Target radar signal enhancement method and device based on DPRNN

Abstract

The invention discloses a DPRNN-based target radar signal enhancement method and device, and the method comprises the steps: firstly constructing a data set, carrying out the sliding window segmentation processing of a mixed signal of a training set in the data set, obtaining a mixed time domain signal, building a target radar signal enhancement model based on a DPRNN network, and carrying out the sliding window segmentation processing of the mixed signal, performing signal enhancement processing on the mixed time domain signal to obtain a reconstructed time domain signal, inputting the reconstructed time domain signal into a target radar signal enhancement model, and training the target radar signal enhancement model by using a training set in the data set to obtain a trained target radar signal enhancement model; and finally, the trained target radar signal enhancement model is used to process the online intercepted mixed time domain signal to obtain an enhanced target time domain signal. The context features of the radar signals are extracted through the constructed target radar signal enhancement model, the target signal enhancement problem under the time-frequency overlapping condition is solved, and the signal enhancement effect is improved.

Description

Technical Field

[0001] The present application relates to the technical field of signal enhancement, and in particular to a method and device for enhancing target radar signals based on DPRNN. Background Art

[0002] In the field of signal processing, especially for low signal-to-noise ratio (SNR) signal enhancement in complex environments, both traditional signal processing methods and emerging deep learning technologies have certain limitations. Traditional signal enhancement methods mainly include wavelet transform (WT), empirical mode decomposition (EMD) based on Hilbert-Huang transform (HHT), singular spectrum analysis (SSA), and independent component analysis (ICA).

[0003] Although traditional methods have achieved certain results in specific application scenarios, they perform poorly in low signal-to-noise ratio environments; and due to the lack of prior information, it is difficult to accurately distinguish target signals from background noise in complex backgrounds, which can easily lead to the loss of key pulse signals. With the development of deep learning technology, some studies have begun to explore its application in signal enhancement. However, current signal enhancement schemes based on deep learning still face some challenges: on the one hand, these models often ignore the contextual information in radar signals, which limits their generalization ability in actual environments; on the other hand, when faced with overlapping target signals and background signals in the time-frequency domain, existing deep learning methods rely on the assumption of signal independence in time and frequency, making it difficult to effectively separate the two, which can easily lead to the loss of pulses of target signals.

[0004] Therefore, there is an urgent need for a new method that can overcome the shortcomings of existing technologies. This method should not only maintain good robustness under low signal-to-noise ratio conditions, but also have the ability to accurately extract and enhance target signals from complex background environments. Summary of the invention

[0005] The purpose of this application is to overcome the existing technical defects and provide a target radar signal enhancement method and device based on DPRNN. By constructing a target radar signal enhancement model to extract the contextual features of the radar signal, the problem of target signal enhancement in the case of time-frequency overlap is solved, and the effect of signal enhancement is improved.

[0006] The purpose of this application is achieved through the following technical solutions:

[0007] In the first aspect, the present application proposes a target radar signal enhancement method based on DPRNN, the method comprising:

[0008] Step S1, constructing a data set, performing sliding window segmentation processing on the mixed signal of the training set in the data set to obtain a mixed time domain signal;

[0009] Step S2: construct a target radar signal enhancement model based on the DPRNN network, and perform signal enhancement processing on the mixed time domain signal to obtain a reconstructed time domain signal;

[0010] Step S3, inputting the reconstructed time domain signal into the target radar signal enhancement model, and using the training set in the data set to train the target radar signal enhancement model to obtain a trained target radar signal enhancement model;

[0011] Step S4: Use the trained target radar signal enhancement model to process the mixed time domain signal intercepted online to obtain an enhanced target time domain signal.

[0012] In a possible implementation, step S1 includes:

[0013] Step S11, obtaining a plurality of original mixed signals and corresponding target radar signal time domain sample pairs, and constructing a data set;

[0014] Step S12: Divide the data set into a training set, a validation set, and a test set according to the ratio;

[0015] Step S13: Perform sliding window segmentation processing on the mixed signal in the training set to obtain a mixed time domain signal.

