Repetition frequency type identification method and device based on graph domain mapping and feature enhancement

Abstract

The invention provides a repetition frequency type identification method and device based on graph domain mapping and feature enhancement, and relates to the technical field of radar signal processing. The method comprises the following steps: determining a pulse repetition frequency sequence corresponding to screened pulse repetition frequencies; taking the reciprocal of the pulse repetition frequency sequence to obtain a pulse repetition interval sequence; segmenting the pulse repetition interval sequence into a plurality of subsequences according to a preset length; according to a graph domain mapping algorithm, converting the plurality of subsequences into a Gramer difference angle field image, and dividing the image into a training sample and a test sample; constructing a repetition frequency type identification network, inputting the training sample into the repetition frequency type identification network, and training the repetition frequency type identification network to obtain a trained repetition frequency type identification network; and inputting a test sample into the trained repetition frequency type identification network, and identifying a plurality of repetition frequency types. In this way, the recognition accuracy and robustness of the repetition frequency type are improved in a complex non-ideal scene with high-proportion leakage pulses and false pulses.

Description

Technical Field

[0001] The present invention relates to the field of radar signal processing technology, and in particular to a method and device for identifying repetition frequency types based on image domain mapping and feature enhancement. Background Art

[0002] Analyzing the pulse repetition frequency (PRF) is the core part of radar emitter analysis. PRF is defined as the number of pulses emitted per second, which is the reciprocal of the pulse repetition interval (PRI). It is one of the most complex parameters in radar emitters. PRF usually includes data of six types of repetition frequencies, including fixed, jitter, staggered, sliding, group variable, and sine. The diversity of repetition frequency types reflects the various working modes, uses, and technical characteristics of radars, and fully characterizes the timing structure of radar signals. Accurate analysis of repetition frequency types is an important part of in-depth exploration of radar signal characteristics and inference of radar working modes. In addition, the analysis of repetition frequency types helps to invert radar signal models, provide interference basis and feedback information for radar signal prediction and radar countermeasure strategy formulation, so it is of great significance to study the repetition frequency type recognition of radar emitters. The recognition of repetition frequency types combined with working mode labels can be used to identify radar working modes. Repetition frequency type recognition focuses on the identification of signal repetition frequency characteristics and can be targeted at various types of radars.

[0003] At present, the recognition method based on supervised neural network is usually used to realize the recognition of repetition type. However, this method needs to start with sequence processing. In complex non-ideal scenarios with a high proportion of missed pulses and false pulses, the difference of sequences is not large. Directly inputting sequences with small differences into supervised neural network recognition effect is not good, resulting in low recognition accuracy and poor robustness of repetition type in complex non-ideal scenarios with a high proportion of missed pulses and false pulses. Summary of the invention

[0004] The purpose of the embodiments of the present invention is to provide a method and device for identifying repetition frequency types based on image domain mapping and feature enhancement, so as to solve the problem of low recognition accuracy and poor robustness of repetition frequency types in complex non-ideal scenarios with a high proportion of missed pulses and false pulses.

[0005] To solve the above technical problems, the embodiments of the present invention provide the following technical solutions:

[0006] A first aspect of the present invention provides a method for identifying repetition types based on image domain mapping and feature enhancement, the method comprising:

[0007] Screening a plurality of pulse repetition frequencies according to a preset threshold range, and determining a pulse repetition frequency sequence corresponding to the screened pulse repetition frequencies;

[0008] The pulse repetition frequency sequence is processed by taking the inverse thereof to obtain a pulse repetition interval sequence;

[0009] The pulse repetition interval sequence is divided according to a preset length to obtain a plurality of subsequences;

[0010] According to the image domain mapping algorithm, multiple subsequences are transformed into corresponding Grammar difference angular field images, and the multiple Grammar difference angular field images are divided into training samples and test samples;

[0011] Construct a repetition type recognition network, which includes a convolutional neural sub-network and an attention mechanism layer;

[0012] Inputting the training samples into the repetition type recognition network, training the repetition type recognition network, and obtaining a trained repetition type recognition network;

[0013] The test sample is input into the trained repetition type recognition network to identify multiple repetition types.

