A method and apparatus for pulse stream segmentation and signal pattern recognition

Abstract

This invention relates to the field of radiation source identification technology, providing a method and apparatus for pulse stream segmentation and signal pattern recognition. The invention performs outlier detection on a radar pulse stream sequence to determine anchor points; segments the radar pulse stream sequence into multiple pulse subsequences based on the anchor points; extracts features from the multiple pulse subsequences to obtain original features; reweights the original features to obtain features to be classified; predicts the signal pattern of the features to be classified to calculate the classification loss, and iteratively updates the model to obtain a target recognition model, which is then used to perform signal pattern recognition on the segmented multiple pulse subsequences. This invention addresses the problem that outliers in radar pulse stream sequences cause performance degradation in radar signal pattern recognition algorithms.

Description

Technical Field

[0001] This invention relates to the field of radiation source identification technology, and in particular to a method and apparatus for pulse stream segmentation and signal pattern identification. Background Technology

[0002] Existing technologies typically acquire full-pulse data from radar and identify radar radiation sources based on this data. Specifically, full-pulse data is obtained by measuring the carrier frequency, arrival time, pulse width, pulse amplitude, and pulse repetition period of each radar signal entering the receiver, and then outputting and displaying the results sequentially in chronological order in the form of a data table.

[0003] Radar signal pattern recognition is a crucial component of radar radiation source identification. During the radar target search phase and when the radar pattern changes, the periodic scanning characteristics of the radar antenna and the time required for radar pattern switching cause significant jumps in the pulse repetition period of the full pulse data. Therefore, when using radar pulse stream sequences to represent the characteristics of the pulse repetition period changing over time, this jump manifests in the radar pulse stream sequence as the appearance of obvious outlier data points in the pulse repetition period.

[0004] On the one hand, the outlier data has certain reference value for analyzing information such as the radar's target search revisit mechanism and the radar's working mode switching rules; existing technologies often use classification and recognition algorithms to identify radar signal patterns, but because the pulse repetition period value of the outlier data is much larger than the pulse repetition period value at other times in the pulse sequence, it is not conducive to the convergence of the classification and recognition algorithm, which seriously affects the radar signal pattern recognition effect.

[0005] On the other hand, due to the different coherent processing cycles of different radar patterns, the corresponding radar pulse stream sequence lengths are not equal. In addition, in practical applications, the signal acquisition time set manually cannot guarantee consistency, so the length of the radar pulse stream sequence is likely to be different. Therefore, even if the modulation pattern (i.e., signal pattern) of the radar pulse stream sequence is the same, the length will be different, which brings many challenges to radar signal pattern recognition and seriously affects the effectiveness of radar radiation source identification.

[0006] Therefore, overcoming the shortcomings of the existing technology is an urgent problem to be solved in this technical field. Summary of the Invention

[0007] The technical problem to be solved by this invention is to provide a method and apparatus for pulse stream segmentation and signal pattern recognition. The purpose is to detect outliers in a radar pulse stream sequence as anchor points, and segment the radar pulse stream sequence into multiple pulse subsequences according to the anchor points, which is beneficial to the convergence of the classification and recognition algorithm. Furthermore, by reweighting the feature vectors of the pulse subsequences and using feature aggregation methods, the feature unification of pulse subsequences of different lengths is achieved, thereby improving the radar signal pattern recognition effect and solving the problem that the existence of outliers in the radar pulse stream sequence leads to a decrease in the performance of the radar signal pattern recognition algorithm.

[0008] The present invention adopts the following technical solution:

[0009] In a first aspect, the present invention provides a method for pulse stream segmentation and signal pattern recognition, comprising:

[0010] Outlier detection is performed on the radar pulse stream sequence to determine anchor points in the radar pulse stream sequence; based on the anchor points, the radar pulse stream sequence is divided into multiple pulse subsequences;

[0011] The initial recognition model extracts features from each of the pulse subsequences to obtain original features; the original features are reweighted to obtain the features to be classified for the multiple pulse subsequences; the signal pattern of the features to be classified is predicted to obtain the predicted classification result.

[0012] The classification loss is calculated using the predicted classification results; based on the classification loss, the weight parameters of the initial recognition model are iteratively updated until the number of iterations reaches the total number of iterations, thus obtaining the target recognition model, which is then used to perform signal pattern recognition on multiple segmented pulse subsequences.

[0013] Furthermore, the reweighting of the original features to obtain the classification features of the multiple pulse subsequences includes:

[0014] The original features are linearly transformed, the result of the linear transformation is nonlinearly activated, and the result of the nonlinear activation is weighted and summed to generate attention weights.

[0015] The product of the attention weight and the original feature is determined as the feature to be aggregated, so as to complete the reweighting of the original feature;

[0016] The features to be aggregated in each pulse subsequence are summed along the length to obtain the features to be classified.

