Multifunction radar operating mode recognition method based on incomplete pulse sequence

Abstract

The application relates to the technical field of multifunctional radars, and provides a multifunctional radar working mode recognition method based on an incomplete pulse sequence. The method constructs a multifunctional radar working mode recognition model to form a new framework for adaptive consistency processing under attack defense and incomplete conditions, models full pulse sequence data under incomplete conditions in a way of resisting attack with no boundary constraint, so that the model can perform consistent processing on two different task scenes during attack training, and the robustness of the multifunctional radar working mode recognition model under attack and the adaptability of the model under incomplete conditions are improved. The method has the abilities of robust recognition of attack samples and self-adaptation of the model under incomplete conditions caused by pulse loss, false pulses and noise conditions.

Description

Technical Field

[0001] This application relates to the field of multi-functional radar technology, and in particular to a method for recognizing the operating mode of multi-functional radar based on incomplete pulse sequences. Background Technology

[0002] Due to their flexible beam pointing and large parameter variation space, multi-functional radars present incomplete conditions such as pulse loss and fragmented observations during target detection by space-based and air-based observation platforms. Furthermore, the complex electromagnetic environment, with its widespread presence of co-frequency radiation sources and high-density pulse currents, makes signal sorting difficult, resulting in varying degrees of false pulses and observation noise within the multi-functional radar sequence, thus leading to incomplete observations of the entire pulse sequence.

[0003] Traditional methods relying on expert knowledge, template matching, and machine learning are no longer adequate for the complex and incomplete conditions of multifunctional radar, resulting in low model recognition accuracy and poor robustness. Therefore, intelligent recognition methods such as deep learning and representation learning have been introduced into this field, achieving some improvement in recognition accuracy and adaptability to incomplete conditions. However, the introduction of deep learning also brings the problem of adversarial example attacks. Deep neural networks are susceptible to adversarial attacks; even small perturbations from well-designed adversarial examples can cause a sharp decline in the model's recognition accuracy, thereby compromising the model's security and robustness.

[0004] Significant progress has been made in research on adversarial attacks and defenses. Existing techniques primarily focus on improving the efficiency of adversarial training, generating high-quality adversarial examples, and developing effective adversarial training strategies. This results in more efficient training of adversarial defense models and higher robustness and recognition accuracy. In recent years, methods utilizing regularization techniques to enhance the robustness of models during adversarial training have also emerged, such as the introduction of sample augmentation techniques. These augmented samples optimize the model's regularity, further improving its adversarial defense capabilities. However, the aforementioned work mainly focuses on the generalization and adaptability of operating mode recognition models under various complex noise conditions. Robust recognition capabilities for multi-functional radar operating modes under incomplete conditions remain weak, requiring further in-depth research targeting these complex adversarial conditions. Summary of the Invention

[0005] In view of this, embodiments of this application provide a method for recognizing the operating mode of a multi-functional radar based on incomplete pulse sequences, in order to solve the problem of weak robust recognition capability of multi-functional radar operating modes under incomplete conditions in the prior art.

[0006] A first aspect of this application provides a method for recognizing the operating mode of a multi-functional radar based on an incomplete pulse sequence, comprising:

[0007] Acquire a pre-trained multi-functional radar operating mode recognition model and initial full-pulse sequence samples;

[0008] Incomplete conditional adversarial samples are generated using an incomplete conditional attack method to generate initial full-pulse sequence samples, and adversarial samples are generated using a gradient perturbation-based method to generate initial full-pulse sequence samples.

[0009] Using initial full-pulse sequence samples, incomplete conditional adversarial samples, and adversarial samples as inputs to a pre-trained multi-functional radar operating mode recognition model, the Hyers-Ulam regularization loss function, classification loss function, and contrastive loss function of the pre-trained multi-functional radar operating mode recognition model are determined.

[0010] The total loss function of the pre-trained multi-functional radar operating mode recognition model is determined based on the Hyers-Ulam regularized loss function, the classification loss function, and the contrastive loss function. The total loss function is then used to train the pre-trained multi-functional radar operating mode recognition model to obtain the trained multi-functional radar operating mode recognition model.

