Radar interference type recognition and parameter estimation method, electronic device and storage medium

Abstract

The present disclosure provides a method for radar interference type identification and parameter estimation, an electronic device, and a storage medium. By converting the acquired radar signal into a time-frequency spectrogram and extracting its amplitude information, then normalizing this amplitude information and adjusting the image scale through bicubic linear interpolation to generate a pseudo-color image in a preset format. Next, the YOLO object detection network is used to identify and locate the interference pattern in the pseudo-color image, and the corresponding image region is extracted as an input sample. Then, these input samples are input into a disentangled variational autoencoder (DVAE) to optimize the network parameters to maximize the mutual information between different dimensions inside the latent variable and minimize the reconstruction error, obtaining the target latent variable. Finally, the target latent variable is input into a multi-layer perceptron (MLP) for decoding to obtain specific interference parameter estimation values. It improves the accuracy of interference type identification, enhances the precision of interference parameter estimation, and effectively improves the anti-interference ability of the radar system.

Description

Technical Field

[0001] The present invention relates to the technical field of radar interference type recognition, and particularly relates to a method for radar interference type recognition and parameter estimation, an electronic device, and a storage medium. Background Art

[0002] Radar interference parameter estimation is an important research direction in the field of radar signal processing, aiming to identify the interference type and further identify the interference parameters by analyzing the characteristic parameters of the interference signal.

[0003] Jiaxiang Zhang et al. proposed a composite interference cognition method based on a deep learning object detection network for identifying and detecting single interference elements and estimating their key parameters. First, the time-frequency distribution (TFD) is used to characterize the characteristics of the interference in multiple dimensions (time, frequency, and energy). Then, YOLOv5 is selected as the backbone network for the entire process because of its strong flexibility, fast processing speed, and high mean average precision (mAP). Then, the object detection network is applied to identify and locate the interference in the time-frequency domain. Finally, according to the type of interference parameters, the corresponding parameter estimation method is given.

[0004] Based on the above method, the present invention further adjusts the module order of data preprocessing, improves the classification and localization accuracy, adds feature disentanglement encoding to facilitate the development of downstream tasks, adds feature decoding according to parameter estimation, and the parameters are given by the decoder instead of manual calculation, improving the real-time performance of the estimation. Summary of the Invention

[0005] In a first aspect of the present disclosure, there is provided a method for radar interference type recognition and parameter estimation, including the following steps:

[0006] S101: Obtain and process the interference signal received by the radar, specifically including: performing short-time Fourier transform on the one-dimensional complex signal of the interference signal to obtain a time-frequency spectrogram, extracting amplitude information therefrom, performing maximum-minimum normalization processing on the amplitude information to generate a normalized amplitude feature, using bicubic linear interpolation to adjust the image scale of the normalized amplitude feature to form an image in a preset format, and converting the image into a pseudo-color image;

[0007] S102: Use a pre-trained YOLO object detection network to recognize and locate the interference pattern in the pseudo-color image, and output information including the interference category, position, size, and probability;

[0008] S103: Extract the corresponding image of the information in the pseudo-color image as an input sample, encode and decode the input sample through a disentangled variational autoencoder, optimize the network parameters, and obtain a target latent variable;

[0009] S104: Input the target latent variable into a multi-layer perceptron to decode and obtain a specific interference parameter estimation value.

[0010] Combined with the first aspect, the short-time Fourier transform is performed on the one-dimensional complex signal of the interference signal, and the time-frequency spectrogram is calculated through the following formula:

[0011] ，

[0012] where and represent the time window and frequency index respectively, is the window function, is the sampling interval, represents the discrete time point, is the complex exponential part in the Fourier transform, where is the imaginary unit, is the frequency index, is the sampling frequency, is the time variable.

[0013] Combined with the first aspect, the extraction of the amplitude information is calculated through the following formula:

[0014] .

[0015] Combined with the first aspect, the maximum-minimum normalization processing of the amplitude information is calculated through the following formula:

[0016] .

