Dithered pri sequence retrieval method and device based on a fully convolutional network

Abstract

The application relates to the technical field of radar signal sorting, and provides a jitter PRI sequence retrieval method and device based on a full convolution network. The method introduces a PP sequence as a supplementary feature of a DTOA sequence in pulse sequence retrieval, uses a trained jitter PRI sequence retrieval model based on a full convolution network to retrieve the DTOA sequence and the PP sequence, effectively retrieves the jitter PRI sequence without needing an a priori jitter upper bound, and significantly improves sequence retrieval performance.

Description

Technical Field

[0001] This application relates to the field of radar signal sorting technology, and in particular to a jitter PRI sequence retrieval method and apparatus based on a fully convolutional network. Background Technology

[0002] Radar signal sorting is a core component of electronic support measures (ESR) for identifying multiple radiation sources. It supports radar attribute analysis, threat assessment, and electronic countermeasures decision-making by separating interleaved pulse streams. The sorting process typically employs a two-stage architecture: the pre-sorting stage performs preliminary clustering of pulses based on parameters such as carrier frequency, pulse width, and angle of arrival (Angle of Arrival), filtering out pulse sets with similar characteristics; the main sorting stage recursively performs pulse separation operations by analyzing the Time of Arrival (TOA) sequence, sequentially separating each pulse sequence until very few unlabeled pulses remain. The main sorting process typically includes two parts: Pulse Repetition Interval (PRI) estimation and pulse sequence retrieval. PRI estimation indicates potential PRIs within the TOA sequence, while pulse sequence retrieval extracts the corresponding pulse sequences based on the estimated PRI.

[0003] Existing pulse sequence retrieval algorithms mainly include pulse sequence retrieval algorithms based on PRI tolerance and global programming pulse sequence retrieval algorithms. Pulse sequence retrieval algorithms based on PRI tolerance employ a pulse-by-pulse approach, meaning that each time a potential subsequent pulse is searched based only on the current reference pulse, and the reference pulse is updated accordingly. Global programming pulse sequence retrieval algorithms, considering that the reliability of pulse segments is higher than that of individual pulses, typically use theories such as directed acyclic graphs and maximum likelihood estimation to extract long pulse segments at once to achieve pulse sequence retrieval.

[0004] When PRI jitter, interference pulses, and pulse missing occur, the linear progression strategy employed by PRI-tolerance-based pulse sequence retrieval methods—which involves no backtracking, no branching, and step-by-step execution—is easily misled by interference pulses, resulting in spurious sequences. While globally programming-based pulse sequence retrieval algorithms achieve high sequence extraction accuracy in the context of jitter and interference pulses, their performance depends on the precise setting of the PRI jitter upper limit. If the set value is lower than the actual jitter amplitude, sequence construction may be interrupted, leading to retrieval failure; conversely, if the limit is set too loosely, non-target pulses or interference pulses may be incorrectly associated, thus reducing retrieval accuracy. Summary of the Invention

[0005] In view of this, embodiments of this application provide a method and apparatus for dithering PRI sequence retrieval based on a fully convolutional network, in order to solve the problem in the prior art that dithering PRI sequence retrieval relies on a priori dithering upper bounds.

[0006] A first aspect of this application provides a dithered PRI sequence retrieval method based on a fully convolutional network, comprising:

[0007] Based on the received radar signal pulse data, determine N TOA and the average PRI of the jitter sequence to be retrieved; N is a positive integer;

[0008] Based on N TOA and the average PRI, determine the Differential Time of Arrival (DTOA) sequence and the Pulse Phase (PP) sequence; where the k-th pulse phase in the PP sequence is used to characterize the normalized phase of the k-th pulse relative to the average PRI, and k is a positive integer greater than or equal to 1 and less than or equal to N.

[0009] By inputting the DTOA sequence, PP sequence, and average PRI into the trained fully convolutional network-based jitter PRI sequence retrieval model, pulse sequence retrieval results are obtained.

