Method and device for extracting multi-level features of radar signal

Abstract

The invention relates to the field of radar signal processing, and provides a method and a device for extracting multi-level features of radar signals. Performing convolution on a to-be-extracted time-frequency graph of the original DRFM forwarding signal on a plurality of feature extraction scales to extract local detail features of each position in the to-be-extracted time-frequency graph to obtain a first distinguishing feature; modeling global information in the first distinguishing feature to obtain a first global feature; fusing the first distinguishing feature and the first global feature to obtain a time-frequency graph feature; processing a to-be-extracted bispectrogram of the original DRFM forwarding signal by using the same method to obtain bispectrogram features; according to the method, the time-frequency graph features and the double-spectrum graph features are fused to obtain the multi-level features of the original DRFM forwarding signals, and the problem that the recognition accuracy of the types of the DRFM forwarding signals is sharply reduced due to the fact that the DRFM forwarding signals are seriously aliasing and artificial features are not obvious is solved.

Description

Technical Field

[0001] The present invention relates to the field of radar signal processing, and particularly to a method and device for extracting multi-level features of radar signals. Background Art

[0002] With the continuous development of Digital Radio Frequency Memory (DRFM) technology, various DRFM forwarding signal patterns have emerged continuously. The electromagnetic environment faced by radars has become increasingly complex. The fast and effective DRFM forwarding signals pose a serious threat to the detection ability of radars, which requires radars to have stronger environmental perception capabilities.

[0003] In order to achieve stronger environmental perception capabilities of radars, it is first necessary to accurately identify the category to which the DRFM forwarding signal belongs. Only by extracting the effective features of the DRFM forwarding signal can a better recognition effect be achieved.

[0004] In a complex electromagnetic environment, due to the severe aliasing of DRFM forwarding signals and the lack of obvious artificial features, the traditional methods of the prior art have greatly reduced the ability to extract features, resulting in a sharp decline in the recognition accuracy of the categories of DRFM forwarding signals. As a result, in a complex electromagnetic environment, radar systems cannot accurately identify and filter out unnecessary DRFM forwarding signals. Among them, artificial features refer to the features of DRFM forwarding signals extracted manually by traditional methods, that is, based on prior knowledge or experience of DRFM forwarding signals, to describe certain characteristics of DRFM forwarding signals.

[0005] In view of this, overcoming the defects of the prior art is an urgent problem to be solved in this technical field. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method and device for extracting multi-level features of radar signals, aiming to extract local detail features of the original DRFM forwarding signal at multiple feature extraction scales, model the corresponding global information, and extract more hierarchical feature information to obtain the multi-level features of the original DRFM forwarding signal, thereby solving the problem of the sharp decline in the recognition accuracy of signal categories due to severe signal aliasing and lack of obvious artificial features.

[0007] The present invention adopts the following technical solutions:

[0008] In a first aspect, the present invention provides a method for extracting multi-level features of radar signals, including:

[0009] Obtaining a time-frequency diagram to be extracted and a bispectrum diagram to be extracted of the original DRFM forwarding signal;

[0010] Convolve the time-frequency map to be extracted at multiple feature extraction scales to extract the local detailed features at each position in the time-frequency map to be extracted, obtaining a first discriminative feature; model the global information in the time-frequency map to be extracted, obtaining a first global feature; fuse the first discriminative feature and the first global feature to obtain a time-frequency map feature;

[0011] Convolve the bispectrum map to be extracted at multiple feature extraction scales to extract the local detailed features at each position in the bispectrum map to be extracted, obtaining a second discriminative feature; model the global information in the bispectrum map to be extracted, obtaining a second global feature; fuse the second discriminative feature and the second global feature to obtain a bispectrum map feature;

[0012] Fuse the time-frequency map feature and the bispectrum map feature to obtain the multi-level features of the original DRFM forwarding signal.

[0013] Further, the convolving the time-frequency map to be extracted at multiple feature extraction scales to extract the local detailed features at each position in the time-frequency map to be extracted, obtaining a first discriminative feature includes:

[0014] Parallelly extract local detailed features at multiple spatial resolutions of the time-frequency map to be extracted, obtaining intermediate features of multiple different scales;

[0015] Perform feature splicing on the intermediate features of the multiple different scales to extract the signal category information of the first feature to be processed at each spatial resolution, obtaining a first discriminative feature.

[0016] Further, the parallelly extracting local detailed features at multiple spatial resolutions of the time-frequency map to be extracted, obtaining intermediate features of multiple different scales includes:

[0017] Use a convolution kernel of the first size to convolve the time-frequency map to be extracted, obtaining an intermediate feature of the first scale;

[0018] Use a convolution kernel of the second size to convolve the time-frequency map to be extracted, obtaining a first feature; use a convolution kernel of the first size to convolve the first feature, obtaining an intermediate feature of the second scale;

[0019] Use a convolution kernel of the first size to convolve the time-frequency map to be extracted, obtaining a second feature; use a convolution kernel of the third size to convolve the second feature, obtaining an intermediate feature of the third scale;

[0020] Perform convolution on the to-be-extracted time-frequency map using a convolution kernel of the first size to obtain a third feature; perform convolution on the third feature using a convolution kernel of the second size to obtain a fourth feature; perform convolution on the fourth feature using a convolution kernel of the second size to obtain an intermediate feature of the fourth scale.

[0021] Further, the modeling of the global information in the to-be-extracted time-frequency map to obtain a first global feature includes:

[0022] Divide the to-be-extracted time-frequency map into tiles of a fixed size; convert each tile into a to-be-processed vector of a fixed dimension;

[0023] Determine the attention weights of the to-be-processed vectors, and perform attention calculation on the to-be-processed vectors using the attention weights to obtain a first encoded feature;

[0024] Perform linear transformation and non-linear activation on the first encoded feature to learn the first encoded feature using a feed-forward neural network to obtain a second encoded feature;

[0025] Perform an addition operation on the first encoded feature and the second encoded feature to obtain a first global feature.

