A multifunction radar working mode recognition method based on GRUTCN network

Abstract

The application discloses a multifunctional radar working mode recognition method based on a GRUTCN network, first analyzes the working mode characteristics of an airborne phased array radar, generates multifunctional radar pulse group data, inputs the data set into a data set generation program and carries out data preprocessing, divides the data set into a training set and a test set, then constructs a multifunctional radar GRUTCN working mode recognition model, inputs the training set data after preprocessing into the model for training, finally sends the test set data into the trained model, and completes multifunctional radar working mode recognition test. The method combines GRU and TCN, proposes a network model based on an encoder-decoder GRUTCN, solves the problems that existing working mode recognition methods are not comprehensive in working mode feature analysis and have low recognition rate, and can keep the recognition accuracy much higher than other existing networks under non-ideal conditions, and the robustness is enhanced.

Description

Technical Field

[0001] This invention belongs to the field of electronic countermeasures technology, specifically relating to a multi-functional radar operating mode recognition method based on the GRUTCN network. Background Technology

[0002] Classifying intercepted radar signals and determining their intent is fundamental to cognitive electronic warfare. With the development of electronic scanning and phased array systems, radars can change beam direction in real time to achieve different objectives, and this flexible combination of tasks forms operational modes. Radar operational mode recognition is the basis and criterion for the reconnaissance side to take further action. If the reconnaissance side can accurately determine the radar's operational mode, it can adjust its countermeasures in real time to gain the initiative in electronic warfare. The GRUTCN network's multi-functional radar operational mode recognition method is developed to address this need.

[0003] The Gate Recurrent Unit (GRU) network, a type of temporal recurrent neural network, combines the forget gate and input gate of an LSTM network. By using the same gating information z, it selectively remembers important information while forgetting less important information. This improves the network's training efficiency while preserving memory capabilities. Compared to Long Short-Term Memory (LSTM) networks, GRU achieves faster convergence with fewer parameters. For sequence recognition, GRU networks can split the training set into mini-batches during training and pad the input data to have uniform lengths, allowing for the full extraction of feature information from the input data. However, GRU networks suffer from problems such as gradient explosion.

[0004] TCN stands for Temporal Convolutional Network. It consists of dilated, causal 1D convolutional layers with the same input and output lengths. Its convolutions operate on the input data only at one time.

[0005] Existing methods for identifying the operating modes of multi-function radars have made some progress. Existing methods based on template matching or statistical histogram sorting perform well for modulated radars with fixed signal parameters. However, due to the complexity of multi-function radar signal parameters and the ambiguous correspondence between them and tasks / operating modes, existing methods are unable to extract complex dynamic features and thus yield limited results. For multi-function radars, the paper "Hidden Markov models for radar pulse train analysis in electronic warfare, Proceedings. IEEE International Conference on Acoustics, Speech, and Signal Processing, 2005" proposes a radar signal representation method based on a syntactic model. This method reconstructs radar signals from the perspectives of "radar words," "radar phrases," and "radar sentences" to represent the complex changes in MFR signals. Radar words represent a fixed arrangement of a finite number of pulses, and radar phrases are arrangements of a finite number of radar words, representing the operating mode layer. This syntactic model combines abstract radar tasks, operating modes, and functions with the intercepted signals, representing the radar signal hierarchy at the linguistic level. It is concise, easy to understand, and efficient, and has been widely applied in research related to multi-function radar operating mode recognition.

[0006] Subsequently, a sequence-to-class (seq2one) working mode recognition method was proposed, taking the pulse sequence as input and outputting a class label. However, due to the inter-class partitioning problem of the intercepted pulse stream, the seq2one method struggles to generate correct classifications for mixed input sequences of multiple classes. The paper "Work modes recognition and boundary identification of MFR pulse sequences with a hierarchical seq2seq LSTM", IETradar Sonar & Navigation, vol.14, no.9, pp.1343-1353, Jul.2020, proposed a pulse-level seq2seq method, using a hierarchical LSTM model to identify multi-functional radar pulse sequences. However, this method is time-consuming and lacks feature analysis of working modes, resulting in a lack of theoretical support for the dataset.