[0010] A second aspect of this application provides a jitter PRI sequence retrieval device based on a fully convolutional network, comprising:

[0011] The pre-sorting and estimation module is configured to determine N TOA and the average PRI of the jitter sequence to be retrieved based on the received radar signal pulse data; N is a positive integer.

[0012] The preprocessing module is configured to determine the DTOA sequence and PP sequence based on N TOA and the average PRI; wherein, the phase of the k-th pulse in the PP sequence is used to characterize the normalized phase of the k-th pulse relative to the average PRI, and k is a positive integer greater than or equal to 1 and less than or equal to N;

[0013] The retrieval module is configured to take DTOA sequences, PP sequences, and average PRI sequences as inputs to a fully convolutional network-based jitter PRI sequence retrieval model to obtain pulse sequence retrieval results.

[0014] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0015] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method.

[0016] The beneficial effects of this application embodiment compared with the prior art are as follows: This application embodiment introduces PP sequence as a supplementary feature of DTOA sequence in pulse sequence retrieval, and uses a trained jitter PRI sequence retrieval model based on a fully convolutional network to retrieve DTOA sequence and PP sequence. Without the need for a priori jitter upper bound, it achieves effective retrieval of jitter PRI sequence and significantly improves sequence retrieval performance. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a jittery PRI sequence retrieval method based on a fully convolutional network provided in an embodiment of this application.

[0019] Figure 2 This is a schematic diagram of the network structure of the jittery PRI sequence retrieval model based on a fully convolutional network, provided in an embodiment of this application.

[0020] Figure 3 This is a performance comparison chart of different pulse sequence retrieval methods provided in the embodiments of this application when the pulse missing rate is 20%.

[0021] Figure 4 This is a performance comparison chart of different pulse sequence retrieval methods provided in the embodiments of this application when the pulse missing rate is 40%.

[0022] Figure 5 This is a schematic diagram of a jitter PRI sequence retrieval device based on a fully convolutional network provided in an embodiment of this application.

[0023] Figure 6 This is a schematic diagram of the electronic device provided in the embodiments of this application. Detailed Implementation

[0024] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0025] The following describes in detail, with reference to the accompanying drawings, a jittery PRI sequence retrieval method and apparatus based on a fully convolutional network according to embodiments of this application.

[0026] As mentioned above, radar signal sorting includes a pre-sorting stage and a main sorting stage. In the pre-sorting stage, pulses are initially clustered based on parameters such as TOA, pulse width, radio frequency, pulse amplitude, and angle of arrival to form pulse sets with similar characteristics, thus diluting the pulse stream and facilitating subsequent main sorting steps. The main sorting stage recursively separates pulses from each radiation source by analyzing the TOA sequence until only a small number of pulses remain unseparated. The main sorting process mainly includes two key steps: PRI estimation and pulse sequence retrieval. PRI estimation is used to identify potential pulse repetition interval values ​​in the TOA sequence, while pulse sequence retrieval extracts the corresponding pulse sequence based on the estimated PRI.

[0027] Existing pulse sequence retrieval algorithms mainly include pulse sequence retrieval algorithms based on PRI tolerance and pulse sequence retrieval algorithms based on global planning. When PRI jitter, interference pulses, and pulse missing exist, the above retrieval methods all have drawbacks such as limited retrieval accuracy and dependence on the precise setting of the PRI jitter upper limit.

[0028] In view of this, this application provides a jitter PRI sequence retrieval method based on a fully convolutional network. By introducing PP sequences as supplementary features to DTOA sequences in pulse sequence retrieval, the trained jitter PRI sequence retrieval model based on a fully convolutional network is used to retrieve DTOA sequences and PP sequences. Without the need for a priori upper bound on jitter, effective retrieval of jitter PRI sequences is achieved, and sequence retrieval performance is significantly improved.