[0016] In a possible implementation, the target radar signal enhancement model includes an encoder, a masking estimation module, and a decoder, and step S2 includes:

[0017] Step S21, constructing an encoder using a one-dimensional convolutional layer, and mapping the mixed time domain signal into a high-dimensional feature tensor through the encoder;

[0018] Step S22, using a mask estimation module to extract context information of the high-dimensional feature tensor to generate a high-dimensional spatial mask;

[0019] Step S23, multiplying the high-dimensional spatial mask and the high-dimensional feature tensor point by point to obtain an enhanced high-dimensional representation;

[0020] Step S24, constructing a decoder using one-dimensional transposed convolution, and reconstructing the original one-dimensional time domain signal waveform from the enhanced high-dimensional representation through the decoder to obtain a reconstructed time domain signal.

[0021] In a possible implementation, the mapping rule in step S21 is: w = H (conv1D (xk )), where conv1D is a one-dimensional convolution filter, H(·) is a PReLU activation function, and x k is the small block divided into the mixed time domain signal, w is the k The transformed n-dimensional representation.

[0022] In a possible implementation, the mask estimation module includes a normalization layer, a convolutional layer, a DPRNN network, a gated linear unit, and a rectified linear unit, and step S22 includes:

[0023] Step S22-1, using a normalization layer to normalize the high-dimensional feature tensor to obtain a normalized high-dimensional feature tensor;

[0024] Step S22-2, using a convolutional layer to adjust the number of channels or feature dimensions of the normalized high-dimensional feature tensor to obtain a processed high-dimensional feature tensor;

[0025] Step S22-3: Use the DPRNN network to extract context information from the processed high-dimensional feature tensor to generate a high-dimensional spatial mask.

[0026] In a possible implementation, step S22-3 includes:

[0027] Perform sliding window segmentation on the processed high-dimensional feature tensor to obtain multiple overlapping area blocks and fill them with zeros at both ends, and then splice all the overlapping area blocks into a three-dimensional tensor;

[0028] Based on the three-dimensional tensor, the long short-term memory network is used to perform intra-block and inter-block processing respectively to extract the context information of the signal;

[0029] The overlapping area blocks of context information are overlapped and added, restored to the original dimension, and a high-dimensional mask is generated.

[0030] In a possible implementation, step S3 includes:

[0031] Initialize the target radar signal enhancement model, and input the reconstructed time domain signal into the initialized target radar signal enhancement model;

[0032] Input the mixed signal samples in the training set, based on the constructed loss function SI-SNR Loss Calculate the loss through forward propagation;

[0033] The model parameters are updated iteratively through back-propagation. When the loss on the validation set does not decrease after ten consecutive rounds, the training is stopped and the optimal model parameters are saved to obtain the trained target radar signal enhancement model.

[0034] In a possible implementation, the loss function SI-SNRLoss for: X E =X * -X T , where X * represents the network output, Indicates the true target signal.

[0035] In the second aspect, the present application proposes a target radar signal enhancement device based on DPRNN, the device comprising:

[0036] A construction module is used to construct a data set, and perform sliding window segmentation processing on the mixed signal of the training set in the data set to obtain a mixed time domain signal;

[0037] The signal enhancement module is used to build a target radar signal enhancement model based on the DPRNN network and perform signal enhancement processing on the mixed time domain signal to obtain the reconstructed time domain signal;

[0038] A training module is used to input the reconstructed time domain signal into the target radar signal enhancement model, and train the target radar signal enhancement model using the training set in the data set to obtain a trained target radar signal enhancement model;

[0039] The generation module is used to use the trained target radar signal enhancement model to process the mixed time domain signal intercepted online to obtain the enhanced target time domain signal.

[0040] In a possible implementation, a building block is used to:

[0041] Obtain multiple original mixed signals and corresponding target radar signal time domain sample pairs to construct a data set;

[0042] Divide the data set into training set, validation set and test set according to the proportion;

[0043] The mixed signal in the training set is processed by sliding window segmentation to obtain a mixed time domain signal.

[0044] The above-mentioned main scheme of the present application and its further options can be freely combined to form multiple schemes, all of which are schemes that can be adopted and claimed for protection in the present application; and in the present application, (non-conflicting options) options and other options can also be freely combined. After understanding the scheme of the present application, those skilled in the art can understand that there are multiple combinations based on the prior art and common knowledge, all of which are technical schemes to be protected by the present application, and they are not exhaustively listed here.

