Radar target noise reduction identification method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of signal processing, and provides a radar target noise reduction identification method and device, electronic equipment and a storage medium, and the method comprises the steps: inputting a source radar signal into a frequency domain filter, and obtaining a source radar signal without out-of-band clutter; inputting the out-of-band clutter filtered source radar signal into an in-band clutter noise reduction model to obtain an in-band clutter filtered target radar signal output by the in-band clutter noise reduction model; and performing target identification on a to-be-identified target based on the target radar signal. Since the in-band clutter noise reduction model is obtained by training based on the out-of-band clutter filtered sample radar signal and the corresponding tag radar signal, the mapping from the out-of-band clutter filtered sample radar signal to the corresponding tag radar signal is realized in the training process. Therefore, in-band clutters can be further filtered out of the source radar signal after the out-of-band clutters are filtered out during model application reasoning, and effective noise reduction of the source radar signal is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and in particular, to a method, device, electronic device and storage medium for radar target noise reduction and recognition. Background Art

[0002] Radar target recognition is one of the important applications in the field of radar signal processing. Among them, the accuracy of the received radar signal plays a key role in the final recognition effect. However, in real scenarios, both the radar receiver and some interference factors in the natural environment will generate clutter interference to the radar signal.

[0003] At present, the method of frequency domain filter is one of the mainstream methods to realize clutter suppression of radar signals, that is, to filter out the clutter of radar signals through a frequency domain filter to achieve noise reduction of radar signals. The core idea of the frequency domain filter is: since the frequency of the radar signal is usually distributed in a certain fixed frequency band, while the frequency of the clutter is more dispersed in the frequency domain interval, therefore, a frequency domain filter with a flat passband and high stopband characteristics can filter out the clutter outside the frequency band. However, the in-band clutter overlaps with the radar signal in the frequency domain, and the frequency domain filter cannot effectively filter out the in-band clutter and cannot effectively reduce the noise of the radar signal. Summary of the Invention

[0004] The present invention provides a method, device, electronic device and storage medium for radar target noise reduction and recognition, which are used to solve the problem that the frequency domain filter in the prior art cannot effectively filter out the in-band clutter and cannot effectively reduce the noise of the radar signal.

[0005] The present invention provides a method for radar target noise reduction and recognition, including the following steps: Input the source radar signal into a frequency domain filter to obtain a source radar signal with out-of-band clutter filtered out; Input the source radar signal with out-of-band clutter filtered out into an in-band clutter noise reduction model to obtain a target radar signal with in-band clutter filtered out output by the in-band clutter noise reduction model; Perform target recognition on the target to be recognized based on the target radar signal; Wherein, the in-band clutter noise reduction model is trained based on the sample radar signal with out-of-band clutter filtered out and the labeled radar signal with in-band clutter filtered out corresponding to the sample radar signal.

[0006] According to the method for radar target noise reduction and recognition provided by the present invention, the frequency domain filter includes: a plurality of sub-band frequency domain filters, different sub-band frequency domain filters correspond to different frequency bands, and the frequency bands of all sub-band frequency domain filters cover the frequency band range of the source radar signal; Inputting the source radar signal into a frequency domain filter to obtain a source radar signal with out-of-band clutter filtered out includes: Input the source radar signal into each of the sub-band frequency domain filters respectively to obtain the filtered sub-band signals; Merge the sub-band signals to obtain the source radar signal with out-of-band clutter filtered out.

[0007] According to a radar target noise reduction and recognition method provided by the present invention, the multiple sub-band frequency domain filters are generated in the following manner: Perform sub-band decomposition on the prototype filter using the discrete Fourier transform with linear phase to obtain the sub-band frequency domain filters with different frequency bands after decomposition.

[0008] According to a radar target noise reduction and recognition method provided by the present invention, the prototype filter is a causal linear phase low-pass filter of order 2L, where L is an integer greater than 0, and the impulse response of the prototype filter satisfies: h ( n ) = h (2 L - n ), n = 0, 1, …, 2 L ; h ( n ) is the time-domain representation of the prototype filter.

[0009] According to a radar target noise reduction and recognition method provided by the present invention, the in-band clutter noise reduction model includes: an encoder and a decoder; The encoder includes: a first convolution module, a second convolution module, a third convolution module, a flattening module, and a first fully connected layer. The first convolution module is connected to the second convolution module, the second convolution module is connected to the third convolution module, the third convolution module is connected to the flattening module, and the flattening module is connected to the first fully connected layer; The decoder includes: a first transposed convolution module, a second transposed convolution module, a third transposed convolution module, an inverse flattening module, and a second fully connected layer. The first fully connected layer is connected to the second fully connected layer, the second fully connected layer is connected to the inverse flattening module, the inverse flattening module is connected to the third transposed convolution module, the third transposed convolution module is connected to the second transposed convolution module, and the second transposed convolution module is connected to the first transposed convolution module; The i-th convolution module and the i-th transposed convolution module have the same number of channels and convolution kernels of equal size, i = 1, 2, 3; The first convolution module, the second convolution module, and the third convolution module are used to perform convolution on the source radar signal with out-of-band clutter filtered out in sequence to obtain the first feature map after three convolutions; The flattening module flattens the first feature map into a first one-dimensional vector; The first fully connected layer is used to reduce the dimension of the first one-dimensional vector into a second one-dimensional vector and output the second one-dimensional vector to the second fully connected layer. The second fully connected layer is used to increase the dimension of the second one-dimensional vector into a third one-dimensional vector and output the third one-dimensional vector to the inverse flattening module; The inverse flattening module is used to inverse flatten the third one-dimensional vector into a second feature map and output the second feature map to the third transposed convolutional module. The third one-dimensional vector has the same dimension as the first one-dimensional vector; The third transposed convolutional module, the second transposed convolutional module, and the first transposed convolutional module are used to perform transposed convolution on the second feature map in sequence to obtain the target radar signal after three times of transposed convolution.