[0029] Figure 1 This is a flowchart illustrating a jittery PRI sequence retrieval method based on a fully convolutional network provided in an embodiment of this application. Figure 1 As shown, the method includes the following steps:

[0030] In step S101, N TOA and the average PRI of the jitter sequence to be retrieved are determined based on the received radar signal pulse data.

[0031] Where N is a positive integer.

[0032] In step S102, the pulse arrival time difference (DTOA) sequence and the pulse phase (PP) sequence are determined based on N TOAs and the average PRI.

[0033] In this context, the phase of the kth pulse in the PP sequence is used to characterize the normalized phase of the kth pulse relative to the average PRI, where k is a positive integer greater than or equal to 1 and less than or equal to N.

[0034] In step S103, the DTOA sequence, PP sequence, and average PRI are input into the trained jitter PRI sequence retrieval model based on a fully convolutional network to obtain the pulse sequence retrieval result.

[0035] In some embodiments of this application, the method may be executed by a server or by a terminal device with certain processing capabilities.

[0036] In some embodiments of this application, N TOA (Time of Arrival) and the average PRI (Primary Rank) of the jitter sequence to be retrieved can be determined based on the received radar signal pulse data. The average PRI of the jitter sequence to be retrieved can be obtained based on existing PRI estimation algorithms, and this is not limited here. That is, the received radar signal pulse data can first undergo pre-sorting and PRI estimation processing to obtain the average PRI of the jitter sequence to be retrieved.

[0037] In some embodiments of this application, the pulse arrival time difference (DTOA) sequence and the pulse phase (PP) sequence can be determined based on N TOAs and the average PRI.

[0038] In some implementations, if the k-th TOA among the N TOA is represented as Then the DTOA sequence can be represented as Meanwhile, this PP sequence can be represented as PP ,in, This represents the k-th pulse phase in the PP sequence.

[0039] In some embodiments of this application, the pulse sequence retrieval result can be obtained by using the DTOA sequence, PP sequence, and average PRI input trained on a fully convolutional network-based jitter PRI sequence retrieval model.

[0040] In other words, a jittery PRI sequence retrieval model based on a fully convolutional network can be pre-trained, and then the DTOA sequence, PP sequence, and average PRI can be used as model inputs, with the pattern output as the jittery PRI sequence retrieval result.

[0041] According to the technical solution provided in the embodiments of this application, by introducing PP sequences as supplementary features of DTOA sequences in pulse sequence retrieval, and using a trained jitter PRI sequence retrieval model based on a fully convolutional network to retrieve DTOA sequences and PP sequences, effective retrieval of jitter PRI sequences is achieved without the need for a priori upper bound on jitter, and the sequence retrieval performance is significantly improved.

[0042] In some embodiments of this application, the phase of the kth pulse in the PP sequence can be determined in the following manner: ;in, For the k-th pulse phase, Let be the principal argument of the complex number. It is an exponential function. The imaginary unit, The average PRI.

[0043] In other words, without considering jitter, the estimated average PRI is a fixed value. As the first The arrival time of each pulse can be expressed as: Among them, the expression It can be considered as a pulse repetition frequency The phase is the "frequency", which describes time. Relative to the cycle The normalized phase.

[0044] The PP sequence is used as a supplement to the DTOA sequence. Both the DTOA and PP sequences are used together for feature extraction and classification prediction, thereby improving the retrieval accuracy of jittery PRI sequences. Simultaneously, a trained jittery PRI sequence retrieval model based on a fully convolutional network is used for retrieval, enabling effective retrieval of jittery PRI sequences without requiring a priori upper bound on jitter.

[0045] In some embodiments of this application, the trained jitter PRI sequence retrieval model based on a fully convolutional network includes at least a DTOA branch, a PP branch, and an output layer. The DTOA branch determines a first feature based on the DTOA sequence and the average PRI; the PP branch determines a second feature based on the PP sequence; and the output layer determines the pulse sequence retrieval result based on the fused first and second features.