[0011] Acquire the full pulse sequence of the target multi-function radar, and use the trained multi-function radar operating mode recognition model to identify the operating mode of the multi-function radar based on the full pulse sequence of the target multi-function radar.

[0012] A second aspect of this application provides a multi-functional radar operating mode recognition device based on incomplete pulse sequences, comprising:

[0013] The acquisition module is configured to acquire a pre-trained multi-functional radar operating mode recognition model and initial full-pulse sequence samples;

[0014] The generation module is configured to generate incomplete conditional adversarial samples of the initial full-pulse sequence samples using an incomplete conditional attack method, and to generate adversarial samples of the initial full-pulse sequence samples using a gradient perturbation-based method.

[0015] The determination module is configured to take the initial full-pulse sequence samples, incomplete conditional adversarial samples, and adversarial samples as inputs to the pre-trained multi-functional radar operating mode recognition model, and determine the Hyers-Ulam regularization loss function, classification loss function, and contrastive loss function of the pre-trained multi-functional radar operating mode recognition model.

[0016] The training module is configured to determine the total loss function of the pre-trained multi-functional radar operating mode recognition model based on the Hyers-Ulam regularized loss function, the classification loss function, and the contrastive loss function, and to use the total loss function to train the pre-trained multi-functional radar operating mode recognition model to obtain the trained multi-functional radar operating mode recognition model.

[0017] The identification module is configured to acquire the target multi-function radar full pulse sequence and use the trained multi-function radar operating mode identification model to identify the operating mode of the multi-function radar based on the target multi-function radar full pulse sequence.

[0018] 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.

[0019] 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.

[0020] The beneficial effects of this application's embodiments compared to existing technologies are as follows: This application's embodiments construct a novel framework for adversarial attack defense and incomplete condition adaptive consistency processing of a multi-functional radar operating mode recognition model. Based on the physical meaning of the incomplete conditions of full-pulse sequence data, it models full-pulse sequence data under incomplete conditions using an adversarial attack method without boundary constraints. This enables the model to handle two different task scenarios consistently during adversarial training, improving the robustness of the multi-functional radar operating mode recognition model under adversarial attacks and its adaptability under incomplete conditions. This method possesses robust recognition capabilities against adversarial attack samples and adaptive capabilities of the model under incomplete conditions caused by pulse loss, false pulses, and noise.

[0021] Meanwhile, this method optimizes the generation of adversarial attack samples and incomplete condition samples to assist the model in adversarial training, taking into account the characteristics of full-pulse data serialization. It also introduces a Hyers-Ulam type stability regularization loss function and a contrastive loss function to extract stability features against training and contrast features between samples, further improving the stability and robustness of model training. This achieves a consistent improvement in robust accuracy for multi-functional radar working mode recognition under adversarial attack and incomplete conditions. Attached Figure Description

[0022] 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.

[0023] Figure 1 This is a flowchart illustrating a multi-functional radar operating mode recognition method based on incomplete pulse sequences provided in an embodiment of this application.

[0024] Figure 2 This is a flowchart illustrating the method for determining incomplete condition adversarial examples of an initial full-pulse sequence sample using an incomplete condition attack method, as provided in this application embodiment.

[0025] Figure 3 This is a schematic diagram of a multi-functional radar operating mode recognition device based on an incomplete pulse sequence, provided in an embodiment of this application.

[0026] Figure 4 This is a schematic diagram of the electronic device provided in the embodiments of this application. Detailed Implementation

[0027] 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.

[0028] The following describes in detail, with reference to the accompanying drawings, a method for recognizing the operating mode of a multi-functional radar based on an incomplete pulse sequence according to an embodiment of this application.

[0029] As mentioned above, in related technologies, the work on multi-functional radar operating mode recognition mainly focuses on the generalization and adaptability of operating mode recognition models under various complex noise conditions. The robust recognition capability of multi-functional radar operating modes under incomplete conditions is still weak, and further in-depth research is needed for the aforementioned complex adversarial conditions.