[0026] Further, the determining the attention weights of the to-be-processed vectors and performing attention calculation on the to-be-processed vectors using the attention weights to obtain a first encoded feature includes:

[0027] Calculate the product of the to-be-processed vector and the query weight parameter to obtain a query vector; calculate the product of the to-be-processed vector and the key weight parameter to obtain an original key vector; calculate the product of the to-be-processed vector and the value weight parameter to obtain an original value vector;

[0028] Perform a convolution operation on the original key vector in the spatial dimension according to a reduction ratio to obtain a first reduced feature; perform a linear projection operation on the first reduced feature using the reduction ratio to compress the channel dimension of the first reduced feature to obtain a target key vector;

[0029] Perform a convolution operation on the original value vector in the spatial dimension according to a reduction ratio to obtain a second reduced feature; perform a linear projection operation on the second reduced feature using the reduction ratio to compress the channel dimension of the second reduced feature to obtain a target value vector;

[0030] Calculate the dot product of the query vector and the target key vector to obtain attention weights;

[0031] Use the attention weights to perform weighted summation on the target value vector to generate a first encoded feature.

[0032] Further, the obtaining of the time-frequency map feature by fusing the first discrimination feature and the first global feature includes:

[0033] Extracting local detail features of the first discrimination feature in the channel dimension and the spatial dimension respectively to obtain local attention weights;

[0034] Extracting global dependency relationships between different positions in the first global feature to obtain channel attention features;

[0035] Performing feature splicing on the local attention weights and the channel attention features to obtain time-frequency map features.

[0036] Further, the extracting local detail features of the first discrimination feature in the channel dimension and the spatial dimension respectively to obtain local attention weights includes:

[0037] Performing average pooling on the first discrimination feature in the channel dimension to obtain a first channel global feature; performing max pooling on the first discrimination feature in the channel dimension to obtain a second channel global feature; inputting the first channel global feature and the second channel global feature into a shared fully connected layer to generate channel attention weights;

[0038] Performing average pooling on the first discrimination feature in the spatial dimension to obtain a first spatial global feature; performing max pooling on the first discrimination feature in the spatial dimension to obtain a second spatial global feature; inputting the first spatial global feature and the second spatial global feature into a shared fully connected layer to generate spatial attention weights;

[0039] Performing an addition operation on the channel attention weights and the spatial attention weights to obtain local attention weights.

[0040] Further, the obtaining of the time-frequency map to be extracted and the bispectrum to be extracted of the original DRFM forwarding signal includes:

[0041] Converting the original time-frequency map of the original DRFM forwarding signal into a grayscale map to obtain the time-frequency map to be extracted;

[0042] Converting the original bispectrum of the original DRFM forwarding signal into a grayscale map to obtain the bispectrum to be extracted.

[0043] In a second aspect, the present invention further provides an apparatus for extracting multi-level features of radar signals, including:

[0044] At least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions, when executed by the processor, are used to execute the method for extracting multi-level features of radar signals in the first aspect.

[0045] In a third aspect, the present invention further provides a non-volatile computer storage medium storing computer-executable instructions that, when executed by one or more processors, are used to implement the method for extracting multi-level features of radar signals described in the first aspect.

[0046] In a fourth aspect, there is provided a computer program product containing instructions that, when run on a computer or a processor, cause the computer or the processor to execute the method for extracting multi-level features of radar signals as described in the first aspect.

[0047] In a fifth aspect, the present invention further provides a system for extracting multi-level features of radar signals, including the device for extracting multi-level features of radar signals as described in the second aspect, and using the method for extracting multi-level features of radar signals as described in the first aspect to complete the interaction of the device for extracting multi-level features of radar signals described in the second aspect.

[0048] Distinct from the prior art, the present invention has at least the following beneficial effects:

[0049] The present invention uses the time-frequency diagram and bispectrum diagram of the DRFM forwarding signal to extract the local detail features of the DRFM forwarding signal at multiple feature extraction scales, and models the global information of the DRFM forwarding signal. By fusing the local detail features and the global information, the features of the original DRFM forwarding signal at a deeper level are captured, and finally multi-level features are obtained, thereby improving the robustness and accuracy of signal recognition in a complex electromagnetic environment, and solving the problem that the recognition accuracy of signal categories drops sharply due to severe signal aliasing and unclear artificial features. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required to be used in the embodiments of the present invention. Obviously, the following described drawings are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0051] Figure 1 is a schematic flowchart of a method for extracting multi-level features of radar signals provided by an embodiment of the present invention;

[0052] Figure 2 is a schematic diagram of a specific example of the neural network structure of an embodiment of the present invention;

[0053] Figure 3 is a schematic flowchart of the first step 20 provided by an embodiment of the present invention;

[0054] Figure 4 It is a schematic flowchart of the first step 201a provided by an embodiment of the present invention;

[0055] Figure 5 It is a schematic diagram of a specific example of a multi-scale module provided by an embodiment of the present invention;

[0056] Figure 6 It is a schematic diagram of a specific example of another neural network structure of an embodiment of the present invention provided by an embodiment of the present invention;

[0057] Figure 7 It is a schematic flowchart of the second step 20 provided by an embodiment of the present invention;

[0058] Figure 8 It is a schematic flowchart of a step 202b provided by an embodiment of the present invention;

[0059] Figure 9 It is a schematic flowchart of the third step 20 provided by an embodiment of the present invention;

[0060] Figure 10 It is a schematic diagram of a specific example of respectively extracting first discrimination features in the channel dimension and the spatial dimension provided by an embodiment of the present invention;

[0061] Figure 11 It is a schematic diagram of a specific example of obtaining channel attention features provided by an embodiment of the present invention;

[0062] Figure 12 It is a schematic diagram of the architecture of a device for extracting multi-level features of radar signals provided by an embodiment of the present invention. Detailed implementation manners

[0063] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0064] Unless the context otherwise requires, throughout the specification and claims, the term "comprising" is to be construed in an open - inclusive sense, i.e., "including, but not limited to". In the description of the specification, terms such as "one embodiment", "some embodiments", "exemplary embodiments", "examples", "specific examples", or "some examples" are intended to indicate that specific features, structures, materials, or characteristics related to the embodiment or example are included in at least one embodiment or example of the present disclosure. The schematic representations of the above - mentioned terms do not necessarily refer to the same embodiment or example. In addition, the specific features, structures, materials, or characteristics may be included in any one or more embodiments or examples in any appropriate manner. That is, although they may be carried in the embodiments or examples of the above - mentioned terms due to reasons such as the order and position of appearance, it is not limited that they can be carried by one embodiment or example in a combined manner.