[0046] Furthermore, the first feature can be determined as follows: a high-dimensional semantic feature representation of the DTOA sequence is obtained through a first convolutional learning layer; difference correction is performed by subtracting the high-dimensional semantic feature representation from the average PRI; and the high-dimensional semantic feature representation after difference correction is subjected to nonlinear transformation and abstraction through a second convolutional learning layer to obtain the first feature.

[0047] The second feature can be determined as follows: obtain the high-dimensional semantic feature representation of the PP sequence through the third convolutional learning layer; determine the high-dimensional semantic feature representation of the PP sequence as the second feature.

[0048] Furthermore, the first and second features can be fused by concatenating along the channel dimension. The fused features are then processed by the fourth convolutional layer and transmitted to the output layer. The output layer determines the predicted probability value of the PRI pulse at each TOA position based on the received features, and identifies PRI pulses at TOA positions with predicted probability values ​​greater than a preset probability threshold as pulse sequence retrieval results.

[0049] The specific number of the first, second, third, and fourth convolutional learning layers can be set according to actual needs and is not limited here. In one example, the first convolutional learning layer can be set to four layers, the second convolutional learning layer to one layer, the third convolutional learning layer to four layers, and the fourth convolutional learning layer to four layers.

[0050] Figure 2 This is a schematic diagram of the network structure of the trained jittery PRI sequence retrieval model based on a fully convolutional network, provided in an embodiment of this application. For example... Figure 2 As shown, the TOA sequence can be input into the trained jittery PRI sequence retrieval model based on a fully convolutional network, and the preprocessing module converts the TOA sequence into a DTOA sequence and a PP sequence, which are then used as input to a dual-branch network. One branch is the DTOA branch, and the other branch is the PP branch.

[0051] In the DTOA branch, a high-dimensional semantic feature representation of the DTOA sequence can first be learned through four convolutional learning layers. Then, the average PRI is subtracted from the extracted high-dimensional semantic feature representation for difference correction, thereby incorporating the relevant information of the average PRI of the jitter sequence to be retrieved into the extracted high-dimensional semantic features. This average PRI is represented by the purple block in the figure. This is the average PRI value. Next, a convolutional learning layer is used to perform nonlinear transformation and abstraction on the corrected high-dimensional semantic features to extract more discriminative high-dimensional semantic features, which are used as the output of the DTOA branch. This output can be denoted as the first feature.

[0052] In the PP branch, the corresponding high-dimensional semantic features can be extracted through four convolutional layers and used as the output of the PP branch. This output can be denoted as the second feature.

[0053] The outputs of the two branches, namely the first and second features, can be concatenated along the channel dimension and mapped to the output layer via multiple convolutional learning layers. The output layer can use the sigmoid activation function to predict a probability value between 0 and 1 for each time position in the input TOA sequence, indicating whether the pulse at the corresponding position belongs to the target PRI sequence.

[0054] In some implementations, pulses at positions where the predicted probability value is greater than a preset probability threshold can be used as the target PRI sequence. The preset probability threshold can be set according to actual needs and is not limited here. In one example, the preset probability threshold can be set to 0.5.

[0055] In some embodiments of this application, the trained jittery PRI sequence retrieval model based on a fully convolutional network is trained using supervised learning. Furthermore, this trained jittery PRI sequence retrieval model based on a fully convolutional network can use a binary cross-entropy loss function as its model loss function during training. This binary cross-entropy loss function is... ;in, The pulse number; This represents the total number of pulses. It is the first The true label of the pulse, when the... When each pulse belongs to the target pulse ,otherwise ; The model considers the first The probability that a pulse belongs to the target pulse.

[0056] To verify the practicality and effectiveness of the technical solutions provided in the embodiments of this application, the following simulation experiments were designed:

[0057] The simulation parameters for the TOA sequence are shown in Table 1. Forty different test conditions were created by combining various PRI jitter rates, pulse loss rates, and interference pulse ratios. 1000 test samples were generated under each condition, for a total of 40,000 samples, to comprehensively evaluate the model's performance in different scenarios.