[0010] According to a radar target noise reduction and recognition method provided by the present invention, the training method of the in-band clutter noise reduction model is as follows: Obtain a sample radar signal and a labeled radar signal; Input the sample radar signal into a frequency domain filter to obtain a sample radar signal with out-of-band clutter filtered out; Input the sample radar signal with out-of-band clutter filtered out into the in-band clutter noise reduction model to obtain a radar prediction signal output by the in-band clutter noise reduction model. Substitute the radar prediction signal and the labeled radar signal into a loss function. When the loss function converges, the model training is completed.

[0011] According to a radar target noise reduction and recognition method provided by the present invention, obtaining a sample radar signal and a labeled radar signal includes: Based on the simulation of the first simulation model, obtain the labeled radar signal. Based on the simulation of the second simulation model, obtain the clutter signal; Superimpose the labeled radar signal and the clutter signal to obtain the sample radar signal.

[0012] The present invention also provides a radar target noise reduction and recognition device, including the following modules: An out-of-band clutter filtering unit, which is used to input the source radar signal into a frequency domain filter to obtain a source radar signal with out-of-band clutter filtered out; An in-band clutter filtering unit, which is used to input the source radar signal with out-of-band clutter filtered out into the in-band clutter noise reduction model to obtain a target radar signal with in-band clutter filtered out output by the in-band clutter noise reduction model; A target recognition unit, which is used to perform target recognition on the target to be recognized based on the target radar signal; Wherein, the in-band clutter noise reduction model is trained based on the sample radar signal with out-of-band clutter filtered out and the labeled radar signal with in-band clutter filtered out corresponding to the sample radar signal.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and running on the processor. When the processor executes the program, the radar target noise reduction and recognition method as described in any one of the above is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the radar target noise reduction and recognition method as described in any one of the above is implemented.

[0015] The radar target noise reduction and recognition method, device, electronic device, and storage medium provided by the present invention first input the source radar signal into a frequency domain filter to filter out out-of-band clutter, and then input the source radar signal after filtering out the out-of-band clutter into an in-band clutter noise reduction model. Since the in-band clutter noise reduction model is trained based on the sample radar signal after filtering out the out-of-band clutter and the labeled radar signal corresponding to the sample radar signal after filtering out the in-band clutter, a mapping from the sample radar signal after filtering out the out-of-band clutter to the corresponding labeled radar signal is realized during the training process. Therefore, when the model is applied for inference, the in-band clutter in the source radar signal after filtering out the out-of-band clutter can be further filtered out, thereby effectively reducing the noise of the source radar signal, improving the signal-to-clutter ratio of the source radar signal, and then using the target radar signal to perform target recognition on the target to be recognized, which can improve the accuracy of target recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 is a schematic flowchart of the radar target noise reduction and recognition method provided by the present invention.

[0018] Figure 2 is a schematic structural diagram of the frequency domain filter in the radar target noise reduction and recognition method provided by the present invention.

[0019] Figure 3 is a schematic structural diagram of the in-band clutter noise reduction model in the radar target noise reduction and recognition method provided by the present invention.

[0020] Figure 4 is a schematic diagram of the effect after the in-band clutter noise reduction model reduces the noise of the sample radar signal in the radar target noise reduction and recognition method provided by the present invention.

[0021] Figure 5It is one of the comparison charts of the recognition accuracies of radar signals before and after noise reduction using the radar target noise reduction and recognition method provided by the present invention for different radar target recognition models.

[0022] Figure 6 It is the second of the comparison charts of the recognition accuracies of radar signals before and after noise reduction using the radar target noise reduction and recognition method provided by the present invention for different radar target recognition models.

[0023] Figure 7 It is the third of the comparison charts of the recognition accuracies of radar signals before and after noise reduction using the radar target noise reduction and recognition method provided by the present invention for different radar target recognition models.

[0024] Figure 8 It is the fourth of the comparison charts of the recognition accuracies of radar signals before and after noise reduction using the radar target noise reduction and recognition method provided by the present invention for different radar target recognition models.

[0025] Figure 9 It is a schematic structural diagram of the radar target noise reduction and recognition device provided by the present invention.

[0026] Figure 10 It is a schematic structural diagram of the electronic device provided by the present invention. Detailed implementation manners

[0027] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts shall fall within the protection scope of the present invention.

[0028] The radar target noise reduction and recognition method provided by the embodiment of the present invention, as Figure 1 shown, includes the following steps S110 to step S130.

[0029] Step S110: Input the source radar signal into a frequency domain filter to obtain a source radar signal with out-of-band clutter filtered out. Herein, the source radar signal refers to the received original radar signal, which contains clutter signals generated due to receiver and external environmental factors, specifically including out-of-band clutter and in-band clutter.

