Radar signal processing method and system
The autoencoder deep neural network is used to reduce the dimensionality and feature splicing of radar signals, and combined with the DBSCAN clustering algorithm, solves the problem of poor applicability of existing radar signal processing methods in complex environments, and achieves efficient and robust radar signal analysis. Select the effect.
Patent Information
- Application Number
- CN202411457470.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-18
- Publication Date
- 2025-01-21
AI Technical Summary
The existing radar signal processing methods are poor in unknown and dynamically changing radar environments, especially when facing complex signal environments, noise and interference, it is difficult to effectively sort radar signals.
The self -encoder deep neural network uses the pulse description of the radar signal to reduce the dimension, and stitch the features after dimension reduction with other signal features. Finally, the radar signal is assigned through the DBSCAN cluster algorithm.
It achieves efficient and reliable analysis of radar signals in unknown environments, has good robustness, can effectively resist noise interference, and improves the accuracy of radar signal sorting.
Smart Images
Figure CN119335526A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of radar signal processing, and in particular to a radar signal processing method and system. Background Art
[0002] Radar signals are different from other signals. They are random, nonlinear, and non-stationary. In addition, with the development of electronic warfare technology and the widespread application of radar systems, battlefield environments are filled with a large number of complex and diverse radar signals. These signals not only come from a wide range of sources and are difficult to distinguish between true and false, but also have extremely complex and variable characteristic parameters. In this case, how to efficiently classify and identify these unknown radar signals has become an important research direction in the field of radar signal processing.
[0003] Radar signal processing generally includes two aspects: feature extraction and target sorting. The main task of radar signal sorting is to extract the individual signals of each radar from a mixed signal environment. This process plays a vital role in electronic countermeasures, intelligence analysis and electronic warfare systems. Effective radar signal sorting technology can not only improve the reconnaissance capability of the radar system, but also provide important technical support for interference and anti-interference in electronic warfare.
[0004] In the prior art, researchers have proposed a variety of traditional radar signal processing methods, which mainly rely on the analysis and statistics of pulse descriptor parameters. For example, the pulse arrival time analysis method analyzes and classifies based on the radar pulse arrival time series. This method distinguishes different radar signal sources by analyzing the distribution and pattern of arrival time. However, this method is sensitive to multipath propagation and noise environment, and is easily interfered, resulting in reduced sorting accuracy. The pulse repetition interval transformation method performs signal sorting by analyzing the frequency characteristics of the pulse repetition interval. Common pulse repetition interval transformation methods include fast Fourier transform and autocorrelation transform. This method works well when processing periodic pulse repetition interval signals, but has limited effect on complex pulse repetition interval patterns. The parameter clustering method uses a clustering algorithm to perform cluster analysis on multiple parameters of radar signals. This method achieves signal sorting by grouping pulses with similar characteristics. However, this method relies heavily on the performance and parameter selection of the clustering algorithm, and has poor robustness to noise and interference.
[0005] Although traditional radar signal processing methods have achieved certain results in certain application scenarios, they generally have the following limitations: traditional methods are less applicable in unknown and dynamically changing radar environments. Due to the development of modern radar technology, the signal environment has become more complex, the mutual interference between signals and environmental noise have increased significantly, and the resistance to noise and interference is weak, and it is easily affected by external factors. For radar signals with complex characteristics in modern electronic warfare environments, the robustness of traditional methods is increasingly challenged and difficult to effectively sort. Summary of the invention
[0006] To achieve the above and other purposes, the present invention provides a radar signal processing method, comprising the steps of:
[0007] S1: Collect radar signals in target environment;
[0008] S2: extracting a pulse description word of the radar signal, wherein the pulse description word includes arrival time, arrival angle, pulse width, pulse repetition interval and carrier frequency;
[0009] S3: preprocessing the pulse description words, sorting them by arrival time, and normalizing them to obtain radar signal characteristics;
[0010] S4: The deep neural network based on the autoencoder performs dimensionality reduction processing on the radar signal features of pulse width, pulse repetition interval and carrier frequency;
[0011] S5: splicing the radar signal features of pulse width, pulse repetition interval and carrier frequency after dimensionality reduction with the radar signal features of arrival time and arrival angle to obtain spliced radar signal features;
[0012] S6: Perform clustering processing on the spliced radar signal features to obtain clustered radar signals.