[0007] The paper "Multi-Function Radar Work Mode Recognition Based on Encoder-Decoder Model, IGARSS 2022-2022 IEEE International Geoscience and Remote Sensing Symposium, Kuala Lumpur, Malaysia, Jul. 2022, pp. 1189-1192" proposes an encoding-decoding model for radar working mode recognition and achieves good performance. However, the corresponding dataset is relatively simple, and the basis for its generation is not explained. Furthermore, the model consumes significant time resources. As previously mentioned, existing research largely focuses on network analysis, lacking analysis of the dataset itself. Summary of the Invention

[0008] To address the aforementioned technical problems, this invention proposes a multi-functional radar operating mode recognition method based on the GRUTCN network, which can maintain a recognition accuracy far higher than other existing networks under non-ideal conditions.

[0009] The technical solution adopted in this invention is: a multi-functional radar operating mode recognition method based on GRUTCN network, the specific steps of which are as follows:

[0010] S1. Analyze the working mode characteristics of airborne phased array radar, generate multi-functional radar pulse group data, input the dataset generation program and perform data preprocessing, and divide the dataset into training set and test set.

[0011] S2. Construct a multi-functional radar GRUTCN working mode recognition model;

[0012] S3. Input the training set data after preprocessing in step S1 into the model for training.

[0013] S4, Multifunctional Radar Operating Mode Recognition Test.

[0014] Furthermore, step S1 is specifically as follows:

[0015] First, based on the characteristic parameters of different operating modes, the pulse carrier frequency RF, the distance R between the reconnaissance aircraft and the adversary's radar carrier aircraft, and the reconnaissance aircraft speed V are set respectively. a and the speed V of the radar carrier t Pulse width, light speed, number of working mode groups.

[0016] Then generate data for each working mode:

[0017] The specific PRI values ​​generated for each working mode are as follows:

[0018] (1) The specific expression for generating the PRI value under the working mode of the medium-repetition-frequency transformer is as follows:

[0019]

[0020] Among them, R c Indicates the maximum clear distance of the corresponding mid-repetition frequency design; T A Let c represent an intermediate variable, where c = 2.9979 × 10 8 m / s represents the speed of light; τ p Indicates the transmit pulse width; PRI k ,PRI j This represents the required pulse repetition interval for each group within the variable group, specifically the pulse repetition interval between the k-th and j-th groups. k and j represent parameters indicating the required number of repetition groups; C1 is odd, and C2 is even. δ j This indicates that a small perturbation is obtained that increases the target visibility by more than 96% at the corresponding repetition frequency.

[0021] (2) The expression for generating the PRI value under the high-repetition-frequency transformer operating mode is as follows:

[0022]

[0023] Among them, PRI A PRI1 and PRI2 represent the three sets of high repetition rate PRI values ​​required for generation; V a and V t These represent the speed of the reconnaissance aircraft and the speed of the radar carrier aircraft, respectively; and the number of pulse carrier frequency (RF) groups and the number of pulse repetition interval (PRI) groups are the same, λ represents the pulse signal wavelength, ceil represents rounding up the specific value, and I represents an intermediate variable.

[0024] In each working mode, the number of group variables for RF and PW is the same as the number of group variables for PRI, and the specific value is a random number within a certain range.

[0025] Based on the data generated in each of the above working modes, working modes are transferred according to certain working mode transfer relationships to generate corresponding working mode data with working mode transfer relationships.

[0026] Next, the pulse datasets generated under different operating modes are distinguished by operating mode labels. Then, the pulse datasets are padded with data, and considering the contamination caused by non-ideal conditions in the real electromagnetic environment, a receiver pulse sequence interception model is proposed to process the padded dataset.

[0027] The actual pulse sequence intercepted by the receiver may be contaminated by: measurement noise, pulse loss, and false pulses. The specific model for the receiver's intercepted pulse sequence is as follows:

[0028]

[0029] in, This refers to a pulse sequence representing the characterization operating mode under non-ideal conditions such as measurement noise, pulse loss, and spurious pulses. A pulse group indicating the presence of measurement noise. This represents a spurious pulse group, where L represents the length of the pulse sequence in the ideal operating mode. s This indicates the sequence length of the corresponding spurious pulse. This indicates a lost pulse, along with corresponding measurement noise.