[0030] In this step, the source radar signal containing clutter signals is input into the frequency domain filter, and the frequency domain filter filters out the out-of-band clutter from the source radar signal to obtain a source radar signal with out-of-band clutter filtered out. In this embodiment, the frequency domain filter can be any filter capable of filtering out out-of-band clutter, and no specific limitation is imposed on the frequency domain filter.

[0031] Step S120: Input the source radar signal with out-of-band clutter filtered into the in-band clutter reduction model to obtain the target radar signal with in-band clutter filtered, which is output by the in-band clutter reduction model. Wherein, the in-band clutter reduction model is trained based on the sample radar signal with out-of-band clutter filtered and the labeled radar signal with in-band clutter filtered corresponding to the sample radar signal.

[0032] Step S130: Perform target recognition on the target to be recognized based on the target radar signal. Since the target radar signal filters out both out-of-band clutter and in-band clutter and there is no clutter interference, using the target radar signal to perform target recognition on the target to be recognized can improve the accuracy of target recognition.

[0033] Specifically, the labeled radar signal can be understood as the ideal signal obtained by filtering out-of-band clutter and in-band clutter from the sample radar signal. The in-band clutter reduction model is used to learn the feature differences between the sample radar signal with out-of-band clutter filtered and the labeled radar signal during the training stage, and realize the mapping from the sample radar signal with out-of-band clutter filtered to the corresponding labeled radar signal. Thus, during the inference stage, the in-band clutter reduction model can infer the corresponding target radar signal with in-band clutter removed according to the source radar signal with out-of-band clutter filtered received in real time. Using the target radar signal for subsequent radar target recognition can more accurately identify the target.

[0034] Moreover, in this embodiment, the input sample during the training of the in-band clutter reduction model is the sample radar signal with out-of-band clutter filtered, which avoids the interference of out-of-band clutter on signal feature learning during training, enabling the trained in-band clutter reduction model to better filter out in-band clutter.

[0035] For the radar target noise reduction and recognition method of this embodiment, first input the source radar signal into the frequency domain filter to filter out out-of-band clutter, and then input the source radar signal with out-of-band clutter filtered into the in-band clutter reduction model. Since the in-band clutter reduction model is trained based on the sample radar signal with out-of-band clutter filtered and the labeled radar signal with in-band clutter filtered corresponding to the sample radar signal, and realizes the mapping from the sample radar signal with out-of-band clutter filtered to the corresponding labeled radar signal during the training process, it can further filter out in-band clutter from the source radar signal with out-of-band clutter filtered during the model application and inference, thereby effectively reducing the noise of the source radar signal and improving the signal-to-clutter ratio (SCR, the ratio of signal power to clutter power) of the source radar signal. Furthermore, using the target radar signal to perform target recognition on the target to be recognized can improve the accuracy of target recognition.

[0036] For source radar signals with strong clutter, i.e., source radar signals with a low signal-to-clutter ratio (e.g., less than or equal to 0 dB), for example, in the field of sea surface target recognition, sea surface clutter has strong non-stationarity (dynamic changes) and spatial heterogeneity, and out-of-band clutter often distributes in different frequency bands of the radar signal. Existing frequency domain filters are usually limited to fixed frequency domain filtering and are difficult to take into account the dynamic changes of out-of-band clutter in different frequency bands, resulting in poor out-of-band clutter removal effect. Especially for source radar signals with a low signal-to-clutter ratio, out-of-band clutter cannot be effectively removed.

[0037] For source radar signals with strong clutter, in order to better remove out-of-band clutter, in some embodiments, as Figure 2 shown, the frequency domain filter includes: a plurality of sub-band frequency domain filters. Different sub-band frequency domain filters correspond to different frequency bands, and the frequency bands of all sub-band frequency domain filters cover the frequency band range of the source radar signal.

[0038] Based on this, step S110 specifically includes: Input the source radar signal into each of the sub-band frequency domain filters respectively to obtain the filtered sub-band signals. For each sub-band frequency domain filter, its corresponding sub-band belongs to the passband, and other sub-bands belong to the stopband. Therefore, after the source radar signal is input into any sub-band frequency domain filter, the frequency band corresponding to the any sub-band frequency domain filter passes through the filter, and other frequency bands are filtered out to obtain the corresponding sub-band signal.

[0039] Merge the sub-band signals to obtain the source radar signal with out-of-band clutter removed. Specifically, the sub-band signals obtained after frequency domain filtering of the signal are restored to the time domain, and the sub-band signals are merged by superposition in the time domain.

[0040] In this embodiment, the frequency domain filter includes a plurality of sub-band frequency domain filters, which can effectively remove out-of-band clutter from source radar signals with strong clutter and a low signal-to-clutter ratio. Moreover, compared with traditional frequency domain filters, each sub-band frequency domain filter can effectively retain the components of the target radar signal that are obvious in the corresponding sub-band but not obvious in the whole frequency band, making the filtered target radar signal more accurate.