[0017] Combined with the first aspect, the use of bicubic linear interpolation to adjust the image scale of the normalized amplitude features to form an image in a preset format is calculated through the following formula:

[0018] ，

[0019] where and are the offsets relative to the point, , , are the cubic polynomial coefficients obtained by least squares fitting, and represent the order of the polynomial respectively;

[0020] The conversion of the image into a pseudo-color image is calculated through the following formula:

[0021] ,

[0022] Among them, and respectively represent the row and column indices of the image, represents the grayscale value of the corresponding pixel point, is the output pseudo-color image matrix, represents the color channel, .

[0023] Combined with the first aspect, the method of using the pre-trained YOLO object detection network to identify and locate the interference pattern in the pseudo-color image includes:

[0024] Identifying the pseudo-color image through the following formula:

[0025] ,

[0026] Among them, is the total number of prediction boxes, is the true label. For each prediction box , if the box contains the target, then , otherwise , is the probability distribution output by the th prediction box, indicating the probability that the box belongs to a certain category;

[0027] Locating the interference pattern in the pseudo-color image through the following formula:

[0028] ,

[0029] Among them, is the ratio of the overlapping area between the prediction box and the true box to the total area of the two boxes, is the Euclidean distance between the center points of the prediction box and the true box, is the diagonal distance of the closed area of the two rectangular boxes, is the weight coefficient used to balance the importance of different terms, measures the consistency of the relative ratio of the two rectangular boxes, , , , , respectively represent the widths and heights of the label and the predicted anchor box.

[0030] In combination with the first aspect, the disentangled variational autoencoder includes an encoder and a decoder. The encoder includes a first basic block, a second basic block, a third basic block, and a fourth basic block. The input and output channels of each basic block increase sequentially in the order of the first basic block, the second basic block, the third basic block, and the fourth basic block. The input sample is input from the first basic block and output to the image convolution layer through the fourth basic block.

[0031] The decoder includes a first reverse basic block, a second reverse basic block, a third reverse basic block, and a fourth reverse basic block. The first reverse basic block has the fewest input channels and the most output channels. The output channels of each reverse basic block decrease sequentially in the order of the first reverse basic block, the second reverse basic block, the third reverse basic block, and the fourth reverse basic block.

[0032] In combination with the first aspect, encoding and decoding the input sample by the disentangled variational autoencoder and optimizing network parameters to obtain the target latent variable include:

[0033] S1031: Extract the corresponding region in the pseudo-color image as the input sample according to the position information output by the YOLO object detection network;

[0034] S1032: Input the input sample into the encoder to reduce the spatial dimension of the input sample, increase the number of channels, and convert it into a target latent variable.

[0035] In a second aspect of the present disclosure, there is provided an electronic device, including:

[0036] One or more processors;

[0037] A storage unit for storing one or more programs, which, when executed by the one or more processors, enable the one or more processors to implement the radar jamming type recognition and parameter estimation method.

[0038] In a third aspect of the present disclosure, there is provided a computer-readable storage medium, on which a computer program is stored, characterized in that the computer program, when executed by a processor, can implement the radar jamming type recognition and parameter estimation method.

[0039] Beneficial effects: A method for radar interference type recognition and parameter estimation provided by the present invention converts the acquired one-dimensional complex radar signal into a time-frequency spectrogram through short-time Fourier transform, and extracts its amplitude information. Subsequently, maximum-minimum normalization processing is performed on these amplitude information, and bicubic linear interpolation is used to adjust the image scale to generate a pseudo-color image in a preset format. Then, the pre-trained YOLO object detection network is used to recognize and locate the interference pattern in the pseudo-color image, and the corresponding image region is extracted as an input sample. Then, these input samples are input into a disentangled variational autoencoder (DVAE), and by optimizing the network parameters to maximize the mutual information between different dimensions inside the latent variable and minimize the reconstruction error, a target latent variable with a good disentanglement effect is obtained. Finally, these target latent variables are input into a multi-layer perceptron (MLP) for decoding to obtain specific interference parameter estimation values. This method not only improves the accuracy of interference type recognition but also enhances the precision of interference parameter estimation, thus effectively improving the anti-jamming ability of the radar system. Brief Description of the Drawings

[0040] Figure 1 It is a schematic flowchart of a method for radar interference type recognition and parameter estimation according to an embodiment of the present disclosure;

[0041] Figure 2 It is an electronic device according to an embodiment of the present disclosure. Detailed Embodiments

[0042] Here, the exemplary embodiments will be described in detail, and the examples are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present disclosure.