[0013] Furthermore, the autoencoder deep neural network includes a deep neural network consisting of an encoder and a decoder, wherein:
[0014] The structure of the encoder is: composed of 5 linear fully connected layers connected in series, with activation functions connected between each fully connected layer, the first linear layer is a 3*16 fully connected layer, the second linear layer is a 16*64 fully connected layer, the third linear layer is a 64*64 fully connected layer, the fourth linear layer is a 64*16 fully connected layer, and the fifth linear layer is a 16*2 fully connected layer;
[0015] The decoder has the same structure as the encoder.
[0016] Furthermore, the autoencoder deep neural network includes a deep neural network consisting of an encoder and a decoder, wherein:
[0017] The structure of the encoder is: composed of 5 linear fully connected layers connected in series, with activation functions connected between each fully connected layer, the first linear layer is a 3*16 fully connected layer, the second linear layer is a 16*32 fully connected layer, the third linear layer is a 64*32 fully connected layer, the fourth linear layer is a 64*16 fully connected layer, and the fifth linear layer is a 16*3 fully connected layer;
[0018] The decoder has the same structure as the encoder.
[0019] Furthermore, the autoencoder deep neural network includes a deep neural network consisting of an encoder and a decoder, wherein:
[0020] The structure of the encoder is: composed of 3 linear fully connected layers connected in series, with activation functions connected between each fully connected layer, the first linear layer is a 3*16 fully connected layer, the second linear layer is a 16*16 fully connected layer, and the third linear layer is a 64*2 fully connected layer;
[0021] The decoder has the same structure as the encoder.
[0022] Furthermore, the preprocessing in step S3 is: adopting the standard deviation normalization method, subtracting the mean of all sample data from the sample data and dividing it by the standard deviation of all sample data, so that the data conforms to the standard normal distribution with a mean of 0 and a variance of 1.
[0023] Furthermore, the clustering process in step S6 is specifically as follows:
[0024] S61: Randomly select a point. If the point has at least the minimum number of neighbors in the neighborhood, mark it as a core point and generate a new cluster.
[0025] S62: Continue to expand all core points in the cluster, adding all points directly reachable by density within the neighborhood range;
[0026] S63: Repeat steps S61 and S62 until no new points can be added to the current cluster;
[0027] S64: For points that do not belong to any cluster, mark them as noise points.
[0028] The present invention also provides a radar signal processing system, comprising an acquisition unit, a pulse description word extraction unit, a signal feature extraction unit, a preprocessing unit, a dimensionality reduction processing unit, a splicing unit and a clustering processing unit, wherein:
[0029] The acquisition unit is used to collect radar signals in the target environment;
[0030] The pulse description word extraction unit is used to extract the pulse description word of the radar signal, wherein the pulse description word includes arrival time, arrival angle, pulse width, pulse repetition interval and carrier frequency;
[0031] The preprocessing unit is used to preprocess the pulse description words, sort them according to arrival time, and perform normalization processing to obtain radar signal characteristics;
[0032] The dimension reduction processing unit is used to perform dimension reduction processing on radar signal characteristics of pulse width, pulse repetition interval and carrier frequency based on a deep neural network of an autoencoder;
[0033] The splicing unit is used to splice the radar signal characteristics of pulse width, pulse repetition interval and carrier frequency after dimension reduction with the radar signal characteristics of arrival time and arrival angle to obtain the spliced radar signal characteristics;
[0034] The clustering processing unit is used to perform clustering processing on the spliced radar signal features to obtain clustered radar signals.
[0035] Furthermore, the autoencoder deep neural network includes a deep neural network consisting of an encoder and a decoder, wherein:
[0036] The structure of the encoder is: composed of 5 linear fully connected layers connected in series, with activation functions connected between each fully connected layer, the first linear layer is a 3*16 fully connected layer, the second linear layer is a 16*64 fully connected layer, the third linear layer is a 64*64 fully connected layer, the fourth linear layer is a 64*16 fully connected layer, and the fifth linear layer is a 16*2 fully connected layer;
[0037] The decoder has the same structure as the encoder.
[0038] The present invention also provides a storage medium, wherein the storage medium stores a computer program, and when the computer program is run on a computer, the computer executes the radar signal processing method.
[0039] The present invention also provides an electronic device, comprising: a processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to execute the radar signal processing method.
