Radar signal state classification method and device based on time sequence labeling, and storage medium

Abstract

The application discloses a radar signal state classification method and device based on time sequence labeling and a storage medium, wherein according to the time of arrival (TOA) characteristics of two data points before and after a tracking while scanning (TWS) radar signal, the radar signal data is columned; the trend characteristics and seasonal characteristics of the radar signal are extracted through a Transform module to generate a target sequence; a double-tower encoder is used to perform representation learning on the target sequence, and the representation results are fused through a gating unit; each data point in the time sequence is labeled to determine the state of each data point; a regularization decoder is used to predict the consistency of adjacent labels in the time sequence, and the incoherent conversion of the labels in the time sequence is determined through a regularization term and a total loss function to classify the radar signal.

Description

Technical Field

[0001] This invention relates to the field of time series and signal classification technology, and in particular to a radar signal state classification method, device and storage medium based on time series labeling. Background Technology

[0002] With the continuous development of radar technology, multi-functional radar systems, such as Track-while-Scan (TWS) radar, have been widely used on the battlefield. To maintain an advantage in radar countermeasures, accurately predicting the operational state of target radars (e.g., tracking, searching, etc.) has become an urgent need. Since radars in different states emit different signal patterns, research on radar signal state classification has gradually emerged in the radar field. This research aims to classify received radar signals by labeling each signal point in a sub-segment with its corresponding state. Considering the difficulty of human interpretation of radar signals, radar signal state classification methods based on artificial intelligence models have emerged. However, due to the special nature of radar countermeasures, real radar signal data is scarce due to its unique characteristics. Therefore, model training requires a large amount of simulation data, but simulation data does not possess the imbalanced characteristics of real data. Thus, existing datasets suffer from a lack of real data and low-quality simulation data. Furthermore, the duration of each state in a radar signal is not uniform, making it impossible to classify the sequence using fixed-length windows. Therefore, how to perform length-adaptive analysis on the sequence is another problem that needs to be solved.

[0003] In summary, radar signal state classification tasks present the following challenges: 1. The radar signal data volume is small, resulting in low-quality simulation sequences; 2. Radar signals have non-equilibrium characteristics, making it difficult for models to adaptively segment them; Existing technologies cannot effectively solve the above problems, making radar signal state classification a challenging task. Summary of the Invention

[0004] The purpose of this invention is to provide a radar signal state classification method, device, and storage medium based on time-series annotation, in order to solve the problems of dataset quality and sequence length adaptive analysis, and to improve the classification performance of the model in real-world scenarios.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A radar signal state classification method based on time-series labeling includes: Data augmentation module; Based on the Time of Arrival (TOA) characteristics of the two data points before and after the scanning and tracking radar signal (TWS), the radar signal data is divided into columns. The Transformer module is used to extract trend and seasonal features of radar signals to generate target sequences. Time series annotation module; A dual-tower encoder is used to learn the representation of the target sequence, and the representation results are fused through a gating unit; Label each data point in the time series and determine the status of each data point; The consistency of adjacent labels in a time series is predicted using a regularized decoder, and the incoherent transitions of labels in the time series are judged by the regularization term and the total loss function, thus classifying radar signals.

[0006] Preferably, the power amplification feature (PA) of the radar signal is extracted using the following expression, and the difference between the two data points in the time of arrival (TOA) feature is calculated based on the power amplification feature PA. ; in, ; ; ; In the formula, It is Gaussian noise with a mean of 0 and a standard deviation of 3; The effective angle is calculated from the pitch and azimuth angles; Antenna gain; Gain loss due to distance; For distance; The pitch angle; It is the azimuth angle; Radio frequency; The imaginary unit; The wavelength of the signal; The distance between adjacent array elements; This represents the beam pointing angle.

[0007] Preferably, the Transformer module includes an encoder and a decoder; In the encoder, a diffusion model is used to jointly model and encode the target sequence and the conditional sequence through a cross-attention mechanism; The decoder adopts a multi-layer structure, with each layer containing a Transformer block, a feedforward neural network, and a set of feature extraction layers. The feature extraction layer includes a trend extractor and a seasonality extractor.