[0014] A second aspect of the present application provides a device for identifying repetition types based on image domain mapping and feature enhancement, the device comprising:

[0015] A screening module, used to screen a plurality of pulse repetition frequencies according to a preset threshold range, and determine a pulse repetition frequency sequence corresponding to the screened pulse repetition frequencies;

[0016] A processing module, used for performing reciprocal processing on the pulse repetition frequency sequence to obtain a pulse repetition interval sequence;

[0017] A segmentation module, used for segmenting the pulse repetition interval sequence according to a preset length to obtain a plurality of subsequences;

[0018] A transformation module, used for transforming a plurality of subsequences into corresponding Grammar difference angular field images according to an image domain mapping algorithm, and dividing the plurality of Grammar difference angular field images into training samples and test samples;

[0019] A building block for building a repetition type recognition network, which includes a convolutional neural sub-network and an attention mechanism layer;

[0020] A training module, used for inputting training samples into the repetition type recognition network, training the repetition type recognition network, and obtaining a trained repetition type recognition network;

[0021] The recognition module is used to input the test sample into the trained repetition frequency type recognition network to identify multiple repetition frequency types.

[0022] A third aspect of the present application provides a computer-readable storage medium, the storage medium including a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute the method for repetition type recognition based on image domain mapping and feature enhancement according to the above-mentioned first aspect or any optional embodiment of the first aspect.

[0023] Compared with the prior art, the repetition type recognition method and device based on image domain mapping and feature enhancement provided by the present invention converts a PRF sequence with unclear differences in complex non-ideal scenarios with a high proportion of missed pulses and false pulses into a Grammar difference angle field map with more significant feature differences. The repetition type recognition network constructed by the convolutional neural sub-network and the attention mechanism layer helps to highlight the important features of the image and strengthen the role of important features in classification. In complex non-ideal scenarios with a high proportion of missed pulses and false pulses, the Grammar difference angle field map with more significant feature differences is used to strengthen the repetition type recognition network with important features, thereby improving the recognition accuracy and robustness of the repetition type. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] By reading the detailed description below with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0025] Figure 1 The schematic diagram shows the process of the repetition type recognition method based on image domain mapping and feature enhancement. Figure 1 ;

[0026] Figure 2 The schematic diagram shows the process of the repetition type recognition method based on image domain mapping and feature enhancement. Figure 2 ;

[0027] Figure 3 Schematically showing the PRI sequence distribution diagram corresponding to the repetition frequency type and the corresponding Gram difference angle field diagram;

[0028] Figure 4 The confusion matrix of the test for all non-ideal samples is schematically shown;

[0029] Figure 5 The structure of a device for identifying repetition types based on image domain mapping and feature enhancement is schematically shown. DETAILED DESCRIPTION

[0030] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0031] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in the present invention should have the common meanings understood by those skilled in the art to which the present invention belongs.

[0032] The method in the embodiment of the present invention is described in detail below.

[0033] Figure 1 The flowchart of the method for identifying the repetition type based on image domain mapping and feature enhancement in the embodiment of the present invention is schematically shown. Figure 1 As shown, the method may include:

[0034] S101. Screen multiple pulse repetition frequencies according to a preset threshold range, and determine a pulse repetition frequency sequence corresponding to the screened pulse repetition frequencies.

[0035] The threshold ranges of the PRFs emitted by different radars are different. The preset threshold range can be determined according to the model of the radar. Different models of radars correspond to different preset threshold ranges, and the preset threshold range is not specifically limited here.

[0036] The pulse repetition frequencies exceeding the preset threshold range are eliminated, and the pulse repetition frequencies within the preset threshold range are retained to determine the screened PRF, and a PRF sequence is constructed for the screened PRF.

[0037] S102, performing reciprocal processing on the pulse repetition frequency sequence to obtain a pulse repetition interval sequence.

[0038] By taking the reciprocal of the PRF sequence, the PRF sequence can be processed into a PRI sequence, which is convenient for subsequent processing.

[0039] S103 . Divide the pulse repetition interval sequence according to a preset length to obtain a plurality of subsequences.

[0040] Using non-overlapping sliding windows, the PRI sequence is segmented according to a preset length to obtain a plurality of subsequences. A longer PRI sequence can be segmented into a plurality of PRI subsequences of a preset length.

[0041] The preset length can be 64 or 128, and there is no specific limitation on the preset length.

[0042] S104. According to an image domain mapping algorithm, the multiple subsequences are transformed into corresponding Grammar difference angular field images, and the multiple Grammar difference angular field images are divided into training samples and test samples.

[0043] The image domain mapping algorithm can be used to transform multiple one-dimensional PRI subsequences into two-dimensional images, namely Gram difference angular field images.

[0044] One subsequence corresponds to one Gram difference angular field image, and the number of Gram difference angular field images is the same as the number of subsequences.