[0030] Finally, the dataset after data pollution treatment is divided into corresponding training and testing sets according to actual needs, thus completing the preprocessing steps of the dataset.

[0031] Furthermore, step S2 is specifically as follows:

[0032] By combining GRU and TCN networks, a network model for encoder and decoder is constructed, namely the multi-functional radar GRUTCN working mode recognition model. This model uses a bidirectional two-layer GRU neural network as the corresponding encoder architecture. The bidirectional GRU extracts the temporal features of the working mode pulse input sequence and performs quantization and dimensionality reduction to form the corresponding working mode features. After the GRU network extracts the features of the data, the decoder composed of the TCN network decodes the extracted features into the required output.

[0033] Furthermore, step S3 is specifically as follows:

[0034] The training set data generated in step S1 is input into the model for training. The learning rate and batch size are adjusted according to the accuracy of the training set until the loss function tends to stabilize and the model converges, thus completing the model training. The expression of the loss function is as follows:

[0035]

[0036] Where, N train This indicates the number of samples in the training set. L represents the network output at time t, i.e., the probability of each operating mode determined by the network at time t; i s represents the pulse sequence length of the i-th training sample. t λ represents the true label at time t; λ represents the corresponding regularization factor; and ω2 represents the L2 regularization weight vector.

[0037] Furthermore, step S4 is specifically as follows:

[0038] The test set data generated in step S1 is fed into the trained model. The model outputs a radar operating mode probability label. The corresponding operating mode recognition accuracy is calculated based on the label, and the corresponding pulse-level multi-function radar operating mode recognition result is obtained, thus completing the multi-function radar operating mode recognition test.

[0039] The accuracy calculation formula is as follows:

[0040]

[0041]

[0042] Where, N test This indicates the number of samples in the test set. y represents the identification tag of the t-th pulse output by the network. t This indicates its true label. L i This represents the pulse sequence length for the i-th test sample.

[0043] The beneficial effects of this invention are as follows: This invention discloses a multi-functional radar operating mode recognition method based on a GRUTCN network. First, the operating mode characteristics of airborne phased array radar are analyzed, multi-functional radar pulse group data is generated, input into a dataset generation program, and preprocessed. The dataset is then divided into training and test sets. Next, a multi-functional radar GRUTCN operating mode recognition model is constructed. The preprocessed training set data is input into the model for training. Finally, the test set data is fed into the trained model to complete the multi-functional radar operating mode recognition test. This invention combines GRU and TCN, proposing a network model for encoder and decoder based on GRU-TCN. This solves the problems of incomplete analysis of operating mode features and low recognition rates in existing operating mode recognition methods. It can maintain a recognition accuracy far higher than other existing networks under non-ideal conditions, enhancing robustness. Attached Figure Description

[0044] Figure 1 This is a flowchart of a multi-functional radar operating mode recognition method based on a GRUTCN network according to the present invention.

[0045] Figure 2 This is a schematic diagram illustrating the working mode switching pattern of the multi-functional radar in this embodiment.

[0046] Figure 3 This is a structural diagram of the GRUTCN multi-functional radar working mode recognition model in an embodiment of the present invention.

[0047] Figure 4 This is a diagram illustrating the effect of multi-functional radar working mode recognition in an embodiment of the present invention. Detailed Implementation

[0048] The method of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0049] like Figure 1 The flowchart of a multi-functional radar operating mode recognition method based on a GRUTCN network according to the present invention is shown below. The specific steps are as follows:

[0050] S1. Analyze the working mode characteristics of airborne phased array radar, generate multi-functional radar pulse group data, input the dataset generation program and perform data preprocessing, and divide the dataset into training set and test set.

[0051] S2. Construct a multi-functional radar GRUTCN working mode recognition model;

[0052] S3. Input the training set data after preprocessing in step S1 into the model for training.

[0053] S4, Multifunctional Radar Operating Mode Recognition Test.

