Radar scanning mode recognition method based on pulse sequence structure features and abnormality discrimination mechanism

Abstract

The application discloses a radar scanning mode recognition method based on pulse sequence structure features and abnormality discrimination mechanism, takes TOA and PA information in PDW as input, models the pulse sequence through a sliding window, maps the windowed pulse amplitude sequence into a horizontal visibility graph structure, and introduces a joint weighting strategy under the constraint of visibility topology to construct a weighted horizontal visibility graph; on this basis, multi-dimensional topological features reflecting overall connectivity, local energy distribution and global structure stability of the pulse sequence are extracted from the weighted graph structure, a one-class support vector machine discrimination model based on background samples is further constructed, and automatic recognition of the radar scanning mode is realized. The application effectively improves the stability and reliability of the scanning mode recognition under complex conditions such as noise interference and pulse loss, and better meets the engineering application requirements of the electronic reconnaissance system.

Description

Technical Field

[0001] This invention relates to the field of electronic reconnaissance and radar signal processing technology, specifically to an automatic identification method for radar scanning modes (mechanical scanning and phased array scanning) based on time of arrival (TOA) and pulse amplitude (PA) sequences. Background Technology

[0002] In electronic reconnaissance systems, radar scanning mode is one of the key parameters characterizing the working mechanism and radiation characteristics of radar antennas. Its identification results provide important basis for radiation source type identification, threat assessment, and subsequent parameter processing. Existing technologies primarily rely on parameters such as arrival time, amplitude, pulse width (PW), and radio frequency (RF) contained in the pulse descriptor word (PDW). These parameters are used to distinguish between typical scanning modes such as mechanical scanning and phased array scanning through statistical analysis, time-series modeling, or pattern matching.

[0003] Existing scanning pattern identification methods typically analyze the statistical patterns of arrival time or amplitude changes over time to extract periodic indicators, modulation features, or spectral characteristics, which are then used as the basis for scanning pattern identification. While these methods can reflect the periodic changes during the scanning process under ideal conditions, they are susceptible to noise pulses, pulse loss, and measurement errors in real-world complex electromagnetic environments, leading to decreased feature stability and difficulty in guaranteeing identification accuracy. Meanwhile, some methods incorporate machine learning or deep learning models to construct supervised or semi-supervised classifiers for scanning pattern identification. These methods usually rely on sufficient and accurately labeled training samples, requiring high-quality prior information and data. When sample class imbalance or limited labeling information, the model's generalization ability and engineering applicability are limited.

[0004] Therefore, in practical application scenarios where sample annotation information is limited, existing radar scanning identification methods still suffer from problems such as insufficient identification stability and limited applicability in complex electromagnetic environments, making it difficult to meet the reliability requirements in practical applications. Summary of the Invention

[0005] The purpose of this invention is to provide a radar scanning mode identification method based on pulse sequence structural features and anomaly discrimination mechanism, in order to solve the problems of limited sample labeling information and insufficient identification stability in complex electromagnetic environments of existing radar scanning mode identification methods.

[0006] To achieve the above objectives, the present invention employs the following technical solution: A radar scanning mode identification method based on pulse sequence structure features and anomaly detection mechanism includes: Acquire radar radiation source signals, extract the arrival time and pulse amplitude of each pulse, construct arrival time sequence and pulse amplitude sequence, and preprocess the pulse amplitude sequence; The preprocessed pulse amplitude sequence is segmented by a sliding window. For each segmented pulse amplitude subsequence, the pulse amplitude is used as a node to first construct a horizontal visible graph. Then, a joint weighting strategy is introduced to assign edge weights to each undirected edge, thereby constructing a weighted horizontal visible graph. Based on the weighted horizontal visibility map, three feature indicators are extracted for each sliding window pulse amplitude subsequence: average side span, side weight variance, and normalized Laplace gap ratio, and a three-dimensional feature vector is constructed. For radar radiation source signals with known mechanical scanning method, the three-dimensional feature vector corresponding to each sliding window is used as a training sample to construct a training set and then the training samples are standardized. A classification model is constructed using radial basis function kernels, and the model is trained using a training set; during the training process, a discriminant function is established through dual solution. For a segment of radar radiation source signal to be tested, the three-dimensional feature vector corresponding to each sliding window is used as a sample to be tested and standardized. The discriminant function is used to calculate the discriminant output, and the scanning mode identification result of each sample to be tested is obtained based on the preset discriminant rules. The global scanning mode identification result is obtained by using the majority voting strategy.