[0045] The multiple Grammar difference angle field images are divided into training samples and test samples, and a part of all the Grammar difference angle field images can be randomly selected as training samples and another part as test samples. The number of Grammar difference angle field images included in the training samples and the test samples can be manually set.

[0046] S105, constructing a repetition frequency type recognition network.

[0047] Among them, the repetition type recognition network includes a convolutional neural sub-network and an attention mechanism layer.

[0048] By introducing the attention mechanism layer in the convolutional neural sub-network, it helps to highlight the important features of the Grammar difference angular field image, suppress the influence of minor features, and improve the classification effect.

[0049] S106: Input the training samples into the repetition frequency type recognition network, train the repetition frequency type recognition network, and obtain a trained repetition frequency type recognition network.

[0050] The Gram difference angle field image in the training sample is input into the repetition type recognition network, the repetition type recognition network is trained, and the loss function is iteratively optimized until convergence to obtain a trained repetition type recognition network.

[0051] S107: Input the test sample into the trained repetition frequency type recognition network to identify multiple repetition frequency types.

[0052] Based on the above Figure 1It can be seen from the implementation method that the embodiment of the present invention screens multiple pulse repetition frequencies according to a preset threshold range, determines a pulse repetition frequency sequence corresponding to the screened pulse repetition frequencies; takes the inverse of the pulse repetition frequency sequence to obtain a pulse repetition interval sequence; divides the pulse repetition interval sequence according to a preset length to obtain multiple subsequences; according to the image domain mapping algorithm, transforms the multiple subsequences into corresponding Grammar difference angular field images, and divides the multiple Grammar difference angular field images into training samples and test samples; constructs a repetition type recognition network, inputs the training samples into the repetition type recognition network, trains the repetition type recognition network, and obtains a trained repetition type recognition network; inputs the test samples into the trained repetition type recognition network to identify multiple repetition types. In this way, the PRF sequence with unclear difference in complex non-ideal scenes with a high proportion of missed pulses and false pulses is converted into a Grammar difference angular field map with more significant feature difference. The repetition type recognition network constructed by the convolutional neural sub-network and the attention mechanism layer helps to highlight the important features of the image and strengthen the role of important features in classification. In complex non-ideal scenes with a high proportion of missed pulses and false pulses, the Grammar difference angular field map with more significant feature difference is used to strengthen the repetition type recognition network of important features, so as to improve the recognition accuracy and robustness of the repetition type.

[0053] As a refinement and expansion of the above embodiment, Figure 2 The flow chart of a method for identifying repetition types based on image domain mapping and feature enhancement in an embodiment of the present invention is as follows: Figure 2 , see Figure 2 As shown, a method for identifying repetition types based on image domain mapping and feature enhancement provided by an embodiment of the present invention may include:

[0054] S201. Screen multiple pulse repetition frequencies according to a preset threshold range, and determine a pulse repetition frequency sequence corresponding to the screened pulse repetition frequencies.

[0055] The threshold ranges of the PRFs emitted by different radars are different. The preset threshold range can be determined according to the model of the radar. Different models of radars correspond to different preset threshold ranges, and the preset threshold range is not specifically limited here.

[0056] The presence of a certain number of outliers in the input PRF may cause the feature space after subsequent domain mapping to be distorted, making the originally dissimilar data points closer in the transformed feature space, making the differences smaller, thus causing the performance of the subsequent model to deteriorate, the stability to decrease, and ultimately affecting the recognition effect. In order to improve the final recognition accuracy, the preset threshold range is set based on the radar model in the preprocessing stage, and the outliers that exceed the preset threshold range are removed. After removing the outliers, the PRF that does not exceed the preset threshold range can be retained to determine the screened PRF, and a PRF sequence is constructed for the screened PRF.

[0057] S202, performing reciprocal processing on the pulse repetition frequency sequence to obtain a pulse repetition interval sequence.

[0058] By taking the reciprocal of the PRF sequence, the PRF sequence can be processed into a PRI sequence, which is convenient for subsequent processing.

[0059] S203 , dividing the pulse repetition interval sequence according to a preset length to obtain a plurality of subsequences.

[0060] The preset length can be 64 or 128, and there is no specific limitation on the preset length.

[0061] Using non-overlapping sliding windows, the PRI sequence is segmented according to a preset length to obtain a plurality of subsequences. A longer PRI sequence can be segmented into a plurality of PRI subsequences of a preset length.