[0017] Furthermore, the step of performing a linear transformation on the original features, applying nonlinear activation to the result of the linear transformation, and weighted summing of the results of the nonlinear activation to generate attention weights includes:

[0018] The product of the original feature and the first parameter is determined as the first intermediate vector;

[0019] The first intermediate vector is processed using the hyperbolic tangent function to obtain the second intermediate vector;

[0020] The product of the second parameter and the second intermediate vector is used to determine the third intermediate vector;

[0021] The third intermediate vector is processed using a preset activation function to obtain attention weights.

[0022] Furthermore, the anchor points include hard threshold anchor points and soft threshold anchor points;

[0023] The step of outlier detection in the radar pulse stream sequence to determine anchor points in the radar pulse stream sequence includes:

[0024] The pulse repetition period of each data point at each moment is obtained from the radar pulse stream sequence;

[0025] When the pulse repetition period at a certain moment is greater than the hard threshold, the data point at that moment is used as the hard threshold anchor point.

[0026] According to each moment The pulse repetition period is related to the previous moment The pulse repetition period, and the next moment. The pulse repetition period is used to determine the soft threshold anchor point.

[0027] Furthermore, the statement based on each moment The pulse repetition period is related to the previous moment The pulse repetition period, and the next moment. The pulse repetition period is used to determine the soft threshold anchor point, which includes:

[0028] The time The pulse repetition period is related to the previous moment. The ratio of the pulse repetition periods is determined as the first ratio value;

[0029] The time The pulse repetition period and its subsequent moment The ratio of the pulse repetition periods is determined as the second ratio value;

[0030] When the first ratio is greater than the soft threshold and the second ratio is greater than the soft threshold, the data point at that moment is taken as the soft threshold anchor point.

[0031] Further, the step of dividing the radar pulse stream sequence into multiple pulse subsequences based on the anchor point includes:

[0032] The radar pulse stream sequence is segmented according to the hard threshold anchor point and the soft threshold anchor point, and the data points corresponding to the hard threshold anchor point and the soft threshold anchor point are deleted to obtain multiple original sub-sequences;

[0033] The length of each original subsequence is counted, and original subsequences with a length less than a length threshold are removed from the multiple original subsequences to obtain multiple intermediate subsequences;

[0034] The multiple intermediate subsequences are adjusted to a uniform length to obtain multiple pulse subsequences.

[0035] Furthermore, the expression for calculating the classification loss using the predicted classification result is as follows:

[0036]

[0037] in, For classifying losses, This represents the total number of signal pattern categories. In the true classification results, the signal pattern belongs to the first... The probability of a class To predict which signal pattern belongs to the classification result. The probability of a class.

[0038] Furthermore, the initial recognition model extracts features from each of the pulse sub-sequences to obtain the original features, including:

[0039] Convert the plurality of pulse subsequences into an initial feature vector;

[0040] The initial feature vector is convolved to obtain a first feature vector; the first feature vector is processed using an activation function to obtain a second feature vector; the second feature vector is convolved to obtain a third feature vector.

[0041] The third feature vector is added to the initial feature vector to obtain the fourth feature vector;

[0042] The fourth feature vector is processed using an activation function to obtain the original features.

[0043] Secondly, the present invention also provides an apparatus for pulse stream segmentation and signal pattern recognition, comprising:

[0044] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor for performing the pulse stream segmentation and signal pattern recognition method described in the first aspect.

[0045] Thirdly, the present invention also provides a non-volatile computer storage medium storing computer-executable instructions that are executed by one or more processors to perform the pulse stream segmentation and signal pattern recognition method described in the first aspect.

[0046] Fourthly, a computer program product containing instructions is provided that, when executed on a computer or processor, causes the computer or processor to perform the pulse stream segmentation and signal pattern recognition method as described in the first aspect.

[0047] Fifthly, the present invention also provides a system for pulse stream segmentation and signal pattern recognition, including the apparatus for pulse stream segmentation and signal pattern recognition as described in the second aspect, and using the method for pulse stream segmentation and signal pattern recognition as described in the first aspect to complete the interaction of the apparatus for pulse stream segmentation and signal pattern recognition in the second aspect.

[0048] Unlike existing technologies, the present invention has at least the following beneficial effects:

[0049] This invention effectively removes and segments radar pulse stream sequences by detecting outliers, eliminating their impact on radar signal pattern recognition. This facilitates model convergence and improves the robustness of the recognition model. Furthermore, by reweighting the feature vectors of pulse subsequences and utilizing feature aggregation methods, it achieves feature unification for pulse subsequences of different lengths. This effectively avoids the problem of long processing times caused by the inability of recurrent neural networks to process long sequence data in parallel, thereby improving the radar signal pattern recognition effect and solving the problem of performance degradation caused by outliers in radar pulse stream sequences. Attached Figure Description

[0050] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments of the present invention will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0051] Figure 1 This is a flowchart illustrating a method for pulse stream segmentation and signal pattern recognition provided in an embodiment of the present invention;

[0052] Figure 2 This is a flowchart illustrating step 10 provided in an embodiment of the present invention;

[0053] Figure 3 This is a schematic diagram of a process for determining an anchor point according to an embodiment of the present invention;