[0007] Furthermore, the construction process of the weighted level visibility graph is as follows: Each pulse amplitude in the pulse amplitude subsequence Mapped to a node in the horizontally visible graph; for any two nodes in the horizontally visible graph, the index... If for all indexes located between the two All meet Then the node is considered and If the graphs are mutually visible in a horizontal sense, construct an undirected edge; all undirected edges constitute the edge set. The horizontal visible graph is then represented as: ;in A set of nodes; A joint weighting strategy is introduced to assign weights to each undirected edge that reflect both amplitude contrast and temporal proximity; let... The edge weights are defined as follows: ; in, Indicates pulse amplitude contrast. The arrival time span of the pulse amplitude. Indicates the first pulse amplitude Corresponding arrival time This is the absolute deviation of the median; for invisible nodes, it is assumed that: Based on this, a symmetric weighted adjacency matrix can be constructed. Thus, a weighted level visible graph is obtained.

[0008] Furthermore, the time spans of all undirected edges in the weighted level visible graph are statistically analyzed, and their average value is calculated as the average edge span: ; in, This represents the total number of undirected edges; Statistical analysis of the distribution of all edge weights in the weighted level visible graph is performed, and the edge weight variance is calculated: ; in, The average edge weight; To characterize the overall connectivity and structural stability of the weighted horizontally visible graph, a weighted degree matrix is ​​constructed: ; in, Represents a diagonal matrix. It is a node The weighting degree; The total number of nodes; And based on this, the normalized Laplace matrix is ​​obtained: ; in for 3D identity matrix; Perform spectral decomposition on the normalized Laplace matrix, and calculate the normalized Laplace spectral gap ratio using the eigenvalues ​​obtained from the decomposition: ; in, The second smallest eigenvalue obtained after spectral decomposition. It is the largest eigenvalue; To prevent constants with a denominator of zero.

[0009] Furthermore, for the first The pulse amplitude subsequences within each sliding window have the following corresponding three-dimensional feature vectors: ; in, , , They represent the first The average side span, side weight variance, and normalized Laplace gap ratio corresponding to the pulse amplitude subsequences in each sliding window.

[0010] Furthermore, the classification model is as follows: ; in The superscript represents the normal vector of the decision hyperplane, used to define the classification boundary; Indicates transpose; This represents a nonlinear mapping function defined by the radial basis function kernel. For hyperparameters; The number of training samples in the training set; Indicates the sequence number of the training sample; This represents the standardized training samples. For bias terms; These are slack variables; , , These are all self-learning parameters during the training process.

[0011] Furthermore, the classification model is trained using the training set until it converges; during the training process, the discriminant function is obtained through dual solution as follows: ; in, The standardized feature vector corresponding to the sample to be tested; These are Lagrange multipliers, obtained by training a classification model; This is the RBF kernel function.

[0012] Furthermore, the preset discrimination rule is as follows: It was determined to be a mechanical scan. It was determined to be a phased array scan.

[0013] Furthermore, let the signal of the radar radiation source to be measured be... The decision result of the test sample in each sliding window is denoted as a binary variable: ; A majority voting strategy is used to provide sequence-level decisions: ; in, This represents the number of sliding windows; The voting threshold; When the value is 1, the global scanning method identification result is determined to be mechanical scanning; when the value is 0, it is determined to be phased array scanning.

[0014] A terminal device includes a processor, a memory, and a computer program stored in the memory; when the processor executes the computer program, it implements the radar scanning mode identification method based on pulse sequence structure features and anomaly discrimination mechanism.

[0015] A computer-readable storage medium storing a computer program; when the computer program is executed by a processor, it implements the radar scanning mode identification method based on pulse sequence structural features and anomaly discrimination mechanism.