[0041] In some embodiments, the plurality of sub-band frequency domain filters are generated in the following manner: The prototype filter (preferably the prototype filter designed by the Kaiser window) is modulated by the discrete Fourier transform with linear phase to obtain each of the sub-band frequency domain filters with different frequency bands. Among them, the prototype filter is the basic filter for generating each sub-band frequency domain filter. Modulating the prototype filter by the discrete Fourier transform with linear phase means moving the center frequency of the prototype filter forward and backward by a predetermined frequency step respectively, so as to obtain a plurality of sub-band frequency domain filters. Each sub-band frequency domain filter has a different center frequency, that is, has a different frequency band (that is, also has a different passband). The process of moving the center frequency of the above prototype filter to obtain a plurality of sub-band frequency domain filters is expressed by the following formula: The frequency response function of the prototype filter is H ( ω ): (1).

[0042] Among them, h ( n ) is the time domain representation of the prototype filter, N is the order of the prototype filter, ω represents the frequency, j is the imaginary unit, representing the complex exponential in the Fourier transform. The frequency response function of each sub-band frequency domain filter is: (2).

[0043] It can be seen from the above formula that when k takes values from -K to K according to the value range of the above formula, 2K + 1 sub-band frequency domain filters are formed, represents the k-th sub-band frequency domain filter.

[0044] When the prototype filter simultaneously satisfies the causality, linear phase, and finite impulse response (FIR) conditions, then all the sub-band frequency domain filters represented by formula (2) satisfy the above conditions. In the target recognition scenario with strong clutter interference, for example: in the sea surface target recognition task scenario, it is necessary to suppress the sea surface clutter outside the band while retaining the useful information inside the band as much as possible. Therefore, the design of the prototype filter needs to satisfy a flat passband to ensure that there is only a small amplitude distortion in the target echo, and at the same time, it should have a high stopband attenuation to suppress the clutter outside each sub-band. Based on this, the specified interval of the passband of the prototype filter can be designed as: (3).

[0045] Its stopband can be designed as: (4).

[0046] In some embodiments, the prototype filter is a causal linear-phase low-pass filter of order 2L, where L is an integer greater than 0, U represents taking the union, and the finite impulse response (FIR) of the prototype filter satisfies: h ( n ) = h (2 L - n ), n = 0, 1, …, 2 L (5).

[0047] h ( n ) is the time-domain representation of the prototype filter. The linear phase can ensure that the waveform is not distorted, and the causality has physical realizability. The prototype filter that satisfies both linear phase and causality has the symmetric property shown in formula (5).

[0048] Exemplarily, the stopband attenuation is -40 dB, the number of filters is set to M = 2 K + 1, the filter order is N = 2 L where K = 10, N = 96. Through frequency modulation, a subband frequency-domain filter bank formed by subband frequency-domain filters can be obtained (that is, a subband frequency-domain filter bank composed of all subband frequency-domain filters): (6).

[0049] In this embodiment, the prototype filter is a causal linear-phase low-pass filter of order 2L. The linear phase can ensure that the phase structure of the high-resolution range profile (HRRP) corresponding to the final target radar signal remains unchanged, and the phase delays generated by target radar signal components of different frequencies are the same, without causing signal distortion. It should be noted that: HRRP is a commonly used data form in the radar automatic target recognition scenario, and the radar signal can be converted into the corresponding HRRP data.

[0050] In some embodiments, as Figure 3 shown, the in-band clutter noise reduction model includes: an encoder and a decoder.

[0051] The encoder includes: a first convolutional module, a second convolutional module, a third convolutional module, a flattening module, and a first fully connected layer. The first convolutional module is connected to the second convolutional module, the second convolutional module is connected to the third convolutional module, the third convolutional module is connected to the flattening module, and the flattening module is connected to the first fully connected layer.

[0052] The decoder includes: a first transposed convolution module, a second transposed convolution module, a third transposed convolution module, a de-flattening module, and a second fully-connected layer. The first fully-connected layer is connected to the second fully-connected layer, the second fully-connected layer is connected to the de-flattening module, the de-flattening module is connected to the third transposed convolution module, the third transposed convolution module is connected to the second transposed convolution module, and the second transposed convolution module is connected to the first transposed convolution module; the i-th convolution module and the i-th transposed convolution module have the same number of channels and convolution kernels of equal size, where i = 1, 2, 3. That is Figure 3 in, the corresponding convolution module and transposed convolution module of each layer have the same number of channels and convolution kernels of equal size.

[0053] The first convolution module, the second convolution module, and the third convolution module are used to sequentially perform convolution on the source radar signal with out-of-band clutter filtered out to obtain a first feature map after three convolutions.

[0054] The flattening module flattens the first feature map into a first one-dimensional vector.

[0055] The first fully-connected layer is used to reduce the dimension of the first one-dimensional vector into a second one-dimensional vector and output the second one-dimensional vector to the second fully-connected layer. The second fully-connected layer is used to increase the dimension of the second one-dimensional vector into a third one-dimensional vector and output the third one-dimensional vector to the de-flattening module.

[0056] The de-flattening module is used to de-flatten the third one-dimensional vector into a second feature map and output the second feature map to the third transposed convolution module. The third one-dimensional vector has the same dimension as the first one-dimensional vector.

[0057] The third transposed convolution module, the second transposed convolution module, and the first transposed convolution module are used to sequentially perform transposed convolution on the second feature map to obtain a target radar signal after three transposed convolutions.