[0040] Compared with the prior art, the present application proposes an improved technical solution based on the complexity of radar signals and combined with the learning ability and feature extraction ability of deep neural networks. The radar signals are first preprocessed, and then part of the radar signals are subjected to dimensionality reduction feature processing, which is then fused and spliced with other signals. Finally, the radar signals are obtained through an innovative clustering algorithm. This can effectively realize effective, efficient and reliable analysis of radar signals in unknown environments, and has good robustness to noise. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 The figure is a flow chart of a radar signal processing method of the present invention.
[0042] Figure 2 This is a schematic diagram of the autoencoder network structure of this application.
[0043] Figure 3 This is a structural diagram of performing dimensionality reduction, splicing and clustering processing on radar signals in an embodiment of the present application.
[0044] Figure 4 A schematic diagram of the actual classification of a radar signal.
[0045] Figure 5 This is a schematic diagram of the visualization results after implementing the technical solution of the present application for the real classification of the radar. DETAILED DESCRIPTION
[0046] The following describes the implementation of the present invention through specific examples and in conjunction with the accompanying drawings. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific examples, and the details in this specification can also be modified and changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention.
[0047] This application addresses the fact that traditional sorting methods that rely on prior knowledge and predefined features are greatly limited in unknown and dynamically changing radar environments. Although deep neural networks can learn features from data without relying on prior knowledge and manually designed features, and have strong analytical capabilities, good adaptability and robustness, the current deep learning analysis of complex signals has increased the complexity of signal types.
[0048] Therefore, according to the complexity characteristics of radar signals, the present application proposes an improved technical solution in combination with the learning ability and feature extraction ability of deep neural networks. The radar signals are first preprocessed, and then part of the radar signals are subjected to dimensionality reduction feature processing, which is then fused and spliced with other signals. Finally, the radar signals are obtained through a flexible clustering algorithm. This can effectively realize effective, efficient and reliable analysis of radar signals in unknown environments, and has good robustness to noise.
[0049] like Figure 1 As shown, Figure 1 The following is a flow chart of a radar signal processing method according to the present invention. The following is a radar signal processing method according to the present invention, comprising:
[0050] S1: Collect radar signals in target environment;
[0051] S2: extracting a pulse description word of the radar signal, wherein the pulse description word includes arrival time, arrival angle, pulse width, pulse repetition interval and carrier frequency;
[0052] S3: preprocessing the pulse description words, sorting them by arrival time, and normalizing them to obtain radar signal characteristics;
[0053] S4: The deep neural network based on the autoencoder performs dimensionality reduction processing on the radar signal features of pulse width, pulse repetition interval and carrier frequency;
[0054] S5: splicing the radar signal features of pulse width, pulse repetition interval and carrier frequency after dimensionality reduction with the radar signal features of arrival time and arrival angle to obtain spliced radar signal features;
[0055] S6: Perform clustering processing on the spliced radar signal features to obtain clustered radar signals.
[0056] The technical solution of the present application is explained below in conjunction with various specific implementations.
[0057] S1: Collect radar signals in the target environment.
[0058] Step S1 refers to setting up the radar equipment for collecting radar signals, debugging and calibrating the radar equipment, selecting a suitable collection environment, minimizing interference, and starting to collect radar signals and record data.
[0059] Radar signals contain a lot of information, generally including pulse descriptor word (PDW), signal strength, Doppler frequency shift, phase information, signal-to-noise ratio, antenna radiation pattern, target characteristics, etc.
[0060] S2: extracting a pulse description word of the radar signal, wherein the pulse description word includes arrival time, arrival angle, pulse width, pulse repetition interval and carrier frequency;
[0061] In the technical solution of this application, the pulse description word is analyzed. The pulse description word is composed of a series of parameters that describe the characteristics of the radar pulse signal. The arrival time (TOA) indicates the time when the radar pulse arrives at the receiver, the arrival angle (AOA) indicates the direction of the radar signal arriving at the receiver, the pulse width (Pulse Width, PW) indicates the time width of the radar pulse, the pulse repetition interval (PRI) indicates the time interval between two consecutive pulses, and the carrier frequency (CF) indicates the center frequency of the radar signal. These parameters constitute the characteristic vector of the radar signal, which becomes the basis for subsequent classification and recognition, specifically:
[0062] Arrival time: It indicates the time delay from the transmission to the reception of the radar signal, and is usually used to calculate the distance between the target and the radar.