[0058] Table 1 Simulation parameters of the TOA sequence

[0059]

[0060] This application uses the PRism metric to measure the performance of impulse retrieval, which is defined as follows: Where TP represents the number of successfully detected target pulses, FP represents the number of other pulses falsely detected as target pulses, and FN represents the number of missed target pulses. The PRism metric is the product of precision and recall. Precision... Recall measures the reliability of detection results; a higher value indicates fewer false positives. This measures the completeness of coverage of the true target; a higher value indicates fewer missed detections. The PRism metric, which multiplies the two, requires the model to strike a good balance between reducing false positives and avoiding false negatives; a higher value indicates better overall performance.

[0061] The performance of different pulse sequence retrieval methods under different pulse loss rates is as follows: Figure 3 and Figure 4As shown in the figure, DAG refers to the method in the published literature “M. Xie, C. Zhao, X. Han, Z. Wang and D. Hu, "Separation of Interleaved Pulse Stream Based on Directed Acyclic Graphs," in IEEE Signal Processing Letters, vol. 30, pp. 613-617, 2023, doi: 10.1109 / LSP.2023.3254439.”. FSC refers to the method in the published literature “Z. -M. Liu, "Online Pulse Deinterleaving With Finite Automata," in IEEE Transactions on Aerospace and Electronic Systems, vol. 56, no. 2, pp. 1139-1147, April 2020, doi: 10.1109 / TAES.2019.2925447.”. Ours refers to the method proposed in the embodiments of this application.

[0062] As can be seen, compared with DAG and FSC, the method proposed in this application demonstrates superior retrieval performance under various test conditions. Specifically, the method proposed in this application can effectively separate the target pulse sequence from the pulse stream interfered with by stray pulses under various combinations of jitter boundaries, pulse missing rates, and spurious pulse ratios, exhibiting strong robustness. Only under conditions of low jitter rate and high spurious pulse ratio does the performance of the proposed method slightly lag behind DAG. These results verify that the proposed method can accurately retrieve jittered PRI sequences without prior knowledge of jitter boundaries, demonstrating strong adaptability and broad application potential.

[0063] All of the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this application, and will not be described in detail here.

[0064] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0065] Figure 5 This is a schematic diagram of a jittery PRI sequence retrieval device based on a fully convolutional network provided in an embodiment of this application. Figure 5 As shown, the device includes:

[0066] The pre-sorting and estimation module 501 is configured to determine the arrival times (TOA) of N pulses and the average pulse repetition interval (PRI) of the jitter sequence to be retrieved based on the received radar signal pulse data; N is a positive integer.

[0067] The preprocessing module 502 is configured to determine the pulse arrival time difference (DTOA) sequence and the pulse phase (PP) sequence based on N TOA and the average PRI; wherein the k-th pulse phase in the PP sequence is used to characterize the normalized phase of the k-th pulse relative to the average PRI, and k is a positive integer greater than or equal to 1 and less than or equal to N.

[0068] The retrieval module 503 is configured to input the DTOA sequence, PP sequence and average PRI into the trained jitter PRI sequence retrieval model based on a fully convolutional network to obtain the pulse sequence retrieval result.

[0069] According to the technical solution provided in the embodiments of this application, by introducing PP sequences as supplementary features of DTOA sequences in pulse sequence retrieval, and using a trained jitter PRI sequence retrieval model based on a fully convolutional network to retrieve DTOA sequences and PP sequences, effective retrieval of jitter PRI sequences is achieved without the need for a priori upper bound on jitter, and the sequence retrieval performance is significantly improved.

[0070] In some implementations, the phase of the k-th pulse in the PP sequence is determined in the following manner: ;in, For the k-th pulse phase, Let be the principal argument of the complex number. It is an exponential function. The imaginary unit, The average PRI.