[0058] Exemplarily, each convolution module includes a convolutional layer, a ReLU layer, and a BN layer (Batch Normalization). Each transposed convolution module includes a transposed convolutional layer, a ReLU layer, and a BN layer. Since the source radar signal is a one-dimensional vector, for example: in the scenario of automatic target recognition, the source radar signal is represented by one-dimensional HRRP. Therefore, the convolution kernels in each convolution module and transposed convolution module are all one-dimensional convolution kernels. The number of channels and the size of the convolution kernels in each convolution module and transposed convolution module, as well as the number of channels in each fully-connected layer, can be set according to the actual situation. For example: Figure 3 in, the convolution kernels in each convolution module and transposed convolution module are all 3×1 convolution kernels. Among them, the first convolution module and the second convolution module have 64 channels, and the third convolution module has 128 channels. Correspondingly, the first transposed convolution module and the second transposed convolution module have 64 channels, and the third transposed convolution module has 128 channels.

[0059] As Figure 3 shown, after three convolutions through three convolution modules, a first feature map of 128×25 is obtained. The flattening module flattens the 128×25 first feature map to obtain a first one-dimensional vector of 1×3200. The first fully connected layer reduces the dimension of the 1×3200 first one-dimensional vector to a second one-dimensional vector of 1×256 and outputs it to the second fully connected layer. The parameter matrix of the first fully connected layer is 3200×256. The second fully connected layer raises the dimension of the 1×256 second one-dimensional vector to a third one-dimensional vector of 1×3200. The parameter matrix of the second fully connected layer is 256×3200, and then the 1×3200 third one-dimensional vector is output to the inverse flattening module. The inverse flattening module inverse-flattens the 1×3200 third one-dimensional vector into a second feature map of 128×25 and outputs the second feature map to the third transposed convolution module. The second feature map of 128×25 undergoes transposed convolution through three transposed convolution modules to obtain the target radar signal.

[0060] As Figure 3 shown, during the encoding process, the input data is the 1×200 HRRP data corresponding to the source radar signal. The feature map obtained through the 3×1 convolution kernel with 64 channels of the first convolution module is 64×100. 100 is calculated through the convolution operation formula when the stride is set to 2 and the padding is 1. The same applies to the subsequent convolution modules. With such settings, the feature length is divided by 2 each time of convolution. Since this model trains the mapping from the training source radar signal to the label radar signal during training, therefore, during decoding, the feature map obtained by the encoder is reconstructed into the 1×200 HRRP data of the target radar signal after filtering out in-band clutter through each transposed convolution module.

[0061] Since HRRP data has translational sensitivity, in the in-band clutter reduction model of this embodiment, the main modules for feature extraction are each convolution module, and the main modules for feature reconstruction are each transposed convolution module. Each convolution module and transposed convolution module have translational invariance of features, effectively solving the problem of translational sensitivity of HRRP data, thereby being able to better filter out in-band clutter and obtain more accurate HRRP data corresponding to the target radar signal. As Figure 4 shown, it shows the recovery situation of a certain HRRP data under the condition that the signal-to-clutter ratio is -8dB. It can be seen that Figure 3 the in-band clutter reduction model can effectively recover the peak amplitude and position distribution characteristics submerged by clutter, enabling the reconstructed data to retain as much useful information as possible while reducing the influence of in-band clutter.

[0062] In some embodiments, the training method of the in-band clutter reduction model is as follows: Obtain a sample radar signal and a label radar signal. The sample radar signal is a radar signal containing out-of-band clutter and in-band clutter.

[0063] Input the sample radar signal into the frequency domain filter to obtain the sample radar signal with out-of-band clutter removed.

[0064] Input the sample radar signal with out-of-band clutter removed into the in-band clutter noise reduction model to obtain the radar prediction signal output by the in-band clutter noise reduction model. Substitute the radar prediction signal and the labeled radar signal into the loss function. When the loss function converges, the model training is completed.

[0065] Specifically, the loss function can be the Mean Squared Error (MSE) loss function. By calculating the mean squared error between the radar prediction signal and the labeled radar signal, perform gradient descent and backpropagation on the in-band clutter noise reduction model according to the mean squared error to update the model parameters. When the mean squared error converges, the model training is completed.

[0066] Since the process of manually creating the sample radar signal and calibrating the labeled radar signal from the actual radar signal is relatively complex, in some embodiments, obtaining the sample radar signal and the labeled radar signal includes: Based on the simulation of the first simulation model, obtain the labeled radar signal. Based on the simulation of the second simulation model, obtain the clutter signal.

[0067] Superimpose the labeled radar signal and the clutter signal to obtain the sample radar signal.

[0068] Specifically, the first simulation model and the second simulation module are determined according to the actual application scenario. For example: for the application scenario of sea target recognition, the first simulation model can be a combined model of a ship CAD model and a FEKO electromagnetic simulation model. Obtain the ideal radar signal, i.e., the labeled radar signal, through the ship CAD model and the FEKO electromagnetic simulation model. The second simulation model can be a Pareto distribution sea clutter model. The Pareto distribution can use two parameters to achieve a better fit to the measured data. The Pareto distribution P ( z ) The formula is as follows: (7).

[0069] Where θ represents the shape parameter of the generalized Pareto distribution, λ represents the scale parameter of the generalized Pareto distribution, z represents the random variable of the clutter signal amplitude.