[0062] The following steps S204 to S207 are specific operations of transforming multiple subsequences into corresponding Grammar difference angular field images according to the image domain mapping algorithm, and dividing the multiple Grammar difference angular field images into training samples and test samples.

[0063] S204, scaling the multiple subsequences to a range of [-1, 1] to obtain a scaled value in each subsequence.

[0064] Specifically, multiple subsequences are scaled to the range of [-1, 1] to obtain the scaled values ​​in each subsequence, including:

[0065] Using the following first formula, multiple subsequences are scaled to the range of [-1,1] to obtain the scaled values ​​in each subsequence:

[0066]

[0067] in, The corresponding p n The scaled value, that is, the value after the nth value in each subsequence is scaled, p nis the nth value in each subsequence, P is a number of subsequences, and N is the length of the subsequence.

[0068] S205 , converting each value into a polar coordinate system to obtain the angle and radius corresponding to each value.

[0069] The scaled value of the nth value in each subsequence is converted to the polar coordinate system according to the following formula, that is, the value is regarded as the angle in the polar coordinate system, and the timestamp corresponding to each value is regarded as the radius in the polar coordinate system.

[0070] Specifically, use the following formula to convert each value into a polar coordinate system to obtain the angle and radius corresponding to each value:

[0071]

[0072] in, is the angle corresponding to the nth value in the polar coordinate system, The corresponding p n The scaled value, that is, the value after the nth value in each subsequence is scaled, r n is the radius corresponding to the nth value, t n is the timestamp corresponding to the nth value, and N is the length of the subsequence.

[0073] S206: Determine a Gram difference angular field image according to the angle difference between every two values.

[0074] The angular difference between points corresponding to different values ​​is considered to identify the event correlation at different time points, and the Gramian Angular Difference Field (GADF) image can be determined using the difference angle formula.

[0075] Specifically, the GADF image is determined according to the angle difference between each two values, including:

[0076] Step A1: Determine the corresponding matrix according to the angle difference between each two values ​​and the following second formula:

[0077]

[0078] Among them, GADF is the matrix corresponding to the angle between each two values, is the angle corresponding to the nth value in the polar coordinate system, is the angle corresponding to the mth value in the polar coordinate system; I is the unit matrix of the same size as the subsequence P, is the normalized subsequence, for The transposed matrix of .

[0079] Finally, the GADF calculation is converted into an inner product-like operation in the rectangular coordinate system.

[0080] Through the above processing, the PRF sequence with unclear difference shape in complex non-ideal scenes with a high proportion of missed pulses and false pulses can be converted into a Gram difference angular field image with significant corresponding features. Figure 3 The PRI sequence distribution diagram corresponding to the repetition frequency type and the corresponding Gram difference angle field diagram are schematically shown. Specifically, Figure 3 (a) is the PRI sequence diagram corresponding to the fixed type PRF. Figure 3 (d) is the Gram difference angle field diagram corresponding to the fixed type PRF, Figure 3 (b) is the PRI sequence diagram corresponding to the jitter type PRF, Figure 3 (e) is the Gram difference angle field diagram corresponding to the jitter type PRF, Figure 3 (c) is the PRI sequence diagram corresponding to the different types of PRF, Figure 3 (f) is the Gram difference angle field diagram corresponding to the staggered type PRF, Figure 3 (g) is the PRI sequence diagram corresponding to the slip type PRF, Figure 3 (j) is the Gram difference angle field diagram corresponding to the slip type PRF, Figure 3 (h) is the PRI sequence diagram corresponding to the group edge type PRF, Figure 3 (k) is the Gram difference angle field diagram corresponding to the group edge type PRF, Figure 3 (i) is the PRI sequence diagram corresponding to the sinusoidal PRF, Figure 3 (l) is the Gram difference angle field diagram corresponding to the sinusoidal type PRF. Figure 3 (a)- Figure 3 (l) shows the original PRI sequence distribution of 6 different repetition frequency types and their corresponding Grammar difference angle field diagrams. By comparing the PRI sequence diagrams of the same repetition frequency type with the Grammar difference angle field diagrams, Figure 3 (a) and Figure 3 (d) For comparison, Figure 3 (b) with Figure 3 (e) For comparison, Figure 3 (c) with Figure 3 (f) For comparison, Figure 3 (g) with Figure 3 (j) for comparison, Figure 3 (h) with Figure 3 (k) for comparison, Figure 3 (i) with Figure 3By comparing (l), it can be seen that after converting each PRI sequence into the corresponding Gram difference angle field map, the redundancy of data information is effectively reduced, the aggregation of differential features of different repetition frequency types is enhanced, and it is helpful for the learning process of deep neural network.