[0008] Preferably, the joint modeling coding includes the following steps: First, a linear transformation is performed on the two sets of data points in the radar signal, and the two sets of data points are assigned query vector, key vector, and value vector, as shown in the following formula: ; ; ; In the formula, , They are respectively , , The weight matrix; For target sequence data; The number of data points; Feature dimensions for each data point; For conditional data; The feature dimensions for each conditional data point; Secondly, the attention score is calculated using the query vector (Query) and the key vector (Key), as shown in the following formula: ; Finally, the joint representation output is obtained by combining the Value and the attention score. , It serves as a joint encoding of the target sequence and the conditional sequence for subsequent feature extraction.

[0009] Preferably, the calculation expression for the trend extractor is: ; In the formula, For the first The average value of the output of each decoder block; For the first Each decoder block at time step The input to the trend feature extraction module includes the encoded information of the target sequence and conditional sequence through a cross-attention mechanism; Linear represents a linear transformation. It is a slowly changing polynomial space matrix; The degree of the polynomial is the highest degree term. The calculation expression for the seasonal extractor is: ; ; ; In the formula, For the first Each decoder block at time step The Amplitude of the seasonal component of frequency ; It is the Fourier transform function; For frequency; For the first Each decoder block at time step The The phase of the seasonal component of the frequency; This is a function used to calculate the phase angle of the frequency components; Before selection indivual; For the first Each decoder block at time step The The pre-seasonal component of frequency Large amplitude ; The length of one period in the time series; For the first Each decoder block at time step Reconstruction of seasonal patterns; For the first Each decoder block at time step The former The amplitude of large seasonal components; The positive frequency component represents the frequency of a specific seasonal pattern in the seasonality analysis of time series data. The normalized frequency is used to adjust the frequency components to match the actual periodicity of the time series data. For the first Each decoder block at time step The former The phase of major seasonal components; The negative frequency component is the complex conjugate of the positive frequency component in the Fourier transform of a real signal, and together they constitute the complete spectrum of the signal. The phase of the negative frequency component.

[0010] Preferably, the dual-tower encoder includes a time-step feature encoder and a multivariate feature encoder; the dual-tower encoder performs representation learning on the target sequence including the following steps: Time-step feature encoder: The time series is positionally encoded. The time series data obtained by position encoding and nonlinear transformation encoding are input into the time-step feature encoder. Self-attention and masking mechanisms are used to focus on each point on a single channel. Paired attention weights between all time steps are calculated through QKV pattern (query vector, key vector, value vector). Multivariate feature encoder: Captures the correlation between multiple variables based on an attention mechanism across all channels at the current time step using a mask; A fully connected layer with non-linear activation is used to process the time steps or sequences in order, mapping the time step feature outputs to a step-wise manner and processing each feature channel independently; the multivariate feature outputs are mapped to a channel-wise manner; the time steps and multivariate features are mapped to a vector h through concatenation and linear projection, and each dimension in the vector h is mapped to the weights of the two towers through Softmax. Finally, the final representation is obtained by combining the weights and the outputs of the two towers.

[0011] Preferably, the regularization term is the sum of the transformation costs between adjacent labels: ; in, ; In the formula, The cost function is used to quantize the state. Transform to state Inconsistency; It is the Kronecker function if and only if Its value is 1 if it is true, and 0 otherwise. The total loss function is a weighted sum of the prediction loss and the regularization term: ; In the formula, For model-based prediction and real labels The predicted loss; The regularization weight is used to control the degree of influence of the regularization term on the total loss.

[0012] A computer device includes a memory and a processor, wherein the memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the aforementioned time-series labeled radar signal state classification method.

[0013] A computer storage medium for storing program data, which, when executed by a processor, is used to implement the aforementioned radar signal state classification method based on time-series annotation.