[0065] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by terms such as "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present disclosure and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation to the present disclosure.

[0066] In the description of the present invention, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the embodiments of the present disclosure, unless otherwise specified, the meaning of "a plurality" is two or more. In addition, for example, in the description, for the same type of nouns, the method of adding "A" and "B" at the end is used to describe them as two independent individuals. In this case, the features defined with "A" and "B" are only used for the purpose of distinguishing similar individuals in the description and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features.

[0067] In the description of some embodiments, expressions such as "coupled", "coupling", and "connected" and their derivatives may be used. For example, in the description of some embodiments, the term "connected" may be used to indicate that two or more components have direct physical or electrical contact with each other. Another example is that in the description of some embodiments, the term "coupled" may be used to indicate that two or more components have direct physical or electrical contact. However, the term "connected" or "coupled" may also mean that two or more components do not have direct contact with each other, but still cooperate or interact with each other, such as "optical path coupling", "wireless connection", etc. The embodiments disclosed herein are not necessarily limited to the content of the present invention.

[0068] In the description of the present invention, the expression "A and / or B" (where A and B are used to formally represent specific feature contents) will be involved, and the corresponding expression includes the following three combinations: only A, only B, and the combination of A and B.

[0069] The "about", "substantially", or "approximate" used in the present invention includes the stated value and the average value within an acceptable deviation range of the specific value, where the acceptable deviation range is determined by those of ordinary skill in the art considering the measurement being discussed and the errors associated with the measurement of the specific quantity (i.e., the limitations of the measurement system).

[0070] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0071] Embodiment 1:

[0072] In a complex electromagnetic environment, due to the development of modulation technology, multiple DRFM forwarding signals can enter the radar receiver simultaneously, and the received composite DRFM forwarding signal exacerbates the difficulty of identifying the DRFM forwarding signal. In strongly overlapping DRFM forwarding signals, the features of the DRFM forwarding signals cover each other, and useless noise will contaminate a large part of the feature image. Ordinary convolutional neural networks cannot capture useful features, and it is an urgent need to improve the feature extraction and characterization ability of the convolutional neural network.

[0073] The short-time Fourier transform captures the instantaneous frequency change through a localized time-frequency window. Since the frequency distributions of different types of DRFM forwarding signals are different at different time points, different DRFM forwarding signals can be better displayed in the time-frequency domain, and the differential features between the DRFM forwarding signals are intuitively visible.

[0074] However, for several types of DRFM retransmitted signals with gate pulling, such as Range Gate Pull-Off (RGPO), Range-Velocity Gate Pull-Off (RGPO-VGPO), Velocity Gate Pull-Off (VGPO), and Azimuth Gate Pull-Off (AGPO), the frequency components of the corresponding signals evolve over time in a similar process, making it difficult to distinguish them in the time-frequency domain. For this type of signal, designing methods for analyzing and extracting its effective features is particularly important for signal recognition.

[0075] In some cases, the differences between DRFM signals and real echoes are reflected in the modulation information carried by the DRFM retransmitted signals themselves, which generally originate from the fingerprint features retained during phase quantization or amplitude quantization when DRFM performs modulation; in other cases, the differences between DRFM retransmitted signals and real echoes will be manifested in the time domain, frequency domain, time-frequency domain, or other transform domains. Based on these subtle differences, some effective features can be extracted in each domain to improve the recognition effect of DRFM retransmitted signals.

[0076] In some solutions of the prior art, due to the fact that phase quantization will introduce a delay quantization effect, by utilizing this delay quantization effect, the fractional Fourier transform is used to detect the chirp rate of the received signal, and the difference between the DRFM forwarded signal and the real echo is compared to extract features and achieve the recognition of the DRFM forwarded signal. There are also some solutions that propose a velocity modulation DRFM forwarded signal recognition algorithm for digital radio frequency memories based on singular spectrum analysis. By using the difference in the singular value distributions of the real echo and the DRFM forwarded signal reflected in the harmonic components, the singular spectrum analysis is performed on the received signal, and the corresponding singular value statistical histogram is obtained by decomposition. Relevant features are extracted through this singular value statistical histogram for the recognition and classification of the DRFM forwarded signal. There are also some solutions that extract features of the DRFM forwarded signal in the time-frequency domain. For example, the time-frequency image of the received signal is obtained through the smoothed pseudo-Wigner-Ville distribution, and the texture features of the time-frequency image are extracted by using the local binary pattern algorithm of the image, realizing the recognition and classification of two types of signals, namely, the spectrum-diffused DRFM forwarded signal and the sliced combination DRFM forwarded signal. There are also some solutions that propose an identification algorithm for distance modulation signals, velocity modulation signals, and angle modulation signals based on bispectrum. By performing bispectrum estimation on these three DRFM forwarded signals, the variance and information entropy of the diagonal slices of the corresponding bispectrum distributions are taken as characteristic parameters to complete the classification of these three signals. There are also solutions that perform bispectrum transformation on the real target echo and three types of towing DRFM forwarded signals. After processing the DRFM forwarded signal by using the dimension reduction method of the vertical projection of the Y axis and normalization, the Renyi entropy, exponential entropy, and box dimension are extracted to form a feature set and put into a v-support vector classifier to obtain the corresponding feature vectors and perform recognition.