[0030] In view of this, this application provides a method for recognizing the operating mode of a multi-functional radar based on incomplete pulse sequences. By constructing a new framework for adversarial attack defense and consistent handling of incomplete conditions in the multi-functional radar operating mode recognition model, and based on the physical meaning of the incomplete conditions of the full pulse sequence data, the method models the full pulse sequence data under incomplete conditions using an adversarial attack without boundary constraints. This allows the model to handle two different task scenarios consistently during adversarial training, improving the robustness of the multi-functional radar operating mode recognition model under adversarial attacks and its adaptability under incomplete conditions. This method possesses robust recognition capabilities against adversarial attack samples and adaptive capabilities under incomplete conditions caused by pulse loss, false pulses, and noise.

[0031] Figure 1 This is a flowchart illustrating a multi-functional radar operating mode recognition method based on incomplete pulse sequences provided in an embodiment of this application. Figure 1 As shown, the method includes the following steps:

[0032] In step S101, a pre-trained multi-functional radar operating mode recognition model and initial full-pulse sequence samples are obtained.

[0033] In step S102, an incomplete conditional adversarial sample of the initial full-pulse sequence sample is generated using an incomplete conditional attack method, and an adversarial sample of the initial full-pulse sequence sample is generated using a gradient perturbation-based method.

[0034] In step S103, the initial full-pulse sequence samples, incomplete conditional adversarial samples, and adversarial samples are used as inputs to the pre-trained multi-functional radar operating mode recognition model to determine the Hyers-Ulam regularized loss function, classification loss function, and contrastive loss function of the pre-trained multi-functional radar operating mode recognition model.

[0035] In step S104, the total loss function of the pre-trained multi-functional radar operating mode recognition model is determined based on the Hyers-Ulam regularized loss function, the classification loss function, and the contrastive loss function. The total loss function is then used to train the pre-trained multi-functional radar operating mode recognition model, resulting in the trained multi-functional radar operating mode recognition model.

[0036] In step S105, the full pulse sequence of the target multi-function radar is acquired, and the operating mode of the multi-function radar is identified based on the target multi-function radar full pulse sequence using the trained multi-function radar operating mode recognition model.

[0037] In some embodiments of this application, the method may be executed by a server or by a terminal device with certain processing capabilities.

[0038] In some embodiments of this application, a pre-trained multi-functional radar operating mode recognition model and initial full-pulse sequence samples can be obtained. These full-pulse sequence samples can be clean samples, i.e., samples that have not undergone adversarial processing.

[0039] Full pulse sequence samples can include pulse arrival time (TOA), carrier frequency (RF), pulse width (PW), and pulse amplitude (PA), and their main characteristics are inter-pulse modulation and serialization.

[0040] In some embodiments of this application, incomplete conditional adversarial samples of the initial full-pulse sequence can be generated using an incomplete conditional attack method, and adversarial samples of the initial full-pulse sequence can be generated using a gradient perturbation-based method. In one example, an incomplete conditional adversarial sample of the initial full-pulse sequence can be generated first using an incomplete conditional attack method, and then an adversarial sample of the initial full-pulse sequence can be generated using a gradient perturbation-based method on the incomplete conditional adversarial sample.

[0041] Incomplete conditions can include incomplete conditions for spurious pulses, incomplete conditions for lost pulses, and incomplete conditions for measurement noise.

[0042] In some embodiments of this application, initial full-pulse sequence samples, incomplete conditional adversarial samples, and adversarial samples can be used as inputs to a pre-trained multi-functional radar operating mode recognition model to determine the Hyers-Ulam regularization loss function, classification loss function, and contrastive loss function of the pre-trained multi-functional radar operating mode recognition model.

[0043] The Hyers-Ulam regularized loss function is based on the work of American mathematician Donald Holmes Hyers and Polish-American mathematician... The proposed Hyers-Ulam stability theory constructs a regularized loss function that provides specific stability regularization to make model training more stable.

[0044] In some embodiments of this application, the total loss function of the pre-trained multi-function radar operating mode recognition model can be determined based on the Hyers-Ulam regularized loss function, the classification loss function, and the contrastive loss function, and the pre-trained multi-function radar operating mode recognition model can be trained using the total loss function to obtain the trained multi-function radar operating mode recognition model.

