteristics; wherein, the expression of the pulse waveform parameter characteristics is:

[0029]

[0030] Among them, p ij Indicates the value of the jth parameter corresponding to the i-th pulse description word, i represents the index of the pulse description word, j represents the parameter index, minp j Indicates the minimum value of the jth parameter corresponding to all pulse description words, maxp j Indicates the maximum value of the jth parameter corresponding to all pulse description words;

[0031] According to the characteristics of multiple pulse waveform parameters, a pulse waveform parameter characteristic sequence is constructed, and its expression is:

[0032] P = {p1 T ,p2 T ,…,p n T} T ;

[0033] Among them, p k represents the parameter vector corresponding to the kth pulse description word, k∈[1,n], n represents the length of the radar pulse sequence, and P represents a matrix of size n×L.

[0034] S103, performing pulse repetition frequency pattern feature transformation on the pulse description words in the radar pulse sequence to obtain a pulse repetition frequency pattern feature sequence.

[0035] Specifically, in this embodiment, the pulse repetition frequency pattern characteristic sequence is obtained through the following process.

[0036] According to the pulse arrival time parameters corresponding to the pulse description words in the radar pulse sequence, the pulse description words are time-series encoded to obtain the pulse repetition frequency pattern characteristics, which are expressed as follows:

[0037]

[0038] Where i represents the dimension index of the pulse arrival time vector, i∈[0,L), TOA k represents the pulse arrival time of the kth pulse description word, k∈[0,L+1 / 2), maxt represents the maximum pulse arrival time corresponding to all pulse description words, t oak Indicates the pulse arrival time of the kth pulse description word;

[0039] According to multiple pulse repetition frequency pattern characteristics, a pulse repetition frequency pattern characteristic sequence is constructed, and its expression is:

[0040] Te={te1 T ,te2 T ,…,te n T} T ;

[0041] Where n is the length of the radar pulse sequence.

[0042] S104, adding the pulse waveform parameter characteristic sequence and the pulse repetition rate pattern characteristic sequence to obtain an unlabeled pulse description graph.

[0043] Specifically, in this embodiment, the expression of the unlabeled pulse description graph is:

[0044] X={x1,x2,…,x n},x k =p k +te k ,k∈[1,n];

[0045] Among them, p k Indicates the pulse waveform parameter characteristics, te k Indicates the pulse repetition rate pattern characteristics.

[0046] S105 , performing bilinear interpolation and normalization processing on the unlabeled pulse description graph to obtain a processed unlabeled pulse description graph.

[0047] Specifically, in this embodiment, the processed unlabeled pulse description graph is obtained through the following process.

[0048] The unlabeled pulse description map is scaled using bilinear interpolation;

[0049] The scaled unlabeled pulse description graph is normalized to obtain a processed unlabeled pulse description graph; wherein the normalization expression is:

[0050]

[0051] Among them, p′ represents the value of the normalized pixel, p represents the value of the pixel in the unlabeled pulse description graph X, a represents the minimum value, and b represents the maximum value.

[0052] S106: Input the processed unlabeled pulse description graph into the trained working mode recognition model for processing to obtain the predicted current working mode of the radar.

[0053] Specifically, in this embodiment, the trained working mode recognition model is obtained by training the initial working mode recognition model using data of a preset category as a training data set; wherein the training of the initial working mode recognition model includes:

[0054] Acquire data of a preset category and construct a training data set, wherein the training data set includes a plurality of training samples, and the training samples include a pulse description diagram of a single working mode;

[0055] The training samples are input into the initial typical contrastive learning model for iterative training. During a finite number of iterations, the value of the loss function reaches a minimum, the parameters of the typical contrastive learning model are determined, and a trained working pattern recognition model is obtained.

[0056] In this embodiment, obtaining data of a preset category and constructing a training data set includes:

[0057] Determine the radar operating mode that needs to be identified;

[0058] Obtain a single radar pulse sequence corresponding to each radar operating mode;

[0059] Performing pulse waveform parameter characteristic transformation on pulses in a plurality of single radar pulse sequences to obtain a plurality of single pulse waveform parameter characteristic sequences;

[0060] Perform pulse repetition frequency mode characteristic transformation on multiple single pulse waveform parameter characteristic sequences to obtain multiple single pulse repetition frequency mode characteristic sequences;

[0061] Adding a single pulse waveform parameter characteristic sequence to its corresponding single pulse repetition frequency mode characteristic sequence to obtain a single working mode pulse description diagram;

[0062] A training data set is constructed based on multiple single working mode pulse description diagrams.