[0016] Compared with the prior art, the present invention has the following technical features: This invention maps radar arrival time and amplitude sequences to a horizontally visible graph, achieving a graphical model representation of the structural characteristics of pulse sequences. This allows for the characterization of distributional differences between different scanning methods at the overall structural level. Furthermore, by introducing a weighted strategy under the topological constraints of the horizontally visible graph, this invention integrates pulse amplitude differences and time-series span information into the graph structure modeling process. This enhances the ability to express the structural differences of pulse sequences from different scanning methods and reduces the impact of noise interference and pulse loss on recognition performance. Finally, by extracting multi-dimensional topological features and constructing a support vector machine discriminant model based on background samples, this invention achieves effective recognition of radar scanning methods even with only single-class or weakly labeled samples. This reduces reliance on prior information and large-scale labeled samples while ensuring controllable computational complexity. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of TOA-PA distribution; where (a) is the mechanical scan TOA-PA data distribution, and (b) is the phase scan TOA-PA data distribution. Figure 3 This is a schematic diagram of the construction of the weighted horizontal visibility map of the present invention; wherein (a) is a schematic diagram of the mechanically scanned weighted horizontal visibility map, and (b) is a schematic diagram of the phase scanned weighted horizontal visibility map; Figure 4 This is a schematic diagram of the distribution of mechanical scanning and phased array scanning in the three-dimensional topological feature space (MES, EWV, LSG); Figure 5 This is a schematic diagram of the decision score distribution based on OC-SVM; Figure 6 This is a schematic diagram of the ROC curve of the method of the present invention. Detailed Implementation

[0018] This invention provides a radar scanning mode identification method based on pulse sequence structural features and anomaly discrimination mechanism. It models a horizontally visible graph of radar arrival time and amplitude sequences, and introduces a weighted strategy under topological constraints to enhance the representation of pulse sequence structural characteristics. Furthermore, it extracts multi-dimensional topological features reflecting graph structural differences and constructs a one-class support vector machine (OC-SVM) discrimination model based on background samples. This enables effective identification of different scanning modes, thereby improving the stability and reliability of scanning mode identification under complex conditions such as noise interference and pulse loss, even when relying solely on single-class or weakly labeled samples, thus meeting the engineering application requirements of electronic reconnaissance systems.

[0019] See Figure 1 The basic idea behind this invention is as follows: Using the TOA and PA information from the PDW as input, a sliding window model is applied to the pulse sequence, mapping the windowed pulse amplitude sequence to a horizontal visibility graph structure. A joint weighting strategy is introduced under visibility topological constraints to construct a weighted horizontal visibility graph. Based on this, multi-dimensional topological features reflecting the overall connectivity, local energy distribution, and global structural stability of the pulse sequence are extracted from the weighted graph structure. Furthermore, a support vector machine discriminant model based on background samples is constructed to achieve automatic identification of radar scanning methods. The specific steps of this invention are as follows: Step 1: Pulse parameter extraction and sequence construction.

[0020] In an electronic reconnaissance system, the receiver intercepts and detects radar radiation source signals, extracts the arrival time (TOA) and pulse amplitude (PA) from the pulse pulse's pulse distance (PDW), sorts them in chronological order to form an arrival time sequence and a pulse amplitude sequence, and preprocesses the pulse amplitude sequence.

[0021] Suppose that a total of [number] results were obtained during the observation period. If there are 1 effective pulse, it can be expressed as: (1); in For the first Time of Arrival (TOA) The corresponding pulse amplitude PA; Indicates arrival time series, This represents a pulse amplitude sequence.

[0022] To reduce the impact of dimensional drift caused by different batches and receiving conditions, the pulse amplitude sequence is preprocessed (e.g., normalized / standardized) to obtain: (2); in, Indicates to Perform normalization / standardization operations.

[0023] This yields the pulse amplitude sequence used for subsequent graph modeling: (3); Step 2: Segmented modeling using a sliding window.

[0024] To eliminate the impact of uneven pulse number and sequence length differences on the statistical stability of features, the preprocessed pulse amplitude sequence was... Perform sliding window segmentation.

[0025] Let the window length be Step size is Then the first The pulse amplitude subsequences corresponding to each window are: (4); in And the total number of windows satisfies: (5); in, This indicates rounding down to the nearest integer.

[0026] Sliding window processing ensures that each pulse amplitude subsequence has a consistent dimension. This facilitates batch training and inference of subsequent classifiers.

[0027] Step 3: Construct the weighted horizontal visibility graph.

