MFI-QRAM-based lpi radar modulation pattern identification method

Abstract

The application discloses an LPI radar modulation pattern identification method based on MFI-QRAM, and the method comprises the following steps: constructing a fuzzy function image, a covariance matrix image and a short-time Fourier transform image of an LPI radar signal to be identified; generating a three-dimensional feature fusion image according to the fuzzy function image, the covariance matrix image and the short-time Fourier transform image through a multi-feature fusion model; performing identity mapping on the three-dimensional feature fusion image and residual mapping on the three-dimensional feature fusion image; and identifying the mapped identity feature image and the mapped residual image through a target modulation pattern identification model. In the foregoing manner, the three-dimensional feature fusion image is generated by using the multi-feature fusion model, then the identity mapping and the residual mapping are performed respectively, and finally the modulation pattern identification is performed by using the target modulation pattern identification model trained by the attention module and the residual module, so that the accuracy of identifying the modulation pattern can be effectively improved.

Description

Technical Field

[0001] This invention relates to the field of equipment construction image recognition technology, and in particular to an LPI radar modulation pattern recognition method based on MFI-QRAM. Background Technology

[0002] LPI radar modulation pattern identification based on intra-pulse features mainly involves two aspects: selecting the identification object and determining the specific identification strategy. Regarding object selection, considerations include representational capability, extraction complexity, sensitivity to signal-to-noise ratio (SNR), intra-class convergence, and inter-class separation ability. Furthermore, based on the time-frequency resolution requirements of the radar radiation source pulse design, fuzzy function analysis methods that comprehensively consider signal delay and Doppler shift have become the main methods for LPI radar modulation pattern identification in recent years. However, the above identification methods suffer from drawbacks such as difficulty in effectively extracting comprehensive intra-pulse features and a lack of strong discriminative intra-pulse features. Additionally, the extracted features exhibit significant performance degradation at low SNR levels and cannot be applied to scenarios with high SNR, ultimately resulting in low accuracy in modulation pattern identification.

[0003] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main objective of this invention is to provide an LPI radar modulation pattern recognition method based on MFI-QRAM, which aims to solve the technical problem of low accuracy in modulation pattern recognition in existing technologies.

[0005] To achieve the above objectives, this invention provides an LPI radar modulation pattern recognition method based on MFI-QRAM, the method comprising the following steps:

[0006] Acquire the LPI radar signal to be identified, and construct the fuzzy function image, covariance matrix image, and short-time Fourier transform image of the LPI radar signal to be identified;

[0007] A three-dimensional feature fusion image is generated based on the fuzzy function image, covariance matrix image, and short-time Fourier transform image using a multi-feature fusion model.

[0008] Identity mapping is performed on the 3D feature fusion image, and residual mapping is performed on the 3D feature fusion image;

[0009] The modulation pattern of the LPI radar signal to be identified is obtained by recognizing the mapped identity feature image and the mapped residual image through the target modulation pattern recognition model.

[0010] Optionally, constructing the fuzzy function image of the LPI radar signal to be identified includes:

[0011] Acquire the LPI radar signal to be identified, and determine the sampling data and signal ambiguity function of the LPI radar signal to be identified;

[0012] The joint distribution data of the LPI radar signal to be identified in the time and frequency domains are determined based on the signal ambiguity function.

[0013] Construct a target-dimensional image matrix based on the sampled data and the joint distribution data;

[0014] The target dimension image matrix is ​​normalized to obtain a blurred function image.

[0015] Optionally, constructing the covariance matrix image of the LPI radar signal to be identified includes:

[0016] Determine the signal interception coefficient, and segment the LPI radar signal to be identified according to the signal interception coefficient;

[0017] The segmented LPI radar signals to be identified are arranged sequentially according to the column vector arrangement strategy to generate a target signal matrix.

[0018] The covariance matrix is ​​solved based on the target signal matrix, and a covariance matrix image is constructed based on the covariance matrix.

[0019] A short-time Fourier transform image is constructed based on the LPI radar signal to be identified.

[0020] Optionally, constructing the short-time Fourier transform image of the LPI radar signal to be identified includes:

[0021] Perform complex domain transformation on the LPI radar signal to be identified;

[0022] The transformed LPI radar signal to be identified is extracted using a time-domain function;

[0023] The intercepted LPI radar signal to be identified is subjected to Fourier transform to obtain the frequency of each segment of the LPI radar signal to be identified.

[0024] The signal frequency distribution is determined based on the frequency of each segment of the LPI radar signal to be identified.

[0025] A short-time Fourier transform feature carrier is constructed based on the signal frequency distribution, and parameters are set for the short-time Fourier transform feature carrier to construct a short-time Fourier transform image.

[0026] Optionally, the step of generating a three-dimensional feature fusion image based on the fuzzy function image, covariance matrix image, and short-time Fourier transform image using a multi-feature fusion model includes:

[0027] The blurred function image, covariance matrix image, and short-time Fourier transform image are converted respectively using a resolution conversion strategy;

[0028] Noise reduction is performed on the transformed blurred function image, covariance matrix image, and short-time Fourier transform image, respectively;

[0029] Channel conversion is performed on the denoised blurred function image, covariance matrix image, and short-time Fourier transform image using a target channel conversion strategy.