[0045] In some implementations, the full pulse sequence of the target multi-function radar can be acquired, and the operating mode of the multi-function radar can be identified based on the target multi-function radar full pulse sequence using a trained multi-function radar operating mode recognition model.

[0046] According to the technical solution provided in the embodiments of this application, a new framework for adversarial attack defense and incomplete condition adaptive consistency processing of a multifunctional radar operating mode recognition model is constructed. Based on the physical meaning of the incomplete conditions of full pulse sequence data, the full pulse sequence data under incomplete conditions is modeled in a way that avoids boundary constraints during adversarial attacks. This enables the model to handle two different task scenarios consistently during adversarial training, improving the robustness of the multifunctional radar operating mode recognition model under adversarial attacks and its adaptability under incomplete conditions. This method has robust recognition capability for adversarial attack samples and adaptive capability of the model under incomplete conditions caused by pulse loss, false pulses, and noise.

[0047] In some embodiments of this application, the full-pulse sequence samples exhibit typical characteristics of a one-dimensional time series. The TOA dimension is an important parameter in the full-pulse sequence samples, possessing both time-domain and modulation-domain characteristics. For example, the first-order difference represents the Pulse Repetition Interval (PRI) information, reflecting information such as the transmission time interval of the full-pulse sequence samples and the duty cycle of radar operation, and has a clear physical meaning. For ease of explanation and description, this application will use TOA dimension data from full-pulse sequence samples as an example for illustration. It is understood that the processing methods for PW and RF dimension data are similar.

[0048] Due to the time-series characteristics of TOA (Transient Attack) sequences, adversarial attacks primarily manifest in two modes: time-domain attacks and modulation-domain attacks, which are equivalent in the first-order difference sense. That is, incomplete conditional modulation-domain attacks can be used to determine incomplete conditional adversarial samples of the initial full-pulse sequence, and incomplete conditional time-domain attacks can also be used. The former has a slightly smaller impact than the latter; in practical applications, either method can be chosen based on the specific needs.

[0049] Figure 2 This is a flowchart illustrating a method for determining incomplete conditional adversarial examples of an initial full-pulse sequence sample using an incomplete conditional attack method, as provided in an embodiment of this application. Figure 2 As shown, the method includes the following steps:

[0050] In step S201, the incomplete conditional modulation domain attack method is used to determine the incomplete conditional adversarial sample of the initial full-pulse sequence sample.

[0051] In step S202, the incomplete conditional adversarial sample of the initial full-pulse sequence sample is determined using the incomplete conditional temporal attack method.

[0052] Among them, the incomplete conditional adversarial sample for determining the initial full-pulse sequence sample using the incomplete conditional modulation domain attack method can be, using the formula... Incomplete conditional modulated domain attack method is used to determine incomplete conditional adversarial samples of initial full-pulse sequence samples.

[0053] in, For incomplete conditional adversarial samples, This indicates a pulse sorting operation. This refers to the initial full-pulse sequence sample, for example, the initial full-pulse sequence sample with TOA dimension. , It is a positive integer; To find the difference between sets, Indicates according to The proportion is Lost pulse Indicates according to The proportion is Add fake pulses, For the lost pulse rate, This is a false pulse rate. It is random noise.

[0054] The incomplete conditional adversarial example for determining the initial full-pulse sequence sample using the incomplete conditional temporal attack method can be obtained by using the formula. Incomplete conditional adversarial examples are determined using an incomplete conditional temporal attack method on the initial full-pulse sequence samples; among which, For incomplete conditional adversarial samples, This means inversely mapping the repetition interval (PRI) sequence of the initial full-pulse sequence sample to the pulse arrival time (TOA) sequence of the initial full-pulse sequence sample. Represents the initial full-pulse sequence sample The first difference, Indicates according to The proportion is Lost pulse Indicates according to The proportion is Add fake pulses.

[0055] In one example, the lost pulse rate and spurious pulse rate can be set according to actual needs, but the maximum values ​​of both the lost pulse rate and spurious pulse rate should not exceed 50%.