[0063] Pulse Amplitude: Indicates the received signal strength, reflecting the reflection characteristics and distance of the target.
[0064] Pulse duration: Indicates the length of the radar pulse, which affects the signal resolution and target recognition capabilities.
[0065] Frequency: Indicates the frequency characteristics of a signal and may be used to analyze the speed of a target (Doppler effect).
[0066] In addition, phase and other characteristics may also be included.
[0067] Phase: Indicates the phase information of the signal, which helps to further analyze the motion state of the target.
[0068] Other characteristics: may include the noise level of the signal, the signal type (such as pulse, continuous wave, etc.), and other target-related characteristics.
[0069] Most radar signal analysis is based on pulse descriptors. The combination and analysis of pulse descriptors can help radar systems identify and classify different targets, perform target tracking and environmental perception. By clustering and analyzing pulse descriptors, radar systems can effectively process complex signal environments and extract useful information.
[0070] The technical solution in this application is analyzed based on the above five signals, namely, arrival time, arrival angle, pulse width, pulse repetition interval and carrier frequency.
[0071] S3: Preprocess the pulse description words, sort them by arrival time, and perform normalization processing to obtain radar signal characteristics.
[0072] As a preferred implementation, the preprocessing method of the present application is: adopting the standard deviation normalization method, subtracting the mean of all sample data from the sample data and dividing it by the standard deviation of all sample data, so that the data conforms to the standard normal distribution with a mean of 0 and a variance of 1. The specific standard deviation normalization can be expressed as:
[0073]
[0074] Where μ is the mean of all sample data, and σ is the standard deviation of all sample data.
[0075] In this embodiment, the preprocessed pulse width, pulse repetition interval and carrier frequency are input into the autoencoder to obtain low-dimensional embedding features. The autoencoder can be expressed as:
[0076] z=f(x;θ e )
[0077] Where f is the nonlinear mapping function of the encoder, θ e are the parameters of the encoder. The decoder reconstructs the low-dimensional feature representation into high-dimensional data. Assume that the output of the decoder is Its output can be expressed as:
[0078]
[0079] Among them, g is the nonlinear mapping function of the decoder, θ d are the parameters of the decoder.
[0080] S4: The deep neural network based on the autoencoder performs dimensionality reduction processing on the radar signal features of pulse width, pulse repetition interval and carrier frequency.
[0081] Radar signals are complex and changeable, so it is critical to quickly analyze radar signals and extract features that can reflect the characteristics of the radar itself. Although a single pulse description word has only five features, due to the changeable radar signals, true and false signals, and interference signals in today's complex electronic information confrontation, the type of data collected is very complex and the amount of data is very large, making it difficult for existing technologies to quickly and effectively analyze them.
[0082] Therefore, this application is based on a variety of tests and studies on many types of radars and summarizes that the arrival time and arrival angle are greatly affected by the environment and position changes, which greatly increases the complexity of signal analysis, and these two signal types do not directly reflect the inherent characteristics of the radar. Therefore, in the feature extraction process, we choose to extract features from parameters related to the radar itself, namely pulse width, pulse repetition interval and carrier frequency, which can better reflect the physical and technical characteristics of the radar while reducing the amount of calculation.
[0083] This application uses the learning ability and powerful analytical ability of the deep neural network constructed by the autoencoder to reduce the dimensionality of some of the features and obtain low-dimensional features.
[0084] See also Figure 2 , Figure 2 This is a schematic diagram of the autoencoder network structure of the present application. The autoencoder includes an encoder and a decoder. The structures of the encoder and the decoder are the same. As a preferred embodiment, the encoder structure of the present application is: composed of 5 linear fully connected layers in series, and activation functions are connected between each fully connected layer. The activation function introduces nonlinearity. In this embodiment, the ReLU activation function is selected. Among them, the first linear layer is a 3*16 fully connected layer, the second linear layer is a 16*64 fully connected layer, the third linear layer is a 64*64 fully connected layer, the fourth linear layer is a 64*16 fully connected layer, and the fifth linear layer is a 16*2 fully connected layer. The low-dimensional features (also called embedded features) of the radar signal are output from the fifth linear layer. The structure of the decoder is the same as that of the encoder. The decoder maps the embedded features to 16 dimensions, then gradually increases to 64 dimensions, and finally reduces to the dimension of the input features.