[0030] A three-dimensional feature fusion image is obtained by fusing the blurred function image, covariance matrix image, and short-time Fourier transform image after channel conversion using a multi-feature fusion model.

[0031] Optionally, before identifying the mapped identity feature image and the mapped residual image through the target modulation pattern recognition model to obtain the modulation pattern of the LPI radar signal to be identified, the method further includes:

[0032] A quasi-residual structure is generated based on a residual mapping module containing three convolutional layers, an identity mapping module with one convolutional layer, and the weights of the identity mapping module.

[0033] The target attention module is embedded in each mapping module of the quasi-residual structure to obtain the current residual structure;

[0034] A spatial AM is embedded at the input of the current residual structure, and a channel AM is embedded at the output of the current residual structure;

[0035] Based on the current embedded residual structure, a target modulation pattern recognition model is trained using a set of modulation pattern samples from historical radar signals.

[0036] Optionally, after training the target modulation pattern recognition model based on the modulation pattern sample set of historical radar signals according to the embedded current residual structure, the method further includes:

[0037] Obtain the modulation pattern recognition test set;

[0038] The target modulation pattern recognition model is tested according to the modulation pattern recognition test set to obtain the current test result;

[0039] Determine the current test accuracy based on the current test results;

[0040] When the current test accuracy is greater than or equal to a preset accuracy threshold, the mapped identity feature image and the mapped residual image are identified by the target modulation style recognition model.

[0041] Furthermore, to achieve the above objectives, the present invention also proposes an LPI radar modulation pattern recognition device based on MFI-QRAM, wherein the MFI-QRAM-based LPI radar modulation pattern recognition device comprises:

[0042] The acquisition module is used to acquire the LPI radar signal to be identified and construct the fuzzy function image, covariance matrix image and short-time Fourier transform image of the LPI radar signal to be identified.

[0043] The generation module is used to generate a three-dimensional feature fusion image based on the fuzzy function image, covariance matrix image, and short-time Fourier transform image using a multi-feature fusion model;

[0044] The mapping module is used to perform identity mapping on the 3D feature fusion image and residual mapping on the 3D feature fusion image.

[0045] The identification module is used to identify the mapped identity feature image and the mapped residual image through a target modulation pattern identification model to obtain the modulation pattern of the LPI radar signal to be identified.

[0046] Furthermore, to achieve the above objectives, the present invention also proposes an LPI radar modulation pattern recognition device based on MFI-QRAM. The LPI radar modulation pattern recognition device based on MFI-QRAM includes: a memory, a processor, and an LPI radar modulation pattern recognition program based on MFI-QRAM stored in the memory and executable on the processor. The LPI radar modulation pattern recognition program based on MFI-QRAM is configured to implement the LPI radar modulation pattern recognition method based on MFI-QRAM as described above.

[0047] Furthermore, to achieve the above objectives, the present invention also proposes a storage medium storing an LPI radar modulation pattern recognition program based on MFI-QRAM. When the LPI radar modulation pattern recognition program based on MFI-QRAM is executed by a processor, it implements the LPI radar modulation pattern recognition method based on MFI-QRAM as described above.

[0048] The proposed method for LPI radar modulation pattern recognition based on MFI-QRAM involves acquiring the LPI radar signal to be identified and constructing its fuzzy function image, covariance matrix image, and short-time Fourier transform image. A multi-feature fusion model generates a three-dimensional feature fusion image based on these images. Identity mapping and residual mapping are then performed on the 3D feature fusion image. Finally, a target modulation pattern recognition model identifies the mapped identity feature image and the mapped residual image to obtain the modulation pattern of the LPI radar signal. By constructing the fuzzy function image, covariance matrix image, and short-time Fourier transform image of the LPI radar signal, a multi-feature fusion model generates a three-dimensional feature fusion image, followed by identity mapping and residual mapping. Finally, a target modulation pattern recognition model trained by an attention module and a residual module is used to identify the modulation pattern, thereby effectively improving the accuracy of modulation pattern recognition. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the structure of an LPI radar modulation pattern recognition device based on MFI-QRAM in the hardware operating environment of the embodiment of the present invention;

[0050] Figure 2 This is a flowchart illustrating the first embodiment of the LPI radar modulation pattern recognition method based on MFI-QRAM of the present invention.

[0051] Figure 3 This is a schematic diagram of the target modulation pattern recognition model in an embodiment of the LPI radar modulation pattern recognition method based on MFI-QRAM of the present invention.

[0052] Figure 4 This is a flowchart illustrating the second embodiment of the LPI radar modulation pattern recognition method based on MFI-QRAM of the present invention;

[0053] Figure 5 This is a functional module diagram of the first embodiment of the LPI radar modulation pattern recognition device based on MFI-QRAM of the present invention.

