Passive perception radar target classification method and system based on electromagnetic detection feature extraction

Through the method of multi-dimensional signal decomposition and dynamic weight allocation, the problem of feature fusion inaccuracy in passive radar target classification is solved, and high-precision target recognition in complex electromagnetic environments is achieved.

CN120724248AActive Publication Date: 2025-09-30BEIJING JUNDE INTELLIGENT COMPUTING TECHNOLOGY CO LTD

Patent Information

Application Number
CN202510849668.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-06-24
Publication Date
2025-09-30
Estimated Expiration
2045-06-24

AI Technical Summary

Technical Problem

Existing passive radar target classification methods have difficulty distinguishing targets of different materials with similar polarization responses in complex electromagnetic environments, and the static weight mechanism cannot adapt to environmental noise fluctuations, resulting in inaccurate feature fusion. Traditional classifiers lack electromagnetic physical mechanism modeling and find it difficult to analyze effective features in weak scattered signals.

Method used

Through multi-dimensional signal decomposition processing, the time-frequency distribution characteristics and polarization response characteristics are obtained. Combined with the dynamic weight allocation strategy of electromagnetic physics mechanism, a fusion feature vector is generated and input into the adaptive classification network to achieve high-precision distinction of target scatterers.

Benefits of technology

High-precision distinction of multiple types of scatterers is achieved in complex electromagnetic environments, the robustness of feature representation and the environmental adaptability of the classification process are enhanced, and the reliability and computational efficiency of target recognition are improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120724248A_ABST
    Figure CN120724248A_ABST
Patent Text Reader

Abstract

The invention provides a passive perception radar target classification method and system based on electromagnetic detection feature extraction, and the method comprises the steps: carrying out the multi-dimensional signal decomposition of an original scattered signal sequence through obtaining an electromagnetic radiation signal set of a target region; obtaining a time-frequency distribution feature set and a polarization response feature set corresponding to each signal acquisition period, extracting a target electromagnetic feature combination from the time-frequency distribution feature set and the polarization response feature set according to a preset electromagnetic feature association rule, generating a weight distribution parameter, and performing weighted fusion processing on the target electromagnetic feature combination to obtain a target electromagnetic feature combination; and the fusion feature vector set is input into an adaptive classification network, hierarchical classification operation is performed on the fusion feature vector set, the category identifier of the target scatterer and the corresponding classification confidence are output, and a radar detection optimization instruction is generated based on the category identifier and the corresponding classification confidence. According to the method, high-precision distinguishing of multiple types of scatterers is achieved under the passive sensing condition, and meanwhile the environmental adaptability and the calculation efficiency of the classification process are guaranteed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a passive perception radar target classification method and system based on electromagnetic detection feature extraction. Background Art

[0002] In the field of electromagnetic sensing, passive radar target classification methods, which analyze ambient electromagnetic radiation to achieve target identification, have gradually become a key technology in electronic reconnaissance and remote sensing monitoring. Existing methods often employ fixed-pattern feature extraction strategies, such as separately analyzing the time-frequency energy distribution or polarization response characteristics of the target's scattered signal, fusing them with static weights, and then inputting them into a general classifier for classification. Such methods have inherent limitations in complex electromagnetic environments: time-frequency features make it difficult to distinguish targets of different materials with similar polarization responses, while relying solely on polarization features cannot capture the dynamic scattering characteristics of the target. Static weighting mechanisms cannot adapt to changes in feature reliability caused by fluctuations in ambient noise, resulting in inaccurate feature fusion. Traditional classifiers lack embedded modeling of electromagnetic physics mechanisms, making it difficult to extract effective features from weak scattered signals under passive sensing conditions. These shortcomings severely restrict the ability to accurately identify camouflaged and composite-material targets in complex battlefield environments and civilian remote sensing scenarios, limiting the practical application of passive sensing technology. Summary of the Invention

[0003] The present invention provides a passive perception radar target classification method and system based on electromagnetic detection feature extraction.

[0004] In a first aspect, an embodiment of the present invention provides a passive perception radar target classification method based on electromagnetic detection feature extraction, comprising: Acquire an electromagnetic radiation signal set of a target area, wherein the electromagnetic radiation signal set includes a sequence of original scattered signals collected during multiple signal collection cycles; Performing multi-dimensional signal decomposition processing on the original scattered signal sequence to obtain a time-frequency distribution feature set and a polarization response feature set corresponding to each signal acquisition period; Extracting a target electromagnetic feature combination from the time-frequency distribution feature set and the polarization response feature set according to a preset electromagnetic feature association rule, and generating a weight allocation parameter; Performing weighted fusion processing on the target electromagnetic feature combination based on the weight distribution parameters to generate a fused feature vector set, and inputting the fused feature vector set into an adaptive classification network; A hierarchical classification operation is performed on the fused feature vector set through the adaptive classification network, a category identifier of the target scatterer and a corresponding classification confidence are output, and a radar detection optimization instruction is generated according to the classification confidence.

[0005] In a second aspect, an embodiment of the present invention provides a computer system, including: a memory storing a computer program; A processor is used to load the computer program to implement the passive perception radar target classification method based on electromagnetic detection feature extraction as described above.