[0045] The present application discloses a method and device for enhancing target radar signals based on DPRNN. First, a data set is constructed, and a mixed signal of a training set in the data set is subjected to sliding window segmentation processing to obtain a mixed time domain signal. Secondly, a target radar signal enhancement model is constructed based on a DPRNN network, and the mixed time domain signal is subjected to signal enhancement processing to obtain a reconstructed time domain signal. Then, the reconstructed time domain signal is input into the target radar signal enhancement model, and the target radar signal enhancement model is trained using the training set in the data set to obtain a trained target radar signal enhancement model. Finally, the trained target radar signal enhancement model is used to process the mixed time domain signal intercepted online to obtain an enhanced target time domain signal. The constructed target radar signal enhancement model extracts the contextual features of the radar signal, solves the problem of target signal enhancement in the case of time-frequency overlap, and improves the effect of signal enhancement. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 A flow chart of a target radar signal enhancement method based on DPRNN proposed in an embodiment of the present application is shown.

[0048] Figure 2 A schematic diagram of a target radar signal enhancement model proposed in an embodiment of the present application is shown.

[0049] Figure 3 A schematic diagram of the DPRNN network proposed in an embodiment of the present application is shown.

[0050] Figure 4a A schematic diagram of a target signal proposed in an embodiment of the present application.

[0051] Figure 4b This is a schematic diagram of a mixed signal before suppression proposed in an embodiment of the present application.

[0052] Figure 4c This is a schematic diagram of an enhanced signal under low signal-to-noise ratio and time-frequency overlap conditions proposed in an embodiment of the present application.

[0053] Figure 5a This is a time-frequency diagram of the corresponding target signal proposed in the embodiment of the present application.

[0054] Figure 5b This is a time-frequency diagram of the mixed signal proposed in the embodiment of the present application.

[0055] Figure 5c This is a time-frequency diagram of the target signal enhancement result proposed in the embodiment of the present application. DETAILED DESCRIPTION

[0056] The following describes the embodiments of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. The present application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0057] Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of this application.

[0058] In the existing technology, traditional signal enhancement methods are sensitive to noise and difficult to adapt to low signal-to-noise ratio environments; existing deep learning methods lack effective utilization of signal context information, rely on the independence of signal time and frequency, and have difficulty distinguishing target signals from background signals in the case of time-frequency overlap, which easily leads to pulse loss of target signals.

[0059] Therefore, in order to solve the above technical problems, the embodiment of the present application proposes a method and device for enhancing target radar signals based on DPRNN. Starting from the intermediate frequency long sequence data, a set of convolutional codecs are used to obtain a high-dimensional optimized representation of a one-dimensional signal, and a stacked DPRNN network is used to enhance the long sequence modeling capability of the model, extract the context information of the signal, and distinguish the target signal from the background signal. Through deep learning technology, especially the application of convolution and recursive networks, the recognition and enhancement capabilities of radar target signals are significantly improved, which will be described in detail below.

[0060] Please refer to Figure 1 , Figure 1 A schematic diagram of a target radar signal enhancement method based on DPRNN proposed in an embodiment of the present application is shown, and the method includes:

[0061] Step S1, constructing a data set, performing sliding window segmentation processing on the mixed signal of the training set in the data set to obtain a mixed time domain signal.

[0062] Multiple time-domain sample pairs of original mixed signals and corresponding target radar signals were obtained, and an initial data set was constructed. The data set was divided into training set, validation set, and test set in a ratio of 6:2:2. The mixed signals in the training set were segmented by sliding window to obtain mixed time-domain signal fragments suitable for network training.

[0063] The steps of step S1 include:

[0064] Step S11: obtaining a plurality of original mixed signals and corresponding target radar signal time domain sample pairs to construct a data set.

[0065] In order to collect a sufficient number of original mixed signals and their corresponding target radar signal samples to build a high-quality data set. First, obtain the original mixed signal samples containing background noise and other interference signals. The signals can come from the actual environment or simulated generation. Then, simultaneously obtain the target radar signal samples corresponding to each mixed signal. The target signal is the pure radar signal that is expected to be extracted from the mixed signal. Each pair of original mixed signals and their corresponding target radar signals are combined into a sample pair. All sample pairs constitute the original data set. Where D is the original mixed signal sample set, is the corresponding target radar signal sample set.

[0066] Step S12: Divide the data set into a training set, a validation set, and a test set according to the proportion.