[0054] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0055] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0056] Reference Figure 1 , Figure 1This is a schematic diagram of the structure of an LPI radar modulation pattern recognition device based on MFI-QRAM, which is part of the hardware operating environment of the embodiment of the present invention.

[0057] like Figure 1 As shown, the MFI-QRAM-based LPI radar modulation pattern recognition device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wireless-Fidelity (Wi-Fi) interface). The memory 1005 may be high-speed random access memory (RAM) or stable non-volatile memory (NVM), such as a disk storage device. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0058] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the LPI radar modulation pattern recognition device based on MFI-QRAM, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0059] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a network communication module, a user interface module, and an LPI radar modulation pattern recognition program based on MFI-QRAM.

[0060] exist Figure 1In the MFI-QRAM-based LPI radar modulation pattern recognition device shown, the network interface 1004 is mainly used for data communication with the network integrated platform workstation; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and memory 1005 in the MFI-QRAM-based LPI radar modulation pattern recognition device of the present invention can be set in the MFI-QRAM-based LPI radar modulation pattern recognition device. The MFI-QRAM-based LPI radar modulation pattern recognition device calls the MFI-QRAM-based LPI radar modulation pattern recognition program stored in the memory 1005 through the processor 1001, and executes the MFI-QRAM-based LPI radar modulation pattern recognition method provided in the embodiment of the present invention.

[0061] Based on the above hardware structure, an embodiment of the LPI radar modulation pattern recognition method based on MFI-QRAM of the present invention is proposed.

[0062] Reference Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the LPI radar modulation pattern recognition method based on MFI-QRAM of the present invention.

[0063] In the first embodiment, the LPI radar modulation pattern recognition method based on MFI-QRAM includes the following steps:

[0064] Step S10: Obtain the LPI radar signal to be identified, and construct the fuzzy function image, covariance matrix image, and short-time Fourier transform image of the LPI radar signal to be identified.

[0065] It should be noted that the execution subject of this embodiment is an LPI radar modulation pattern recognition device based on MFI-QRAM, but it can also be other devices that can achieve the same or similar functions, such as recognition controllers. This embodiment does not limit this, and in this embodiment, the recognition controller is used as an example for explanation.

[0066] It should be understood that the LPI radar signal to be identified refers to the radar signal that requires modulation pattern identification. Commonly used radar signals are generally real narrowband signals, and their expression is:

[0067] s(t)=a(t)cos[2πf c t+θ(t)].

[0068] Here, it is usually assumed that a(t) is an ideal rectangular pulse envelope with amplitude A and time length τ; f c This is the radar carrier frequency; when θ(t) is a constant, the signal s(t) is called a conventional pulse signal, and when θ(t) is a function of t, it is called a modulating signal. Hand f L To meet the frequency threshold for signal transmission, B is the signal operating bandwidth. When θ(t) is considered as a whole, then θ is used. m This is represented as a phase-modulated signal, when f m When expressed as a frequency modulation signal, it is called a frequency-modulated signal. LPI radar mainly achieves ultra-wide bandwidth and ultra-low sidelobes through various modulation methods.

[0069] It is understandable that after obtaining the LPI radar signal to be identified, the fuzzy function image, covariance matrix image, and short-time Fourier transform image of the LPI radar signal to be identified are constructed respectively. The fuzzy function image refers to the image represented by the fuzzy function, the covariance matrix image refers to the image represented by the covariance matrix of the LPI radar signal to be identified, and the short-time Fourier transform image refers to the image constructed after the short-time Fourier transform.

[0070] Further, step S10 includes: acquiring the LPI radar signal to be identified, determining the sampling data and signal ambiguity function of the LPI radar signal to be identified; determining the joint distribution data of the LPI radar signal to be identified in the time domain and frequency domain according to the signal ambiguity function; constructing a target dimension image matrix according to the sampling data and the joint distribution data; and performing normalization processing on the target dimension image matrix to obtain the ambiguity function image.

[0071] It is understandable that the sampling data includes, but is not limited to, sampling duration, number of sampling points, and sampling frequency. The signal ambiguity function can be an important means of characterizing the range and frequency resolution of the LPI radar signal to be identified, and is defined as:

[0072]

[0073] Among them, s * (t) is the conjugate of s(t), τ is the time delay, and f d This is due to the Doppler frequency shift.

[0074] It should be understood that the signal ambiguity function is the inverse Fourier transform of the instantaneous autocorrelation function of the LPI radar signal to be identified with respect to time, that is, the two-dimensional joint distribution data of the autocorrelation function in the time and frequency domains. Then, the target-dimensional image matrix is ​​constructed by combining the sampled data. Specifically, for a duration of T, N sampling points, and a sampling frequency of f... s For an LPI radar signal to be identified with a working bandwidth of B, a uniform bandwidth B = f can be selected. s / 10, define τ as the equidistant sampling N within the interval [-T, T], f d For each equidistant sample N within the interval [-B / 2, B / 2], construct an N-dimensional square matrix AF. NThen, the target dimension image matrix is ​​normalized to generate a blurred function image of the LPI radar signal to be identified.