[0014] Compared with the prior art, the present invention has the following beneficial effects: This invention augments radar signal simulation data using a diffusion model and Transformer module. The columnar features of the radar signal are represented by a sequence columnar segmentation method, resulting in more accurate generation of target sequences. Subsequently, by introducing sequence labeling into the field of time series classification, the state classification task of unbalanced radar signals is completed with high accuracy, and the model performance is improved through effective regularization. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the overall workflow of the present invention; Figure 2 This is a schematic diagram of a radar signal sequence according to an embodiment of the present invention; Figure 3 This is a schematic diagram comparing radar signal training batches in an embodiment of the present invention; Figure 4 This is a schematic diagram of the decoder structure according to an embodiment of the present invention; Figure 5 This is a schematic diagram of time-series labeling regularization based on Transformer in an embodiment of the present invention; Figure 6 This is a schematic diagram illustrating an example of implementing the present invention. Detailed Implementation

[0016] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0017] Example 1: Please see Figures 1-5 A radar signal state classification method based on time-series labeling includes the following steps; First, such as Figure 1 As shown, based on the diffusion model, the data augmentation module is used to perform style transfer on the signal simulation sequence to generate a target sequence that is closer to the real data in terms of distribution, so as to improve the quality and quantity of the dataset.

[0018] like Figure 2As shown, the scanning and tracking radar signal (TWS) can be considered as a multivariate time series, including two main features: power amplification (PA) and time of arrival (TOA). The TOA feature exhibits a distinct step-like characteristic; therefore, the sequence plotted with TOA as the horizontal axis and PA as the vertical axis shows an unbalanced characteristic, exhibiting a multi-bar feature. Within each bar, PA shows an envelope trend of first rising and then falling, while TOA shows a gradually rising step-like trend. Due to the large numerical differences between bars, the gradual increase within a bar and the rapid increase between bars cannot be well learned simultaneously by the model. Therefore, a bar-splitting strategy is adopted, such as... Figure 3 As shown, the model learns the temporal variation characteristics in each column to better fit the sequence variation trend and periodicity.

[0019] The power amplification characteristic (PA) involves multiple physical parameters such as radio frequency, elevation angle, azimuth angle, and range. The PA reflects the state change trend and periodicity of the radar signal, and its calculation expression is as follows: ; in, ; ; ; In the formula, It is Gaussian noise with a mean of 0 and a standard deviation of 3; The effective angle is calculated from the pitch and azimuth angles; Antenna gain; Gain loss due to distance; For distance; The pitch angle; It is the azimuth angle; Radio frequency; The imaginary unit; The wavelength of the signal; The distance between adjacent array elements; This represents the beam pointing angle.

[0020] Specifically, a threshold for abrupt changes in the TOA (Time of Arrival) value is established based on the signal values. When the difference between two consecutive data points in the TOA dimension exceeds this threshold, the two data points are divided into different pillars. After the division, the batch size for training the diffusion model is designed based on the amount of data in each pillar, ensuring that the individual signal pillars are located within the same batch. This allows the model to focus on learning and simulating the data features within a single pillar, better adapting to the problem of sudden increases in data between unbalanced sequences.

[0021] Second, by decoupling and learning temporal features through the encoder-decoder-based Transformer module, the model's ability to capture global correlations, temporal trends, and periodic features is enhanced.

[0022] In the encoder, the target sequence and the conditional sequence are jointly modeled and encoded through a cross-attention mechanism, and a diffusion model is adopted.

[0023] Specifically, First, based on the attention mechanism, a linear transformation is performed on the two sets of data points in the radar signal, and the two sets of data points are assigned query vector, key vector, and value vector, as shown in the following formula: ; ; ; In the formula, , They are respectively , , The weight matrix; For target sequence data; The number of data points; Feature dimensions for each data point; For conditional data; The feature dimension for each conditional data point.

[0024] Secondly, the attention score is calculated using the query vector (Query) and the key vector (Key), as shown in the following formula: ; These attention scores determine how each target sequence data point is implicitly subject to attentional constraints, such as the smoothness of state changes and the correlation between different state segments.

[0025] Finally, the joint representation output is obtained by combining the Value and the attention score. , It serves as a joint encoding of the target sequence and the conditional sequence for subsequent feature extraction.

[0026] like Figure 4 As shown, the decoder adopts a multi-layer structure, with each layer containing a Transformer block, a feedforward neural network, and a set of feature extraction layers; the feature extraction layers include a trend extractor and a seasonality extractor.