[0028] For any divided pulse amplitude subsequence: (6); Construct its horizontal visibility graph; extract the amplitude of each pulse in the pulse amplitude subsequence. Mapped to a node in the horizontally visible graph , .

[0029] For any two node indices in the horizontal visible graph If for all indexes located between the two All satisfy: (7); Then the node is considered and They are mutually visible in a horizontal sense. Establish an undirected edge in the graph. ;in Indicates that the index is Pulse amplitude at the location; This indicates finding the minimum value; It is a universal quantifier symbol.

[0030] Form an edge set from the undirected edges of all node pairs that satisfy equation (7) above. The basic unweighted level visible graph is obtained as follows: (8); in It is a set of nodes.

[0031] To enhance the characterization of energy distribution patterns under different scanning methods, this invention introduces a joint weighting strategy while satisfying the visibility requirement of equation (7), assigning weights reflecting amplitude contrast and temporal proximity to each undirected edge; let... The edge weights are defined as follows: (9); in, Indicates pulse amplitude contrast. The arrival time span of the pulse amplitude is determined using the arrival time series. Obtain and calculate; Indicates the first pulse amplitude The corresponding arrival time (TOA); This represents the absolute deviation of the median, used to suppress the influence of outliers; for invisible nodes, it is assumed that: (10); Based on this, a symmetric weighted adjacency matrix can be constructed. This allows for the construction of a weighted, leveled visibility graph.

[0032] The median absolute deviation refers to the median of the absolute deviations of a given node from the medians of all nodes, i.e. ; This indicates the calculation of the median; see the schematic diagram of the weighted level for different scanning methods in the embodiment. Figure 3 .

[0033] Step 4: Weighted graph topological feature extraction.

[0034] After constructing a weighted horizontal visibility graph, topological features that can distinguish different radar scanning methods are extracted from the graph structure; this invention designs the following three types of feature indices: (1) Mean Edge Span (MES).

[0035] The average value of all undirected time spans in the weighted level visible graph is calculated to describe the overall connectivity and extensibility of the graph structure. It is defined as follows: (11); in, express The modulus here refers to the total number of undirected edges; express The undirected edges between them belong to .

[0036] The periodicity of mechanical scanning pulses ensures visibility between distant nodes, resulting in a larger MES value; while the distribution of phase scanning pulses is more random, leading to a relatively smaller MES value.

[0037] (2) Edge Weight Variance (EWV).

[0038] The distribution of all edge weights in the weighted visible graph is statistically analyzed, and the variance of the edge weights is calculated. This variance is used to characterize the dispersion of the edge weight distribution in the graph structure and is defined as: (12); in, This represents the average edge weight.

[0039] Mechanical scanning results in a significant difference in amplitude between the main lobe and side lobes, leading to a larger EWV; phase-scanned sequences have a more uniform energy distribution and a lower EWV value.

[0040] (3) Normalized Laplacian Spectral Gap (LSG).

[0041] To characterize the overall connectivity and structural stability of the weighted horizontally visible graph, a weighted degree matrix is ​​constructed: (13); in, Represents a diagonal matrix. It is a node The weighted degree is defined as the degree of the node. The sum of the weights of all connected edges.

[0042] And based on this, the normalized Laplace matrix is ​​obtained: (14); in for 3D identity matrix; edge weight The weighted adjacency matrix is ​​formed.

[0043] The normalized Laplace matrix is ​​a symmetric positive semi-definite matrix, and its spectral properties reflect the connectivity and structural characteristics of the graph; The eigenvalues ​​(obtained through spectral decomposition) are arranged in ascending order as follows: (15); The normalized Laplace spectral gap ratio is then defined as: (16); in, The second smallest eigenvalue obtained after spectral decomposition. It is the largest eigenvalue; To prevent small constants with a denominator of zero, for example The normalized Laplace gap ratio (LSG) is used to measure the global connectivity and stability of a graph; the larger the value, the closer the graph is to an ideal connectivity structure (such as strongly connected and non-bisection), and vice versa, it may be close to bisection or weak connectivity.

[0044] Mechanical scanning sequences have a compact structure and relatively high LSG values; phase scanning sequences have a loose structure and lower LSG values.

[0045] The structural characteristics of windowed pulse sequences can be characterized from different levels using the three types of topological features MES, EWV, and LSG, forming feature vectors for scanning mode identification.