[0075] Further, step S10 includes: determining signal interception coefficients; segmenting the LPI radar signal to be identified according to the signal interception coefficients; arranging the segmented LPI radar signal segments to be identified sequentially according to the column vector arrangement strategy to generate a target signal matrix; solving the covariance matrix according to the target signal matrix; and constructing a covariance matrix image according to the covariance matrix.

[0076] It should be understood that the signal truncation coefficient refers to the coefficients of the segmented LPI radar signal to be identified. Since different modulation styles of the LPI radar signal to be identified have different autocorrelation functions at different times of delay and frequency shift, the correlation between segments within the signal also varies. In this case, the covariance matrix can be used to characterize the correlation between segments within the signal. Specifically: for an LPI radar signal s(t) with N sampling points, a signal truncation coefficient n≥2 is selected. To reduce the Gibbs phenomenon, a Hamming window of length s = floor(N / n) (floor represents rounding down) is used for truncation. To refine the characterization of the correlation between segments within the LPI radar signal to be identified, the overlap length between adjacent segments is l = s - n + 1. The truncated segments of the LPI radar signal to be identified are arranged sequentially as column vectors, ultimately forming an s×s target signal matrix A. s Since the covariance matrix can effectively characterize the correlation between segments within the LPI radar signal to be identified, and this correlation is less sensitive to noise, the covariance matrix is ​​solved based on the target signal matrix.

[0077] Further, step S10 includes: performing a complex domain transformation on the LPI radar signal to be identified; truncating the transformed LPI radar signal to be identified using a time-domain function; performing a Fourier transform on the truncated LPI radar signal to be identified to obtain the frequency of each segment of the LPI radar signal to be identified; determining the signal frequency distribution based on the frequency of each segment of the LPI radar signal to be identified; constructing a short-time Fourier transform feature carrier based on the signal frequency distribution, and setting parameters for the short-time Fourier transform feature carrier to construct a short-time Fourier transform image.

[0078] It is understandable that since the LPI radar signals to be identified are distinguished by constant phase when the difference parameter is the same, it is necessary to perform complex domain transformation on the LPI radar signals to be identified. Linear time-frequency analysis mainly uses a time-domain window function to extract the signal and uses Fourier transform to solve for the frequency of each segment. The value is the frequency of each segment of the signal, rather than the difference. The purpose of this embodiment is to characterize the difference in frequency distribution of signals with different modulation patterns. That is, a short-time Fourier transform feature carrier can be constructed by using a fixed window length and step size according to the signal frequency distribution.

[0079] It should be understood that after constructing the short-time Fourier transform image, according to the working principle of STFT, its main parameters are the window length l and the sliding step size S. Through the analysis of the polyphase coded signal in TABLE 1, it can be seen that when they have the same carrier frequency and compression rate, the absolute values ​​of the phase difference between P3 and P4 signals are the same when they have the same time delay, and their instantaneous autocorrelation functions are the same. Therefore, the obtained ambiguity function values ​​are also the same. The phase change rate difference gradually increases in the first half of the signal. Considering both frequency resolution and time resolution, S = 4 and l = round(N / 10) are selected as the parameters set for the short-time Fourier transform feature carrier of the generated signal to construct the short-time Fourier transform image.

[0080] Step S20: Generate a three-dimensional feature fusion image based on the fuzzy function image, covariance matrix image, and short-time Fourier transform image using a multi-feature fusion model.

[0081] It is understandable that a multi-feature fusion model refers to a model that performs multi-feature fusion. This multi-feature fusion model can be an MFI model. In order to solve the shortcomings of fuzzy function images, covariance matrix images and short-time Fourier transform images in signal recognition, it is necessary to generate a fusion image with multiple features as the input of the target modulation pattern recognition model. The fusion image with multiple features generated at this time is a three-dimensional feature fusion image. The size of the fuzzy function image, covariance matrix image and short-time Fourier transform image is 32×32×1, and the size of the multi-feature fusion model is 32×32×3.

[0082] Step S30: Perform identity mapping on the three-dimensional feature fusion image and perform residual mapping on the three-dimensional feature fusion image.

[0083] It should be understood that after obtaining the 3D feature fusion image, identity mapping and residual mapping are performed on the 3D feature fusion image respectively. The identity mapping module, as input, retains almost complete information about the fuzzy function image, based on its strong discriminative ability for most LPI radar signals to be identified. The covariance matrix image and short-time Fourier transform image are inputs to the residual mapping module. The residual mapping module extracts features to help the fuzzy function image identify specific LPI radar signals to be identified, without considering the loss of some information.

[0084] Step S40: The mapped identity feature image and the mapped residual image are identified by the target modulation pattern recognition model to obtain the modulation pattern of the LPI radar signal to be identified.