[0077] However, the above methods can only extract the shallow features of the DRFM forwarded signal reflected in the time domain, frequency domain, time-frequency domain, or other transform domains, and do not fully utilize the deep features of the signal, resulting in the radar system being unable to accurately identify and filter out unnecessary signals according to the extracted features in a complex electromagnetic environment.

[0078] To solve the above problems, as Figure 1 shown, the present invention provides a method for extracting multi-level features of radar signals, including:

[0079] Step 10: Obtain the time-frequency graph to be extracted and the bispectrum graph to be extracted of the original DRFM forwarded signal.

[0080] Among them, the time-frequency graph to be extracted is obtained after preprocessing the time-frequency graph of the original DRFM forwarded signal, and the time-frequency graph is the time-frequency domain image of the original DRFM forwarded signal; the bispectrum graph to be extracted is obtained after preprocessing the time-frequency graph of the original DRFM forwarded signal, and the bispectrum graph is the bispectrum domain image of the original DRFM forwarded signal.

[0081] In one embodiment, the original time-frequency diagram of the original DRFM forwarding signal can be converted into a grayscale diagram to obtain a time-frequency diagram to be extracted; the original bispectrum diagram of the original DRFM forwarding signal can be converted into a grayscale diagram to obtain a bispectrum diagram to be extracted.

[0082] In a complex electromagnetic environment, the features of the signals received by the radar are diverse and complex. It is difficult to distinguish the differences between composite DRFM forwarding signals using features in a single dimension. Therefore, it is necessary to extract multi-dimensional and multi-domain features to describe the DRFM forwarding signal, which helps with the subsequent identification of the type of DRFM forwarding signal. Among them, the composite DRFM forwarding signal refers to a composite signal composed of multiple types of DRFM forwarding signals, that is, the original DRFM forwarding signal in the embodiments of the present invention.

[0083] The embodiments of the present invention use the time-frequency diagram and bispectrum diagram of the DRFM forwarding signal to obtain the features of the signal. The projection of the bispectrum diagram of the DRFM forwarding signal onto two frequency components forms a plane. The corresponding plane can reflect the phase coupling relationship between different frequency components in the DRFM forwarding signal. The brightness or color represents the phase coupling degree of the frequency pair, and the nonlinear interaction between different frequency components is revealed through the phase relationship. Since the bispectrum diagram can effectively suppress Gaussian noise while retaining the amplitude and phase information of the signal, for DRFM forwarding signals with low distinguishability in the time-frequency domain, the embodiments of the present invention introduce the corresponding bispectrum diagram to capture the nonlinear features of the DRFM forwarding signal, and through multi-level feature learning and enhancement, achieve effective complementarity and deep fusion of cross-domain (i.e., time-frequency domain and bispectrum domain) features.

[0084] Step 20: Convolve the time-frequency diagram to be extracted at multiple feature extraction scales to extract the local detail features at each position in the time-frequency diagram to be extracted, obtaining a first distinguishing feature; model the global information in the time-frequency diagram to be extracted, obtaining a first global feature; fuse the first distinguishing feature and the first global feature to obtain the time-frequency diagram feature.

[0085] On the one hand, although different types of DRFM forwarding signals are difficult to distinguish in the time-frequency domain, there are essential differences in the subtle features and energy distributions in the time-frequency domain. Therefore, the embodiments of the present invention use multi-scale convolution to extract the local detail features of the time-frequency diagram to be extracted or the bispectrum diagram to be extracted at multiple feature extraction scales, so as to enhance the sensitivity of the neural network in the embodiments of the present invention to the local subtle features of the DRFM forwarding signal. In one embodiment, as Figure 2 shown, after the time-frequency diagram to be extracted is input into the multi-scale module, the multi-scale module convolves the time-frequency diagram to be extracted at multiple feature extraction scales to extract the local detail features at each position in the time-frequency diagram to be extracted, obtaining a first distinguishing feature.

[0086] On the other hand, in the prior art, since the position of the features important for the DRFM forwarding signal recognition task cannot be known in the signal, and in the DRFM forwarding signal, the context information representing the essence of different signal types is larger than the area occupied by the signal itself, the features helpful for recognizing the DRFM forwarding signal cannot be well extracted by only extracting artificial features from the signal. In the embodiments of the present invention, the global information in the time-frequency diagram to be extracted is modeled, and then the key features helpful for recognizing the DRFM forwarding signal, that is, the first global features, are extracted; and the key features are used as the context, combined with the local detail features (that is, the first discrimination features or the second discrimination features in the following text), to obtain the features helpful for recognizing the DRFM forwarding signal (that is, the time-frequency domain features or the bispectrum domain features in the following text) in the time-frequency diagram to be extracted or the bispectrum diagram to be extracted from the DRFM forwarding signal. In one embodiment, as Figure 2 shown, after the time-frequency diagram to be extracted is input into the multi-scale module, the global feature module models the global information in the time-frequency diagram to be extracted to obtain the first global features.

[0087] It should be noted that the multi-scale module and the global feature module are two independent neural network modules, and there is no sharing of network parameters between them.

[0088] Step 30: Convolve the bispectrum diagram to be extracted at multiple feature extraction scales to extract the local detail features at each position in the bispectrum diagram to be extracted, and obtain the second discrimination features; model the global information in the bispectrum diagram to be extracted to obtain the second global features; fuse the second discrimination features and the second global features to obtain the bispectrum diagram features.

[0089] As Figure 2 shown, the processing flow of the time-frequency diagram to be extracted and the bispectrum diagram to be extracted in the embodiments of the present invention is the same.

[0090] Step 40: Fuse the time-frequency diagram features and the bispectrum diagram features to obtain the multi-level features of the original DRFM forwarding signal.

[0091] The present invention uses the time-frequency diagram and the bispectrum diagram of the original DRFM forwarding signal to extract the local detail features of the signal at multiple feature extraction scales, and models the global information of the signal. By fusing the local detail features and the global information, the features of the original DRFM forwarding signal at a deeper level are captured, and finally multi-level features are obtained, so as to improve the robustness and accuracy of signal recognition in a complex electromagnetic environment, and solve the problem that the recognition accuracy of signal categories drops sharply due to severe signal aliasing and unclear artificial features.