[0071] In some implementations, the trained fully convolutional network-based jitter PRI sequence retrieval model includes at least a DTOA branch, a PP branch, and an output layer;

[0072] The DTOA branch determines the first feature based on the DTOA sequence and the average PRI.

[0073] PP branch determines the second feature based on the PP sequence;

[0074] The output layer determines the pulse sequence retrieval result based on the fused first and second features.

[0075] In some implementations, the first feature is determined in the following manner:

[0076] The high-dimensional semantic feature representation of the DTOA sequence is obtained through the first convolutional learning layer;

[0077] Difference correction is performed by subtracting the high-dimensional semantic feature representation from the average PRI.

[0078] The first feature is obtained by performing nonlinear transformation and abstraction on the high-dimensional semantic feature representation after difference correction through the second convolutional learning layer.

[0079] In some implementations, the second feature is determined in the following manner:

[0080] The high-dimensional semantic feature representation of the PP sequence is obtained through the third convolutional learning layer;

[0081] The high-dimensional semantic features of the PP sequence are represented as the second feature.

[0082] In some implementations, the first and second features are fused by splicing along the channel dimension, and the fused features are processed by the fourth convolutional layer and then transmitted to the output layer.

[0083] The output layer determines the predicted probability value of the PRI pulse at each TOA position based on the received features, and determines the PRI pulse at the TOA position with the predicted probability value greater than the preset probability threshold as the pulse sequence retrieval result.

[0084] In some implementations, the trained jitter PRI sequence retrieval model based on a fully convolutional network is trained using supervised learning.

[0085] The trained jittery PRI sequence retrieval model based on a fully convolutional network uses a binary cross-entropy loss function as its loss function during training. The binary cross-entropy loss function is: ;in, The pulse number; This represents the total number of pulses. It is the first The true label of the pulse, when the... When each pulse belongs to the target pulse ,otherwise ; The model considers the first The probability that a pulse belongs to the target pulse.

[0086] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0087] Figure 6 This is a schematic diagram of the electronic device provided in an embodiment of this application. For example... Figure 6As shown, the electronic device 6 of this embodiment includes a processor 601, a memory 602, and a computer program 603 stored in the memory 602 and executable on the processor 601. When the processor 601 executes the computer program 603, it implements the steps in the various method embodiments described above. Alternatively, when the processor 601 executes the computer program 603, it implements the functions of each module / unit in the various device embodiments described above.

[0088] Electronic device 6 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 6 may include, but is not limited to, processor 601 and memory 602. Those skilled in the art will understand that... Figure 6 This is merely an example of electronic device 6 and does not constitute a limitation on electronic device 6. It may include more or fewer components than shown, or different components.

[0089] The processor 601 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0090] The memory 602 can be an internal storage unit of the electronic device 6, such as a hard disk or RAM of the electronic device 6. The memory 602 can also be an external storage device of the electronic device 6, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc., equipped on the electronic device 6. The memory 602 can also include both internal and external storage units of the electronic device 6. The memory 602 is used to store computer programs and other programs and data required by the electronic device.