[0081] Step A2: Perform grayscale conversion on the matrix to obtain a Gram difference angular field image.

[0082] One subsequence corresponds to one Gram difference angular field image, and the number of Gram difference angular field images is the same as the number of subsequences.

[0083] S207, dividing the multiple Gram difference angle field images into training samples and test samples.

[0084] The multiple Grammar difference angle field images are divided into training samples and test samples, and a part of all the Grammar difference angle field images can be randomly selected as training samples and another part as test samples. The number of Grammar difference angle field images included in the training samples and the test samples can be manually set.

[0085] S208, constructing a repetition frequency type recognition network.

[0086] The repetition type recognition network includes a convolutional neural subnetwork and an attention mechanism layer. By introducing the attention mechanism layer into the convolutional neural subnetwork, it helps to highlight the important features of the Gram difference angular field image, suppress the influence of minor features, and improve the classification effect.

[0087] The convolutional neural network consists of 2 convolutional layers, 2 normalization layers, 2 activation layers, 2 pooling layers, and 2 fully connected layers. Among them, the convolutional layer and the pooling layer are responsible for feature extraction and dimensionality reduction; the activation layer increases the nonlinearity of the network and enhances the feature representation ability of the model; the normalization layer is usually after the activation layer to prevent overfitting; the fully connected layer converts the two-dimensional features into one-dimensional features.

[0088] The attention mechanism layer can be called a channel feature enhancement module. The channel feature enhancement module adopts the attention mechanism and is an end-to-end network structure. The module mainly includes two operations, compression and excitation, for processing input features.

[0089] The repetition type recognition network consists of 1 convolution layer, 1 normalization layer, 1 activation layer, 1 pooling layer, 1 attention mechanism layer, 1 convolution layer, 1 normalization layer, 1 activation layer, 1 pooling layer and 2 fully connected layers.

[0090] In order to solve the problem of low recognition rate of repetition type in complex non-ideal scenarios with high proportion of missed pulses and false pulses, the constructed repetition type recognition network adds an attention mechanism layer between the two convolutional layers of the convolutional neural sub-network, compresses the output feature map of the first convolutional layer into a feature vector through global average pooling operation, and then generates a channel weight vector, which allows the neural network to focus on feature channels with large differences through weighting. Then, the adaptively weighted feature map is input into the second convolutional layer, which is conducive to screening and strengthening the features before the convolutional layer, thereby improving the expression ability of the entire network.

[0091] S209: Input the training samples into the repetition frequency type recognition network, train the repetition frequency type recognition network, and obtain a trained repetition frequency type recognition network.

[0092] The Gram difference angle field image in the training sample is input into the repetition type recognition network, the repetition type recognition network is trained, and the loss function is iteratively optimized until convergence to obtain a trained repetition type recognition network.

[0093] S210: Input the test sample into the trained repetition frequency type recognition network to identify multiple repetition frequency types.

[0094] The trained repetition type recognition network consists of 1 convolution layer, 1 normalization layer, 1 activation layer, 1 pooling layer, 1 attention mechanism layer, 1 convolution layer, 1 normalization layer, 1 activation layer, 1 pooling layer and 2 fully connected layers.

[0095] Specifically, the test sample is input into the trained repetition type recognition network to identify multiple repetition types, including:

[0096] Step B1: Input the test sample into 1 convolution layer, 1 normalization layer, 1 activation layer, and 1 pooling layer in sequence, and output the input features corresponding to the test sample.

[0097] Step B2: Input the input features into 1 attention mechanism layer and output the final feature representation.

[0098] Specifically, the input features are input into an attention mechanism layer, and the final feature representation is output, including:

[0099] Step B21: compress the input features to obtain channel-level global features.

[0100] Step B22: Excite the global features at the channel level to obtain the weights of each channel.

[0101] Step B23: Multiply the weight of each channel by the input feature to obtain the final feature representation.

[0102] Step B3: The final feature representation is sequentially input into 1 convolution layer, 1 normalization layer, 1 activation layer, 1 pooling layer, and 2 fully connected layers to output multiple repetition frequency types.

[0103] In order to study the influence of different measurement errors, missed pulses and wrong pulse levels on the repetition type identification, the effect of the method of the present invention is verified by simulation experiments as follows.