[0085] See also Figure 3 As shown in the figure, during training, the mean square error (MSE) of the input features and output features is used as the loss function, and the Adam optimizer is used with a learning rate of 0.001 for training. The mean square error formula is:
[0086]
[0087] As another variation of the implementation, the structure of the autoencoder can be: composed of 5 linear fully connected layers in series, each fully connected layer is connected with a ReLU activation function, the first linear layer is a 3*16 fully connected layer, the second linear layer is a 16*32 fully connected layer, the third linear layer is a 64*32 fully connected layer, the fourth linear layer is a 64*16 fully connected layer, and the fifth linear layer is a 16*3 fully connected layer.
[0088] In addition, as another variation of the embodiment, a simplified autoencoder structure is composed of three linear fully connected layers connected in series, each fully connected layer is connected with a ReLU activation function, the first linear layer is a 3*16 fully connected layer, the second linear layer is a 16*16 fully connected layer, and the third linear layer is a 64*2 fully connected layer.
[0089] After dimensionality reduction processing of some radar signals through autoencoding, the feature dimension is greatly reduced, making the model more efficient and improving the generalization ability, and also reducing the amount of calculation and complexity of subsequent clustering processing.
[0090] S5: Splice the radar signal features of pulse width, pulse repetition interval and carrier frequency after dimensionality reduction with the radar signal features of arrival time and arrival angle to obtain spliced radar signal features.
[0091] See also Figure 3 In step S4, after reducing the dimension of the radar signal features of pulse width (PW), pulse repetition interval (PRI) and carrier frequency (CF), the reduced dimension features are obtained and spliced with the radar signal features of arrival time (TOA) and arrival angle (AOA). As a preferred splicing method, the splicing method of the embodiment of the present application is:
[0092] Combine the time of arrival (TOA) and angle of arrival (AOA) into a two-dimensional feature;
[0093] The pulse width (PW), pulse repetition interval (PRI) and carrier frequency (CF) are combined into three-dimensional features, which are then processed by the above autoencoder for dimensionality reduction to obtain two-dimensional features;
[0094] The two-dimensional features of the combination of time of arrival (TOA) and angle of arrival (AOA) are combined with the two-dimensional features obtained by the above-mentioned dimensionality reduction process to obtain the spliced radar signal features of the four-dimensional features.
[0095] S6: Perform clustering processing on the spliced radar signal features to obtain clustered radar signals.
[0096] As a preferred implementation, the clustering method of the present application clusters radar signal features based on the DBSCAN clustering algorithm, as follows:
[0097] S61: Randomly select a point. If the point has at least the minimum number of neighbors in the neighborhood, mark it as a core point and generate a new cluster.
[0098] The present application embodiment is based on the DBSCAN algorithm, which identifies clusters in areas with high data point density and treats sparse areas as noise. It works through two main parameters: neighborhood and minimum number of points.
[0099] The neighborhood defines a radius around a point, and the minimum number of points represents the minimum number of neighbors required within this radius to consider a point as a core point. The algorithm first randomly selects a point and checks whether it has at least the minimum number of neighbors in its neighborhood. If so, the point is marked as a core point and a new cluster is started.
[0100] S62: Continue to expand all core points in the cluster, adding all points directly reachable by density within the neighborhood range;
[0101] S63: Repeat steps S61 and S62 until no new points can be added to the current cluster;
[0102] S64: For points that do not belong to any cluster, mark them as noise points.
[0103] The specific implementation is expressed as follows:
[0104] Given a point p, its ∈ domain is defined as:
[0105] N ∈ (p)={q∈D|dist(p,q)≤∈}
[0106] Where D is the dataset and dist(p,q) is the distance between point p and point q.
[0107] If a point p's ∈ neighborhood contains at least MinPts points (including point p itself), then p is a core point. The formula is expressed as:
[0108] |N ∈ (p)|≥MinPts
[0109] If point p is a core point and point q is in the ∈ neighborhood of p, then q is directly density-reachable from p. The formula is:
[0110] q∈N ∈ (p)
[0111] If there exists a series of points p1, p2, ..., p n , where p1 = p and p n =q, so that every p i+1 From p i If q is directly density-reachable, then q is density-reachable from p.