[0043] The terms used in the embodiments of the present disclosure are only for the purpose of describing specific embodiments, and are not intended to limit the embodiments of the present disclosure. The singular forms "a", "the", and "said" used in the embodiments of the present disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0044] It should be understood that although the terms first, second, third, etc. may be used in the embodiments of the present disclosure to describe various kinds of information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of the embodiments of the present disclosure, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "when" or "while" or "in response to determining".

[0045] As Figure 1 shown, it is a schematic flow diagram of a method for radar interference type identification and parameter estimation according to an embodiment of the present disclosure, including:

[0046] S101: Obtain and process the interference signal received by the radar, specifically including: performing a short-time Fourier transform on the one-dimensional complex signal of the interference signal to obtain a time-frequency spectrogram, extracting amplitude information therefrom, performing maximum-minimum normalization processing on the amplitude information to generate a normalized amplitude feature, using bicubic linear interpolation to adjust the image scale of the normalized amplitude feature to form an image in a preset format, and converting the image into a pseudo-color image.

[0047] Specifically, since the radar data fluctuates greatly, and the neural network has poor fitting, generalization, and abstraction capabilities for data with a large order of magnitude, it is necessary to compress and transform the radar data, including:

[0048] Perform a short-time Fourier transform on the one-dimensional complex signal of the interference signal through the following formula to obtain a time-frequency spectrogram:

[0049] ，

[0050] where and represent the time window and the frequency index respectively, the complex signal received by the radar is J(t), is the window function, is the sampling interval, represents the discrete time point, is the complex exponential part in the Fourier transform, where is the imaginary unit, is the frequency index, is the sampling frequency, is the time variable.

[0051] Next, extract the amplitude information therefrom through the following formula:

[0052] .

[0053] This step is from Extract the amplitude information from the results. For each time-frequency point, calculate the modulus (i.e., the absolute value) of the corresponding complex value to obtain the amplitude at that point. Here, is denoted as , representing the energy magnitude of the signal at different times and frequencies.

[0054] Furthermore, the fluctuations of the interference amplitude information are relatively large, and maximum-minimum normalization processing is required:

[0055] ,

[0056] Beneficial effects: Normalization can help the model converge faster and improve its performance and stability. It makes the amplitude information of the radar signal more suitable for input into the deep learning model for further analysis.

[0057] Then, in order to balance the aspect ratio of the image for the network to capture features, bicubic linear interpolation is used for image adjustment:

[0058] Form an image in a preset format, calculated by the following formula:

[0059] ，

[0060] where and are the offsets relative to the point, , , are the cubic polynomial coefficients obtained by least squares fitting, and represent the orders of the polynomial respectively;

[0061] The conversion of the image into a pseudo-color image is calculated by the following formula:

[0062] ,

[0063] where and represent the row and column indices of the image respectively, represents the grayscale value of the corresponding pixel point, is the output pseudo-color image matrix, represents the color channel, .

[0064] The image in the preset format here includes a two-dimensional grayscale image.

[0065] Beneficial effects: The main purpose of bicubic interpolation is to provide a smoother result than nearest-neighbor interpolation or bilinear interpolation when resizing or scaling an image. By considering more information from neighboring pixels and using higher-order polynomial interpolation, it can better preserve the details and smoothness of the image.

[0066] Furthermore, it is necessary to convert the above two-dimensional grayscale image into a pseudo-color image that can be processed by a deep learning model (YOLO object detection network).

[0067] The conversion of the image into a pseudo-color image is calculated by the following formula:

[0068] ,

[0069] is an image in a preset format, where, and represent the row and column indices of the image respectively, represents the grayscale value of the corresponding pixel point, is the output pseudo-color image matrix, represents the color channel, .

[0070] S102: Use the pre-trained YOLO object detection network to identify and locate the interference pattern in the pseudo-color image, and output information including the interference category, location, size, and probability.