[0054] In this embodiment, step S1 is specifically as follows:

[0055] In this embodiment, under the Matlab 2022b environment, the pulse carrier frequency RF, the distance R between the reconnaissance aircraft and the adversary's radar carrier aircraft, and the reconnaissance aircraft speed V are set according to the characteristic parameters of different working modes. a and the speed V of the radar carrier t Pulse width, light speed, number of working mode groups.

[0056] This embodiment includes four working modes: VS, RWS, TAS, and STT. VS mode refers to the velocity search mode, while RWS mode (search-while-ranging mode) can be divided into two sub-modes: HRWS (High-PRF Range While Search) mode and MRWS (Medium-PRF Range While Search) sub-mode. TAS (search and track mode) mode can also be divided into two sub-states: search sub-state (TAS(S)) and track sub-state (TAS(T)). STT refers to the single-target tracking mode.

[0057] The working modes are defined and the corresponding modulation methods are shown in Table 1 (Working Mode Modulation Table):

[0058] Table 1

[0059]

[0060] The specific values ​​for each working mode are shown in Table 2 (Working Mode Value Table), which shows the range of values ​​for the corresponding data under each working mode.

[0061] Table 2

[0062]

[0063] Where U represents a uniform distribution.

[0064] Then, generate data for each working mode (i.e., calculate the corresponding PRI data based on the information set above):

[0065] The specific PRI values ​​generated for each working mode are as follows:

[0066] (1) The specific expression for generating the PRI value under the working mode of the medium-repetition-frequency transformer (i.e., the MRWS, TAS, STT and other working modes of the transformer) is as follows:

[0067]

[0068] Among them, R c Indicates the maximum clear distance of the corresponding mid-repetition frequency design; T A Let c represent an intermediate variable, where c = 2.9979 × 10 8 m / s represents the speed of light; τ p Indicates the transmit pulse width; PRI k ,PRI j This represents the required pulse repetition interval for each group within the variable group, specifically the pulse repetition interval between the k-th and j-th groups. k and j represent parameters indicating the required number of repetition groups, which in this embodiment are 0, ..., 4; C1 is an odd number, taken as 9 in this embodiment, and C2 is an even number, taken as 12 in this embodiment. δ j This indicates a small perturbation that results in a target visibility greater than 96% at the corresponding repetition frequency. In this embodiment, the values ​​are 0.1, ..., 0.3.

[0069] The pulse width value is determined according to Table 2. The maximum clear distance in the (MRWS) sub-mode is 150-200km, while the maximum clear distance in the TAS mode is 50-200km, and the maximum clear distance in the STT working mode is also 50-200km.

[0070] (2) The expression for generating the PRI value in the high repetition rate transformer working mode (i.e., HRWS mode) is as follows:

[0071]

[0072] Among them, the maximum clear distance designed for HRWS mode is 300km, and its pulse width value is determined according to Table 2; PRIA PRI1 and PRI2 represent the three sets of high repetition rate PRI values ​​required for generation; V a and V t These represent the speed of the reconnaissance aircraft and the speed of the radar carrier aircraft, respectively; and the number of pulse carrier frequency (RF) groups and the number of pulse repetition interval (PRI) groups are the same, λ represents the pulse signal wavelength, ceil represents rounding up the specific value, and I represents an intermediate variable.

[0073] In each working mode, the number of group variables for RF and PW is the same as the number of group variables for PRI, and the specific value is a random number within a certain range.

[0074] Based on the data generated within each of the above-mentioned working modes, working modes are transferred according to certain working mode transfer relationships, generating corresponding working mode data with these relationships. The specific working mode transfer relationships are as follows: Figure 2 As shown, in this embodiment, the transition relationship between the four working modes is that VS and RWS alternately search to achieve spatial domain search and target detection. When a target is detected, it switches to the corresponding TAS mode or the corresponding STT working mode, as detailed below:

[0075] Because the VS mode suffers from severe range ambiguity, it needs to be alternated with the RWS mode to achieve target de-ambiguity. Simultaneously, alternating between VS and RWS modes also provides full-attitude target detection capability. Once a target is detected in VS mode, the HRWS waveform and corresponding MRWS waveform in RWS mode must be transmitted to establish tracking that meets the accuracy requirements for track updates.