[0054] Figure 4 This is a flowchart illustrating step 103a provided in an embodiment of the present invention;

[0055] Figure 5 This is a flowchart illustrating another step 10 provided in an embodiment of the present invention;

[0056] Figure 6 This is a schematic diagram illustrating a specific example of the original data before radar pulse stream sequence segmentation provided in an embodiment of the present invention;

[0057] Figure 7 This is a schematic diagram illustrating a specific example of the first pulse subsequence after radar pulse stream sequence segmentation provided in an embodiment of the present invention;

[0058] Figure 8 The diagram shown is a specific example of a second pulse subsequence after radar pulse stream sequence segmentation provided by an embodiment of the present invention;

[0059] Figure 9 This is a schematic diagram illustrating a specific example of a third pulse subsequence after radar pulse stream sequence segmentation provided in an embodiment of the present invention;

[0060] Figure 10 This is a schematic diagram illustrating a specific example of a fourth pulse subsequence after radar pulse stream sequence segmentation provided in an embodiment of the present invention;

[0061] Figure 11 This is a schematic diagram of a specific example of the network structure of a recognition model provided in an embodiment of the present invention;

[0062] Figure 12 This is a flowchart illustrating step 20 provided in an embodiment of the present invention;

[0063] Figure 13 This is a flowchart illustrating another step 20 provided in an embodiment of the present invention;

[0064] Figure 14 This is a flowchart illustrating step 201b provided in an embodiment of the present invention;

[0065] Figure 15 This is a schematic diagram illustrating a specific example of the accuracy of radar signal pattern recognition during the training and verification phases provided by an embodiment of the present invention.

[0066] Figure 16 This is a schematic diagram illustrating a specific example of a classification loss value for radar signal pattern recognition during the training and verification phases, provided by an embodiment of the present invention.

[0067] Figure 17 This is a schematic diagram of the architecture of a pulse stream segmentation and signal pattern recognition device provided in an embodiment of the present invention. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0069] Unless the context otherwise requires, throughout the specification and claims, the term "comprising" is interpreted as openly inclusive, meaning "including, but not limited to." In the description of the specification, terms such as "one embodiment," "some embodiments," "exemplary embodiment," "example," "specific example," or "some examples" are intended to indicate that a particular feature, structure, material, or characteristic associated with that embodiment or example is included in at least one embodiment or example of this disclosure. The illustrative representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics mentioned may be included in any suitable manner in any one or more embodiments or examples; that is, although they may be incorporated into embodiments or examples using the above terms for reasons such as order and position, it does not limit them to be incorporated in combination by a single embodiment or example.

[0070] In the description of this invention, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this disclosure and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this disclosure.

[0071] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of embodiments of this disclosure, unless otherwise stated, "a plurality of" means two or more. Furthermore, for example, the description may use the prefix "A" or "B" to describe the same type of nouns as two independent entities. In this case, the corresponding features defined with "A" and "B" are used only to distinguish between similar entities and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features.

[0072] In describing some embodiments, the terms "coupled," "coupled," and "connected," and their derivative expressions, may be used. For example, the term "connected" may be used in describing some embodiments to indicate that two or more components have direct physical or electrical contact with each other. Similarly, the term "coupled" may be used in describing some embodiments to indicate that two or more components have direct physical or electrical contact. However, the terms "connected" or "coupled" may also refer to two or more components that do not have direct contact with each other but still cooperate or interact with each other, such as "optical coupling," "wireless connection," etc. The embodiments disclosed herein are not necessarily limited to the scope of this invention.

[0073] In the description of this invention, the expression “A and / or B” (where A and B are used to formally represent specific features) will be used. The corresponding expression includes the following three combinations: only A, only B, and a combination of A and B.

[0074] As used in this invention, “about,” “approximately,” or “approximately” includes the stated value and the average value within an acceptable range of deviation from a particular value, wherein the acceptable range of deviation is determined by a person skilled in the art taking into account the measurement under discussion and the error associated with the measurement of the particular quantity (i.e., the limitations of the measurement system).

[0075] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0076] Example 1:

[0077] To solve the above problems, such as Figure 1 As shown, this embodiment of the invention provides a method for pulse stream segmentation and signal pattern recognition, including:

[0078] Step 10: Perform outlier detection on the radar pulse stream sequence to determine the anchor points in the radar pulse stream sequence; based on the anchor points, divide the radar pulse stream sequence into multiple pulse subsequences.

[0079] The main data information used to identify radar radiation sources includes basic radar radiation source data, signal parameter data, radiation characteristic data, and the status of the collected data. Signal parameter data can be divided into frequency domain parameters, time domain parameters, energy domain parameters, and modulation domain parameters. Frequency domain parameters mainly refer to the radar's operating frequency; time domain parameters mainly include the radar's pulse width and pulse repetition period; energy domain parameters mainly refer to the radar radiation source's transmit power, with the collected signal power reflected in the signal amplitude; and modulation domain parameters refer to the modulation method within the intermediate frequency pulse.