[0067] Divide the dataset into the training set DataSet according to the ratio of 6:2:2 train , validation set DataSet val And the test set DataSet test , of which the training set accounts for 60% for model training, the validation set accounts for 20% for adjusting model parameters and preventing overfitting, and the test set accounts for 20% for the final evaluation of model performance. In addition, it can ensure that the division of the data set is random to avoid bias and ensure the representativeness of each subset. By reasonably dividing the data set, we can ensure that there are enough representative samples in the model training, validation and testing stages.

[0068] Step S13: Perform sliding window segmentation processing on the mixed signal in the training set to obtain a mixed time domain signal.

[0069] Select 20 milliseconds as the sliding window size, and set an appropriate overlap ratio (e.g., 50%) as needed to ensure that there is enough information overlap between adjacent segments. Each mixed signal is segmented according to a 20ms time window to obtain multiple time domain segments, each of which is used as an independent input sample. Each segmented time domain segment is used as the input for network training to form a new training set. The mixed signal time domain signal samples in the training set are cut into segments of fixed length as the input for network training.

[0070] Step S2: construct a target radar signal enhancement model based on the DPRNN network, and perform signal enhancement processing on the mixed time domain signal to obtain a reconstructed time domain signal.

[0071] The target radar signal enhancement model based on the DPRNN network converts the mixed time domain signal into a high-dimensional feature representation through a one-dimensional convolutional layer, and then uses the DPRNN to extract context information to generate a masking matrix to suppress noise and retain the target signal. Finally, a clear time domain signal is reconstructed through a transposed convolutional layer, thereby effectively improving the quality of the radar signal.

[0072] Please refer to Figure 2 , Figure 2 A schematic diagram of a target radar signal enhancement model proposed in an embodiment of the present application is shown. The target radar signal enhancement model includes an encoder, a mask estimation module, and a decoder. The encoder is used to process an input signal to generate a mask estimation. The mask estimation output by the encoder is used to enhance the input signal to generate an enhanced signal. The decoder further processes the enhanced signal to generate a final output signal. In the mask estimation module, 1×Conv represents a convolutional layer, which performs preliminary processing on the input signal, a deep recursive neural network is used to process and enhance the signal, a gated linear unit (GLU) is used to control the information flow, and ReLU is an activation function for introducing nonlinear characteristics.

[0073] Step S2 includes:

[0074] Step S21, constructing an encoder using a one-dimensional convolutional layer, and mapping the mixed time domain signal into a high-dimensional feature tensor through the encoder.

[0075] The encoder is constructed using a one-dimensional convolutional block to map the input one-dimensional signal x into a high-dimensional feature tensor. The convolutional layer divides x into several overlapping small blocks (kernel_size indicates the filter length) and is converted to n (number of filters) dimensions In step S21, the mapping rule is: w = H (conv1D (x k )), where conv1D is a one-dimensional convolution filter, H(·) is a PReLU activation function, and x k is the small block divided into the mixed time domain signal, w is the k The transformed n-dimensional representation.

[0076] Step S22: Use the mask estimation module to extract context information of the high-dimensional feature tensor to generate a high-dimensional spatial mask.

[0077] The masking estimation module extracts the context information of the high-dimensional feature tensor through the DPRNN network and generates a high-dimensional spatial masking matrix, which is used to identify and enhance the target signal components while suppressing background noise and other interference, thereby improving the signal quality.

[0078] The mask estimation module includes a normalization layer, a convolutional layer, a DPRNN network, a gated linear unit, and a rectified linear unit. Step S22 includes:

[0079] Step S22-1, using a normalization layer to normalize the high-dimensional feature tensor to obtain a normalized high-dimensional feature tensor;

[0080] The normalization layer is used to standardize the input high-dimensional feature tensor to ensure that its numerical distribution is consistent and obtain a normalized high-dimensional feature tensor.

[0081] Step S22-2, using a convolutional layer to adjust the number of channels or feature dimensions of the normalized high-dimensional feature tensor to obtain a processed high-dimensional feature tensor;

[0082] A 1×1 convolution layer is used to adjust the number of channels or feature dimensions of the normalized high-dimensional feature tensor to make it suitable for input into the subsequent DPRNN network to obtain the processed high-dimensional feature tensor.

[0083] Step S22-3: Use the DPRNN network to extract context information from the processed high-dimensional feature tensor to generate a high-dimensional spatial mask.

[0084] The processed high-dimensional feature tensor is input into the DPRNN (Dual-Path RNN) network. In the DPRNN network, the gated linear unit (GLU) is used to retain the timing information and improve the accuracy of the output masking. The rectified linear unit (ReLU) is used to ensure the non-negativity of the output masking, and finally a high-dimensional spatial mask is generated.