[0076] Therefore, upon target detection, VS mode always switches to RWS mode for distance measurement and corresponding deblurring. If the target threat level is low, it switches to TAS tracking mode for target tracking. If the target threat level is high, it switches to STT mode for high-precision tracking. If the target is lost during tracking, it switches back to the corresponding search mode for re-search. Similarly, when the target threat level decreases, the tracking mode switches from STT mode to the corresponding TAS mode.

[0077] Next, the pulse datasets generated under different operating modes are distinguished by operating mode labels. Then, the pulse datasets are padded with data, and considering the contamination caused by non-ideal conditions in the real electromagnetic environment, a receiver pulse sequence interception model is proposed to process the padded dataset.

[0078] The pulse sequence interception model of the receiver is as follows: Due to the complexity of the electromagnetic environment, the pulse sequence intercepted by the actual receiver may be contaminated by: measurement noise, pulse loss, and false pulses, as detailed below:

[0079] (1) Noise measurement

[0080] During parameter measurement, the captured pulse sequence may contain measurement errors due to factors such as device thermal noise and electromagnetic environment. For the captured pulse group, the actual value of the captured signal parameter is the true value of the transmitted pulse signal plus the measurement error.

[0081] (2) Pulse loss

[0082] Because phased array radar beams have a high degree of directionality, it is difficult for reconnaissance receivers to continuously capture the main lobe of the radar beam, which may lead to the loss of the corresponding pulse sequence.

[0083] (3) False pulses

[0084] In real combat environments, multiple radars are often present. During the sorting process, the complexity of the radiation source signals and the performance limitations of the sorting algorithm can lead to the inclusion of spurious pulses in the pulse sequence group. Based on the non-ideal conditions such as lost pulses, measurement noise, and spurious pulses discussed in this section, the following pulse sequence model intercepted by the receiver is derived:

[0085]

[0086] in, This refers to a pulse sequence representing the characterization operating mode under non-ideal conditions such as measurement noise, pulse loss, and spurious pulses. A pulse group indicating the presence of measurement noise. This represents a spurious pulse group, where L represents the length of the pulse sequence in the ideal operating mode. s This indicates the sequence length of the corresponding spurious pulse. This indicates a lost pulse, along with corresponding measurement noise.

[0087] Finally, in this embodiment, the dataset after data contamination is divided into training and testing sets in a 7:3 ratio to complete the preprocessing steps of the dataset.

[0088] In this embodiment, step S2 is specifically as follows:

[0089] This embodiment combines the GRU network and TCN network according to the above-described working mode transfer method to build an encoder-decoder network model based on GRUTCN, namely, a multi-functional radar GRUTCN working mode recognition model, as shown in the specific model below. Figure 3 As shown.

[0090] This embodiment uses a bidirectional two-layer GRU neural network as the corresponding encoder architecture. The timing features of the working mode pulse input sequence are extracted by the bidirectional GRU and then quantized and reduced in dimensionality to form the corresponding working mode features. After the GRU network extracts the features of the data, the decoder composed of the TCN network decodes the extracted features into the required output.

[0091] in, Figure 3 In this context, g represents a neuron in a linear fully connected layer, and d represents the corresponding inflation factor in the TCN network. This represents the i-th captured pulse sequence.

[0092] In this embodiment, step S3 is specifically as follows:

[0093] In this embodiment, a multi-functional radar GRUTCN working mode recognition model is built in the PyCharm environment. The training set data generated in step S1 is input into the model for training. The learning rate and batch size are adjusted according to the accuracy of the training set until the loss function tends to stabilize and the model converges, thus completing the model training. The expression of the loss function is as follows:

[0094]

[0095] Where, N train This indicates the number of samples in the training set. L represents the network output at time t, i.e., the probability of each operating mode determined by the network at time t; i s represents the pulse sequence length of the i-th training sample. t ωt represents the true label at time t; λ represents the corresponding regularization factor; and ω2 represents the L2 regularization weight vector. In this embodiment, the training set contains 2400 training samples, and the pulse sequence length for each mode in each sample is 100. The corresponding GRUTCN parameters are: batchsize = 64, and the learning rate is 10t. -3 level.