[0071] Specifically, the pseudo-color image is identified by the following formula:

[0072] ,

[0073] where, is the total number of prediction boxes, is the ground truth label. For each prediction box , if the box contains the target, then , otherwise , is the probability distribution output for the th prediction box, indicating the probability that the box belongs to a certain category;

[0074] The above calculation is used to predict the category of interference. For each prediction box, the model outputs a probability distribution indicating the likelihood that the interference belongs to each category.

[0075] The interference pattern in the pseudo-color image is located by the following formula:

[0076] ,

[0077] where, is the ratio of the overlapping area between the predicted box and the ground truth box to the total area of the two boxes. is the Euclidean distance between the centers of the predicted box and the ground truth box. is the diagonal distance of the closed area of the two rectangular boxes. is the weight coefficient used to balance the importance of different terms. measures the consistency of the relative ratio of the two rectangular boxes. , , , , respectively represent the widths and heights of the labeled and predicted anchor boxes.

[0078] The above calculations are used to optimize the position and size of the predicted box. It involves the regression problems of the center point coordinates, width, and height of the predicted box.

[0079] Finally, the interference category, horizontal and vertical coordinates, width, length, size information, and corresponding probabilities labeled by YOLO are obtained. 。

[0080] S103: Extract the corresponding image of the information in the pseudo-color image as the input sample, and encode and decode the input sample through the disentangled variational autoencoder to optimize the network parameters and obtain the target latent variable.

[0081] Specifically, the disentangled variational autoencoder includes an encoder and a decoder. The encoder includes a first basic block, a second basic block, a third basic block, and a fourth basic block. The input and output channels of each basic block increase sequentially in the order of the first basic block, the second basic block, the third basic block, and the fourth basic block. The input sample is input from the first basic block and output to the image convolution layer through the fourth basic block.

[0082] The decoder includes a first reverse basic block, a second reverse basic block, a third reverse basic block, and a fourth reverse basic block. The input channel of the first reverse basic block is the least, and the output channel is the most. The output channels of each reverse basic block decrease sequentially in the order of the first reverse basic block, the second reverse basic block, the third reverse basic block, and the fourth reverse basic block.

[0083] Specifically, the encoder is composed of several Basic_blocks (basic blocks) with different parameters.

[0084] Exemplarily, the encoder includes a first basic block, a second basic block, a third basic block, and a fourth basic block. The corresponding input and output channel numbers are: (3, 32), (32, 64), (64, 128), (128, 256). Each Basic_block contains a convolutional layer, a batch normalization layer, and a ReLU activation function. These layers gradually reduce the spatial dimension of the input samples while increasing the number of channels, and finally convert the input samples into a low-dimensional latent variable representation.

[0085] The decoder consists of several Inverse_Basic_blocks (inverse basic blocks) with different parameters.

[0086] Exemplarily, the decoder includes a first inverse basic block, a second inverse basic block, a third inverse basic block, and a fourth inverse basic block. The corresponding input and output channel numbers are: (10, 256), (256, 128), (128, 64), (64, 32). Each Inverse_Basic_block contains a transposed convolutional layer, a batch normalization layer, and a ReLU activation function. These layers gradually restore the spatial dimension of the input samples and finally reconstruct the input samples.

[0087] Exemplarily, denote an instance of the sample as , and the derivation of the disentanglement and reconstruction losses of this network is introduced below.

[0088] A common variational lower bound is:

[0089] ,

[0090] where is a latent variable, is the parameterized posterior distribution, the parameterized reconstruction distribution of is The prior of is usually a standard normal distribution. and is the KL (Kullback-Leibler) divergence of two continuous random variables with distribution functions . is a distribution function artificially introduced, and different forms of distribution functions can be constructed according to different needs.

[0091] The above formula is used to measure the difference between the approximate posterior probability distribution function and the prior distribution, and overfitting can be prevented through this term;

[0092] is the logarithmic likelihood estimation, which is equivalent to the reconstruction loss function in autoencoders and is used to reconstruct the original sample data. In this way, can be regarded as the encoding process, while can be regarded as the decoding process. Regarding the output encoding for a given sample which is the data re-expression , which is also equivalent to the reconstructed sample of the decoding output .