[0080] In one embodiment, radio frequency (RF), pulse width (PW), and pulse repetition interval (PRI) are selected to characterize the radar pulse stream sequence, then the sampling period... Internal radar signal The radar pulse stream sequence is denoted as:

[0081]

[0082] in, Indicates the sampling period Inside Radio frequency at any moment, Indicates the sampling period Inside The pulse width of time Indicates the sampling period Inside The pulse repetition period at each time point. A radar pulse stream sequence includes data points from multiple time points, each time point... The data points include the radio frequency, pulse width, and pulse repetition period acquired at that moment.

[0083] Among them, the frequency measurement accuracy of the frequency measurement receiver can reach 1 MHz, and the measurement accuracy of pulse width and pulse arrival time can reach 50 nanoseconds.

[0084] Because the pulse repetition period will exhibit significant jumps during the target search phase and when the radar pattern changes, resulting in noticeable outliers in the pulse repetition period's time-varying characteristics before returning to the normal range, this embodiment of the invention uses the variation in the pulse repetition period's value range as the primary basis for pulse segmentation, thereby removing outliers from the radar pulse stream. By segmenting the radar pulse stream sequence into multiple pulse subsequences, it also serves as a preliminary division of the signal pattern changes within the radar pulse stream sequence.

[0085] Step 20: The initial recognition model extracts features from each of the pulse subsequences to obtain the original features; the original features are reweighted to obtain the features to be classified for the multiple pulse subsequences; the signal pattern of the features to be classified is predicted to obtain the predicted classification result.

[0086] The predicted classification result is the classification result of the signal pattern. In this embodiment of the invention, a neural network model (i.e., a recognition model) is used to process the segmented multiple pulse sub-sequences to identify the pattern of the radar signal represented by the radar pulse stream sequence in step 10.

[0087] Step 30: Calculate the classification loss using the predicted classification result; based on the classification loss, iteratively update the weight parameters of the initial recognition model until the number of iterations reaches the total number of iterations, to obtain the target recognition model, so as to use the target recognition model to perform signal pattern recognition on the segmented multiple pulse subsequences.

[0088] The total number of iterations is selected by those skilled in the art based on the specific application scenario; in one embodiment, the total number of iterations can be 200.

[0089] In practical applications, after the radar completes the full-pulse data acquisition, existing technologies only perform data cleaning by removing outliers, without considering pulse signal segmentation. Since radar often needs to switch between different signal patterns to achieve different functions, such as search, tracking, and ranging, the unsegmented radar pulse sequence may contain multiple signal patterns, which is not conducive to effective identification.

[0090] This invention effectively removes and segments radar pulse stream sequences by detecting outliers, eliminating their impact on radar signal pattern recognition. This facilitates model convergence and improves the robustness of the recognition model. Furthermore, by reweighting the feature vectors of pulse subsequences and utilizing feature aggregation methods, it achieves feature unification for pulse subsequences of different lengths. This effectively avoids the problem of long processing times caused by the inability of recurrent neural networks to process long sequence data in parallel, thereby improving the radar signal pattern recognition effect and solving the problem of performance degradation caused by outliers in radar pulse stream sequences.

[0091] The outlier detection and radar pulse stream sequence segmentation process of this invention will now be described:

[0092] like Figure 2 As shown, in step 10, the outlier detection of the radar pulse stream sequence to determine the anchor points in the radar pulse stream sequence includes:

[0093] Step 101a: Obtain the pulse repetition period of each data point at each moment from the radar pulse stream sequence.

[0094] like Figure 3 The diagram shows a process for determining anchor points; wherein the anchor points include hard threshold anchor points and soft threshold anchor points.

[0095] Step 102a: When the pulse repetition period at a certain moment is greater than the hard threshold, the data point at that moment is used as the hard threshold anchor point.

[0096] Due to the significant differences in the repetition period of different pulse patterns, the hard threshold... The choice is made by those skilled in the art based on the specific application scenario; in one embodiment, because the pulse repetition period of the ground-based air defense radar is relatively large, it can achieve... The order of magnitude, therefore, hard threshold The larger value should also be taken accordingly, which can be... Airborne radar has a relatively small pulse repetition period, so the hard threshold... The value can be small, and can be... At a certain moment The pulse repetition period is greater than the hard threshold. When this happens, the data point is used as the hard threshold anchor point.

[0097] Step 103a: Based on each time step The pulse repetition period is related to the previous moment. The pulse repetition period, and the next moment. The pulse repetition period is used to determine the soft threshold anchor point.

[0098] Specifically, such as Figure 4 As shown, step 103a includes:

[0099] Step 1031: Set the time... The pulse repetition period is related to the previous moment. The ratio of the pulse repetition periods is determined as the first ratio.

[0100] Step 1032: Set the time... The pulse repetition period and its subsequent moment The ratio of the pulse repetition periods is determined as the second ratio.

[0101] Step 1033: When the first ratio is greater than the soft threshold and the second ratio is greater than the soft threshold, the data point at that moment is taken as the soft threshold anchor point.