[0085] It should be understood that the target modulation pattern recognition model refers to a model used to identify the modulation pattern of LPI radar signals. The structure of the target modulation pattern recognition model consists of the current residual structure and the target attention module. The target attention module is used for feature enhancement and noise suppression, and there are various types of target attention modules.

[0086] Furthermore, before step S40, the method further includes: generating a quasi-residual structure based on a residual mapping module containing three convolutional layers, an identity mapping module with one convolutional layer, and the weights of the identity mapping module; embedding a target attention module into each mapping module in the quasi-residual structure to obtain the current residual structure; embedding a spatial AM at the input of the current residual structure and embedding a channel AM at the output of the current residual structure; and training a target modulation pattern recognition model based on the modulation pattern sample set of historical radar signals according to the embedded current residual structure.

[0087] It is understandable that a quasi-residual structure refers to a structure composed of a residual mapping module, an identity mapping module, and the weights of the identity mapping module. The residual mapping module contains three convolutional layers, and the identity mapping module contains one convolutional layer. Therefore, the current residual structure represents an improvement over a typical residual structure. Then, by utilizing salient feature extraction and salient region segmentation, the main purpose is to reduce the impact of noise and enhance effective features. Based on the recognition features of each layer, different types of AMs are embedded into the current residual structure. A spatial AM is embedded at the input of the current residual structure for denoising the original image and enhancing effective features. A channel AM is embedded before the output of the current residual structure for feature selection. The three network layers are (16, 1×1), (16, 3×3), and (32, 1×1), and the identity mapping convolutional layer is (32, 3×3). The Dense layer has 32 neurons, and the Dropout parameter is 0.3.

[0088] Furthermore, after training the target modulation pattern recognition model based on the modulation pattern sample set of historical radar signals according to the embedded current residual structure, the method further includes: obtaining a modulation pattern recognition test set; testing the target modulation pattern recognition model according to the modulation pattern recognition test set to obtain the current test result; determining the current test accuracy based on the current test result; and when the current test accuracy is greater than or equal to a preset accuracy threshold, recognizing the mapped identity feature image and the mapped residual image through the target modulation pattern recognition model.

[0089] It should be understood that after training the target modulation pattern recognition model, it is necessary to test the accuracy of the target modulation pattern recognition model in recognizing the modulation pattern of the LPI radar signal to be identified. The target modulation pattern recognition model needs to be tested using a modulation pattern recognition test set, and then it is determined whether the current test accuracy is greater than or equal to the preset accuracy threshold. If so, it indicates that the target modulation pattern recognition model is qualified in the recognition dimension. At this time, the target modulation pattern recognition model is used to identify the mapped identity feature image and the mapped residual image.

[0090] Understandably, it's also necessary to calculate the loss value of the target modulation pattern recognition model. This embodiment uses TensorBoard for parameter optimization, specifically: TensorBoard initialization involves two main parts. The first part is determining the value range of the six parameters. The second part is the parameter initialization for EarlyStopping and ReduceLROnPlateau used for training. Then, using different combinations of the six parameters to train the recognition network, the PSR, loss function value, and the number of iterations required to reach the minimum loss function value are recorded. The optimal combination of parameters is determined by comparing the training results. For example, by comparing the recognition accuracy and loss function value on the validation set under a total of 972 combinations, the optimal training results are: PSR and training set loss of 99.77% and 0.0103, respectively, and 99.51% and 0.0172 on the validation set, respectively. That is, the optimal combination is used to train the target modulation pattern recognition model to ensure the minimum loss value.

[0091] It should be noted that the reference Figure 3 , Figure 3The schematic diagram of the target modulation style recognition model is shown. Specifically, it includes two spatial attention modules, three network layers, one channel attention module, a residual mapping module, an identity mapping module, a Flatten network layer, a Dense network layer, a Dropout network layer, and a Softmax network layer. One spatial attention module is connected to a convolutional layer, and the other spatial attention module is connected to the three network layers. The Flatten network layer, Dense network layer, Dropout network layer, and Softmax network layer are connected sequentially.

[0092] This embodiment acquires the LPI radar signal to be identified and constructs its fuzzy function image, covariance matrix image, and short-time Fourier transform image. A multi-feature fusion model generates a three-dimensional feature fusion image based on these images. Identity mapping and residual mapping are then performed on the three-dimensional feature fusion image. A target modulation pattern recognition model identifies the mapped identity feature image and the mapped residual image to obtain the modulation pattern of the LPI radar signal to be identified. By constructing the fuzzy function image, covariance matrix image, and short-time Fourier transform image of the LPI radar signal to be identified, a multi-feature fusion model generates a three-dimensional feature fusion image, and then identity mapping and residual mapping are performed. Finally, a target modulation pattern recognition model trained by an attention module and a residual module is used to identify the modulation pattern, thereby effectively improving the accuracy of modulation pattern recognition.