[0092] The following is an example of a method for extracting multi-level features of radar signals in an embodiment of the present invention by processing the time-frequency graph to be extracted to obtain dual-spectrum features. The process of obtaining dual-spectrum features is the same as the process of obtaining time-frequency graph features, which will not be repeated here.

[0093] The following is an explanation of the processing flow of the multi-scale module for extracting local detail features of signals at multiple feature extraction scales. Figure 3 As shown, in step 20, the convolution of the time-frequency graph to be extracted at multiple feature extraction scales to extract local detail features of each position in the time-frequency graph to be extracted, and obtaining the first distinguishing feature includes:

[0094] Step 201a: extracting local detail features in parallel at multiple spatial resolutions of the time-frequency graph to be extracted, and obtaining multiple intermediate features of different scales.

[0095] like Figure 5 and Figure 6 As shown, different time and frequency resolutions can be obtained by using convolution kernels of different sizes to perform convolution operations on the time-frequency graph to be processed. Therefore, compared with traditional frequency domain or time domain analysis methods, the embodiments of the present invention can extract information of the original DRFM forwarding signal at different scales.

[0096] In one embodiment, if Figure 6 As shown, the methods of extracting local detail features in parallel at multiple spatial resolutions of the time-frequency graph to be extracted can be:

[0097] (1) Use a 2D convolution kernel of size 1×1 to perform a convolution operation on the time-frequency graph to be extracted (i.e., Figure 6 The “1×1Conv” in ) is used to obtain the intermediate features of the first scale.

[0098] (2) First, a 2D convolution kernel of size 3×3 is used to perform a convolution operation on the time-frequency graph to be extracted (i.e., Figure 6 The “3×3Conv” in the figure is used to obtain a preliminarily extracted feature map; a 2D convolution kernel with a size of 1×1 is then used to perform a convolution operation on the preliminarily extracted feature map (i.e., Figure 6 The “1×1Conv” in ) is used to obtain the intermediate features of the second scale.

[0099] (3) First, use a 2D convolution kernel of size 1×1 to perform a convolution operation on the time-frequency graph to be extracted (i.e., Figure 6 The “1×1Conv” in the figure is used to obtain a preliminarily extracted feature map; a 2D convolution kernel with a size of 5×5 is then used to perform a convolution operation on the preliminarily extracted feature map (i.e., Figure 6 The “5×5Conv” in ) obtains the intermediate features of the third scale.

[0100] (4) First, perform a convolution operation on the time-frequency map to be extracted using a two-dimensional convolution kernel of size 1×1 (i.e., the "1×1Conv" in Figure 6 ) to obtain a preliminarily extracted feature map; then, perform a convolution operation on this preliminarily extracted feature map using a two-dimensional convolution kernel of size 3×3 (i.e., the "3×3Conv" in Figure 6 ) to obtain a further extracted feature map; finally, perform a convolution operation on this further extracted feature map again using a two-dimensional convolution kernel of size 3×3 (i.e., the "3×3Conv" in Figure 6 ) to obtain the intermediate feature of the fourth scale.

[0101] It should be noted that the above convolution operation using a two-dimensional convolution kernel means: performing a standard convolution operation on the corresponding feature map, then performing batch normalization (abbreviated as BN) on the features obtained by convolution, and using a rectified linear unit (abbreviated as ReLu) activation function. Among them, the specific method of the standard convolution operation is selected by those skilled in the art with reference to the prior art and in combination with experience according to the specific usage scenario, and is not limited herein.

[0102] Step 202a: Perform feature splicing on the intermediate features of the multiple different scales to extract the signal category information of the first feature to be processed at each spatial resolution, and obtain a first discrimination feature.

[0103] In one embodiment, as shown in Figure 6 , perform feature splicing on the intermediate features of 4 different scales to obtain the first feature to be processed, and then perform a max pooling operation on this first feature to be processed to obtain the corresponding first discrimination feature.

[0104] In an alternative embodiment, for the multiple different scales of convolution kernels used above, the non-symmetric convolution method can also be adopted for a single convolution kernel. By decomposing a traditional large-size convolution kernel (for example, a convolution kernel of size 7×7) into two smaller non-symmetric convolution kernels (1×7 and 7×1), the feature expression ability and diversity of the first discrimination feature can be effectively enhanced, and the number of model parameters can be significantly reduced during the extraction process, thereby improving the training and inference efficiency.

[0105] The embodiment of the present invention extracts features in parallel at different spatial resolutions of the time-frequency map to be extracted, combines convolution kernels of different sizes and max pooling operations to capture the local detailed features of the DRFM forwarding signal; specifically, as shown in Figure 7 , the step 201a includes:

[0106] Step 2011: Convolve the to-be-extracted time-frequency map using a convolutional kernel of a first size to obtain intermediate features of a first scale.

[0107] Among them, the first size is selected by those skilled in the art according to the specific usage scenario; as Figure 5 shown, in one embodiment, the first size can be 1×1, that is, convolve using a 1×1 convolutional kernel to obtain intermediate features of size .

[0108] Step 2012: Convolve the to-be-extracted time-frequency map using a convolutional kernel of a second size to obtain first features; convolve the first features using a convolutional kernel of a first size to obtain intermediate features of a second scale.

[0109] Among them, the second size is selected by those skilled in the art according to the specific usage scenario; as Figure 5 shown, in one embodiment, the second size can be 3×3, that is, sequentially convolve using a 3×3 convolutional kernel and a 1×1 convolutional kernel to obtain intermediate features of size .

[0110] Step 2013: Convolve the to-be-extracted time-frequency map using a convolutional kernel of a first size to obtain second features; convolve the second features using a convolutional kernel of a third size to obtain intermediate features of a third scale.

[0111] Among them, the third size is selected by those skilled in the art according to the specific usage scenario; as Figure 5 shown, in one embodiment, the third size can be 5×5, that is, sequentially convolve using a 1×1 convolutional kernel and a 5×5 convolutional kernel to obtain intermediate features of size .