[0027] After obtaining the joint representation output, the trend feature is used to capture the overall trend of the non-stationary sequence under smooth change conditions. The trend feature is modeled using a multinomial regressor method, as shown in the following formula: ; In the formula, For the first The average value of the output of each decoder block; For the first Each decoder block at time step The input to the trend feature extraction module includes the encoded information of the target sequence and conditional sequence through a cross-attention mechanism; Linear represents a linear transformation. It is a slowly changing polynomial space matrix; The degree of the polynomial is the highest degree term.

[0028] Seasonality includes periodic features in the sequence and non-periodic error features. Fourier series are used to capture the seasonality of time series in Fourier synthesis layers. Fourier series can transform time series data from the time domain to the frequency domain for better analysis and understanding of periodic patterns in the data, and extract frequency components for synthesizing or reconstructing time series with specific seasonal characteristics. The calculation expression for the seasonality extractor is as follows: ; ; ; In the formula, For the first Each decoder block at time step The Amplitude of the seasonal component of frequency ; It is the Fourier transform function; For frequency; For the first Each decoder block at time step The The phase of the seasonal component of the frequency; This is a function used to calculate the phase angle of the frequency components; Before selection indivual; For the first Each decoder block at time step The The pre-seasonal component of frequency Large amplitude ; The length of one period in the time series; For the first Each decoder block at time step Reconstruction of seasonal patterns; For the first Each decoder block at time step The former The amplitude of large seasonal components; The positive frequency component represents the frequency of a specific seasonal pattern in the seasonality analysis of time series data. The normalized frequency is used to adjust the frequency components to match the actual periodicity of the time series data. For the first Each decoder block at time step The former The phase of major seasonal components; The negative frequency component is the complex conjugate of the positive frequency component in the Fourier transform of a real signal, and together they constitute the complete spectrum of the signal. The phase of the negative frequency component.

[0029] III. Figure 5 As shown, the Transformer-based temporal labeling regularization module represents the time series, the dual-tower encoder module fuses the representation results through a gating unit, and the regularization model guides the model labeling training based on the continuity of the temporal segment state.

[0030] The dual-tower encoder module includes a time-step feature encoder and a multi-feature encoder.

[0031] In the time-step feature encoder, self-attention and masking are used to focus on each point in a single channel in order to encode temporal features, and pairwise attention weights between all time steps are calculated using QKV. Since time-step features are affected by the sequence order, the sequence is first positionally encoded, and the time-series data obtained by positional encoding and nonlinear transform encoding are input into the encoder together. In the multivariate feature encoder, the encoder captures the correlation between multiple variables based on an attention mechanism on all channels of the current time step through masking.

[0032] After implementing dual-tower encoding, a gating mechanism is used to learn the weights of each tower in order to merge the features encoded in the two dimensions. After obtaining the output of each tower, a fully connected layer with non-linear activation is used to map the time-step feature output to step-wise and the multi-dimensional feature output to channel-wise. By concatenation and linear projection, the time-step features and multi-dimensional features are mapped to a vector h. Each dimension in h is mapped to the weights of the two towers through Softmax. Finally, the final representation is obtained by combining the weights and the outputs of the two towers.

[0033] like Figure 6As shown, due to the continuity and smoothness of state transitions in time series, adjacent data points are more likely to belong to the same state category. However, standard loss functions in sequence labeling tasks often fail to adequately consider the characteristics of temporal continuity and smooth state transitions. To effectively model and handle these characteristics, a regularized time series labeling model is established. This regularized time series labeling model focuses on enhancing the consistency of the model in predicting adjacent labels. By penalizing incoherent transitions in the label sequence, it strengthens the regularized time series labeling model's learning of state persistence in the time series, thereby improving the overall performance of the sequence labeling task.

[0034] Specifically, for sequences ,make This represents the output label sequence of sequence labeling, where For time step The state category labels for the data points are defined. This is achieved by introducing an additional regularization term, which is defined as the sum of the transition costs between adjacent labels, quantifying the transition costs between consecutive labels in the sequence. In this way, the time-series labeling model not only learns to identify the correct state of each independent data point, but is also guided to understand and predict state transitions in a way that more closely reflects the dynamics of the actual sequence. This reduces misclassifications and improves overall labeling consistency without sacrificing response speed. The regularization term is: ; in, ; In the formula, The cost function is used to quantize the state. Transform to state Inconsistency; It is the Kronecker function if and only if Its value is 1 if the label changes, and 0 otherwise. This cost function only incurs a cost when the label changes, thus encouraging adjacent labels to remain consistent.