[0093] In one embodiment, such as Figure 4 The second embodiment of the LPI radar modulation pattern recognition method based on MFI-QRAM proposed in the first embodiment of the present invention includes step S20, which includes:

[0094] Step S201: The blurred function image, covariance matrix image, and short-time Fourier transform image are converted respectively using a resolution conversion strategy.

[0095] It should be understood that resolution conversion strategy refers to the strategy of converting the resolution of different images to the same resolution. This resolution conversion strategy can be a linear interpolation conversion strategy or a resampling strategy. After conversion, the blurred function image, covariance matrix image and short-time Fourier transform image are grayscale images with the same resolution.

[0096] Step S202: Denoise reduction is performed on the transformed blurred function image, covariance matrix image, and short-time Fourier transform image, respectively.

[0097] Step S203: Channel conversion is performed on the denoised blurred function image, covariance matrix image, and short-time Fourier transform image respectively using the target channel conversion strategy.

[0098] It should be understood that the target channel conversion strategy refers to the strategy of converting to a three-channel image. After obtaining the denoised blurred function image, covariance matrix image and short-time Fourier transform image, the target channel conversion strategy is used to perform channel conversion. After conversion, the denoised blurred function image, covariance matrix image and short-time Fourier transform image are three-channel images.

[0099] Step S204: The blurred function image, covariance matrix image and short-time Fourier transform image after channel conversion are fused by a multi-feature fusion model to obtain a three-dimensional feature fusion image.

[0100] It is understandable that after obtaining the blurred function image, covariance matrix image, and short-time Fourier transform image after channel conversion, the multi-feature fusion model is used to perform image fusion to obtain a three-dimensional feature fusion image. At this time, the three-dimensional feature fusion image includes all the features of the LPI radar signal to be identified, that is, the modulation pattern of the LPI radar signal to be identified can be identified by using the three-dimensional feature fusion image.

[0101] This embodiment uses a resolution conversion strategy to convert the blurred function image, covariance matrix image, and short-time Fourier transform image respectively; it then performs noise reduction on the converted blurred function image, covariance matrix image, and short-time Fourier transform image respectively; finally, it performs channel conversion on the denoised blurred function image, covariance matrix image, and short-time Fourier transform image respectively using a target channel conversion strategy; and finally, it fuses the channel-converted blurred function image, covariance matrix image, and short-time Fourier transform image using a multi-feature fusion model to obtain a three-dimensional feature fusion image. By using the above method, resolution conversion is performed using a resolution conversion strategy, followed by noise reduction on the converted blurred function image, covariance matrix image, and short-time Fourier transform image respectively, then three-channel conversion is performed using a target channel conversion strategy, and finally, the channel-converted blurred function image, covariance matrix image, and short-time Fourier transform image are fused into a three-dimensional feature fusion image, thereby effectively improving the accuracy of obtaining the three-dimensional feature fusion image.

[0102] Furthermore, this embodiment of the invention also proposes a storage medium storing an LPI radar modulation pattern recognition program based on MFI-QRAM. When the LPI radar modulation pattern recognition program based on MFI-QRAM is executed by a processor, it implements the steps of the LPI radar modulation pattern recognition method based on MFI-QRAM as described above.

[0103] Since this storage medium adopts all the technical solutions of all the above embodiments, it has at least all the beneficial effects brought about by the technical solutions of the above embodiments, which will not be repeated here.

[0104] In addition, refer to Figure 5 This invention also proposes an LPI radar modulation pattern recognition device based on MFI-QRAM, the MFI-QRAM-based LPI radar modulation pattern recognition device comprising:

[0105] The acquisition module 10 is used to acquire the LPI radar signal to be identified and construct the fuzzy function image, covariance matrix image and short-time Fourier transform image of the LPI radar signal to be identified.

[0106] The generation module 20 is used to generate a three-dimensional feature fusion image based on the fuzzy function image, the covariance matrix image, and the short-time Fourier transform image using a multi-feature fusion model.

[0107] The mapping module 30 is used to perform identity mapping on the three-dimensional feature fusion image and residual mapping on the three-dimensional feature fusion image.

[0108] The identification module 40 is used to identify the mapped identity feature image and the mapped residual image through the target modulation pattern identification model to obtain the modulation pattern of the LPI radar signal to be identified.

[0109] This embodiment acquires the LPI radar signal to be identified and constructs its fuzzy function image, covariance matrix image, and short-time Fourier transform image. A multi-feature fusion model generates a three-dimensional feature fusion image based on these images. Identity mapping and residual mapping are then performed on the three-dimensional feature fusion image. A target modulation pattern recognition model identifies the mapped identity feature image and the mapped residual image to obtain the modulation pattern of the LPI radar signal to be identified. By constructing the fuzzy function image, covariance matrix image, and short-time Fourier transform image of the LPI radar signal to be identified, a multi-feature fusion model generates a three-dimensional feature fusion image, and then identity mapping and residual mapping are performed. Finally, a target modulation pattern recognition model trained by an attention module and a residual module is used to identify the modulation pattern, thereby effectively improving the accuracy of modulation pattern recognition.