[0046] For the The topological features of the pulse amplitude subsequences in the sliding window are obtained by calculating using equations (11)-(16): (17); in, , , They represent the first The average side span, side weight variance, and normalized Laplace gap ratio corresponding to the pulse amplitude subsequences in each sliding window.

[0047] And combine them to form a three-dimensional feature vector: (18); Thus, the original pulse amplitude sequence is mapped onto a three-dimensional topological feature space, achieving a compact representation of the structural differences between mechanical scanning and phased array scanning. In the embodiment, the distributions of mechanical scanning and phased array scanning in the three-dimensional feature space are as follows: Figure 4 As shown.

[0048] Step 5: Weak supervision identification based on OC-SVM.

[0049] First, for radar radiation source signals with known radar scanning method of mechanical scanning, the three-dimensional feature vectors corresponding to each sliding window are obtained as training samples using the methods in steps 1 to 4, thereby constructing a training set; here only one scanning method of training samples is needed because the subsequent OC-SVM model can determine that when the sample to be tested is an unknown sample, it is another phased array scanning method based on the learning of the training samples.

[0050] To eliminate the influence of differences in feature dimensions and numerical ranges on the distance metric of the kernel function, the three-dimensional feature vector of each training sample in the training set is standardized (i.e., each type of feature index in the three-dimensional feature vector is standardized separately); let the feature vector of the training set be... Defined as: (19); in, and Let be the standard deviation of each class of feature index in the three-dimensional feature vector of the training samples in the training set, respectively. To prevent small constants with zero denominators, for example, Subsequent training and discrimination are both based on standardized features. .

[0051] A classification model OC-SVM is constructed using a radial basis function (RBF) kernel; the optimization objective of OC-SVM can be expressed as: (20); in, The hyperplane normal vector is used to define the classification boundary and serves as a self-learning parameter during training. It ensures that all training samples in the training set lie on the same side of the hyperplane, thus distinguishing between mechanical scanning and phased array scanning. (Superscript) Indicates transpose; This represents a nonlinear mapping function defined by the radial basis function kernel. Hyperparameters for controlling the proportion of abnormal training samples; The number of training samples in the training set; These are bias terms, which are self-learning parameters during the training process; These are slack variables, reflecting the tolerance for abnormal samples, and are self-learning parameters during the training process.

[0052] The OC-SVM model described above is trained using the training set until it converges. During training, the discriminant function can be obtained through dual solution: (twenty one); in, The standardized feature vector corresponding to the sample to be tested; These are Lagrange multipliers, obtained by training the OC-SVM model; This refers to the bias term obtained during model training; This is the kernel function. This invention uses the RBF kernel: (twenty two); in The kernel width parameter can be set as the median distance between features or through cross-validation. It is a natural exponential function.

[0053] Step 6: Multi-window decision fusion and scanning method output.

[0054] In practical applications, for a segment of radar radiation source signal to be measured, the three-dimensional feature vector corresponding to each sliding window is obtained as the sample to be measured using the methods in steps 1 to 4. The sample to be measured is then standardized to obtain the standardized feature vector. Substituting the discriminant function into equation (21) to calculate the discriminant output The present invention adopts the following decision rules: It was determined to be a mechanical scan. It was determined to be a phased array scan (23); By maximizing the boundary margin in the feature space, the aforementioned OC-SVM model enables mechanically scanned samples to form a compact background distribution in a high-dimensional space, while identifying samples that do not conform to this distribution as phased array scans, thereby achieving reliable scanning method differentiation under weak supervision. In the embodiment, the window-level decision score distribution and the optimal threshold position are as follows: Figure 5 As shown.

[0055] A series of test samples constructed based on the radar radiation source signal can all obtain the decision result through equation (23); in order to obtain the global scanning mode identification result corresponding to the entire observation period, this invention fuses the window-level decision results: Let the signal of the radar radiation source to be measured be the first... The decision result of the test sample in each sliding window is denoted as a binary variable: (twenty four); A majority voting strategy can then be used to give a sequence-level decision: (25); in, This represents the number of sliding windows; As the voting threshold, when This corresponds to a strict majority voting rule; When the value is 1, the global scanning method identification result is determined to be mechanical scanning; when the value is 0, it is determined to be phased array scanning. Through the above steps, this invention achieves automatic identification of mechanical scanning and phased array scanning modes by utilizing only the TOA and PA sequences obtainable by an electronic reconnaissance receiver, without relying on the specific internal structure and signal generation mechanism of the radar. This can effectively improve the accuracy and robustness of radar threat identification and classification in complex electromagnetic environments.