[0063] In this embodiment, see Figure 3 , Figure 3 Schematic diagram of a training working pattern recognition model provided by an embodiment of the present invention, wherein the training sample is input into the initial typical contrast learning model for iterative training, including:

[0064] The training samples are input into the first image enhancement module for processing, and the training samples are enhanced by introducing additive Gaussian white noise; the enhanced training samples are input into the first bilinear interpolation module for processing to scale the training samples; the scaled training samples are input into the first normalization module for processing to obtain the training samples after the first channel processing;

[0065] The training samples are input into the second image enhancement module for processing, and the training samples are enhanced by introducing additive Gaussian white noise; the enhanced training samples are input into the second bilinear interpolation module for processing to scale the training samples; the scaled training samples are input into the second normalization module for processing to obtain the training samples after the second channel processing;

[0066] The training samples processed by the first channel and the training samples processed by the second channel are input into the typical contrast learning model for processing to extract the image features of the training samples.

[0067] In this embodiment, the expression of the loss function is:

[0068]

[0069] Among them, Sim(·) represents the similarity calculation function, stopgrad(·) represents the stop gradient symbol, c1 represents the training sample after the first channel processing, and c2 represents the training sample after the second channel processing.

[0070] In summary, the present invention provides a radar working mode recognition method based on pulse description diagram and contrastive learning. For the acquired radar pulse sequence, pulse waveform parameter feature transformation and pulse repetition frequency mode feature transformation are respectively performed to obtain an unlabeled pulse description diagram, and then the unlabeled pulse description diagram is sequentially subjected to bilinear interpolation and normalization processing, and then the trained working mode recognition model is used to identify the working mode. In this way, it is not limited to a single working mode of the radar, and can recognize more radar working modes with a high recognition accuracy. It can restore the scene of a mixture of multiple radar working modes that are actually acquired, and can still have a high recognition accuracy rate under poor interception conditions, pulse loss and noise pulses. In addition, in the process of training the working mode recognition model in this embodiment, contrastive learning without negative samples is used to extract pulse description diagram features, which is not limited to label data and expert knowledge, and the extracted features are more likely to reflect the actual operating rules of the radar.

[0071] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants are intended to cover non-exclusive inclusion, so that the article or device including a series of elements includes not only those elements, but also other elements that are not explicitly listed. In the absence of more restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the article or device including the elements. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The orientation or position relationship indicated by "up", "down", "left", "right", etc. is based on the orientation or position relationship shown in the accompanying drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention.

[0072] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.

[0073] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A radar working mode recognition method based on pulse description diagram and contrast learning, characterized in that: include: Acquire a radar pulse sequence, wherein the radar pulse sequence includes a plurality of pulse description words arranged in time sequence; Performing pulse waveform parameter feature transformation on the pulse description words in the radar pulse sequence to obtain a pulse waveform parameter feature sequence; Performing pulse repetition frequency pattern feature transformation on the pulse description words in the radar pulse sequence to obtain a pulse repetition frequency pattern feature sequence; Adding the pulse waveform parameter characteristic sequence and the pulse repetition rate mode characteristic sequence to obtain an unlabeled pulse description graph; Performing bilinear interpolation and normalization processing on the unlabeled pulse description graph to obtain a processed unlabeled pulse description graph; The processed unlabeled pulse description graph is input into a trained working mode recognition model for processing to obtain a predicted current working mode of the radar.

2. The radar operating mode recognition method based on pulse description diagram and contrast learning according to claim 1 is characterized in that: The pulse description words in the radar pulse sequence are subjected to pulse waveform parameter feature transformation to obtain a pulse waveform parameter feature sequence, including: The parameters corresponding to each pulse description word in the radar pulse sequence are converted into an L-dimensional vector, and each dimensional vector is linearly normalized, and the value is limited to [-1, 1] to obtain the pulse waveform parameter characteristics; wherein the expression of the pulse waveform parameter characteristics is: Among them, p ij Indicates the value of the jth parameter corresponding to the i-th pulse description word, minp j Indicates the minimum value of the jth parameter corresponding to all pulse description words, maxp j Indicates the maximum value of the jth parameter corresponding to all pulse description words; According to the plurality of pulse waveform parameter features, the pulse waveform parameter feature sequence is constructed, and its expression is: P={p1 T ,p2 T ,…,p n T } T ; Among them, p k Represents the parameter vector corresponding to the kth pulse description word, k∈[1,n], and n represents the length of the radar pulse sequence.