[0110] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this invention. In practical applications, those skilled in the art can select some or all of the workflow to achieve the purpose of this embodiment according to actual needs, and no restrictions are imposed here.

[0111] In addition, for technical details not described in detail in this embodiment, please refer to the LPI radar modulation pattern recognition method based on MFI-QRAM provided in any embodiment of the present invention, which will not be repeated here.

[0112] In one embodiment, the acquisition module 10 is further configured to acquire the LPI radar signal to be identified, determine the sampling data and signal ambiguity function of the LPI radar signal to be identified, determine the joint distribution data of the LPI radar signal to be identified in the time domain and frequency domain according to the signal ambiguity function, construct a target dimension image matrix according to the sampling data and the joint distribution data, and perform normalization processing on the target dimension image matrix to obtain the ambiguity function image.

[0113] In one embodiment, the acquisition module 10 is further configured to determine signal interception coefficients, segment the LPI radar signal to be identified according to the signal interception coefficients, arrange the segmented LPI radar signal to be identified in sequence according to the column vector arrangement strategy to generate a target signal matrix, solve the covariance matrix according to the target signal matrix, and construct a covariance matrix image according to the covariance matrix.

[0114] In one embodiment, the acquisition module 10 is further configured to perform complex domain transformation on the LPI radar signal to be identified; truncate the transformed LPI radar signal to be identified using a time-domain function; perform Fourier transform on the truncated LPI radar signal to be identified to obtain the frequency of each segment of the LPI radar signal to be identified; determine the signal frequency distribution based on the frequency of each segment of the LPI radar signal to be identified; construct a short-time Fourier transform feature carrier based on the signal frequency distribution, and set parameters for the short-time Fourier transform feature carrier to construct a short-time Fourier transform image.

[0115] In one embodiment, the generation module 20 is further configured to convert the blurred function image, covariance matrix image, and short-time Fourier transform image respectively using a resolution conversion strategy; to denoise the converted blurred function image, covariance matrix image, and short-time Fourier transform image respectively; to perform channel conversion on the denoised blurred function image, covariance matrix image, and short-time Fourier transform image respectively using a target channel conversion strategy; and to fuse the channel-converted blurred function image, covariance matrix image, and short-time Fourier transform image using a multi-feature fusion model to obtain a three-dimensional feature fusion image.

[0116] In one embodiment, the recognition module 40 is further configured to generate a quasi-residual structure based on a residual mapping module containing three convolutional layers, an identity mapping module with one convolutional layer, and the weights of the identity mapping module; embed a target attention module into each mapping module in the quasi-residual structure to obtain a current residual structure; embed a spatial AM at the input of the current residual structure and embed a channel AM at the output of the current residual structure; and train a target modulation pattern recognition model based on the modulation pattern sample set of historical radar signals according to the embedded current residual structure.

[0117] In one embodiment, the recognition module 40 is further configured to acquire a modulation pattern recognition test set; test the target modulation pattern recognition model according to the modulation pattern recognition test set to obtain a current test result; determine the current test accuracy according to the current test result; and when the current test accuracy is greater than or equal to a preset accuracy threshold, recognize the mapped identity feature image and the mapped residual image through the target modulation pattern recognition model.

[0118] Other embodiments or implementation methods of the LPI radar modulation pattern recognition device based on MFI-QRAM described in this invention can be referred to the above-described method embodiments, and will not be repeated here.

[0119] It should be understood that although the steps in the flowcharts of this application's embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0120] Furthermore, it should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0121] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0122] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory (ROM) / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, all-in-one platform workstation, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0123] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for identifying LPI radar modulation patterns based on MFI-QRAM, characterized in that, The method includes the following steps: Acquire the LPI radar signal to be identified, and construct the fuzzy function image, covariance matrix image, and short-time Fourier transform image of the LPI radar signal to be identified; A three-dimensional feature fusion image is generated based on the fuzzy function image, covariance matrix image, and short-time Fourier transform image using a multi-feature fusion model. Identity mapping is performed on the 3D feature fusion image, and residual mapping is performed on the 3D feature fusion image; The modulation pattern of the LPI radar signal to be identified is obtained by recognizing the mapped identity feature image and the mapped residual image through the target modulation pattern recognition model. Before identifying the mapping of the identity feature image and the mapping of the residual image through the target modulation pattern recognition model to obtain the modulation pattern of the LPI radar signal to be identified, the method further includes: A quasi-residual structure is generated based on a residual mapping module containing three convolutional layers, an identity mapping module with one convolutional layer, and the weights of the identity mapping module. The target attention module is embedded in each mapping module of the quasi-residual structure to obtain the current residual structure; A spatial AM is embedded at the input of the current residual structure, and a channel AM is embedded at the output of the current residual structure; Based on the current embedded residual structure, a target modulation pattern recognition model is trained using a set of modulation pattern samples from historical radar signals.