[0056] Example: To verify the effectiveness of the algorithm, simulation data for mechanical scanning and phase scanning were first constructed.

[0057] In this implementation example, the total observation time is set to During this time period, the TOA and PA parameters are recorded to form a TOA-PA sequence. To simulate a complex electromagnetic environment, background noise pulses are superimposed in both scanning modes, with the number of noise pulses set to [value missing]. The noise pulse amplitude baseline is set as follows: ; in This represents an exponentially distributed random term.

[0058] In this implementation example, the mechanically scanned radar adopts a periodic scanning model, and its scanning period parameter is set to... The standard deviation of the period jitter is set to , No. The modeling of the strip center time corresponding to the next scan is as follows: ; Within each scan strip, the number of pulses is set to... At the same time, set the strip time extension half-width to The TOA of the pulses inside the strip is: ; in Let the random variable follow a centrally dense distribution. The standard deviation is The time jitter.

[0059] The pulse amplitude of mechanically scanned radar adopts a sinc-like main lobe-side lobe model: ; The main lobe peak value is set as follows: The standard deviation of amplitude noise is set to The TOA-PA distribution generated by the above parameters exhibits a periodic band structure on the time axis, as shown in the attached figure. Figure 2 As shown in (a).

[0060] The phased array scanning radar employs a non-periodic electronic scanning model, and the centers of its scanning strips follow a uniform distribution during the observation time: ; The total number of bands follows a normal distribution with a mean of 90 and a standard deviation of . The number of pulses in each scan strip is set to At the same time, set the strip time extension width to The pulse amplitude is composed of a random baseline and a weak bulge term: ; The baseline amplitude satisfies The amplitude coefficient of the weak convexity satisfies In this configuration, the TOA-PA distribution of the phased array scanning radar lacks a periodic strip structure, and the energy distribution is relatively uniform, such as... Figure 2 As shown in (b).

[0061] In this embodiment, the sliding window length is set. The step size is 128. The parameter is 64. Finally, the detection probability and false alarm probability at different thresholds are statistically analyzed, and ROC curves are plotted, such as... Figure 6 As shown. Experimental results show that, under conditions of superimposed noise pulses and parameter perturbations, the method of the present invention can still stably distinguish between mechanical scanning and phased array scanning modes, verifying the effectiveness of the method of the present invention.

[0062] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A radar scanning mode identification method based on pulse sequence structural features and anomaly discrimination mechanism, characterized in that, include: Acquire radar radiation source signals, extract the arrival time and pulse amplitude of each pulse, construct arrival time sequence and pulse amplitude sequence, and preprocess the pulse amplitude sequence; The preprocessed pulse amplitude sequence is segmented by a sliding window. For each segmented pulse amplitude subsequence, the pulse amplitude is used as a node to first construct a horizontal visible graph. Then, a joint weighting strategy is introduced to assign edge weights to each undirected edge, thereby constructing a weighted horizontal visible graph. Based on the weighted horizontal visibility map, three feature indicators are extracted for each sliding window pulse amplitude subsequence: average side span, side weight variance, and normalized Laplace gap ratio, and a three-dimensional feature vector is constructed. For radar radiation source signals with known mechanical scanning method, the three-dimensional feature vector corresponding to each sliding window is used as a training sample to construct a training set and then the training samples are standardized. A classification model is constructed using radial basis function kernels, and the classification model is trained using the training set; During the training process, a discriminant function is established through dual solution; For a segment of radar radiation source signal to be tested, the three-dimensional feature vector corresponding to each sliding window is used as a sample to be tested and standardized. The discrimination function is used to calculate the discrimination output, and the scanning method identification result of each sample to be tested is obtained based on the preset discrimination rules. The global scanning method is used to obtain the identification results.