[0104] The simulation conditions are as follows: the experiment combines different measurement errors, missed pulses and wrong pulse levels, and simulates data of six repetition frequency types: fixed, jitter, staggered, slip, dwell and switch, and sine. The PRF value ranges of different modulation types are shown in Table 1. Among them, 38,400 pulses are generated for each modulation type; the ratio of missed pulses and false pulses is set between 10% and 60%; the measurement error is achieved by adding Gaussian white noise with a mean of 0 and a variance of 2us to each pulse.

[0105] Table 1 PRF value range for different modulation types

[0106] Repetition Type Value range (KHz) Constant 0.5～10 Jitter 0.5～10(10％) Sine (Sin) 0.5～3.3 Dwell & Switch 2.5～3.3 Stagger 0.5～10 Sliding 5～10

[0107] The experimental settings are as follows: The parameters of the repetition type recognition network used in the present invention are shown in Table 2. In Table 2, Batch size is the number of samples used to update the model weights in each iteration, epoch is the iteration batch, Learningrate is the learning rate, optimizer is the optimizer, Weight decay is the weight decay coefficient, a parameter to prevent overfitting, Dropout is the loss coefficient, a parameter used to prevent model overfitting, Conv1 and Conv2 are both convolution layer sizes, Pool1 and Pool 2 are both pooling layer sizes. 80% of all samples are randomly selected for training, and the remaining 20% ​​are used for verification and testing respectively.

[0108] Table 2 Parameters of the repetition type identification network

[0109]

[0110]

[0111] The results are analyzed specifically as follows: The performance of the method of the present invention was tested under different ratios of missed pulses and false pulses. The results show that as the proportion of false pulses gradually increases, the performance of the algorithm gradually decreases. This is because false pulses have high randomness, causing all repetition types to tend to have characteristics similar to jitter repetition. On the contrary, as the proportion of missed pulses gradually increases, the accuracy of the algorithm remains relatively stable. This shows that the algorithm proposed in this paper has good robustness in the case of missed pulses. It is worth noting that the accuracy in the case of missed pulses is about 10% lower than that in the case of false pulses. This is because as the number of pulse losses increases, the number of pulse samples gradually decreases, resulting in insufficient training.

[0112] Figure 4 The confusion matrix of the test for all non-ideal samples is shown schematically, according to Figure 4 The confusion matrix results in show the sample testing situations under all non-ideal conditions. Figure 4 The horizontal axis is the predicted label, and the vertical axis is the actual label. There are red, orange and yellow boxes in the figure. The darker the color, the higher the accuracy. The accuracy of the method of the present invention is the average of the values ​​in the red box on the diagonal of the matrix, which is 87.72%. That is, the overall recognition accuracy of the method of the present invention reaches 87.72%, and the boundaries of different repetition type features are obvious, with high distinguishability. This result further shows that the method of the present invention has good adaptability in non-ideal conditions.

[0113] The method of the present invention uses a graph domain mapping algorithm to convert a PRF sequence with unclear difference shape in a complex non-ideal scene with a high proportion of leaky pulses and false pulses into a Gram difference angle field image with significant features, thereby reducing information redundancy and enhancing feature aggregation. On this basis, a repetition type recognition network is constructed to address the problem of low recognition rate of repetition type in complex non-ideal scenes with a high proportion of leaky pulses and false pulses. Based on the channel feature enhancement module in the repetition type recognition network, i.e., the attention mechanism layer, the network is helped to give greater weight to channel features with significant differences, thereby improving the leading role of important features in the classification process, strengthening the role of important features in classification, and improving the robustness and recognition accuracy of repetition type in complex non-ideal scenes.

[0114] Based on the same inventive concept, as an implementation of the above-mentioned method for identifying repetition frequency types based on image domain mapping and feature enhancement, an embodiment of the present invention further provides a device for identifying repetition frequency types based on image domain mapping and feature enhancement. Figure 5 is a structural diagram of the device in the embodiment of the present invention, see Figure 5 As shown, the device may include:

[0115] A screening module 501 is used to screen a plurality of pulse repetition frequencies according to a preset threshold range, and determine a pulse repetition frequency sequence corresponding to the screened pulse repetition frequencies;

[0116] The processing module 502 is used to perform reciprocal processing on the pulse repetition frequency sequence determined by the screening module 501 to obtain a pulse repetition interval sequence;

[0117] A segmentation module 503 is used to segment the pulse repetition interval sequence obtained by the processing module 502 according to a preset length to obtain a plurality of subsequences;