[0035] During the training of the time-series labeling model, the total loss function is defined as the weighted sum of the prediction loss and the regularization term: ; In the formula, For model-based prediction and real labels The predicted loss; The regularization weight is used to control the degree of influence of the regularization term on the total loss.

[0036] Regularization strategies not only help time-series labeled models capture more subtle state changes, but also suppress overfitting during model training, especially when labeled data is sparse or imbalanced. By appropriately adjusting the regularization weights, the emphasis on state consistency can be flexibly controlled, making model training more robust and further improving the model's generalization ability to complex time-series data.

[0037] Time series analysis in the radar field essentially involves analyzing time-series data reflecting the interaction between a target object and radar sensors. It has significant application value in accurately predicting the operational status of target radars (e.g., tracking, searching). By classifying the radar's state at different times, it's possible to effectively determine whether the radar is operating normally and accurately identify when it is subject to external interference, facilitating its deployment.

[0038] Example 2: This application also provides a computer storage medium for storing program data, which, when executed by a processor, is used to implement the radar signal state classification method based on time-series annotation as described in the above embodiments.

[0039] This application also provides a computer program product, wherein the computer program product includes a computer program operable to cause a computer to execute the radar signal state classification method based on time-series annotation as described in the embodiments of this application. The computer program product can be a software installation package.

[0040] The radar signal state classification method based on time-series labeling described in the above embodiments of this application, when implemented as a software functional unit and sold or used as an independent product, can be stored in a device, such as a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0041] In one specific embodiment, Table 1 shows the performance evaluation table for radar signal state classification.

[0042] Table 1 Radar Signal State Classification Performance Evaluation Table

[0043] This invention (MEDIAN) significantly outperforms other baseline models across multiple evaluation metrics. In key performance indicators such as accuracy, precision, recall, and F1 score, MEDIAN demonstrates a significant improvement of 41.8% compared to the KNN model, 26.2% compared to the LR model, 36.3% compared to the MLP model, and 8.2% compared to the LSTM model. Even compared to the state-of-the-art TST model, it shows a 7.0% improvement. These results not only highlight the superiority of this invention in radar signal state classification tasks but also demonstrate that data augmentation and the introduction of regularization terms describing temporal state consistency can significantly improve the robustness and generalization ability of the model.

[0044] The above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A radar signal state classification method based on time-series labeling, characterized in that, include: The data augmentation module is used by following these steps: Based on the Time of Arrival (TOA) characteristics of the two data points before and after the scanning and tracking radar signal (TWS), the radar signal data is divided into columns. The power amplification feature (PA) of the radar signal is extracted using the following expression, and the difference between the two data points in the time of arrival (TOA) feature is calculated based on the power amplification feature PA. ； in, ； ； ； In the formula, It is Gaussian noise with a mean of 0 and a standard deviation of 3; The effective angle is calculated from the pitch and azimuth angles; Antenna gain; Gain loss due to distance; For distance; The pitch angle; It is the azimuth angle; Radio frequency; The imaginary unit; The wavelength of the signal; The distance between adjacent array elements; This refers to the beam pointing angle; The Transformer module is used to extract trend and seasonal features of radar signals to generate target sequences. The timing annotation module uses the following steps: A dual-tower encoder is used to learn the representation of the target sequence, and the representation results are fused through a gating unit; Label each data point in the time series and determine the status of each data point; The consistency of adjacent labels in a time series is predicted using a regularized decoder, and the incoherent transitions of labels in the time series are judged by the regularization term and the total loss function, thus classifying radar signals.

2. The radar signal state classification method based on time-series labeling according to claim 1, characterized in that, The Transformer module includes an encoder and a decoder; In the encoder, a diffusion model is used to jointly model and encode the target sequence and the conditional sequence through a cross-attention mechanism; The decoder adopts a multi-layer structure, with each layer containing a Transformer block, a feedforward neural network, and a set of feature extraction layers. The feature extraction layer includes a trend extractor and a seasonality extractor.