[0056] In some embodiments of this application, the adversarial examples for determining the initial full-pulse sequence samples using a gradient perturbation-based method can be achieved by using the formula... Adversarial examples of the initial full-pulse sequence samples are determined using a gradient perturbation-based method.

[0057] in, For adversarial examples, To counteract sample perturbation upper bound, Represents a symbolic function. , It can be any number; For the gradient function, The cross-entropy function, The network model parameters are used to pre-train the multi-functional radar operating mode recognition model. The sample labels are for the initial full-pulse sequence samples.

[0058] Taking TOA dimension data as an example, since the time unit of TOA is usually nanoseconds (ns), the noise level of random noise can be selected with a certain noise boundary upper limit, such as 200ns, which is also called 200ns random noise attack.

[0059] If the initial full-pulse sequence samples are unnormalized sample data, then It needs to be set to absolute time, for example, set to This is referred to as a 5ns gradient adversarial attack.

[0060] In some embodiments of this application, the Hyers-Ulam regularized loss function can be expressed using the formula... Confirmed. Among them, The Hyers-Ulam regularized loss function is used. The network model parameters are used to pre-train the multi-functional radar operating mode recognition model. For adversarial examples, For incomplete conditional adversarial samples, and All are trade-off coefficients, and .

[0061] In some embodiments of this application, the classification loss function can use the formula Confirmed. Among them, For classification loss function, express, For adversarial examples, For incomplete conditional adversarial samples, The sample labels for the initial full-pulse sequence samples. This refers to the temperature parameter. The value can be set according to actual needs. In one example, it can be set to: .

[0062] In some embodiments of this application, the contrastive loss function can be formulated using the formula Confirmed. Among them, To compare loss functions, For adversarial examples, For incomplete conditional adversarial samples, For temperature coefficient, Representing chordal distance similarity, , and Let be any two parameters whose chordal distance similarity needs to be determined.

[0063] In some embodiments of this application, the total loss function of the pre-trained multi-functional radar working mode recognition model can be determined based on the Hyers-Ulam regularized loss function, the classification loss function, and the contrastive loss function by weighted summation of the Hyers-Ulam regularized loss function, the classification loss function, and the contrastive loss function.

[0064] In one example, a formula can be used. Calculate the total loss function, where, For the total loss function, These are weighting coefficients. The value can be set according to actual needs, for example, let... .

[0065] In some embodiments of this application, when training the pre-trained multi-functional radar operating mode recognition model using the total loss function, model training can be stopped when the model is identified as having reached the optimal stopping time condition, and the trained multi-functional radar operating mode recognition model can be determined based on the model parameters at the time of stopping model training.

[0066] The optimal stopping condition may include determining that the model training satisfies the early stopping strategy, or determining that the number of iterations reaches a preset number of iterations without changing the learning rate.

[0067] The early stopping strategy involves monitoring the target model parameters during training and terminating training prematurely when the target model parameters stop improving or begin to deteriorate. The target model parameters can be any one or more model parameters; there are no restrictions here.

[0068] The specific value of the preset number of iterations can be set according to actual needs. In one example, the preset number of iterations can be set to any integer in the range [80, 100]. When the dataset consisting of initial full-pulse sequence samples, incomplete conditional adversarial samples, and adversarial samples is relatively complex, the value of the preset number of iterations can be appropriately increased.

[0069] The technical solution provided in this application optimizes the adversarial example generation method based on the characteristics of multi-functional radar full pulse sequence data and the physical characteristics of its working mode changes. This enables the generated samples to have consistent representation capabilities under adversarial and incomplete conditions. At the same time, it optimizes the training strategy corresponding to the working mode recognition model, thereby improving the adversarial robustness and incomplete condition robustness of the trained model.

[0070] 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.

[0071] 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.

[0072] Figure 3 This is a schematic diagram of a multi-functional radar operating mode recognition device based on an incomplete pulse sequence, provided in an embodiment of this application. Figure 3 As shown, the device includes:

[0073] The acquisition module 901 is configured to acquire a pre-trained multi-functional radar working mode recognition model and initial full-pulse sequence samples.