[0085] The steps of step S22-3 include:

[0086] Perform sliding window segmentation on the processed high-dimensional feature tensor to obtain multiple overlapping area blocks and fill them with zeros at both ends, and then splice all the overlapping area blocks into a three-dimensional tensor;

[0087] Based on the three-dimensional tensor, the long short-term memory network is used to perform intra-block and inter-block processing respectively to extract the context information of the signal;

[0088] The overlapping area blocks of context information are overlapped and added, restored to the original dimension, and a high-dimensional mask is generated.

[0089] Figure 3 The schematic diagram of the DPRNN network proposed in the embodiment of the present application is shown. The DPRNN consists of three parts: block division, block processing and overlap addition. The input is divided into B blocks with overlapping area length P using a sliding window of size Win. The first and last blocks are filled with zeros to ensure that all blocks are of equal size. All blocks are spliced ​​to form a three-dimensional tensor Through stacked long short-term memory networks, intra-block and inter-block processing are performed respectively to obtain signal context information; the outputs are overlapped and added to restore the input dimension.

[0090] Step S23, multiplying the high-dimensional spatial mask and the high-dimensional feature tensor point by point to obtain an enhanced high-dimensional representation.

[0091] The generated high-dimensional mask M (M represents the energy proportion of the target signal in X) is multiplied point by point with the original high-dimensional feature tensor to eliminate other components and retain only the energy proportion of the target signal to obtain the enhanced high-dimensional representation of the target radar signal. This can effectively enhance the mixed time domain signal.

[0092] Step S24, constructing a decoder using one-dimensional transposed convolution, and reconstructing the original one-dimensional time domain signal waveform from the enhanced high-dimensional representation through the decoder to obtain a reconstructed time domain signal.

[0093] The decoder module is constructed using one-dimensional transposed convolution, ensuring that the transposed convolution in the decoder has the same stride and kernel size as the convolution in the encoder to ensure that the size of the reconstructed signal is consistent with the original input. The high-dimensional representation of the target radar signal obtained after masking enhancement is then input into the decoder module. The spatial dimensions of the signal are gradually restored through a series of transposed convolution operations, and finally a reconstructed time domain waveform is generated. After being processed by the decoder, the reconstructed time domain signal is output. The signal is as close to the original input signal as possible in the time domain, but noise and other interference have been effectively suppressed, and the target signal has been significantly enhanced.

[0094] Step S3: input the reconstructed time domain signal into the target radar signal enhancement model, and use the training set in the data set to train the target radar signal enhancement model to obtain a trained target radar signal enhancement model.

[0095] The reconstructed time domain signal is input into the target radar signal enhancement model, and the model is trained using the training set in the data set to optimize the model parameters. During the training process, the model learns how to extract and enhance the target radar signal from the mixed signal, and finally obtains a fully trained and performance-optimized target radar signal enhancement model.

[0096] The steps of step S3 include:

[0097] Initialize the target radar signal enhancement model, and input the reconstructed time domain signal into the initialized target radar signal enhancement model;

[0098] Input the mixed signal samples in the training set, based on the constructed loss function SI-SNR Loss Calculate the loss through forward propagation;

[0099] The model parameters are updated iteratively through back-propagation. When the loss on the validation set does not decrease after ten consecutive rounds, the training is stopped and the optimal model parameters are saved to obtain the trained target radar signal enhancement model.

[0100] The parameters of the target radar signal enhancement model are randomly initialized, and the reconstructed time domain signal is input into the initialized target radar signal enhancement model as the initial input data. The mixed signal samples in the training set are used as the input of the model. Each sample contains background noise and other interference components, as well as the corresponding target radar signal. Through the forward propagation process of the model, the difference between the predicted output and the real target signal is calculated, and the constructed loss function is used to measure the difference.

[0101] According to the calculated loss value, the gradient is calculated by the back propagation algorithm, and the model parameters are iteratively updated using the optimization algorithm to minimize the loss function. During the training process, the model performance is regularly evaluated on the validation set. When the loss on the validation set no longer decreases for ten consecutive rounds, the "early stopping" mechanism is triggered to stop training and save the current optimal model parameters. Finally, the best model parameters obtained after training and optimization are saved, thereby obtaining a well-trained model that can effectively extract and enhance the target radar signal.