[0102] Among them, soft threshold The selection should be made by those skilled in the art based on the specific application scenario; soft threshold The value of is independent of the range of pulse repetition period values ​​for the radar pattern; in one embodiment, the soft threshold... You can take 25 to 50.

[0103] like and Then the The data point at time t is used as the soft threshold anchor point; where, Indicates time The pulse repetition period, Indicates the moment The moment before The pulse repetition period, Indicates the first ratio. Indicates the moment The next moment The pulse repetition period, This indicates the second ratio.

[0104] To illustrate the process of obtaining multiple pulse subsequences, such as Figure 5 As shown, in step 10, dividing the radar pulse stream sequence into multiple pulse sub-sequences based on the anchor point includes:

[0105] Step 101b: Segment the radar pulse stream sequence according to the hard threshold anchor point and the soft threshold anchor point, delete the data points corresponding to the hard threshold anchor point and the soft threshold anchor point, and obtain multiple original subsequences.

[0106] In one embodiment, the radar pulse stream sequence is divided according to anchor points and stored sequentially as new arrays as subsequences after division; from each subsequence after division, the data points at the time of the anchor point are deleted, and finally multiple original subsequences of the radar pulse stream sequence are obtained.

[0107] In one alternative embodiment, the outlier detection method based on soft threshold and hard threshold is implemented sequentially; however, sequential implementation may result in data points at certain times being identified as both soft threshold anchor points and hard threshold anchor points, causing duplicate anchor point settings; therefore, the soft threshold anchor points and hard threshold anchor points can be deduplicated first, and then the data points at the times of the deduplicated anchor points can be deleted.

[0108] Step 102b: Calculate the length of each original subsequence, and remove the original subsequences whose length is less than the length threshold from the multiple original subsequences to obtain multiple intermediate subsequences.

[0109] The length threshold is selected by those skilled in the art based on the specific use case; in one optional embodiment, the length threshold can be 20.

[0110] In one embodiment, the length of each original subsequence is counted, and if the length is less than a length threshold, it is not counted as a valid segment.

[0111] Step 103b: Adjust the multiple intermediate sub-sequences to a uniform length to obtain multiple pulse sub-sequences.

[0112] Because radar operating patterns vary, and false and missing pulses exist in real-world scenarios, the lengths of the segmented pulse subsequences may differ. Furthermore, this embodiment of the invention subsequently employs a convolutional neural network-based recognition algorithm to distinguish the signal patterns of radar signals, and convolutional neural networks require input data to have a uniform size. Therefore, before inputting the recognition model, such as... Figure 11 As shown, each pulse subsequence in the multiple pulse subsequences is first padded to a uniform length using zero elements.

[0113] In one embodiment, such as Figures 6 to 10 The diagram shows a specific example of radar pulse stream sequence segmentation before and after. For example, Figure 6 The diagram shown illustrates a specific example of raw data before radar pulse stream sequence segmentation; where raw data refers to the acquired raw full-pulse data; such as... Figure 7 The diagram shown is a specific example of the first type of pulse subsequence after radar pulse stream sequence segmentation; for example... Figure 8 The diagram shows a specific example of the second type of pulse subsequence after radar pulse stream sequence segmentation; for example... Figure 9 The diagram shows a specific example of the third type of pulse subsequence after radar pulse stream sequence segmentation; for example... Figure 10 The diagram shows a specific example of the fourth type of pulse subsequence after radar pulse stream sequence segmentation. From... Figure 6 It can be seen that the radar pulse sequence corresponding to the original data contains many outliers with large values. Using the outlier detection and pulse segmentation algorithm proposed in this invention, we obtain... Figures 7 to 10 The four pulse subsequences shown reveal that the original data contained multiple signal patterns. Based on their order within the radar pulse sequence, the segmented data is divided into pulse subsequence 1, pulse subsequence 2, pulse subsequence 3, and pulse subsequence 4. Pulse subsequence 1 and pulse subsequence 2 are frequency repetition group modulated signals; pulse subsequence 3 has a constant pulse repetition period. However, pulse loss may occur in pulse subsequence 3 because the pulse repetition period frequently jumps between four values ​​and is a fixed multiple of a certain value. Compared to the original data, the segmented pulse subsequences contain only a single radar signal pattern, which is beneficial for subsequent identification tasks.

[0114] Existing technologies generally use recurrent neural networks to process sequential data. However, since recurrent neural networks can only process input sequence data sequentially and cannot perform parallel operations, the processing efficiency for long sequence data is low. Therefore, this invention uses a network structure based on convolutional neural networks (CNN).

[0115] like Figure 11As shown, this embodiment of the invention constructs a recognition model based on a convolutional neural network with an attention mechanism to perform radar signal pattern recognition. This recognition model includes a feature extraction module, an attention module, and a classification module. The feature extraction module extracts features from multiple pulse subsequences obtained in step 103b. The attention module reweights the feature vectors extracted by the feature extraction module to eliminate the impact of zero-padding elements on the signal pattern recognition task. The weighted feature vectors from the attention module are summed along their length to achieve feature aggregation, ensuring that the feature vector size extracted by the recognition model for sequences of different lengths is uniform. Finally, the aggregated feature vectors are input to the classification module to complete the recognition of radar signal patterns.