[0112] Step 2014: Convolve the to-be-extracted time-frequency map using a convolutional kernel of a first size to obtain third features; convolve the third features using a convolutional kernel of a second size to obtain fourth features; convolve the fourth features using a convolutional kernel of a second size to obtain intermediate features of a fourth scale.

[0113] As Figure 5 shown, in one embodiment, sequentially convolve using a 1×1 convolutional kernel, a 3×3 convolutional kernel, and a 3×3 convolutional kernel to obtain intermediate features of size .

[0114] In the embodiments of the present invention, through multi-scale convolution, rich features with signal category discrimination from fine textures to larger structures are captured, enabling the neural network model of the embodiments of the present invention to efficiently integrate multi-scale information without significantly increasing the computational burden. However, convolution is limited by its local characteristics, and the receptive fields of convolution kernels of different scales are all restricted within local neighborhoods, unable to capture the global context information in the signal, while global context information is often necessary for better identification of complex signal patterns. Therefore, the embodiments of the present invention also capture the local detailed features of the signal and model the global context information, so that the subsequent neural network model of the embodiments of the present invention can use the global context information to guide the learning of each local detailed feature, and thus better learn the features helpful for identifying the original DRFM forwarding signal. The processing flow of the global feature module for modeling the global information of the signal is described below, as Figure 7 shown, in step 20, the modeling of the global information in the to-be-extracted time-frequency diagram to obtain the first global feature includes:

[0115] Step 201b: Segment the to-be-extracted time-frequency diagram into fixed-size patches; convert each patch into a to-be-processed vector with a fixed dimension.

[0116] In one embodiment, as Figure 6 shown, after the input image (i.e., the to-be-extracted time-frequency diagram or the to-be-extracted bispectrum diagram) is segmented into fixed-size patches through a linear layer, each patch is converted into a to-be-processed vector (i.e., token) with a fixed dimension through a linear normalization layer, and a position encoding representing the relative position of each patch in the input image is added.

[0117] Step 202b: Determine the attention weights of the to-be-processed vectors, and perform attention calculation on the to-be-processed vectors using the attention weights to obtain the first encoded feature.

[0118] Then, the encoder is used to learn the features of the to-be-processed vectors, and this process will be described below.

[0119] Step 203b: Perform linear transformation and non-linear activation on the first encoded feature to learn the first encoded feature using a feed-forward neural network to obtain the second encoded feature.

[0120] Among them, the specific manner of learning the first encoded feature using a feed-forward neural network is determined by those skilled in the art with reference to the prior art according to the specific usage scenario, and will not be limited here.

[0121] Step 204b: Perform an addition operation on the first encoded feature and the second encoded feature to obtain the first global feature.

[0122] The global feature module of the embodiment of the present invention can model the global relationship of the entire input image, capture long-range dependencies and global information, and the global features extracted by the global feature module supplement the discriminative features captured by the above multi-scale module.

[0123] As Figure 6 shown, the embodiment of the present invention uses the multi-head attention mechanism to capture the global dependency relationship between different positions in the sequence of the vector to be processed by calculating the weighted sum of the query vector, the key vector and the value vector. Specifically, as Figure 8 shown, the step 202b includes:

[0124] Step 2021: Calculate the product of the vector to be processed and the query weight parameter to obtain a query vector; calculate the product of the vector to be processed and the key weight parameter to obtain an original key vector; calculate the product of the vector to be processed and the value weight parameter to obtain an original value vector.

[0125] Among them, the query weight parameter, the key weight parameter and the value weight parameter are all weight matrices.

[0126] Step 2022: Perform a convolution operation on the original key vector in the spatial dimension according to the reduction ratio to obtain a first reduced feature; perform a linear projection operation on the first reduced feature using the reduction ratio to compress the channel dimension of the first reduced feature to obtain a target key vector.

[0127] Among them, the reduction ratio is selected by those skilled in the art according to the specific usage scenario. In one embodiment, the original key vector is reduced to 1 / R of the original, where 1 / R is the reduction ratio.

[0128] Step 2023: Perform a convolution operation on the original value vector in the spatial dimension according to the reduction ratio to obtain a second reduced feature; perform a linear projection operation on the second reduced feature using the reduction ratio to compress the channel dimension of the second reduced feature to obtain a target value vector.

[0129] For example, for a feature map of size H×W, the spatial dimensions of K and V will be reduced to where H is the length of the feature map and W is the width of the feature map.

[0130] Step 2024: Calculate the dot product of the query vector and the target key vector to obtain an attention weight.

[0131] Step 2025: Use the attention weight to perform a weighted sum on the target value vector to generate a first encoded feature.

[0132] When calculating the weighted sum of the query vector, key vector, and value vector to obtain the first encoded feature, by first calculating the dot product of the query vector and the target key vector to obtain the attention weight, and then calculating the weighted sum of the attention weight and the value vector, the spatial dimension of the calculation is reduced, effectively alleviating the problems of computational complexity and memory consumption caused by high-dimensional data, and can greatly reduce the computational memory overhead; enabling the embodiments of the present invention to utilize the multi-head attention mechanism to capture long-distance interactions without increasing the number of parameters, and further enabling the neural network model of the embodiments of the present invention to more efficiently process large-scale DRFM forwarding signal data, significantly improving the efficiency of extracting global features during the training and inference processes, and without sacrificing the representational ability of the model.

[0133] As Figure 6 shown, the embodiments of the present invention also effectively merge the local detailed features captured by the multi-scale module and the global information extracted by the global feature module through a fusion module to further enhance the overall feature representation. Specifically, as Figure 9 shown, in step 20, the fusing the first discriminative feature and the first global feature to obtain the time-frequency map feature includes:

[0134] Step 201c: Extract the local detailed features of the first discriminative feature in the channel dimension and the spatial dimension respectively to obtain the local attention weight.