[0118] A transformation module 504 is used to transform the multiple subsequences obtained by the segmentation module 503 into corresponding Grammar difference angular field images according to an image domain mapping algorithm, and divide the multiple Grammar difference angular field images into training samples and test samples;

[0119] A construction module 505 is used to construct a repetition type recognition network, wherein the repetition type recognition network includes a convolutional neural sub-network and an attention mechanism layer;

[0120] The training module 506 is used to input the training samples divided by the transformation module 504 into the repetition type recognition network constructed by the construction module 505, train the repetition type recognition network, and obtain a trained repetition type recognition network;

[0121] The identification module 507 is used to input the test samples divided by the transformation module 504 into the repetition type identification network trained by the training module 506 to identify multiple repetition types.

[0122] The transformation module 504 is specifically used to transform multiple subsequences into corresponding Grammar difference angular field images according to the image domain mapping algorithm, including: scaling the multiple subsequences to the range of [-1,1] to obtain the numerical values ​​after numerical scaling in each subsequence; converting each numerical value to the polar coordinate system to obtain the angle and radius corresponding to each numerical value; and determining the Grammar difference angular field image according to the angular difference between each two numerical values.

[0123] The construction module 505 is specifically used for the convolutional neural sub-network including 2 convolutional layers, 2 normalization layers, 2 activation layers, 2 pooling layers and 2 fully connected layers.

[0124] The construction module 505 is specifically used for the repetition type recognition network, which includes 1 convolution layer, 1 normalization layer, 1 activation layer, 1 pooling layer, 1 attention mechanism layer, 1 convolution layer, 1 normalization layer, 1 activation layer, 1 pooling layer and 2 fully connected layers in sequence.

[0125] The recognition module 507 is specifically used to input the test sample into one convolution layer, one normalization layer, one activation layer, and one pooling layer in sequence, and output the input features corresponding to the test sample; input the input features into one attention mechanism layer, and output the final feature representation; input the final feature representation into one convolution layer, one normalization layer, one activation layer, one pooling layer, and two fully connected layers in sequence, and output multiple repetition frequency types.

[0126] The recognition module 507 is specifically used to input the input features into an attention mechanism layer and output the final feature representation, including: compressing the input features to obtain channel-level global features; stimulating the channel-level global features to obtain the weights of each channel; multiplying the weights of each channel by the input features to obtain the final feature representation.

[0127] The transformation module 504 is specifically used to scale the multiple subsequences to the range of [-1, 1] to obtain the numerical value of each subsequence after the numerical value is scaled, including: using the following first formula to scale the multiple subsequences to the range of [-1, 1] to obtain the numerical value of each subsequence after the numerical value is scaled:

[0128]

[0129] in, The corresponding p n The scaled value, that is, the value after the nth value in each subsequence is scaled, p n is the nth value in each subsequence, P is a number of subsequences, and N is the length of the subsequence.

[0130] The transformation module 504 is specifically used to determine the Gram difference angular field image according to the angle difference between each two values, including: determining the corresponding matrix according to the angle difference between each two values ​​and the following second formula:

[0131]

[0132] Among them, GADF is the matrix corresponding to the angle between each two values, is the angle corresponding to the nth value in the polar coordinate system, is the angle corresponding to the mth value in the polar coordinate system;

[0133] The matrix is ​​converted into grayscale to obtain the Gram difference angular field image.

[0134] It should be noted that the above description of the embodiment of the device for identifying the type of repetition based on image domain mapping and feature enhancement is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. For technical details not disclosed in the embodiment of the device for identifying the type of repetition based on image domain mapping and feature enhancement of the embodiment of the present invention, please refer to the description of the method embodiment of the present invention for understanding.

[0135] Based on the same inventive concept, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute the method in one or more of the above embodiments.

[0136] It should be noted here that the description of the above computer-readable storage medium embodiment is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. For technical details not disclosed in the embodiment of the computer-readable storage medium of the embodiment of the present invention, please refer to the description of the method embodiment of the present invention for understanding.

[0137] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for identifying repetition types based on image domain mapping and feature enhancement, characterized in that: The method comprises: Screening a plurality of pulse repetition frequencies according to a preset threshold range, and determining a pulse repetition frequency sequence corresponding to the screened pulse repetition frequencies; Performing reciprocal processing on the pulse repetition frequency sequence to obtain a pulse repetition interval sequence; Dividing the pulse repetition interval sequence according to a preset length to obtain a plurality of subsequences; According to an image domain mapping algorithm, the multiple subsequences are transformed into corresponding Grammar difference angular field images, and the multiple Grammar difference angular field images are divided into training samples and test samples; Constructing a repetition type recognition network, wherein the repetition type recognition network includes a convolutional neural sub-network and an attention mechanism layer; Inputting the training samples into the repetition frequency type recognition network, training the repetition frequency type recognition network, and obtaining a trained repetition frequency type recognition network; The test sample is input into the trained repetition frequency type recognition network to recognize multiple repetition frequency types.