[0116] Specifically, such as Figure 12 As shown, in step 20, the initial recognition model extracts features from each pulse subsequence to obtain the original features, including:

[0117] Step 201a: Convert the plurality of pulse subsequences into an initial feature vector.

[0118] The specific method for converting multiple pulse subsequences into an initial feature vector can be selected by those skilled in the art based on the specific application scenario, and is not limited here.

[0119] Step 202a: Perform a convolution operation on the initial feature vector to obtain a first feature vector; process the first feature vector using an activation function to obtain a second feature vector; perform a convolution operation on the second feature vector to obtain a third feature vector.

[0120] In one embodiment, the activation function is the Rectified Linear Unit (ReLU).

[0121] like Figure 11 As shown, after the initial feature vector is input into the feature extraction module, it is first convolved by a convolutional layer to obtain the first feature vector; the ReLU activation function is applied to the first feature vector to obtain the second feature vector; and the second feature vector is convolved by the next convolutional layer to obtain the third feature vector.

[0122] Step 203a: Add the third feature vector to the initial feature vector to obtain the fourth feature vector.

[0123] This invention uses residual connections to address the gradient vanishing problem caused by increasing network depth, thereby improving the robustness of the model. Figure 11 The plus sign inside the circle indicates that the third feature vector and the initial feature vector of the input are added together.

[0124] Step 204a: Process the fourth feature vector using an activation function to obtain the original features.

[0125] In one embodiment, the activation function is the Rectified Linear Unit (ReLU).

[0126] The feature extraction module is implemented using a convolutional neural network (CNN) based on residual connections. In one embodiment, the expression for the residual connection CNN in the feature extraction module is as follows:

[0127]

[0128] in, It is a pulse subsequence. This is the function of the feature extraction module for processing the input quantity. The model parameters in the feature extraction module are manually initialized and iteratively optimized during training.

[0129] In one embodiment, since the variation pattern of the radar pulse stream sequence is mainly reflected in the time dimension, one-dimensional convolution, pooling, and batch normalization operations are selected for deep feature extraction.

[0130] Next, this embodiment of the invention utilizes an attention mechanism to reweight the original features output by the feature extraction module. This reduces the weight of the zero-padded elements in the original features and increases the weight of data that originally belonged to the intermediate subsequence, thereby reducing the impact of zero-padded elements on the signal pattern recognition task. By combining zero-padded elements and reweighting, the problem of recognizing radar pulse stream sequences of unequal lengths can be effectively solved. Specifically, as... Figure 13 As shown, in step 20, the reweighting of the original features to obtain the unclassified features of the multiple pulse subsequences includes:

[0131] Step 201b: Perform a linear transformation on the original features, perform nonlinear activation on the result of the linear transformation, and perform a weighted summation on the result of the nonlinear activation to generate attention weights.

[0132] The process of sequentially performing linear transformation, nonlinear activation, and weighted summation will be explained below.

[0133] Step 202b: Determine the product of the attention weight and the original feature as the feature to be aggregated, so as to complete the reweighting of the original feature.

[0134] Step 203b: Sum the features to be aggregated along the length of each pulse subsequence to obtain the features to be classified.

[0135] Specifically, such as Figure 14 As shown, step 201b includes:

[0136] Step 2011: Determine the product of the original feature and the first parameter as the first intermediate vector.

[0137] The first parameter is a parameter of the attention module; it is used to map the original features to a new feature space, i.e., to perform a linear projection on the original features. In this embodiment of the invention, the first parameter is used to perform a linear transformation on the original features to facilitate the extraction of features relevant to the signal pattern recognition task, preparing for subsequent nonlinear transformations.

[0138] Step 2012: Process the first intermediate vector using the hyperbolic tangent function to obtain the second intermediate vector.

[0139] In this embodiment of the invention, the hyperbolic tangent function is used to perform nonlinear activation on the first intermediate vector (i.e., the result of the linear transformation); the hyperbolic tangent function is applied to the first intermediate vector to introduce nonlinearity and compress the value range to (-1,1), thereby enhancing the expressive power of the recognition model.

[0140] Step 2013: Determine the product of the second parameter and the second intermediate vector as the third intermediate vector.

[0141] The second parameter is a parameter of the attention module, used to further map the output of the second intermediate vector. In this embodiment of the invention, the second parameter is used to perform a weighted summation on the second intermediate vector (i.e., the result of nonlinear activation), adjusting the features after the nonlinear transformation to generate an intermediate representation related to the attention weights.

[0142] Step 2014: Process the third intermediate vector using a preset activation function to obtain attention weights.

[0143] The preset activation function is selected by those skilled in the art based on the specific application scenario; in one embodiment, the preset activation function may be the softmax function.

[0144] Finally, the result is normalized to a probability distribution using the softmax function to ensure that the output attention weights are non-negative and sum to 1.