[0074] The generation module 902 is configured to generate incomplete condition adversarial samples of the initial full-pulse sequence sample using an incomplete condition attack method, and to generate adversarial samples of the initial full-pulse sequence sample using a gradient perturbation-based method.

[0075] The determination module 903 is configured to take the initial full-pulse sequence samples, incomplete conditional adversarial samples, and adversarial samples as inputs to the pre-trained multi-functional radar operating mode recognition model, and determine the Hyers-Ulam regularization loss function, classification loss function, and contrastive loss function of the pre-trained multi-functional radar operating mode recognition model.

[0076] Training module 904 is configured to determine the total loss function of the pre-trained multi-functional radar operating mode recognition model based on the Hyers-Ulam regularized loss function, the classification loss function, and the contrastive loss function, and use the total loss function to train the pre-trained multi-functional radar operating mode recognition model to obtain the trained multi-functional radar operating mode recognition model.

[0077] The identification module 905 is configured to acquire the full pulse sequence of the target multi-function radar and use the trained multi-function radar operating mode identification model to identify the operating mode of the multi-function radar based on the full pulse sequence of the target multi-function radar.

[0078] According to the technical solution provided in the embodiments of this application, a new framework for adversarial attack defense and incomplete condition adaptive consistency processing of a multifunctional radar operating mode recognition model is constructed. Based on the physical meaning of the incomplete conditions of full pulse sequence data, the full pulse sequence data under incomplete conditions is modeled in a way that avoids boundary constraints during adversarial attacks. This enables the model to handle two different task scenarios consistently during adversarial training, improving the robustness of the multifunctional radar operating mode recognition model under adversarial attacks and its adaptability under incomplete conditions. This method has robust recognition capability for adversarial attack samples and adaptive capability of the model under incomplete conditions caused by pulse loss, false pulses, and noise.

[0079] In some implementations, incomplete conditional attack methods are used to determine incomplete conditional adversarial examples of the initial full-pulse sequence samples, including: using the formula Incomplete conditional adversarial examples are determined from the initial full-pulse sequence samples using an incomplete conditional modulation domain attack method; among which, For incomplete conditional adversarial samples, This indicates a pulse sorting operation. This is the initial full-pulse sequence sample. To find the difference between sets, Indicates according to The proportion is Lost pulse Indicates according to The proportion is Add fake pulses, For the lost pulse rate, This is a false pulse rate. For random noise; or use the formula Incomplete conditional adversarial examples are determined using an incomplete conditional temporal attack method on the initial full-pulse sequence samples; among which, For incomplete conditional adversarial samples, This means inversely mapping the repetition interval (PRI) sequence of the initial full-pulse sequence sample to the pulse arrival time (TOA) sequence of the initial full-pulse sequence sample. Represents the initial full-pulse sequence sample The first difference, Indicates according to The proportion is Lost pulse Indicates according to The proportion is Add fake pulses.

[0080] In some implementations, adversarial examples of the initial full-pulse sequence are determined using a gradient perturbation-based method, including: using the formula Adversarial examples for the initial full-pulse sequence are determined using a gradient perturbation-based method; whereby, For adversarial examples, To counteract sample perturbation upper bound, Represents a symbolic function. For the gradient function, The cross-entropy function, The network model parameters are used to pre-train the multi-functional radar operating mode recognition model. The sample labels are for the initial full-pulse sequence samples.

[0081] In some implementations, the Hyers-Ulam regularized loss function is determined as follows: using the formula Determine the Hyers-Ulam regularized loss function; where, The Hyers-Ulam regularized loss function is used. The network model parameters are used to pre-train the multi-functional radar operating mode recognition model. For adversarial examples, For incomplete conditional adversarial samples, and All are trade-off coefficients, and .

[0082] In some implementations, the classification loss function is determined using the following method: [Formula omitted for brevity] Determine the classification loss function; where, For classification loss function, express, For adversarial examples, For incomplete conditional adversarial samples, The sample labels for the initial full-pulse sequence samples. This refers to the temperature parameter.