[0091] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0092] If an integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. A computer-readable medium may include: any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0093] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A jittery PRI sequence retrieval method based on a fully convolutional network, characterized in that, include: The arrival times (TOA) of N pulses and the average pulse repetition interval (PRI) of the jitter sequence to be retrieved are determined based on the received radar signal pulse data. N is a positive integer; Based on the N TOA and the average PRI, a pulse arrival time difference (DTOA) sequence and a pulse phase (PP) sequence are determined; wherein, the k-th pulse phase in the PP sequence is used to characterize the normalized phase of the k-th pulse relative to the average PRI, and k is a positive integer greater than or equal to 1 and less than or equal to N; The pulse sequence retrieval result is obtained by inputting the DTOA sequence, the PP sequence, and the average PRI into a fully convolutional network-based jitter PRI sequence retrieval model trained on the input. The trained jitter PRI sequence retrieval model based on a fully convolutional network includes at least a DTOA branch, a PP branch, and an output layer. The DTOA branch determines the first feature based on the DTOA sequence and the average PRI; The PP branch determines the second feature based on the PP sequence; The output layer determines the pulse sequence retrieval result based on the fused first and second features. The first feature is determined in the following manner: The high-dimensional semantic feature representation of the DTOA sequence is obtained through the first convolutional learning layer; Difference correction is performed by subtracting the high-dimensional semantic feature representation from the average PRI; The first feature is obtained by performing nonlinear transformation and abstraction on the high-dimensional semantic feature representation after difference correction through the second convolutional learning layer. The second feature is determined in the following manner: The high-dimensional semantic feature representation of the PP sequence is obtained through the third convolutional learning layer; The high-dimensional semantic features of the PP sequence are defined as the second feature.

2. The method according to claim 1, characterized in that, The phase of the kth pulse in the PP sequence is determined in the following manner: ;in, For the k-th pulse phase, For the first TOA of one pulse, Let be the principal argument of the complex number. It is an exponential function. The imaginary unit, The average PRI is mentioned.

3. The method according to claim 1, characterized in that, The first feature and the second feature are fused by concatenating along the channel dimension. The fused feature is then processed by the fourth convolutional layer and transmitted to the output layer. The output layer determines the predicted probability value of the PRI pulse at each TOA position based on the received features, and determines the PRI pulse at the TOA position with a predicted probability value greater than a preset probability threshold as the pulse sequence retrieval result.

4. The method according to claim 1, characterized in that, The trained jitter PRI sequence retrieval model based on a fully convolutional network was trained using a supervised learning method. The trained jittery PRI sequence retrieval model based on a fully convolutional network uses a binary cross-entropy loss function as its loss function during training. The binary cross-entropy loss function is: ;in, The pulse number; This represents the total number of pulses. It is the first The true label of the pulse, when the... When each pulse belongs to the target pulse ,otherwise ; The model considers the first The probability that a pulse belongs to the target pulse.

5. A jittered PRI sequence retrieval device based on a fully convolutional network, characterized in that, include: The pre-sorting and estimation module is configured to determine the arrival times (TOA) of N pulses and the average pulse repetition interval (PRI) of the jitter sequence to be retrieved based on the received radar signal pulse data; N is a positive integer. The preprocessing module is configured to determine a pulse arrival time difference (DTOA) sequence and a pulse phase (PP) sequence based on the N TOAs and the average PRI; wherein the k-th pulse phase in the PP sequence is used to characterize the normalized phase of the k-th pulse relative to the average PRI, and k is a positive integer greater than or equal to 1 and less than or equal to N; The retrieval module is configured to input the DTOA sequence, the PP sequence, and the average PRI into a jitter PRI sequence retrieval model trained on a fully convolutional network, and obtain pulse sequence retrieval results. The trained jitter PRI sequence retrieval model based on a fully convolutional network includes at least a DTOA branch, a PP branch, and an output layer. The DTOA branch determines the first feature based on the DTOA sequence and the average PRI; The PP branch determines the second feature based on the PP sequence; The output layer determines the pulse sequence retrieval result based on the fused first and second features. The first feature is determined in the following manner: The high-dimensional semantic feature representation of the DTOA sequence is obtained through the first convolutional learning layer; Difference correction is performed by subtracting the high-dimensional semantic feature representation from the average PRI; The first feature is obtained by performing nonlinear transformation and abstraction on the high-dimensional semantic feature representation after difference correction through the second convolutional learning layer. The second feature is determined in the following manner: The high-dimensional semantic feature representation of the PP sequence is obtained through the third convolutional learning layer; The high-dimensional semantic features of the PP sequence are defined as the second feature.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 4.

Images