[0145] In one embodiment, the feature vector The dimensions are: [batch size, channel number, length], where each dimension represents the batch size, the number of feature channels, and the length of the original feature, respectively. Set the first parameter. Second parameter Initialize both using a standard Gaussian distribution, and then apply the following expression to the eigenvectors. Reweighting:

[0146]

[0147] in, Indicates the features to be aggregated. Indicates original features, Indicates attention weights, This represents the softmax function. Represents the hyperbolic tangent function; Represents the first intermediate vector. This represents the second intermediate vector. This represents the third intermediate vector.

[0148] In one embodiment, after reweighting, the features to be aggregated for each pulse subsequence can be summed along the length according to the following expression:

[0149]

[0150] in, Indicates the features to be classified. Indicates the first One feature to be aggregated, Indicates the first Attention weights corresponding to the features to be aggregated This represents the softmax function. Indicates the first The first parameter of each pulse subsequence Indicates the first The second parameter of each pulse subsequence.

[0151] After the above calculations, the feature dimension is reduced to [batch size, channel number], which achieves the unification of the dimension of data with different sequence lengths after feature extraction, and at the same time greatly reduces the number of nodes in the fully connected layer of the classifier.

[0152] Finally, the aggregated feature vector is input into the classification module to complete the mapping from feature distribution to classification distribution. In step 20, predicting the signal pattern of the feature to be classified to obtain the predicted classification result includes:

[0153] The classification module consists of two fully connected layers and a non-linear activation function. The number of nodes in the fully connected layers corresponds to the number of channels in the feature vector and the number of signal pattern categories in the samples in the dataset, respectively. The feature vectors are then processed through the fully connected layers and a softmax function to achieve dimensionality reduction, mapping to... indivual The classification module takes a scalar value between two values ​​and selects the category corresponding to the maximum value as the classification result. The expression for the classification module is as follows:

[0154]

[0155] in, For the classification module, These are the parameters for the classification module; the parameters for the classification module are also manually initialized and iteratively optimized during training, which is not limited here.

[0156] In step 30, in one embodiment, the expression for calculating the classification loss using the predicted classification result is:

[0157]

[0158] in, For classifying losses, This represents the total number of signal pattern categories. In the true classification results, the signal pattern belongs to the first... The probability of a class To predict which signal pattern belongs to the classification result. The probability of a class.

[0159] This invention employs gradient descent and backpropagation to iteratively optimize the network parameters of the initial recognition model; in one embodiment, the iterative optimization is performed according to the following expression:

[0160]

[0161]

[0162] in, The learning rate is initially set to... Then, it decreases by 20% every 20 iterations. These are the parameters for the feature extraction module. These are the parameters for the classification module; Represents the loss function Classification module parameters The derivative, Represents the loss function Parameters of the feature extraction module The derivative of . During training, the optimizer is Adam, and the total number of iterations is set to 200.

[0163] like Figure 15The diagram illustrates a specific example of the iterative change curve of the classification loss for radar signal pattern recognition during the training and validation phases when the total number of iterations is set to 200. After outlier detection and removal in step 10, and the radar signal pattern recognition being segmented into multiple pulse subsequences, the initial recognition model's classification loss during the training phase continuously decreases and eventually fluctuates less, indicating that the initial recognition model converges well. Similarly, the target recognition model's classification loss during the validation phase also continuously decreases and eventually fluctuates less, indicating that the target recognition model converges well.

[0164] like Figure 16 This is a schematic diagram illustrating a specific example of the accuracy of a recognition model's predicted classification results during the training and validation phases. Using the pulse stream segmentation and signal pattern recognition method of this invention, the initial recognition model can maintain a high accuracy during the training phase, and the target recognition model can also maintain a high accuracy during the validation phase. It has a good effect on radar pattern recognition and is highly practical.

[0165] Example 2:

[0166] like Figure 17 The diagram shown is an architectural schematic of a pulse stream segmentation and signal pattern recognition apparatus according to an embodiment of the present invention. The apparatus for pulse stream segmentation and signal pattern recognition in this embodiment includes one or more processors 21 and a memory 22. Figure 17 Take a processor 21 as an example.

[0167] Processor 21 and memory 22 can be connected via a bus or other means. Figure 17 Taking the example of a connection between China and Israel via a bus.

[0168] The memory 22, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs and non-volatile computer-executable programs, such as the pulse stream segmentation and signal pattern recognition method in this embodiment. The processor 21 executes the pulse stream segmentation and signal pattern recognition method by running the non-volatile software program and instructions stored in the memory 22.

[0169] Memory 22 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, memory 22 may optionally include memory remotely located relative to processor 21, which can be connected to processor 21 via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0170] The program instructions / modules are stored in the memory 22. When executed by one or more processors 21, they perform the pulse stream segmentation and signal pattern recognition method in the above embodiments, for example, performing each step of the pulse stream segmentation and signal pattern recognition method of the embodiments of the present invention described above.