[0102] Loss function SI-SNR Loss for: X E =X * -X T , where X * represents the network output, Indicates the true target signal.

[0103] Step S4: Use the trained target radar signal enhancement model to process the mixed time domain signal intercepted online to obtain an enhanced target time domain signal.

[0104] In order to segment the mixed time domain signal intercepted online into small segments suitable for model input, the mixed time domain signal intercepted in an online pipeline mode is segmented into segments of 20 milliseconds (ms) to ensure that each segment meets the model input requirements. The segmented 20ms mixed signal segments are used as input and sent to the trained and saved target radar signal enhancement model. The model outputs the enhanced target time domain signal segments through forward propagation calculation. Finally, the output results of all segments are spliced ​​to restore the complete enhanced time domain signal.

[0105] The trained target radar signal enhancement model is used to process the mixed time domain signal intercepted in real time, and finally the enhanced target time domain signal is output, thereby improving the signal quality and resolvability.

[0106] In a possible embodiment, a target signal enhancement scenario is constructed, in which there is 1 target radar and 55 background radars. Table 1 shows the parameters of the target radar and background radar transmission signals that are randomly set in range:

[0107] Table 1

[0108]

[0109] For the constructed scenario simulation data, 2000 samples are generated to build a data set for training.

[0110] Generate a corresponding number of target signals and background signals, randomly superimpose them in the time domain, add Gaussian white noise according to the signal-to-noise ratio of -15 to 20 dB, generate mixed time domain signal samples and save the corresponding target signal samples, generate 2000 such sample pairs, and construct a data set for training Where D is the original mixed signal sample set; is the corresponding target signal sample set.

[0111] Divide the dataset into training set dataset in a ratio of 6:2:2 train , validation set DataSet val And the test set DataSet test , whose sizes are 1200, 400, 400 respectively.

[0112] The mixed signal time domain signal samples in the training set are processed by sliding window segmentation and cut into 20ms long time domain segments as the input of network training.

[0113] The target signal enhancement model is implemented based on Pytorch, including encoder, mask estimation module and decoder.

[0114] Use one-dimensional convolution blocks to build the encoder and divide x into several overlapping small blocks and will be converted to n-dimensional representation w=H(conv1D(x k )). In the formula, conv1D is a one-dimensional convolution filtering process; kernel_size represents the filter length, and in this example, kernel_size = 2; n represents the number of filters, and in this example, n = 32; H(·) is the PReLU activation function.

[0115] The mask estimation module is designed based on the DPRNN network. The high-dimensional tensor X is input to the DPRNN network through a normalization layer and a 1×1 convolution layer; a gated linear unit (GLU) is used to retain the timing information and improve the accuracy of the output mask; a rectified linear unit ReLU is used to ensure the non-negativity of the output mask. Figure 3 The DPRNN network of the present method shown in the figure consists of three parts: block division, block processing and overlap addition: The input is divided into B blocks with overlapping area length P using a sliding window of size Win; in this example, Win = 400; B = 5002; P = 200. The first and last blocks are padded with zeros to ensure that all blocks are of equal size; all blocks are concatenated to form a three-dimensional tensor Through the stacked long short-term memory network, intra-block and inter-block processing are performed respectively to obtain signal context information. The output is overlapped and added to restore the input dimension. The mask estimation module generates a high-dimensional mask M; M represents the energy proportion of the target signal in X, which is multiplied point by point with X to eliminate other components and obtain the enhanced high-dimensional representation of the target signal

[0116] The decoder module is constructed using one-dimensional transposed convolution to reconstruct the time domain waveform from the target high-dimensional representation obtained after masking enhancement:

[0117] Then calculate the loss function, and calculate the loss function SI-SNR based on the network output and the target signal set in the data set Loss , the model parameters are iteratively updated by minimizing the loss function.

[0118] For the target signal enhancement model, the mixed signal samples in the training set are used as the model input, the target signal corresponding to the mixed sample is used as the target, and the SI-SNR Loss The model is trained as the loss function. Adam is selected as the optimizer and the weight is set to 1×10 -4 ; The initial learning rate is set to 1×10 -3 ; Set the maximum number of iterations to 100. In the training phase, first initialize the model parameters, input the mixed signal x, and calculate the SI-SNR through forward propagation Loss The model parameters are updated iteratively through back propagation. When the validation set loss does not decrease for five consecutive rounds, the learning rate is adjusted to half of the current learning rate. The "early stopping" training strategy is adopted. When the loss on the validation set does not decrease for ten consecutive rounds, the training is stopped and the model parameters are saved.