[0093] And in order to further enhance the disentanglement effect, it is further decomposed. In network optimization, this KL divergence term is usually calculated as the mean of the batch:

[0094] , , , ，

[0095] where represents the distribution of the -th sample, represents the joint distribution of the -th sample and its latent variable, is 's -th component distribution.

[0096] In the above formula, is the index-coded mutual information (MI), which is related to the mutual information between the data variable and the latent variable.

[0097] A higher MI indicates a better disentanglement effect.

[0098] is the total correlation (TC), which represents the degree of mutual dependence between variables in the latent variable space and is a measure of redundancy. The total correlation acts as a penalty to make the model look for statistically independent factors in the distribution. A heavier penalty on TC represents stronger statistical independence of semantics in the posterior probability distribution and enhances the ability of disentanglement. When are all independent, this term is 0, which is the most ideal disentanglement effect.

[0099] represents that the dimension of each latent variable cannot deviate too much from the dimension of the prior latent variable. By increasing the weight, the disentanglement effect can be further enhanced.

[0100] However it still cannot be estimated, so assume the total amount of data , sample data, using the technique:

[0101] ,

[0102] where represents is the distribution obtained by sampling according to the conditional distribution .

[0103] In this way, the final loss of the network is obtained :

[0104] ,

[0105] Finally, through the above loss, the network is optimized to obtain a latent variable (target latent variable) with better disentanglement effect.

[0106] S104: Input the target latent variable into a multi-layer perceptron to decode and obtain a specific interference parameter estimate value.

[0107] Specifically, according to the latent variable obtained in step S103 , we directly input it into a multi-layer perceptron (MLP) to decode and obtain the interference parameter.

[0108] Among them, the Linear of the multi-layer perceptron (MLP) is a fully connected layer, in_features is the number of input features, out_features is the number of output features, ReLU is a linear activation unit, Softplus is a softplus function, which aims to adjust the output to a non-negative real number, corresponding to the non-negativity of the estimated parameter. n_features is the number of parameters to be estimated, which corresponds to the number of estimated parameters in the interference modeling. The network design loss is the mean square error (MSE):

[0109] ,

[0110] where and are the th actual interference parameter and the interference parameter estimated by the network, is the number of samples.

[0111] The electronic device 200 can be a desktop computer, a notebook, a palm computer, a cloud server and other electronic devices. The electronic device 200 may include but is not limited to a processor 201 and a memory 202. Those skilled in the art can understand, Figure 2This is only an example of the electronic device 200, which does not constitute a limitation on the electronic device 200. It may include more or fewer components than those shown in the figure, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0112] The processor 201 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or the processor may also be any conventional processor, etc.

[0113] The memory 202 may be an internal storage unit of the electronic device 200. For example, the hard disk or memory of the electronic device 200. The memory 202 may also be an external storage device of the electronic device 200. For example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 200. Further, the memory 202 may also include both the internal storage unit and the external storage device of the electronic device 200. The memory 202 is used to store the computer program 203 and other programs and data required by the electronic device. The memory 202 may also be used to temporarily store the data that has been output or will be output.

[0114] In the embodiments provided in this disclosure, it should be understood that the disclosed apparatus / electronic device and method can be implemented in other ways. For example, the apparatus / electronic device embodiments described above are only illustrative. For example, the division of modules or units is only a logical function division. In actual implementation, there may be other division methods. Multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point, the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical or other forms.

[0115] When the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such understanding, to implement all or part of the processes in the above-described embodiment methods of the present disclosure, it can also be completed by a computer program instructing relevant hardware. The computer program can be stored in the computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-described various method embodiments can be implemented. The computer program can include computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice within the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0116] The above embodiments are only used to illustrate the technical solutions of the present disclosure, rather than to limit them; although the present disclosure 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 recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present disclosure, and should all be included in the protection scope of the present disclosure.