[0112] Points p and q are density connected if there exists a point o such that both points p and q are density reachable from o.
[0113] See also Figure 4 , Figure 5 , Figure 4 This is a real classification diagram of a radar signal. Figure 5 This is a schematic diagram of the visualization result obtained after implementing the technical solution of the present application on the real classification of the radar. In actual tests, the radar signal result obtained by the technical solution of the embodiment of the present application is very close to the real radar signal. By comparison, it can be found that the sorting result of the present invention is very close to the real classification. According to statistics, the sorting accuracy on the test set of this data set reaches 98.84%.
[0114] It can be seen from this that this application, based on the deep autoencoder, selectively extracts features from pulse descriptors, then splices them, and then introduces a clustering algorithm to sort radar signals. By combining the powerful feature extraction capabilities of deep learning and the flexibility of the density clustering algorithm, it effectively solves many limitations of traditional radar signal sorting methods. By selectively selecting part of the radar signal for dimensionality reduction processing, low-dimensional features are obtained for embedding, and feature extraction is performed on the pulse descriptor parameters related to the radar's own characteristics, excluding the arrival time and arrival angle parameters that are greatly affected by external factors, so that it can be applied to the problem of unknown radar signal sorting in various complex environments.
[0115] Based on the above embodiments, the present application also provides a radar signal processing system, including an acquisition unit, a pulse description word extraction unit, a signal feature extraction unit, a preprocessing unit, a dimensionality reduction processing unit, a splicing unit, and a clustering processing unit, wherein:
[0116] The acquisition unit is used to collect radar signals in the target environment;
[0117] The pulse description word extraction unit is used to extract the pulse description word of the radar signal, wherein the pulse description word includes arrival time, arrival angle, pulse width, pulse repetition interval and carrier frequency;
[0118] The preprocessing unit is used to preprocess the pulse description words, sort them according to arrival time, and perform normalization processing to obtain radar signal characteristics;
[0119] The dimension reduction processing unit is used to perform dimension reduction processing on radar signal characteristics of pulse width, pulse repetition interval and carrier frequency based on a deep neural network of an autoencoder;
[0120] The splicing unit is used to splice the radar signal characteristics of pulse width, pulse repetition interval and carrier frequency after dimension reduction with the radar signal characteristics of arrival time and arrival angle to obtain the spliced radar signal characteristics;
[0121] The clustering processing unit is used to perform clustering processing on the spliced radar signal features to obtain clustered radar signals.
[0122] Furthermore, the autoencoder deep neural network includes a deep neural network consisting of an encoder and a decoder, wherein:
[0123] The structure of the encoder is: composed of 5 linear fully connected layers connected in series, with activation functions connected between each fully connected layer, the first linear layer is a 3*16 fully connected layer, the second linear layer is a 16*64 fully connected layer, the third linear layer is a 64*64 fully connected layer, the fourth linear layer is a 64*16 fully connected layer, and the fifth linear layer is a 16*2 fully connected layer;
[0124] The decoder has the same structure as the encoder.
[0125] The present invention also provides a storage medium, wherein the storage medium stores a computer program, and when the computer program is run on a computer, the computer executes the radar signal processing method.
[0126] The present invention also provides an electronic device, comprising: a processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to execute the radar signal processing method.
[0127] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Any person skilled in the art may modify and alter the above embodiments without violating the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be as set forth in the claims.
Claims
1. A radar signal processing method, characterized in that: Includes steps: S1: Collect radar signals in target environment; S2: extracting a pulse description word of the radar signal, wherein the pulse description word includes arrival time, arrival angle, pulse width, pulse repetition interval and carrier frequency; S3: preprocessing the pulse description words, sorting them by arrival time, and normalizing them to obtain radar signal characteristics; S4: The deep neural network based on the autoencoder performs dimensionality reduction processing on the radar signal features of pulse width, pulse repetition interval and carrier frequency; S5: splicing the radar signal features of pulse width, pulse repetition interval and carrier frequency after dimensionality reduction with the radar signal features of arrival time and arrival angle to obtain spliced radar signal features; S6: Perform clustering processing on the spliced radar signal features to obtain clustered radar signals.