2. The method according to claim 1, characterized in that The step of transforming the plurality of subsequences into corresponding Gram difference angular field images according to an image domain mapping algorithm comprises: Scaling the multiple subsequences to a range of [-1, 1] to obtain a numerical value in each subsequence after the numerical value is scaled; Convert each value to the polar coordinate system to obtain the angle and radius corresponding to each value; The Gram difference angular field image is determined according to the angular difference between each two values.

3. The method according to claim 1, characterized in that The convolutional neural sub-network includes 2 convolutional layers, 2 normalization layers, 2 activation layers, 2 pooling layers and 2 fully connected layers.

4. The method according to claim 3, characterized in that The repetition type recognition network includes, in sequence, 1 convolution layer, 1 normalization layer, 1 activation layer, 1 pooling layer, 1 attention mechanism layer, 1 convolution layer, 1 normalization layer, 1 activation layer, 1 pooling layer and 2 fully connected layers.

5. The method according to claim 4, characterized in that The step of inputting the test sample into the trained repetition type recognition network to recognize multiple repetition types includes: Inputting the test sample into one of the convolutional layers, one of the normalization layers, one of the activation layers, and one of the pooling layers in sequence, and outputting the input features corresponding to the test sample; Input the input features into one of the attention mechanism layers and output the final feature representation; The final feature representation is sequentially input into one of the convolutional layers, one of the normalization layers, one of the activation layers, one of the pooling layers, and two of the fully connected layers, and the multiple repetition frequency types are output.

6. The method according to claim 5, characterized in that Inputting the input features into one of the attention mechanism layers and outputting a final feature representation includes: Compressing the input features to obtain channel-level global features; Exciting the channel-level global features to obtain the weights of each channel; The weights of each channel are multiplied by the input features to obtain the final feature representation.

7. The method according to claim 2, characterized in that Scaling the multiple subsequences to a range of [-1, 1] to obtain a numerical value in each subsequence after the numerical value is scaled includes: The multiple subsequences are scaled to the range of [-1, 1] using the following first formula to obtain the scaled value of each subsequence: in, The scaled value for the nth value in each subsequence, p n is the nth value in each subsequence, P is the multiple subsequences, and N is the subsequence length.

8. The method according to claim 2, characterized in that: Determining the Gram difference angular field image according to the angle difference between each two values ​​includes: According to the angle difference between each two values ​​and the following second formula, the corresponding matrix is ​​determined: Among them, GADF is the matrix corresponding to the angle between each two values, is the angle corresponding to the nth value in the polar coordinate system, is the angle corresponding to the mth value in the polar coordinate system; The matrix is ​​subjected to grayscale conversion to obtain the Gram difference angular field image.

9. A device for identifying repetition types based on image domain mapping and feature enhancement, characterized in that: The device comprises: A screening module, used to screen a plurality of pulse repetition frequencies according to a preset threshold range, and determine a pulse repetition frequency sequence corresponding to the screened pulse repetition frequencies; A processing module, used for performing reciprocal processing on the pulse repetition frequency sequence to obtain a pulse repetition interval sequence; A segmentation module, used for segmenting the pulse repetition interval sequence according to a preset length to obtain a plurality of subsequences; A transformation module, used for transforming the plurality of subsequences into corresponding Grammar difference angular field images according to an image domain mapping algorithm, and dividing the plurality of Grammar difference angular field images into training samples and test samples; A construction module is used to construct a repetition type recognition network, wherein the repetition type recognition network includes a convolutional neural sub-network and an attention mechanism layer; A training module, used for inputting the training samples into the repetition type recognition network, training the repetition type recognition network, and obtaining a trained repetition type recognition network; The recognition module is used to input the test sample into the trained repetition frequency type recognition network to recognize multiple repetition frequency types.

10. A computer-readable storage medium, characterized in that: The storage medium includes a stored program, wherein when the program is executed, the device where the storage medium is located is controlled to execute the method for repetition type recognition based on image domain mapping and feature enhancement as described in any one of claims 1 to 8.

Images