Claims

1. A method for radar interference type recognition and parameter estimation, characterized in that Including the following steps: S101: Obtain the interference signal received by the radar for processing, specifically including: performing short-time Fourier transform on the one-dimensional complex signal of the interference signal to obtain a time-frequency spectrogram, extracting amplitude information therefrom, performing maximum-minimum normalization processing on the amplitude information to generate a normalized amplitude feature, using bicubic linear interpolation to adjust the image scale of the normalized amplitude feature to form an image in a preset format, and converting the image into a pseudo-color image; S102: Use a pre-trained YOLO object detection network to identify and locate the interference pattern in the pseudo-color image, and output information including interference category, position, size, and probability; S103: Extract the corresponding image of the information in the pseudo-color image as an input sample, encode and decode the input sample through a disentangled variational autoencoder, and optimize the network parameters to obtain a target latent variable; S104: Input the target latent variable into a multi-layer perceptron, and decode to obtain a specific interference parameter estimation value.

2. The method according to claim 1, wherein The short-time Fourier transform is performed on the one-dimensional complex signal of the interference signal to obtain a time-frequency spectrogram, which is calculated by the following formula: ， Among them, and represent the time window and the frequency index respectively. The complex signal received by the radar is J(t), is the window function, is the sampling interval, represents the discrete time point, is the complex exponential part in the Fourier transform, where is the imaginary unit, is the frequency index, is the sampling frequency, is the time variable.

3. The method according to claim 2, wherein The extraction of the amplitude information therefrom is calculated by the following formula: 。 4. The method according to claim 3, wherein The maximum-minimum normalization processing of the amplitude information is calculated by the following formula: 。 5. The method according to claim 1, characterized in that, The use of bicubic linear interpolation to adjust the image scale of the normalized amplitude feature to form an image in a preset format is calculated by the following formula: ， Among them, and are the offsets relative to point, , , are the coefficients of the cubic polynomial obtained by least squares fitting, and represent the order of the polynomial respectively; The conversion of the image into a pseudo-color image is calculated by the following formula: ， Among them, and represent the row and column indices of the image respectively, represents the grayscale value of the corresponding pixel point, is the output pseudo-color image matrix, represents the color channel, .

6. The method according to claim 1, wherein The use of a pre-trained YOLO object detection network to identify and locate the interference pattern in the pseudo-color image includes: Identifying the pseudo-color image by the following formula: ， Among them, is the total number of prediction boxes, is the ground truth label. For each prediction box , if the box contains the target, then , otherwise , is the probability distribution output by the -th prediction box, representing the probability that the box belongs to a certain class; Locating the interference pattern in the pseudo-color image by the following formula: ， Among them, is the ratio of the overlapping area between the predicted box and the ground truth box to the total area of the two boxes, is the Euclidean distance between the centers of the predicted box and the ground truth box, is the diagonal distance of the closed area of the two rectangular boxes, is the weight coefficient used to balance the importance of different terms, measures the consistency of the relative ratio of the two rectangular boxes, , , , , represent the widths and heights of the labeled and predicted anchor boxes respectively.

7. The method according to claim 1, wherein The disentangled variational autoencoder includes an encoder and a decoder. The encoder includes a first basic block, a second basic block, a third basic block, and a fourth basic block. The input and output channels of each basic block increase sequentially in the order of the first basic block, the second basic block, the third basic block, and the fourth basic block. The input sample is input from the first basic block and output to the image convolution layer through the fourth basic block; The decoder includes a first inverse basic block, a second inverse basic block, a third inverse basic block, and a fourth inverse basic block. The first inverse basic block has the fewest input channels and the most output channels. The output channels of each inverse basic block decrease sequentially in the order of the first inverse basic block, the second inverse basic block, the third inverse basic block, and the fourth inverse basic block.

8. The method according to claim 7, wherein The encoding and decoding of the input sample through the disentangled variational autoencoder and the optimization of the network parameters to obtain a target latent variable include: S1031: Extract the corresponding region in the pseudo-color image as an input sample according to the position information output by the YOLO object detection network; S1032: Input the input sample into the encoder to reduce the spatial dimension of the input sample, increase the number of channels, and convert it into a target latent variable.

9. An electronic device, characterized in that, Including: One or more processors; A storage unit for storing one or more programs, which, when executed by the one or more processors, enable the one or more processors to implement the radar interference type recognition and parameter estimation method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it can implement the radar interference type recognition and parameter estimation method according to any one of claims 1 to 8.

Images