3. The radar working mode recognition method based on pulse description diagram and contrast learning according to claim 1 is characterized in that: The pulse description words in the radar pulse sequence are transformed into pulse repetition frequency mode characteristics, The pulse repetition frequency pattern characteristic sequence is obtained, including: According to the pulse arrival time parameter corresponding to the pulse description word in the radar pulse sequence, the pulse description word is time-series encoded to obtain the pulse repetition frequency mode feature, which is expressed as follows: Where i represents the dimension index of the pulse arrival time vector, i∈[0,L), TOA k represents the pulse arrival time of the kth pulse description word, k∈[0,L+1 / 2), maxt represents the maximum pulse arrival time corresponding to all pulse description words, t oak Indicates the pulse arrival time of the kth pulse description word; According to the plurality of pulse repetition frequency pattern features, the pulse repetition frequency pattern feature sequence is constructed, and its expression is: The={the1 T ,the2 T ,…,the n T } T ; Where n is the length of the radar pulse sequence.

4. The radar working mode recognition method based on pulse description diagram and contrast learning according to claim 1 is characterized in that: The expression of the unlabeled pulse description graph is: X={x1,x2,…,x n },x k =p k +te k ,k∈[1,n]: Among them, p k Indicates the pulse waveform parameter characteristics, te k Indicates the pulse repetition rate pattern characteristics.

5. The radar working mode recognition method based on pulse description diagram and contrast learning according to claim 1 is characterized in that: The unlabeled pulse description graph is subjected to bilinear interpolation and normalization processing to obtain a processed unlabeled pulse description graph, including: Scaling the unlabeled pulse description graph using a bilinear interpolation method; The scaled unlabeled pulse description graph is normalized to obtain the processed unlabeled pulse description graph; wherein the normalized expression is: Among them, p′ represents the value of the normalized pixel, p represents the value of the pixel in the unlabeled pulse description graph X, a represents the minimum value, and b represents the maximum value.

6. The radar working mode recognition method based on pulse description diagram and contrast learning according to claim 1 is characterized in that: The trained working mode recognition model is obtained by training the initial working mode recognition model using data of a preset category as a training data set; wherein the training of the initial working mode recognition model includes: Acquire data of a preset category and construct a training data set, wherein the training data set includes a plurality of training samples, and the training samples include a single working mode pulse description diagram; The training samples are input into the initial typical contrast learning model for iterative training. During a finite number of iterations, the value of the loss function reaches a minimum, the parameters of the typical contrast learning model are determined, and the trained working mode recognition model is obtained.

7. The radar working mode recognition method based on pulse description diagram and contrast learning according to claim 6 is characterized in that: Get data of preset categories and build a training data set, including: Determine the radar operating mode that needs to be identified; Obtain a single radar pulse sequence corresponding to each radar operating mode; Performing pulse waveform parameter characteristic transformation on the pulses in the plurality of single radar pulse sequences to obtain a plurality of single pulse waveform parameter characteristic sequences; Performing pulse repetition frequency mode characteristic transformation on the plurality of single pulse waveform parameter characteristic sequences to obtain a plurality of single pulse repetition frequency mode characteristic sequences; Adding the single pulse waveform parameter characteristic sequence to the corresponding single pulse repetition frequency mode characteristic sequence to obtain a single working mode pulse description diagram; A training data set is constructed based on a plurality of the single working mode pulse description diagrams.

8. The radar working mode recognition method based on pulse description diagram and contrast learning according to claim 6 is characterized in that: The training samples are input into the initial typical contrastive learning model for iterative training, including: The training samples are input into a first image enhancement module for processing, and additive Gaussian white noise is introduced to enhance the training samples; the enhanced training samples are input into a first bilinear interpolation module for processing to scale the training samples; the scaled training samples are input into a first normalization module for processing to obtain training samples after first channel processing; The training samples are input into a second image enhancement module for processing, and additive Gaussian white noise is introduced to enhance the training samples; the enhanced training samples are input into a second bilinear interpolation module for processing to scale the training samples; the scaled training samples are input into a second normalization module for processing to obtain training samples after second channel processing; The training samples processed by the first channel and the training samples processed by the second channel are input into a typical contrast learning model for processing to extract image features of the training samples.

9. The radar working mode recognition method based on pulse description diagram and contrast learning according to claim 8 is characterized in that: The expression of the loss function is: Among them, Sim(·) represents the similarity calculation function, stopgrad(·) represents the stop gradient symbol, c1 represents the training sample after the first channel processing, and c2 represents the training sample after the second channel processing.

Images