[0083] In some implementations, the contrastive loss function is determined using the following method: [Formula omitted for brevity] Determine the contrastive loss function; where, To compare loss functions, For adversarial examples, For incomplete conditional adversarial samples, For temperature coefficient, Representing chordal distance similarity, , and Let be any two parameters whose chordal distance similarity needs to be determined.

[0084] In some implementations, the total loss function is a weighted sum of the Hyers-Ulam regularization loss function, the classification loss function, and the contrastive loss function. ,in, For the total loss function, These are the weighting coefficients.

[0085] In some implementations, when training a pre-trained multi-functional radar operating mode recognition model using the total loss function, in response to determining that the recognition model has reached the optimal stopping condition, model training is stopped, and the trained multi-functional radar operating mode recognition model is determined based on the model parameters at the time of stopping model training. The optimal stopping condition includes determining that the model training satisfies the early stopping strategy, or determining that the number of iterations has reached a preset number of iterations without changing the learning rate. The early stopping strategy involves monitoring the target model parameters during model training and terminating training early when the target model parameters stop improving or begin to deteriorate.

[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 4 This is a schematic diagram of the electronic device provided in an embodiment of this application. For example... Figure 4 As shown, the electronic device 4 of this embodiment includes: a processor 401, a memory 402, and a computer program 403 stored in the memory 402 and executable on the processor 401. When the processor 401 executes the computer program 403, it implements the steps in the various method embodiments described above. Alternatively, when the processor 401 executes the computer program 403, it implements the functions of each module / unit in the various device embodiments described above.

[0088] Electronic device 4 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 4 may include, but is not limited to, processor 401 and memory 402. Those skilled in the art will understand that... Figure 4 This is merely an example of electronic device 4 and does not constitute a limitation on electronic device 4. It may include more or fewer components than shown, or different components.

[0089] The processor 401 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 402 can be an internal storage unit of the electronic device 4, such as a hard disk or RAM of the electronic device 4. The memory 402 can also be an external storage device of the electronic device 4, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc., equipped on the electronic device 4. The memory 402 can also include both internal and external storage units of the electronic device 4. The memory 402 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 method for recognizing the operating mode of a multi-functional radar based on incomplete pulse sequences, characterized in that, include: Acquire a pre-trained multi-functional radar operating mode recognition model and initial full-pulse sequence samples; An incomplete conditional adversarial sample of the initial full-pulse sequence is generated using an incomplete conditional attack method, and an adversarial sample of the initial full-pulse sequence is generated using a gradient perturbation-based method. Using the initial full-pulse sequence samples, the incomplete conditional adversarial samples, and the adversarial samples as inputs to the pre-trained multi-functional radar operating mode recognition model, the Hyers-Ulam regularization loss function, classification loss function, and contrastive loss function of the pre-trained multi-functional radar operating mode recognition model are determined. Based on the Hyers-Ulam regularized loss function, the classification loss function and the contrastive loss function, the total loss function of the pre-trained multi-functional radar operating mode recognition model is determined, and the total loss function is used to train the pre-trained multi-functional radar operating mode recognition model to obtain the trained multi-functional radar operating mode recognition model. Acquire the target multi-function radar full pulse sequence, and use the trained multi-function radar operating mode recognition model to identify the operating mode of the multi-function radar based on the target multi-function radar full pulse sequence.

2. The method according to claim 1, characterized in that, The incomplete conditional adversarial examples of the initial full-pulse sequence sample are determined using an incomplete conditional attack method, including: Use formula Incomplete conditional adversarial examples are determined using an incomplete conditional modulation domain attack method for the initial full-pulse sequence samples; wherein, This refers to the incomplete conditional adversarial sample. This indicates a pulse sorting operation. The initial full-pulse sequence sample, To find the difference between sets, Indicates according to The proportion is Lost pulse Indicates according to The proportion is Add fake pulses, For the lost pulse rate, This is a false pulse rate. It is random noise; Or use a formula Incomplete conditional adversarial examples are determined using an incomplete conditional temporal attack method for the initial full-pulse sequence samples; wherein, This refers to the incomplete conditional adversarial sample. This means that the repetition interval (PRI) sequence of the initial full-pulse sequence sample is inversely mapped to the pulse arrival time (TOA) sequence of the initial full-pulse sequence sample. Represents the initial full-pulse sequence sample The first difference, Indicates according to The proportion is Lost pulse Indicates according to The proportion is Add fake pulses.