[0171] This invention also provides a non-volatile computer storage medium storing computer-executable instructions that are executed by one or more processors, for example... Figure 17 A processor 21 may enable one or more of the processors to execute the pulse stream segmentation and signal pattern recognition method in the specific embodiments of the present invention, for example, to execute the various steps of the pulse stream segmentation and signal pattern recognition method described in the embodiments of the present invention above; it may also implement Figure 17 The various modules and units described above; or the pulse stream segmentation and signal pattern recognition method in the specific embodiments of the present invention, for example, executing the various steps of the pulse stream segmentation and signal pattern recognition method of the embodiments of the present invention described above; can also be implemented Figure 17 The aforementioned modules and units.

[0172] It is worth noting that the information interaction and execution process between the modules and units in the above-mentioned device and system are based on the same concept as the processing method embodiment of the present invention. For details, please refer to the description in the method embodiment of the present invention, and will not be repeated here.

[0173] Those skilled in the art will understand that all or part of the steps in the various methods of the embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, etc.

[0174] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for pulse stream segmentation and signal pattern recognition, characterized in that, include: Obtain the pulse repetition period of each data point at each moment from the radar pulse stream sequence; When the pulse repetition period at a certain moment is greater than the hard threshold, the data point at that moment is used as the hard threshold anchor point. The ratio of the pulse repetition period at the specified moment to the pulse repetition period at the previous moment is determined as the first ratio; the ratio of the pulse repetition period at the specified moment to the pulse repetition period at the next moment is determined as the second ratio; when the first ratio is greater than the soft threshold and the second ratio is greater than the soft threshold, the data point at that moment is taken as the soft threshold anchor point. Based on the hard threshold anchor point and the soft threshold anchor point, the radar pulse stream sequence is divided into multiple pulse subsequences; The initial recognition model extracts features from each of the pulse sub-sequences to obtain the original features; The original features are reweighted to obtain the classification features of the multiple pulse subsequences; Predict the signal pattern of the feature to be classified in order to obtain the predicted classification result; The classification loss is calculated using the predicted classification results; based on the classification loss, the weight parameters of the initial recognition model are iteratively updated until the number of iterations reaches the total number of iterations, thus obtaining the target recognition model, which is then used to perform signal pattern recognition on multiple segmented pulse subsequences.

2. The method for pulse stream segmentation and signal pattern recognition according to claim 1, characterized in that, The method includes: The original features are linearly transformed, the result of the linear transformation is nonlinearly activated, and the result of the nonlinear activation is weighted and summed to generate attention weights. The product of the attention weight and the original feature is determined as the feature to be aggregated, so as to complete the reweighting of the original feature; The features to be aggregated in each pulse subsequence are summed along the length to obtain the features to be classified.

3. The method for pulse stream segmentation and signal pattern recognition according to claim 2, characterized in that, The method includes: The product of the original feature and the first parameter is determined as the first intermediate vector; The first intermediate vector is processed using the hyperbolic tangent function to obtain the second intermediate vector; The product of the second parameter and the second intermediate vector is used to determine the third intermediate vector; The third intermediate vector is processed using a preset activation function to obtain attention weights.

4. The method for pulse stream segmentation and signal pattern recognition according to claim 1, characterized in that, The method includes: The radar pulse stream sequence is segmented according to the hard threshold anchor point and the soft threshold anchor point, and the data points corresponding to the hard threshold anchor point and the soft threshold anchor point are deleted to obtain multiple original sub-sequences; The length of each original subsequence is counted, and original subsequences with a length less than a length threshold are removed from the multiple original subsequences to obtain multiple intermediate subsequences; The multiple intermediate subsequences are adjusted to a uniform length to obtain multiple pulse subsequences.

5. The method for pulse stream segmentation and signal pattern recognition according to any one of claims 1-4, characterized in that, The expression for calculating the classification loss using the predicted classification result is as follows: in, For classifying losses, This represents the total number of signal pattern categories. In the true classification results, the signal pattern belongs to the first... The probability of a class To predict which signal pattern belongs to the classification result. The probability of a class.

6. The method for pulse stream segmentation and signal pattern recognition according to any one of claims 1-4, characterized in that, The method includes: Convert the plurality of pulse subsequences into an initial feature vector; The initial feature vector is convolved to obtain a first feature vector; the first feature vector is processed using an activation function to obtain a second feature vector; the second feature vector is convolved to obtain a third feature vector. The third feature vector is added to the initial feature vector to obtain the fourth feature vector; The fourth feature vector is processed using an activation function to obtain the original features.

7. A device for pulse stream segmentation and signal pattern recognition, characterized in that, The pulse stream segmentation and signal pattern recognition apparatus includes at least one processor and a memory, which are connected via a data bus. The memory stores instructions that can be executed by the at least one processor. After being executed by the processor, the instructions are used to implement the pulse stream segmentation and signal pattern recognition method according to any one of claims 1-6.

8. A non-volatile computer storage medium, characterized in that, The computer storage medium stores computer-executable instructions, which are executed by one or more processors to perform the pulse stream segmentation and signal pattern recognition method according to any one of claims 1-6.

Images