3. The radar signal state classification method based on time-series labeling according to claim 2, characterized in that, The joint modeling coding includes the following steps: First, a linear transformation is performed on the two sets of data points in the radar signal, and the two sets of data points are assigned query vector, key vector, and value vector, as shown in the following formula: ； ； ； In the formula, , They are respectively , , The weight matrix; For target sequence data; The number of data points; Feature dimensions for each data point; For conditional data; The feature dimensions for each conditional data point; Secondly, the attention score is calculated using the query vector (Query) and the key vector (Key), as shown in the following formula: ； Finally, the joint representation output is obtained by combining the Value and the attention score. , It serves as a joint encoding of the target sequence and the conditional sequence for subsequent feature extraction.

4. The radar signal state classification method based on time-series labeling according to claim 2, characterized in that, The calculation expression for the trend extractor is: ； In the formula, For the first The average value of the output of each decoder block; For the first Each decoder block at time step The input to the trend feature extraction module includes the encoded information of the target sequence and conditional sequence through a cross-attention mechanism; Linear represents a linear transformation. This is a normalized time variable, with a value range of [0,1]. It is a slowly changing polynomial space matrix; The degree of the polynomial is the highest-order term. The calculation expression for the seasonal extractor is: ； ； ； In the formula, For the first Each decoder block at time step frequency Amplitude of seasonal components at time ; It is the Fourier transform function; For frequency; For the first Each decoder block at time step frequency The phase of seasonal components at that time; This is a function used to calculate the phase angle of the frequency components; Before selection indivual; For the first Each decoder block at time step frequency The seasonal components of time Large amplitude ; The length of one period in the time series; For the first Each decoder block at time step Reconstruction of seasonal patterns; For the first Each decoder block at time step The former The amplitude of a large seasonal component; The positive frequency component represents the frequency of a specific seasonal pattern in the seasonality analysis of time series data. The normalized frequency is used to adjust the frequency components to match the actual periodicity of the time series data. For the first Each decoder block at time step The former A phase of a major seasonal component; The negative frequency component is the complex conjugate of the positive frequency component in the Fourier transform of a real signal; together they constitute the complete spectrum of the signal. The phase of the negative frequency component.

5. The radar signal state classification method based on time-series labeling according to claim 1, characterized in that, The dual-tower encoder includes a time-step feature encoder and a multi-dimensional feature encoder; The dual-tower encoder performs representation learning on the target sequence, which includes the following steps: Time-step feature encoder: The time series is position-encoded. The time series data obtained by position encoding and nonlinear transform encoding are input together into the time-step feature encoder. Self-attention and masking mechanisms are used to focus on each point on a single channel, and pairwise attention weights between all time steps are calculated. Multivariate feature encoder: Captures the correlation between multiple variables based on an attention mechanism across all channels at the current time step using a mask; A fully connected layer with non-linear activation is used to process the time steps or sequences in order, mapping the time step feature outputs to a step-wise form and processing each feature channel independently; the multivariate feature outputs are mapped to a channel-wise form; the time steps and multivariate features are mapped to a vector h through concatenation and linear projection, and each dimension in the vector h is mapped to the weights of the two towers through Softmax. Finally, the final representation is obtained by combining the weights and the outputs of the two towers.

6. The radar signal state classification method based on time-series labeling according to claim 1, characterized in that, The regularization term is the sum of the transformation costs between adjacent labels: ； in, ； In the formula, The cost function is used to quantize the state. Transform to state Inconsistency; It is the Kronecker function if and only if Its value is 1 if it is true, and 0 otherwise. The total loss function is a weighted sum of the prediction loss and the regularization term: ； In the formula, For model-based prediction and real labels The predicted loss; The regularization weight is used to control the degree of influence of the regularization term on the total loss.

7. A computer device, characterized in that, The device includes a memory and a processor, wherein the memory is coupled to the processor; the memory is used to store program data, and the processor is used to execute the program data to implement the radar signal state classification method based on time-series labeling as described in any one of claims 1-6.

8. A computer storage medium, characterized in that, The computer storage medium is used to store program data, which, when executed by the processor, is used to implement the radar signal state classification method based on time-series labeling as described in any one of claims 1-6.

Images