[0119] For the mixed time domain signal input, it is divided into segments of 20ms in length. For the mixed signal segmented samples x, it is used as the saved network model input to output the enhanced target signal.

[0120] For the mixed signal in the simulation scene, the target signal enhancement model is used to suppress the background signal and noise, and then Figure 4c Schematic diagram of the enhanced signal under low signal-to-noise ratio and time-frequency overlap, where Figure 4a 4a is a schematic diagram of the target signal, and 4b is a schematic diagram of the mixed signal before suppression.

[0121] Figure 5a is the time-frequency diagram of the corresponding target signal, Figure 5b is the time-frequency diagram of the mixed signal, Figure 5c It is a time-frequency diagram of the target signal enhancement result. It can be seen that the present application can achieve target signal enhancement under low signal-to-noise ratio and time-frequency overlap conditions. Table 2 compares the effects of the signal enhancement method proposed in this paper and other signal enhancement methods:

[0122] Table 2

[0123] SI-SNR(dB) This article model 18.66 SSA-ICA 8.06 EMD-ICA -5.37

[0124] It can be seen that the method proposed in this application has achieved a significant improvement in the signal-to-noise ratio, and the signal enhancement performance has achieved significant advantages over other methods.

[0125] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0126] First, the present application uses a one-dimensional convolutional codec instead of time-frequency transform, which can more effectively extract and process time domain features, thereby significantly improving the effect of signal enhancement in a low signal-to-noise ratio environment.

[0127] Second, by introducing a dual-path structure, local and global contextual information can be processed simultaneously, that is, one path focuses on the extraction of short-term local features, and the other path focuses on the learning of long-term dependencies, enabling the model to better separate target signals and background noise in complex time-frequency overlapping situations, thereby achieving more accurate target signal enhancement.

[0128] Third, it performs well under low signal-to-noise ratio and time-frequency overlapping conditions, making up for the shortcomings of existing signal enhancement methods in these complex scenarios.

[0129] Fourth, the one-dimensional convolutional codec and dual-path recurrent network structure adopted are more compact and efficient, which not only reduces the demand for computing resources but also speeds up the inference speed, making it suitable for real-time signal processing applications.

[0130] A possible implementation of a target radar signal enhancement device based on DPRNN is given below, which is used to execute each execution step and corresponding technical effect of the target radar signal enhancement method based on DPRNN shown in the above embodiment and possible implementation. The device includes:

[0131] A construction module is used to construct a data set, and perform sliding window segmentation processing on the mixed signal of the training set in the data set to obtain a mixed time domain signal;

[0132] The signal enhancement module is used to build a target radar signal enhancement model based on the DPRNN network and perform signal enhancement processing on the mixed time domain signal to obtain the reconstructed time domain signal;

[0133] A training module is used to input the reconstructed time domain signal into the target radar signal enhancement model, and train the target radar signal enhancement model using the training set in the data set to obtain a trained target radar signal enhancement model;

[0134] The generation module is used to use the trained target radar signal enhancement model to process the mixed time domain signal intercepted online to obtain the enhanced target time domain signal.

[0135] In a possible embodiment, a building block is used to:

[0136] Obtain multiple original mixed signals and corresponding target radar signal time domain sample pairs to construct a data set;

[0137] Divide the data set into training set, validation set and test set according to the proportion;

[0138] The mixed signal in the training set is processed by sliding window segmentation to obtain a mixed time domain signal.

[0139] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A target radar signal enhancement method based on DPRNN, characterized in that: The method comprises: Step S1, constructing a data set, performing sliding window segmentation processing on the mixed signal of the training set in the data set to obtain a mixed time domain signal; Step S2: construct a target radar signal enhancement model based on the DPRNN network, and perform signal enhancement processing on the mixed time domain signal to obtain a reconstructed time domain signal; Step S3, inputting the reconstructed time domain signal into the target radar signal enhancement model, and training the target radar signal enhancement model using the training set in the data set to obtain a trained target radar signal enhancement model; Step S4: Use the trained target radar signal enhancement model to process the mixed time domain signal intercepted online to obtain an enhanced target time domain signal.