[0135] As Figure 10 shown, in one embodiment, average pooling is performed on the first discriminative feature in the channel dimension to obtain the first channel global feature; max pooling is performed on the first discriminative feature in the channel dimension to obtain the second channel global feature; the first channel global feature and the second channel global feature are input into a shared fully connected layer to generate the channel attention weight. Among them, the shared fully connected layer is used to perform feature fusion on the first channel global feature and the second channel global feature, and the specific implementation manner of the shared fully connected layer is selected by those skilled in the art according to the specific usage scenario. In an optional embodiment, a multi-layer perceptron (MLP for short) can be used to implement it. Similarly, average pooling is performed on the first discriminative feature in the spatial dimension to obtain the first spatial global feature; max pooling is performed on the first discriminative feature in the spatial dimension to obtain the second spatial global feature; the first spatial global feature and the second spatial global feature are input into a shared fully connected layer to generate the spatial attention weight. Finally, an addition operation is performed on the channel attention weight and the spatial attention weight to obtain the local attention weight.

[0136] Since the local detailed features captured by the scale module (i.e., the first discriminative feature or the second discriminative feature) contain the shallow features of the DRFM forwarding signal, and there are often noises in the shallow features, the embodiment of the present invention uses spatial attention as a spatial filter to enhance the local details and suppress the irrelevant regions. The corresponding attention weights are calculated in the spatial and channel dimensions to adaptively enhance the key features that contribute to identifying the DRFM forwarding signal category, so as to guide the neural network model of the embodiment of the present invention to learn more discriminative features during the training and inference processes.

[0137] Step 202c: Extract the global dependencies between different positions in the first global feature to obtain the channel attention feature.

[0138] In one embodiment, as Figure 11 shown, first perform average pooling on the first global feature (i.e., Figure 11 Avg-Pool in Figure 11 ). By average pooling, the average value of the local regions of each channel in the feature map is taken to extract the global information of the feature map and reduce the spatial dimension of the feature map, thereby reducing the spatial resolution of the feature map and the computational amount of subsequent operations. Then, perform a linear transformation on the compressed feature map through a fully connected layer (i.e., Figure 11 Fc in

[0139] ), mapping the input features to another space. Then, connect a non-linear activation function (e.g., Figure 11 the circle containing "S" in

[0139] represents the Sigmoid function) after the fully connected layer. By introducing non-linearity through the non-linear activation function, the expression ability of the neural network model of the embodiment of the present invention is enhanced.

[0140] Step 203c: Perform feature splicing on the local attention weight and the channel attention feature to obtain the time-frequency map feature.

[0141] At the same time, by paying attention to the relationships between the channels in the feature map through the channel attention weight, it is ensured that the most important channels captured by the global feature module can be highlighted, effectively presenting the global information.

[0142] By extracting the time-frequency diagram features and bispectrum features and effectively fusing the two, the present invention can extract deeper features of the DRFM forwarding signal in a complex electromagnetic environment, improve the accuracy and diversity of feature extraction, and thus improve the recognition accuracy of signal categories. It can also greatly enhance the adaptability of the neural network model of the embodiment of the present invention in a complex electromagnetic environment. Especially when facing strong signal overlap, it can extract effective information from the complex background.

[0143] Embodiment 2:

[0144] As Figure 12 shown, it is a schematic structural diagram of a device for extracting multi-level features of radar signals according to an embodiment of the present invention. The device for extracting multi-level features of radar signals in this embodiment includes one or more processors 21 and a memory 22. Among them, Figure 12 One processor 21 is taken as an example herein.

[0145] The processor 21 and the memory 22 can be connected through a bus or other means, Figure 12 Taking the connection through a bus as an example herein.

[0146] The memory 22, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs and non-volatile computer-executable programs, such as the method for extracting multi-level features of radar signals in this embodiment. The processor 21 executes the method for extracting multi-level features of radar signals by running the non-volatile software programs and instructions stored in the memory 22.

[0147] The memory 22 may include a high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the memory 22 may optionally include a memory remotely provided with respect to the processor 21, and these remote memories can be connected to the processor 21 through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0148] The program instructions / modules are stored in the memory 22 and, when executed by the one or more processors 21, execute the method for extracting multi-level features of radar signals in the above embodiment, for example, execute each step of the method for extracting multi-level features of radar signals according to the embodiment of the present invention described above.

[0149] The embodiment of the present invention also provides a non-volatile computer storage medium, and the computer storage medium stores computer-executable instructions, and the computer-executable instructions are executed by one or more processors, for example Figure 12A processor 21 can enable the above-mentioned one or more processors to execute the method for extracting multi-level features of radar signals in the specific implementation manners of the present invention. For example, it can execute each step of the method for extracting multi-level features of radar signals in the embodiments of the present invention described above; it can also implement Figure 12 the various modules and units described above; or execute the method for extracting multi-level features of radar signals in the specific implementation manners of the present invention. For example, it can execute each step of the method for extracting multi-level features of radar signals in the embodiments of the present invention described above; it can also implement Figure 12 the various modules and units described above.

[0150] It should be noted that, regarding the information interaction, execution process, etc. between the modules and units in the above-mentioned device and system, since they are based on the same concept as the method embodiment of the present invention, the specific content can be referred to the description in the method embodiment of the present invention and will not be elaborated here.

[0151] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the embodiments can be completed by instructing relevant hardware through a program. This program can be stored in a computer-readable storage medium, and the storage medium can include: read-only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, etc.

[0152] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for extracting multi-level features of radar signals, characterized in that Including: Obtaining the time-frequency diagram to be extracted and the bispectrum diagram to be extracted of the original DRFM forwarding signal; Convolving the time-frequency diagram to be extracted at multiple feature extraction scales to extract the local detail features at each position in the time-frequency diagram to be extracted, and obtaining the first discrimination feature; Modeling the global information in the time-frequency diagram to be extracted to obtain the first global feature; Fusing the first discrimination feature and the first global feature to obtain the time-frequency diagram feature; Convolving the bispectrum diagram to be extracted at multiple feature extraction scales to extract the local detail features at each position in the bispectrum diagram to be extracted, and obtaining the second discrimination feature; Modeling the global information in the bispectrum diagram to be extracted to obtain the second global feature; Fusing the second discrimination feature and the second global feature to obtain the bispectrum diagram feature; Fusing the time-frequency diagram feature and the bispectrum diagram feature to obtain the multi-level features of the original DRFM forwarding signal.