[0096] In this embodiment, step S4 is specifically as follows:

[0097] The test set data generated in step S1 is fed into the trained model. The model outputs a radar operating mode probability label. The corresponding operating mode recognition accuracy is calculated based on the label, and the corresponding pulse-level multi-function radar operating mode recognition result is obtained, thus completing the multi-function radar operating mode recognition test.

[0098] The accuracy calculation formula is as follows:

[0099]

[0100]

[0101] Where, N test This indicates the number of samples in the test set. y represents the identification tag of the t-th pulse output by the network. t This indicates its true label. L i This represents the pulse sequence length for the i-th test sample.

[0102] The specific recognition effect of this embodiment is as follows: Figure 4 As shown, Figure 4 (a) represents a scenario where only measurement noise exists. Figure 4 (b) represents the scenario where only pulse loss exists. Figure 4 (c) represents a scenario where only spurious pulses exist. Figure 4 (d) is a mixed scene.

[0103] Depend on Figure 4 It can be seen that when only measurement noise exists, the Gaussian noise does not disrupt the transition characteristics and temporal features within the operating mode, as the noise exhibits a Gaussian distribution. Therefore, all networks can effectively extract the temporal features of the data and identify the corresponding operating modes. However, the GRUTCN network has a stronger ability to extract the temporal features of the data, and thus has better recognition capabilities.

[0104] When pulses are lost or false, the temporal sequence of pulses is disrupted, or even erroneous, leading to a decrease in the network's recognition accuracy. However, because the GRUTCN network has stronger capabilities in extracting and decoding temporal features compared to existing temporal recognition networks, its recognition accuracy remains above 80%.

[0105] In complex scenarios, not only are temporal relationships disrupted, but data is also contaminated by noise, damaging data structures and characteristics both within and between operating modes. As this damage intensifies, the performance of all networks further degrades. However, the GRUTCN method still significantly outperforms other methods, further highlighting the effectiveness of the corresponding network.

[0106] This embodiment uses data contaminated by non-ideal conditions such as measurement noise, lost pulses, and spurious pulses for simulation verification. The simulation results show that the method of the present invention can maintain a recognition accuracy much higher than other traditional networks under non-ideal conditions, reflecting the robustness of the method of the present invention.

[0107] In summary, the method of this invention first analyzes the characteristics of various operating modes based on airborne phased array radar. Based on this analysis, a corresponding operating mode transition relationship model is proposed. The possibility of the receiver intercepting pulse sequences under non-ideal conditions is also considered. Finally, an encoding / decoding model based on GRU and TCN networks is proposed, using GRU as the encoder and TCN as the corresponding decoding structure. The encoder extracts intra-class and inter-class features from the input operating mode data, while the decoder reconstructs the decoded information from the extracted features. This method solves the problems of incomplete analysis of operating mode features and low recognition rates in existing operating mode recognition methods. It maintains a recognition accuracy far higher than other existing networks under non-ideal conditions, enhancing robustness.

[0108] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Various modifications and variations can be made to the invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the scope of the claims of the invention.