[0070] According to the measured data of sea clutter in a certain sea area, the two parameters are respectively selected θ = -0.0741, λ= 0.0145, so that the sea clutter simulation signal obeying this distribution can be generated through the above formula (7). Finally, the sea clutter signal and the tagged radar signal are superimposed to obtain the final sample radar signal.

[0071] In this embodiment, the tagged radar signal is obtained by simulating through the first simulation model, and the clutter signal is obtained by simulating through the second simulation model. Then the signals obtained by the two models are superimposed into a sample radar signal, so that the sample radar signal and the tagged radar signal are obtained relatively simply. Moreover, the sample radar signal and its tagged radar signal can correspond better, and can better train the in-band clutter noise reduction model.

[0072] Such as Figure 5 、 Figure 6 、 Figure 7 and Figure 8 shown, which shows the comparison chart of the accuracy of target recognition when the radar signals before and after noise reduction in the radar target noise reduction and recognition method using the above embodiment are respectively used in the scenario of radar target recognition. Figures 5 - 8 Among them, the target recognition modes can be: Model 1 to Model 7, which are Adaptive Auxiliary Learning Net (AALNET), 1 Dimension Convolutional Neural Networks (1dCNN), Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM), Random Forest (RF), Short-Time Fourier Transform Convolutional Neural Network (STFT-CNN), and Naive Bayes (NB).

[0073] Figure 5 is the bar chart of the accuracy of target recognition when the source radar signal with a signal-to-clutter ratio of -10 dB is used for the above 7 models before and after noise reduction, Figure 6 is the bar chart of the accuracy of target recognition when the source radar signal with a signal-to-clutter ratio of -5 dB is used for the above 7 models before and after noise reduction, Figure 7 is the bar chart of the accuracy of target recognition when the source radar signal with a signal-to-clutter ratio of 0 dB is used for the above 7 models before and after noise reduction, Figure 8It is a bar chart of the accuracy of target recognition for the source radar signal with a signal-to-clutter ratio of 5 dB before and after noise reduction using the above 7 models. It can be seen that for the source radar signal with a lower signal-to-clutter ratio, the improvement in the accuracy of target recognition before and after noise reduction is more significant. The radar target noise reduction and recognition method of the above embodiment is particularly suitable for clutter suppression and noise reduction of the source radar signal with a low signal-to-clutter ratio.

[0074] Next, the radar target noise reduction and recognition device provided by the present invention will be described. The radar target noise reduction and recognition device described below can be mutually referred to with the radar target noise reduction and recognition method described above.

[0075] The radar target noise reduction and recognition device according to an embodiment of the present invention, as Figure 9 shown, includes: An out-of-band clutter filtering unit 910, configured to input a source radar signal into a frequency domain filter to obtain a source radar signal with out-of-band clutter filtered.

[0076] An in-band clutter filtering unit 920, configured to input the source radar signal with out-of-band clutter filtered into an in-band clutter noise reduction model to obtain a target radar signal with in-band clutter filtered output by the in-band clutter noise reduction model.

[0077] A target recognition unit 930, configured to perform target recognition on a target to be recognized based on the target radar signal.

[0078] Wherein, the in-band clutter noise reduction model is trained based on a sample radar signal with out-of-band clutter filtered and a labeled radar signal with in-band clutter filtered corresponding to the sample radar signal.

[0079] The radar target noise reduction and recognition device of this embodiment first inputs the source radar signal into a frequency domain filter to filter out-of-band clutter, and then inputs the source radar signal with out-of-band clutter filtered into the in-band clutter noise reduction model. Since the in-band clutter noise reduction model is trained based on a sample radar signal with out-of-band clutter filtered and a labeled radar signal with in-band clutter filtered corresponding to the sample radar signal, a mapping from the sample radar signal with out-of-band clutter filtered to the corresponding labeled radar signal is realized during the training process. Therefore, when the model is applied for inference, the in-band clutter of the source radar signal with out-of-band clutter filtered can be further filtered, so as to effectively reduce the noise of the source radar signal and improve the signal-to-clutter ratio of the source radar signal. Furthermore, using the target radar signal to perform target recognition on the target to be recognized can improve the accuracy of target recognition.

[0080] In some embodiments, the frequency domain filter includes: a plurality of sub-band frequency domain filters, different sub-band frequency domain filters correspond to different frequency bands, and the frequency bands of all the sub-band frequency domain filters cover the frequency band range of the source radar signal.

[0081] The out-of-band clutter filtering unit 910 is specifically configured to input the source radar signal into each of the sub-band frequency domain filters respectively to obtain the filtered sub-band signals; and combine the sub-band signals to obtain the source radar signal with out-of-band clutter filtered out.

[0082] In some embodiments, the multiple sub-band frequency domain filters are generated in the following manner: the prototype filter is modulated by using a linear-phase discrete Fourier transform to obtain each of the sub-band frequency domain filters with different frequency bands after modulation.

[0083] In some embodiments, the prototype filter is a 2L-order causal linear-phase low-pass filter, where L is an integer greater than 0, and the impulse response of the prototype filter satisfies: h ( n ) = h (2 L - n ), n = 0, 1, …, 2 L .