2. The method for extracting multi-level features of radar signals according to claim 1, wherein The step of convolving the time-frequency diagram to be extracted at multiple feature extraction scales to extract the local detail features at each position in the time-frequency diagram to be extracted, and obtaining the first discrimination feature includes: Parallelly extracting local detail features at multiple spatial resolutions of the time-frequency diagram to be extracted to obtain intermediate features of multiple different scales; Performing feature splicing on the intermediate features of multiple different scales to extract the signal category information of the first feature to be processed at each spatial resolution, and obtaining the first discrimination feature.

3. The method for extracting multi-level features of radar signals according to claim 2, wherein The step of parallelly extracting local detail features at multiple spatial resolutions of the time-frequency diagram to be extracted to obtain intermediate features of multiple different scales includes: Using a convolution kernel of the first size to convolve the time-frequency diagram to be extracted to obtain intermediate features of the first scale; Using a convolution kernel of the second size to convolve the time-frequency diagram to be extracted to obtain a first feature; using a convolution kernel of the first size to convolve the first feature to obtain intermediate features of the second scale; Using a convolution kernel of the first size to convolve the time-frequency diagram to be extracted to obtain a second feature; using a convolution kernel of the third size to convolve the second feature to obtain intermediate features of the third scale; Using a convolution kernel of the first size to convolve the time-frequency diagram to be extracted to obtain a third feature; using a convolution kernel of the second size to convolve the third feature to obtain a fourth feature; using a convolution kernel of the second size to convolve the fourth feature to obtain intermediate features of the fourth scale.

4. The method for extracting multi-level features of radar signals according to claim 1, wherein The step of modeling the global information in the time-frequency diagram to be extracted to obtain the first global feature includes: Dividing the time-frequency diagram to be extracted into fixed-size patches; converting each patch into a to-be-processed vector of a fixed dimension; Determining the attention weights of the to-be-processed vectors, and performing attention calculation on the to-be-processed vectors using the attention weights to obtain the first encoded feature; Performing linear transformation and non-linear activation on the first encoded feature to learn the first encoded feature using a feed-forward neural network to obtain the second encoded feature; Performing an addition operation on the first encoded feature and the second encoded feature to obtain the first global feature.

5. The method for extracting multi-level features of radar signals according to claim 4, wherein Determining the attention weights of the vector to be processed and using the attention weights to perform attention calculation on the vector to be processed to obtain the first encoded feature includes: Calculating the product of the vector to be processed and the query weight parameter to obtain a query vector; calculating the product of the vector to be processed and the key weight parameter to obtain an original key vector; calculating the product of the vector to be processed and the value weight parameter to obtain an original value vector; Performing a convolution operation on the original key vector in the spatial dimension according to a reduction ratio to obtain a first reduced feature; using the reduction ratio to perform a linear projection operation on the first reduced feature to compress the channel dimension of the first reduced feature to obtain a target key vector; Performing a convolution operation on the original value vector in the spatial dimension according to a reduction ratio to obtain a second reduced feature; using the reduction ratio to perform a linear projection operation on the second reduced feature to compress the channel dimension of the second reduced feature to obtain a target value vector; Calculating the dot product of the query vector and the target key vector to obtain attention weights; Using the attention weights to perform weighted summation on the target value vector to generate the first encoded feature.

6. The method for extracting multi-level features of radar signals according to claim 1, characterized in that, Fusing the first discriminative feature and the first global feature to obtain a time-frequency map feature includes: Extracting local detail features of the first discriminative feature in the channel dimension and the spatial dimension respectively to obtain local attention weights; Extracting the global dependence relationship between different positions in the first global feature to obtain a channel attention feature; Performing feature concatenation on the local attention weights and the channel attention feature to obtain a time-frequency map feature.

7. The method for extracting multi-level features of radar signals according to claim 6, wherein Extracting local detail features of the first discriminative feature in the channel dimension and the spatial dimension respectively to obtain local attention weights includes: Performing average pooling on the first discriminative feature in the channel dimension to obtain a first channel global feature; performing max pooling on the first discriminative feature in the channel dimension to obtain a second channel global feature; inputting the first channel global feature and the second channel global feature into a shared fully connected layer to generate channel attention weights; Performing average pooling on the first discriminative feature in the spatial dimension to obtain a first spatial global feature; performing max pooling on the first discriminative feature in the spatial dimension to obtain a second spatial global feature; inputting the first spatial global feature and the second spatial global feature into a shared fully connected layer to generate spatial attention weights; Performing an addition operation on the channel attention weights and the spatial attention weights to obtain local attention weights.

8. The method for extracting multi-level features of radar signals according to any one of claims 1-7, characterized in that, Obtaining the time-frequency map to be extracted and the bispectrum to be extracted of the original DRFM forwarding signal includes: Converting the original time-frequency map of the original DRFM forwarding signal into a grayscale image to obtain the time-frequency map to be extracted; Converting the original bispectrum of the original DRFM forwarding signal into a grayscale image to obtain the bispectrum to be extracted.

9. An apparatus for extracting multi-level features of radar signals, characterized in that, The apparatus for extracting multi-level features of radar signals includes at least one processor and a memory, the at least one processor and the memory are connected through a data bus, the memory stores instructions executable by the at least one processor, and after being executed by the processor, the instructions are used to implement the method for extracting multi-level features of radar signals according to any one of claims 1-8.

10. A non-volatile computer storage medium, characterized in that, The computer storage medium stores computer-executable instructions, and the computer-executable instructions are executed by one or more processors to complete the method for extracting multi-level features of radar signals according to any one of claims 1-8.

Images