Claims

1. A multi-functional radar operating mode recognition method based on GRUTCN network, the specific steps of which are as follows: S1. Analyze the working mode characteristics of airborne phased array radar, generate multi-functional radar pulse group data, input the dataset generation program and perform data preprocessing, and divide the dataset into training set and test set. First, based on the characteristic parameters of different operating modes, the pulse carrier frequency (RF) and the distance between the reconnaissance aircraft and the adversary's radar carrier aircraft are set respectively. reconnaissance aircraft speed and the speed of radar carrier Pulse width, speed of light, number of operating mode groups; Then generate data for each working mode: The specific PRI values ​​generated for each working mode are as follows: (1) The specific expression for generating the PRI value under the working mode of the medium-repetition-frequency transformer is as follows: （1）； in, This indicates the maximum clear distance of the corresponding mid-repetition frequency design; Indicates intermediate variables. Represents the speed of light; Indicates the transmission pulse width; This indicates the pulse repetition interval of each group within the required group, i.e., the nth group. Group and No. The pulse repetition interval of the group; The parameter indicates the number of repetition groups required. It is an odd number. It is an even number; This indicates that a small perturbation is obtained that results in a target visibility greater than 96% at the corresponding repetition frequency. (2) The expression for the generation method of PRI value under the high repetition rate group transformer working mode is as follows: （2）； in, , , This represents the three sets of high repetition rate (PRI) values ​​required for the corresponding generation; and These represent the speed of the reconnaissance aircraft and the speed of the radar carrier aircraft, respectively; and the number of pulse carrier frequency (RF) groups and the number of pulse repetition interval (PRI) groups are the same. This represents the wavelength of the pulse signal; ceil indicates rounding up the specific value. Indicates intermediate variables; In each working mode, the number of group variables for RF and PW is the same as the number of group variables for PRI, and the specific value is a random number within a certain range. Based on the data generated in each of the above working modes, the working modes are transferred according to certain working mode transfer relationships to generate corresponding working mode data with working mode transfer relationships. Next, the pulse datasets generated under different working modes are distinguished by working mode labels; then, the pulse datasets are filled with data, and considering the pollution of the dataset by corresponding non-ideal conditions in the real electromagnetic environment, a receiver pulse sequence interception model is proposed to process the data pollution of the filled dataset. Finally, the dataset after data pollution treatment is divided into corresponding training and test sets according to actual needs and a certain ratio, thus completing the preprocessing steps of the dataset. S2. Construct a multi-functional radar GRUTCN working mode recognition model; S3. Input the training set data after preprocessing in step S1 into the model for training. S4, Multifunctional Radar Operating Mode Recognition Test.

2. The multi-functional radar operating mode recognition method based on GRUTCN network according to claim 1, characterized in that, The specific steps of S1 are as follows: The pulse sequences intercepted by actual receivers are susceptible to contamination from measurement noise, pulse loss, and false pulses. Therefore, the specific model for the intercepted pulse sequences by the receiver is as follows: （3）； in, This refers to a pulse sequence representing the characterization operating mode under non-ideal conditions such as measurement noise, pulse loss, and spurious pulses. A pulse group indicating the presence of measurement noise. Indicates a false pulse group, This represents the length of the pulse sequence in the ideal operating mode. Indicates the sequence length of the corresponding spurious pulse; This indicates a lost pulse, along with corresponding measurement noise.

3. The multi-functional radar operating mode recognition method based on GRUTCN network according to claim 1, characterized in that, Step S2 is as follows: By combining GRU and TCN networks, a network model for encoder and decoder is constructed, namely the multi-functional radar GRUTCN working mode recognition model. This model uses a bidirectional two-layer GRU neural network as the corresponding encoder architecture. The bidirectional GRU extracts the temporal features of the working mode pulse input sequence and performs quantization and dimensionality reduction to form the corresponding working mode features. After the GRU network extracts the features of the data, the decoder composed of the TCN network decodes the extracted features into the required output.

4. The multi-functional radar operating mode recognition method based on GRUTCN network according to claim 1, characterized in that, Step S3 is as follows: The training set data generated in step S1 is input into the model for training. The learning rate and batch size are adjusted according to the accuracy of the training set until the loss function tends to stabilize and the model converges, thus completing the model training. The expression of the loss function is as follows: （4）； in, This indicates the number of samples in the training set. express The output of the time-matter network, i.e. The probability of each working mode identified by the time-matter network; Indicates the first The length of the pulse sequence under each training sample express A true label for every moment; This represents the corresponding regularization factor. express Regularized weight vector.

5. The multi-functional radar operating mode recognition method based on GRUTCN network according to claim 1, characterized in that, Step S4 is as follows: The test set data generated in step S1 is fed into the trained model. The model outputs a radar working mode probability label. The corresponding working mode recognition accuracy is calculated based on the label, and the corresponding pulse-level multi-function radar working mode recognition result is obtained, thus completing the multi-function radar working mode recognition test. The accuracy calculation formula is as follows: （5）； in, This indicates the number of samples in the test set. Indicates the network output of the first A pulse identification tag, This indicates its true label; Indicates the first The length of the pulse sequence under each test sample.

Images