[0084] h ( n ) is the time-domain representation of the prototype filter.

[0085] In some embodiments, the in-band clutter noise reduction model includes: an encoder and a decoder.

[0086] The encoder includes: a first convolution module, a second convolution module, a third convolution module, a flattening module, and a first fully-connected layer. The first convolution module is connected to the second convolution module, the second convolution module is connected to the third convolution module, the third convolution module is connected to the flattening module, and the flattening module is connected to the first fully-connected layer.

[0087] The decoder includes: a first transposed convolution module, a second transposed convolution module, a third transposed convolution module, an inverse flattening module, and a second fully-connected layer. The first fully-connected layer is connected to the second fully-connected layer, the second fully-connected layer is connected to the inverse flattening module, the inverse flattening module is connected to the third transposed convolution module, the third transposed convolution module is connected to the second transposed convolution module, and the second transposed convolution module is connected to the first transposed convolution module; the i-th convolution module and the i-th transposed convolution module have the same number of channels and convolution kernels of equal size, i = 1, 2, 3.

[0088] The first convolution module, the second convolution module, and the third convolution module are used to perform convolution on the source radar signal with out-of-band clutter filtered out in sequence to obtain the first feature map after three convolutions.

[0089] The flattening module flattens the first feature map into a first one-dimensional vector.

[0090] The first fully connected layer is used to reduce the dimension of the first one-dimensional vector into a second one-dimensional vector and output the second one-dimensional vector to the second fully connected layer. The second fully connected layer is used to increase the dimension of the second one-dimensional vector into a third one-dimensional vector and output the third one-dimensional vector to the inverse flattening module.

[0091] The inverse flattening module is used to inverse flatten the third one-dimensional vector into a second feature map and output the second feature map to the third transposed convolutional module. The third one-dimensional vector has the same dimension as the first one-dimensional vector.

[0092] The third transposed convolutional module, the second transposed convolutional module, and the first transposed convolutional module are used to perform transposed convolution on the second feature map in sequence to obtain the target radar signal after three times of transposed convolution.

[0093] In some embodiments, the training method of the in-band clutter noise reduction model is as follows: Obtain a sample radar signal and a labeled radar signal.

[0094] Input the sample radar signal into the frequency domain filter to obtain a sample radar signal with out-of-band clutter removed.

[0095] Input the sample radar signal with out-of-band clutter removed into the in-band clutter noise reduction model to obtain the radar prediction signal output by the in-band clutter noise reduction model. Substitute the radar prediction signal and the labeled radar signal into the loss function. When the loss function converges, the model training is completed.

[0096] In some embodiments, obtaining a sample radar signal and a labeled radar signal includes: Based on the simulation of the first simulation model, obtain the labeled radar signal. Based on the simulation of the second simulation model, obtain the clutter signal.

[0097] Superimpose the labeled radar signal and the clutter signal to obtain the sample radar signal.

[0098] Figure 10 Illustrates a schematic diagram of the physical structure of an electronic device, as Figure 10 shown. The electronic device may include: a processor 1010, a communication interface 1020, a memory 1030, and a communication bus 1040. Among them, the processor 1010, the communication interface 1020, and the memory 1030 complete mutual communication through the communication bus 1040. The processor 1010 can call the logical instructions in the memory 1030 to execute the radar target noise reduction and recognition method, which includes: Input the source radar signal into the frequency domain filter to obtain a source radar signal with out-of-band clutter removed.

[0099] Input the source radar signal with out-of-band clutter filtered into the in-band clutter reduction model to obtain the target radar signal with in-band clutter filtered output by the in-band clutter reduction model.

[0100] Perform target recognition on the target to be recognized based on the target radar signal.

[0101] Wherein, the in-band clutter reduction model is trained based on the sample radar signal with out-of-band clutter filtered and the labeled radar signal with in-band clutter filtered corresponding to the sample radar signal.

[0102] In addition, when the logic instructions in the above-mentioned memory 1030 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this 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 for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0103] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the radar target noise reduction and recognition method provided by the above-mentioned various methods. The method includes: Input the source radar signal into a frequency domain filter to obtain the source radar signal with out-of-band clutter filtered.

[0104] Input the source radar signal with out-of-band clutter filtered into the in-band clutter reduction model to obtain the target radar signal with in-band clutter filtered output by the in-band clutter reduction model.

[0105] Perform target recognition on the target to be recognized based on the target radar signal.

[0106] Wherein, the in-band clutter reduction model is trained based on the sample radar signal with out-of-band clutter filtered and the labeled radar signal with in-band clutter filtered corresponding to the sample radar signal.

[0107] In another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the radar target noise reduction and recognition method provided by the above-mentioned various methods. The method includes: Input the source radar signal into a frequency-domain filter to obtain a source radar signal with out-of-band clutter removed.

[0108] Input the source radar signal with out-of-band clutter removed into an in-band clutter noise reduction model to obtain a target radar signal with in-band clutter removed output by the in-band clutter noise reduction model.

[0109] Perform target recognition on the target to be recognized based on the target radar signal.

[0110] Wherein, the in-band clutter noise reduction model is trained based on the sample radar signal with out-of-band clutter removed and the labeled radar signal with in-band clutter removed corresponding to the sample radar signal.

[0111] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative labor.