2. The radar scanning mode identification method based on pulse sequence structural features and anomaly discrimination mechanism according to claim 1, characterized in that, The process of constructing the weighted level visibility graph is as follows: Each pulse amplitude in the pulse amplitude subsequence Mapped to a node in the horizontally visible graph; for any two nodes in the horizontally visible graph, the index... If for all indexes located between the two All meet Then the node is considered and If the graphs are mutually visible in a horizontal sense, construct an undirected edge; all undirected edges constitute the edge set. The horizontal visible graph is then represented as: ;in A set of nodes; A joint weighting strategy is introduced to assign weights to each undirected edge that reflect both amplitude contrast and temporal proximity; let... The edge weights are defined as follows: ； in, Indicates pulse amplitude contrast. The arrival time span of the pulse amplitude. Indicates the first pulse amplitude Corresponding arrival time This is the absolute deviation of the median; for invisible nodes, it is assumed that: Based on this, a symmetric weighted adjacency matrix can be constructed. Thus, a weighted level visible graph is obtained.

3. The radar scanning mode identification method based on pulse sequence structural features and anomaly discrimination mechanism according to claim 1, characterized in that, The time spans of all undirected edges in the weighted level visible graph are statistically analyzed, and their average value is calculated as the average edge span: ； in, This represents the total number of undirected edges; Statistical analysis of the distribution of all edge weights in the weighted level visible graph is performed, and the edge weight variance is calculated: ； in, The average edge weight; To characterize the overall connectivity and structural stability of the weighted horizontally visible graph, a weighted degree matrix is ​​constructed: ； in, Represents a diagonal matrix. It is a node The weighting degree; The total number of nodes; And based on this, the normalized Laplace matrix is ​​obtained: ； in for 3D identity matrix; Perform spectral decomposition on the normalized Laplace matrix, and calculate the normalized Laplace spectral gap ratio using the eigenvalues ​​obtained from the decomposition: ； in, The second smallest eigenvalue obtained after spectral decomposition. It is the largest eigenvalue; To prevent constants with a denominator of zero.

4. The radar scanning mode identification method based on pulse sequence structural features and anomaly discrimination mechanism according to claim 1, characterized in that, For the The pulse amplitude subsequences within each sliding window have the following corresponding three-dimensional feature vectors: ； in, , , They represent the first The average side span, side weight variance, and normalized Laplace gap ratio corresponding to the pulse amplitude subsequences in each sliding window.

5. The radar scanning mode identification method based on pulse sequence structural features and anomaly discrimination mechanism according to claim 1, characterized in that, The classification model is as follows: ； in The superscript represents the normal vector of the decision hyperplane, used to define the classification boundary; Indicates transpose; This represents a nonlinear mapping function defined by the radial basis function kernel. For hyperparameters; The number of training samples in the training set; Indicates the sequence number of the training sample; This represents the standardized training samples. For bias terms; These are slack variables; , , These are all self-learning parameters during the training process.

6. The radar scanning mode identification method based on pulse sequence structural features and anomaly discrimination mechanism according to claim 5, characterized in that, The classification model is trained using the training set until it converges. During the training process, the discriminant function is obtained through dual solution as follows: ； in, The standardized feature vector corresponding to the sample to be tested; These are Lagrange multipliers, obtained by training a classification model; This is the RBF kernel function.

7. The radar scanning mode identification method based on pulse sequence structural features and anomaly discrimination mechanism according to claim 6, characterized in that, The preset discrimination rules are as follows: It was determined to be a mechanical scan. It was determined to be a phased array scan.

8. The radar scanning mode identification method based on pulse sequence structural features and anomaly discrimination mechanism according to claim 1, characterized in that, Let the signal of the radar radiation source to be measured be the first... The decision result of the test sample in each sliding window is denoted as a binary variable: ； A majority voting strategy is used to provide sequence-level decisions: ； in, This represents the number of sliding windows; The voting threshold; When the value is 1, the global scanning method identification result is determined to be mechanical scanning; when the value is 0, it is determined to be phased array scanning.

9. A terminal device, comprising a processor, a memory, and a computer program stored in the memory; characterized in that, When the processor executes the computer program, it implements the radar scanning mode identification method based on pulse sequence structure features and anomaly discrimination mechanism as described in any one of claims 1-8.

10. A computer-readable storage medium storing a computer program; characterized in that, When the computer program is executed by the processor, it implements the radar scanning mode identification method based on pulse sequence structural features and anomaly discrimination mechanism as described in any one of claims 1-8.

Images