3. The method according to claim 2, characterized in that, The adversarial examples of the initial full-pulse sequence samples are determined using a gradient perturbation-based method, including: Use formula Adversarial examples of the initial full-pulse sequence samples are determined using a gradient perturbation-based method. in, For the adversarial example, To counteract sample perturbation upper bound, Represents a symbolic function. For the gradient function, The cross-entropy function, The network model parameters of the pre-trained multi-functional radar operating mode recognition model are... The sample label is the initial full-pulse sequence sample.

4. The method according to claim 1, characterized in that, The Hyers-Ulam regularized loss function is determined in the following manner: Use formula Determine the Hyers-Ulam regularized loss function; in, Let Hyers-Ulam be the regularized loss function. The network model parameters of the pre-trained multi-functional radar operating mode recognition model are... For the adversarial example, This refers to the incomplete conditional adversarial sample. and All are trade-off coefficients, and .

5. The method according to claim 1, characterized in that, The classification loss function is determined in the following manner: Use formula Determine the classification loss function; in, Let the classification loss function be... This represents the label smoothing function. For the adversarial example, This refers to the incomplete conditional adversarial sample. The sample label for the initial full-pulse sequence sample. For temperature parameters, These are the network model parameters of the pre-trained multi-functional radar operating mode recognition model.

6. The method according to claim 1, characterized in that, The contrastive loss function is determined in the following manner: Use the formula: Determine the contrastive loss function; in, Let the contrastive loss function be... For the adversarial example, This refers to the incomplete conditional adversarial sample. For temperature coefficient, Representing chordal distance similarity, , and For any two parameters whose chordal distance similarity needs to be determined, These are the network model parameters of the pre-trained multi-functional radar operating mode recognition model.

7. The method according to claim 1, characterized in that, The total loss function is a weighted sum of the Hyers-Ulam regularization loss function, the classification loss function, and the contrastive loss function. ,in, For the total loss function, For classification loss function, To compare loss functions, The Hyers-Ulam regularized loss function is used. These are the weighting coefficients.

8. The method according to claim 1, characterized in that, When training the pre-trained multi-functional radar operating mode recognition model using the total loss function, in response to determining that the recognition model has reached the optimal stopping time condition, the model training is stopped, and the trained multi-functional radar operating mode recognition model is determined based on the model parameters at the time of stopping the model training. The optimal stopping conditions include determining that the model training meets the early stopping strategy, or determining that the number of iterations reaches a preset number of iterations without changing the learning rate; the early stopping strategy is to monitor the target model parameters during model training and terminate training in advance when the target model parameters stop improving or begin to deteriorate.

9. A multi-functional radar operating mode recognition device based on incomplete pulse sequences, characterized in that, include: The acquisition module is configured to acquire a pre-trained multi-functional radar operating mode recognition model and initial full-pulse sequence samples; The generation module is configured to generate incomplete condition adversarial samples of the initial full-pulse sequence sample using an incomplete condition attack method, and to generate adversarial samples of the initial full-pulse sequence sample using a gradient perturbation-based method. The determination module is configured to use the initial full-pulse sequence samples, the incomplete conditional adversarial samples, and the adversarial samples as inputs to the pre-trained multi-functional radar operating mode recognition model to determine the Hyers-Ulam regularization loss function, classification loss function, and contrastive loss function of the pre-trained multi-functional radar operating mode recognition model. The training module is configured to determine the total loss function of the pre-trained multi-function radar operating mode recognition model based on the Hyers-Ulam regularized loss function, the classification loss function, and the contrastive loss function, and to train the pre-trained multi-function radar operating mode recognition model using the total loss function to obtain the trained multi-function radar operating mode recognition model. The identification module is configured to acquire the target multi-function radar full pulse sequence and use the trained multi-function radar operating mode identification model to identify the operating mode of the multi-function radar based on the target multi-function radar full pulse sequence.

10. 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 8.

11. 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 8.

Images