2. The LPI radar modulation pattern recognition method based on MFI-QRAM as described in claim 1, characterized in that, The construction of the fuzzy function image of the LPI radar signal to be identified includes: Acquire the LPI radar signal to be identified, and determine the sampling data and signal ambiguity function of the LPI radar signal to be identified; The joint distribution data of the LPI radar signal to be identified in the time and frequency domains are determined based on the signal ambiguity function. Construct a target-dimensional image matrix based on the sampled data and the joint distribution data; The target dimension image matrix is ​​normalized to obtain a blurred function image.

3. The LPI radar modulation pattern recognition method based on MFI-QRAM as described in claim 1, characterized in that, The construction of the covariance matrix image of the LPI radar signal to be identified includes: Determine the signal interception coefficient, and segment the LPI radar signal to be identified according to the signal interception coefficient; The segmented LPI radar signals to be identified are arranged sequentially according to the column vector arrangement strategy to generate a target signal matrix. The covariance matrix is ​​solved based on the target signal matrix, and a covariance matrix image is constructed based on the covariance matrix. A short-time Fourier transform image is constructed based on the LPI radar signal to be identified.

4. The LPI radar modulation pattern recognition method based on MFI-QRAM as described in claim 1, characterized in that, The construction of the short-time Fourier transform image of the LPI radar signal to be identified includes: Perform complex domain transformation on the LPI radar signal to be identified; The transformed LPI radar signal to be identified is extracted using a time-domain function; The intercepted LPI radar signal to be identified is subjected to Fourier transform to obtain the frequency of each segment of the LPI radar signal to be identified. The signal frequency distribution is determined based on the frequency of each segment of the LPI radar signal to be identified. A short-time Fourier transform feature carrier is constructed based on the signal frequency distribution, and parameters are set for the short-time Fourier transform feature carrier to construct a short-time Fourier transform image.

5. The LPI radar modulation pattern recognition method based on MFI-QRAM as described in claim 1, characterized in that, The step of generating a three-dimensional feature fusion image based on the fuzzy function image, covariance matrix image, and short-time Fourier transform image using a multi-feature fusion model includes: The blurred function image, covariance matrix image, and short-time Fourier transform image are converted respectively using a resolution conversion strategy; Noise reduction is performed on the transformed blurred function image, covariance matrix image, and short-time Fourier transform image, respectively; Channel conversion is performed on the denoised blurred function image, covariance matrix image, and short-time Fourier transform image using a target channel conversion strategy. A three-dimensional feature fusion image is obtained by fusing the blurred function image, covariance matrix image, and short-time Fourier transform image after channel conversion using a multi-feature fusion model.

6. The LPI radar modulation pattern recognition method based on MFI-QRAM as described in claim 1, characterized in that, After training the target modulation pattern recognition model based on the modulation pattern sample set of historical radar signals according to the embedded current residual structure, the method further includes: Obtain the modulation pattern recognition test set; The target modulation pattern recognition model is tested according to the modulation pattern recognition test set to obtain the current test result; Determine the current test accuracy based on the current test results; When the current test accuracy is greater than or equal to a preset accuracy threshold, the mapped identity feature image and the mapped residual image are identified by the target modulation style recognition model.

7. A device for identifying LPI radar modulation patterns based on MFI-QRAM, characterized in that, The MFI-QRAM-based LPI radar modulation pattern recognition device includes: The acquisition module is used to acquire the LPI radar signal to be identified and construct the fuzzy function image, covariance matrix image and short-time Fourier transform image of the LPI radar signal to be identified. The generation module is used to generate a three-dimensional feature fusion image based on the fuzzy function image, covariance matrix image, and short-time Fourier transform image using a multi-feature fusion model; The mapping module is used to perform identity mapping on the 3D feature fusion image and residual mapping on the 3D feature fusion image. The identification module is used to identify the mapped identity feature image and the mapped residual image through the target modulation pattern identification model to obtain the modulation pattern of the LPI radar signal to be identified. The recognition module is further configured to generate a quasi-residual structure based on a residual mapping module containing three convolutional layers, an identity mapping module with one convolutional layer, and the weights of the identity mapping module; embed a target attention module into each mapping module in the quasi-residual structure to obtain the current residual structure; embed a spatial AM at the input of the current residual structure and embed a channel AM at the output of the current residual structure; and train a target modulation pattern recognition model based on the modulation pattern sample set of historical radar signals according to the embedded current residual structure.

8. An LPI radar modulation pattern recognition device based on MFI-QRAM, characterized in that, The device includes: a memory, a processor, and an MFI-QRAM-based LPI radar modulation pattern recognition program stored in the memory and executable on the processor, wherein the MFI-QRAM-based LPI radar modulation pattern recognition program is configured to implement the MFI-QRAM-based LPI radar modulation pattern recognition method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium stores an LPI radar modulation pattern recognition program based on MFI-QRAM. When the LPI radar modulation pattern recognition program based on MFI-QRAM is executed by the processor, it implements the LPI radar modulation pattern recognition method based on MFI-QRAM as described in any one of claims 1 to 6.

Images