2. The target radar signal enhancement method according to claim 1, characterized in that: The steps of step S1 include: Step S11, obtaining a plurality of original mixed signals and corresponding target radar signal time domain sample pairs, and constructing a data set; Step S12: Divide the data set into a training set, a validation set, and a test set according to the ratio; Step S13: Perform sliding window segmentation processing on the mixed signal in the training set to obtain a mixed time domain signal.

3. The target radar signal enhancement method according to claim 1, characterized in that: The target radar signal enhancement model includes an encoder, a masking estimation module and a decoder. Step S2 includes: Step S21, constructing an encoder using a one-dimensional convolutional layer, and mapping the mixed time domain signal into a high-dimensional feature tensor through the encoder; Step S22, using a mask estimation module to extract context information of the high-dimensional feature tensor to generate a high-dimensional spatial mask; Step S23, multiplying the high-dimensional spatial mask and the high-dimensional feature tensor point by point to obtain an enhanced high-dimensional representation; Step S24, constructing a decoder using one-dimensional transposed convolution, and reconstructing the original one-dimensional time domain signal waveform from the enhanced high-dimensional representation through the decoder to obtain a reconstructed time domain signal.

4. The target radar signal enhancement method as claimed in claim 3, characterized in that: In step S21, the mapping rule is: w = H (conv1D (x k )), where conv1D is a one-dimensional convolution filter, H(·) is a PReLU activation function, and x k is the small block divided into the mixed time domain signal, w is the k The transformed n-dimensional representation.

5. The target radar signal enhancement method as claimed in claim 3, characterized in that: The mask estimation module includes a normalization layer, a convolutional layer, a DPRNN network, a gated linear unit, and a rectified linear unit. Step S22 includes: Step S22-1, using a normalization layer to normalize the high-dimensional feature tensor to obtain a normalized high-dimensional feature tensor; Step S22-2, using a convolutional layer to adjust the number of channels or feature dimensions of the normalized high-dimensional feature tensor to obtain a processed high-dimensional feature tensor; Step S22-3: Use the DPRNN network to extract context information from the processed high-dimensional feature tensor to generate a high-dimensional spatial mask.

6. The target radar signal enhancement method according to claim 5, characterized in that: The steps of step S22-3 include: Perform sliding window segmentation on the processed high-dimensional feature tensor to obtain multiple overlapping area blocks and fill them with zeros at both ends, and then splice all the overlapping area blocks into a three-dimensional tensor; Based on the three-dimensional tensor, the long short-term memory network is used to perform intra-block and inter-block processing respectively to extract the context information of the signal; The overlapping area blocks of context information are overlapped and added, restored to the original dimension, and a high-dimensional mask is generated.

7. The target radar signal enhancement method according to claim 5, characterized in that: The steps of step S3 include: Initialize the target radar signal enhancement model, and input the reconstructed time domain signal into the initialized target radar signal enhancement model; Then input the mixed signal samples in the training set, based on the constructed loss function SI-SNR Loss Calculate the loss through forward propagation; The model parameters are updated iteratively through back-propagation. When the loss on the validation set does not decrease after ten consecutive rounds, the training is stopped and the optimal model parameters are saved to obtain the trained target radar signal enhancement model.

8. The target radar signal enhancement method according to claim 7, characterized in that: Loss function SI-SNR Loss for: X E =X * -X T , where X * represents the network output, Indicates the true target signal.

9. A target radar signal enhancement device based on DPRNN, characterized in that: The device comprises: A construction module is used to construct a data set, and perform sliding window segmentation processing on the mixed signal of the training set in the data set to obtain a mixed time domain signal; The signal enhancement module is used to build a target radar signal enhancement model based on the DPRNN network and perform signal enhancement processing on the mixed time domain signal to obtain the reconstructed time domain signal; A training module is used to input the reconstructed time domain signal into the target radar signal enhancement model, and train the target radar signal enhancement model using the training set in the data set to obtain a trained target radar signal enhancement model; The generation module is used to use the trained target radar signal enhancement model to process the mixed time domain signal intercepted online to obtain the enhanced target time domain signal.

10. The target radar signal enhancement device according to claim 9, characterized in that: Building blocks for: Obtain multiple original mixed signals and corresponding target radar signal time domain sample pairs to construct a data set; Divide the data set into training set, validation set and test set according to the proportion; The mixed signal in the training set is processed by sliding window segmentation to obtain a mixed time domain signal.

Images