[0112] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0113] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A radar target noise reduction and recognition method, characterized in that: include: The source radar signal is input into the frequency domain filter to obtain the source radar signal with out-of-band clutter filtered out; Inputting the source radar signal after filtering out-of-band clutter into an in-band clutter noise reduction model to obtain a target radar signal after filtering out in-band clutter output by the in-band clutter noise reduction model; Performing target identification on the target to be identified based on the target radar signal; The in-band clutter denoising model is obtained by training based on a sample radar signal from which out-of-band clutter is filtered out and a labeled radar signal from which in-band clutter is filtered out corresponding to the sample radar signal.

2. The radar target denoising and identification method according to claim 1, characterized in that: The frequency domain filter includes: a plurality of sub-band frequency domain filters, different sub-band frequency domain filters correspond to different frequency bands, and the frequency bands of all sub-band frequency domain filters cover the frequency band range of the source radar signal; The source radar signal is input into the frequency domain filter to obtain the source radar signal after filtering out-of-band clutter, including: Inputting the source radar signal into each of the sub-band frequency domain filters respectively to obtain each filtered sub-band signal; The sub-band signals are combined to obtain the source radar signal with out-of-band clutter filtered out.

3. The radar target denoising and identification method according to claim 2, characterized in that: The plurality of sub-band frequency domain filters are generated as follows: The prototype filter is modulated by using a linear phase discrete Fourier transform to obtain the sub-band frequency domain filters having different frequency bands after modulation.

4. The radar target denoising and identification method according to claim 3, characterized in that: The prototype filter is a causal linear phase low-pass filter of order 2L, where L is an integer greater than 0, and the impulse response of the prototype filter satisfies: h ( n )= h (2 L - n )， n =0,1,…,2 L ； h ( n ) is the time domain representation of the prototype filter.

5. The radar target denoising and identification method according to any one of claims 1 to 4, characterized in that: The in-band clutter noise reduction model includes: an encoder and a decoder; The encoder comprises: a first convolution module, a second convolution module, a third convolution module, a flattening module and a first fully connected layer, wherein the first convolution module is connected to the second convolution module, the second convolution module is connected to the third convolution module, the third convolution module is connected to the flattening module, and the flattening module is connected to the first fully connected layer; The decoder includes: a first deconvolution module, a second deconvolution module, a third deconvolution module, a deflattening module and a second fully connected layer, wherein the first fully connected layer is connected to the second fully connected layer, the second fully connected layer is connected to the deflattening module, the deflattening module is connected to the third deconvolution module, the third deconvolution module is connected to the second deconvolution module, and the second deconvolution module is connected to the first deconvolution module; the i-th convolution module and the i-th deconvolution module have the same number of channels and convolution kernels of equal size, i=1,2,3; The first convolution module, the second convolution module and the third convolution module are used to sequentially convolve the source radar signal from which out-of-band clutter is filtered out, to obtain a first feature map after three convolutions; The flattening module flattens the first feature map into a first one-dimensional vector; The first fully connected layer is used to reduce the dimension of the first one-dimensional vector into a second one-dimensional vector, and output the second one-dimensional vector to the second fully connected layer; the second fully connected layer is used to increase the dimension of the second one-dimensional vector into a third one-dimensional vector, and output the third one-dimensional vector to the inverse flattening module; The inverse flattening module is used to inversely flatten the third one-dimensional vector into a second feature map, and output the second feature map to the third inverse convolution module, and the third one-dimensional vector has the same dimension as the first one-dimensional vector; The third deconvolution module, the second deconvolution module and the first deconvolution module are used to sequentially perform deconvolution on the second feature map to obtain the target radar signal after three deconvolutions.

6. The radar target denoising and identification method according to claim 5, characterized in that: The training method of the in-band clutter denoising model is as follows: Acquire sample radar signals and label radar signals; The sample radar signal is input into a frequency domain filter to obtain a sample radar signal with out-of-band clutter filtered out; The sample radar signal from which out-of-band clutter is filtered out is input into the in-band clutter denoising model to obtain the radar prediction signal output by the in-band clutter denoising model, and the radar prediction signal and the label radar signal are substituted into the loss function. When the loss function converges, the model training is completed.

7. The radar target denoising and identification method according to claim 6, characterized in that: Obtaining sample radar signals and labeling radar signals includes: Based on the simulation of the first simulation model, the tag radar signal is obtained, and based on the simulation of the second simulation model, the clutter signal is obtained; The tag radar signal and the clutter signal are superimposed to obtain the sample radar signal.

8. A radar target noise reduction and identification device, characterized in that: include: An out-of-band clutter filtering unit is used to input the source radar signal into a frequency domain filter to obtain the source radar signal with out-of-band clutter filtered out; An in-band clutter filtering unit is used to input the source radar signal after filtering out-band clutter into the in-band clutter noise reduction model to obtain the target radar signal after filtering out in-band clutter output by the in-band clutter noise reduction model; A target identification unit, used for identifying a target to be identified based on the target radar signal; The in-band clutter denoising model is obtained by training based on a sample radar signal from which out-of-band clutter is filtered out and a labeled radar signal from which in-band clutter is filtered out corresponding to the sample radar signal.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the radar target denoising and identification method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the radar target denoising and identification method according to any one of claims 1 to 7 is implemented.

Images