2. A radar signal processing method as claimed in claim 1, characterized in that: The autoencoder deep neural network includes a deep neural network consisting of an encoder and a decoder, where: The structure of the encoder is: composed of 5 linear fully connected layers connected in series, with activation functions connected between each fully connected layer, the first linear layer is a 3*16 fully connected layer, the second linear layer is a 16*64 fully connected layer, the third linear layer is a 64*64 fully connected layer, the fourth linear layer is a 64*16 fully connected layer, and the fifth linear layer is a 16*2 fully connected layer; The decoder has the same structure as the encoder.
3. A radar signal processing method as claimed in claim 1, characterized in that: The autoencoder deep neural network includes a deep neural network consisting of an encoder and a decoder, where: The structure of the encoder is: composed of 5 linear fully connected layers connected in series, with activation functions connected between each fully connected layer, the first linear layer is a 3*16 fully connected layer, the second linear layer is a 16*32 fully connected layer, the third linear layer is a 64*32 fully connected layer, the fourth linear layer is a 64*16 fully connected layer, and the fifth linear layer is a 16*3 fully connected layer; The decoder has the same structure as the encoder.
4. A radar signal processing method as claimed in claim 1, characterized in that: The autoencoder deep neural network includes a deep neural network consisting of an encoder and a decoder, where: The structure of the encoder is: composed of 3 linear fully connected layers connected in series, with activation functions connected between each fully connected layer, the first linear layer is a 3*16 fully connected layer, the second linear layer is a 16*16 fully connected layer, and the third linear layer is a 64*2 fully connected layer; The decoder has the same structure as the encoder.
5. A radar signal processing method as claimed in claim 1, characterized in that: The preprocessing in step S3 is: using the standard deviation normalization method, subtracting the mean of all sample data from the sample data and dividing it by the standard deviation of all sample data, so that the data conforms to the standard normal distribution with a mean of 0 and a variance of 1.
6. A radar signal processing method as claimed in claim 1, characterized in that: The clustering process in step S6 is specifically as follows: S61: Randomly select a point. If the point has at least the minimum number of neighbors in the neighborhood, mark it as a core point and generate a new cluster. S62: Continue to expand all core points in the cluster, adding all points directly reachable by density within the neighborhood range; S63: Repeat steps S61 and S62 until no new points can be added to the current cluster; S64: For points that do not belong to any cluster, mark them as noise points.
7. A radar signal processing system, characterized in that: It includes an acquisition unit, a pulse description word extraction unit, a signal feature extraction unit, a preprocessing unit, a dimensionality reduction processing unit, a splicing unit and a clustering processing unit, wherein: The acquisition unit is used to collect radar signals in the target environment; The pulse description word extraction unit is used to extract the pulse description word of the radar signal, wherein the pulse description word includes arrival time, arrival angle, pulse width, pulse repetition interval and carrier frequency; The preprocessing unit is used to preprocess the pulse description words, sort them according to arrival time, and perform normalization processing to obtain radar signal characteristics; The dimension reduction processing unit is used to perform dimension reduction processing on radar signal characteristics of pulse width, pulse repetition interval and carrier frequency based on a deep neural network of an autoencoder; The splicing unit is used to splice the radar signal characteristics of pulse width, pulse repetition interval and carrier frequency after dimension reduction with the radar signal characteristics of arrival time and arrival angle to obtain the spliced radar signal characteristics; The clustering processing unit is used to perform clustering processing on the spliced radar signal features to obtain clustered radar signals.
8. A radar signal processing system as claimed in claim 7, characterized in that: The autoencoder deep neural network includes a deep neural network consisting of an encoder and a decoder, where: The structure of the encoder is: composed of 5 linear fully connected layers connected in series, with activation functions connected between each fully connected layer, the first linear layer is a 3*16 fully connected layer, the second linear layer is a 16*64 fully connected layer, the third linear layer is a 64*64 fully connected layer, the fourth linear layer is a 64*16 fully connected layer, and the fifth linear layer is a 16*2 fully connected layer; The decoder has the same structure as the encoder.
9. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is run on a computer, the computer is enabled to execute the radar signal processing method according to any one of claims 1 to 6.
10. An electronic device, characterized in that: include: A processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the storage medium communicate via the bus, and the processor executes the machine-readable instructions to execute any one of the radar signal processing methods of claims 1-6.