[0006] The passive sensing radar target classification method based on electromagnetic detection feature extraction provided by the present invention synchronously obtains the time-frequency distribution characteristics and polarization response characteristics of the target scatterer through multi-dimensional signal decomposition processing, and realizes the optimized fusion of multi-dimensional features based on the dynamic weight allocation strategy of electromagnetic physics mechanism, effectively overcoming the problem of limited classification accuracy caused by the insufficient characterization capability of traditional methods. The transient electromagnetic response characteristics of the target scatterer are accurately characterized by time-frequency characteristics, and the target material and geometric structure information are deeply analyzed by combining polarization characteristics. The contribution weights of the two types of features are adaptively adjusted under different electromagnetic environments through dynamic weight allocation, significantly enhancing the robustness of feature representation in complex interference scenarios. Furthermore, a hierarchical adaptive classification network is used to make progressive decisions on the fused features, and the physical laws of electromagnetic scattering are embedded in the decision logic of the classification model. This enables the classification process to fully utilize the physical interpretability of electromagnetic features while adapting to the time-varying characteristics of the actual detection environment. Under passive sensing conditions, high-precision distinction of multiple types of scatterers is achieved, while ensuring the environmental adaptability and computational efficiency of the classification process, providing reliable technical support for target recognition in complex electromagnetic environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0008] Figure 1 This is a flowchart of a passive perception radar target classification method based on electromagnetic detection feature extraction provided by an embodiment of the present invention.

[0009] Figure 2 It is a schematic diagram of the composition of a computer system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0010] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0011] See also Figure 1 , Figure 1 A flowchart of a passive perception radar target classification method based on electromagnetic detection feature extraction provided by an embodiment of the present invention is provided. The passive perception radar target classification method based on electromagnetic detection feature extraction can be executed by a computer system. The passive perception radar target classification method based on electromagnetic detection feature extraction may include the following steps: Step S100: Acquire an electromagnetic radiation signal set of a target area, wherein the electromagnetic radiation signal set includes a sequence of original scattered signals collected in multiple signal collection cycles.

[0012] The target area is the spatial range for radar detection. Within this area, various target scatterers, such as aircraft, vehicles, and buildings, may exist. The electromagnetic radiation signal set is the collection of all electromagnetic radiation signals within the target area, acquired during different signal acquisition cycles. The raw scattered signal sequence is a chronological sequence of electromagnetic signals scattered from the target area. These signals contain various information about the target scatterers.

[0013] Acquiring a target area's electromagnetic radiation signal set can be accomplished by deploying radar receiving equipment around the target area. For example, when surveying a target area around an airport, multiple radar receiving nodes can be deployed at different locations within the airport. Each node collects electromagnetic radiation signals from the target area over multiple signal collection cycles, thereby generating a sequence of raw scattered signals. Each signal collection cycle can be set to a specific interval, for example, 1 second. During each 1-second cycle, the radar receiving node continuously collects electromagnetic radiation signals. Ultimately, the signals collected during these cycles are combined into a sequence of raw scattered signals, which in turn forms a set of electromagnetic radiation signals.

[0014] As an embodiment, before the step S100 of acquiring the electromagnetic radiation signal set of the target area, the following signal preprocessing steps S101 to S103 are also included: Step S101: deploy multiple passive radar receiving nodes, configure the frequency band coverage and polarization receiving mode of each receiving node, and form a spatially distributed detection array.

[0015] Passive radar receiving nodes are devices that do not actively transmit electromagnetic signals, but only receive electromagnetic signals scattered from the target area. Frequency band coverage refers to the frequency range of electromagnetic signals that a radar receiving node can receive. Different target scatterers may have more pronounced scattering characteristics in different frequency bands. Therefore, properly configuring frequency band coverage can improve target detection capabilities. Polarization reception mode refers to the way a radar receiving node receives electromagnetic signals with different polarization directions. Common polarization modes include horizontal polarization and vertical polarization. Different polarization modes can obtain different polarization characteristics of target scatterers. A spatially distributed detection array is an array composed of multiple passive radar receiving nodes arranged in a predetermined spatial layout. This array can detect the target area from different angles and positions, thereby obtaining more comprehensive electromagnetic information about the target.

[0016] When deploying multiple passive radar receiving nodes, the number and location of nodes must be determined based on factors such as the size and shape of the target area, as well as the types of targets that may appear. For example, in a target area within an urban environment, passive radar receiving nodes can be deployed in various corners and key locations within the city to provide coverage across the entire city. Frequency band coverage can be configured based on the scattering characteristics of common target scatterers in different frequency bands. For example, aircraft targets may have more pronounced scattering characteristics in certain high-frequency bands, so the frequency band coverage of some receiving nodes can be set to these high-frequency bands. For polarization reception mode configuration, both horizontal and vertical polarization modes can be set simultaneously, allowing for acquisition of scattering information from target scatterers in different polarization directions. By rationally deploying and configuring these receiving nodes, an effective spatially distributed detection array can be formed.

[0017] Step S102: Synchronize the local clock source of each receiving node to eliminate signal phase errors caused by time asynchrony.

[0018] The local clock source is the internal clock system used for timing in each passive radar receiving node. Due to potential deviations in the clock systems of various receiving nodes, the times at which they acquire signals may differ. This asynchrony can cause signal phase errors. This phase error can affect subsequent signal processing and analysis, potentially leading to inaccurate target feature extraction and, consequently, target classification accuracy.

[0019] Various methods can be used to synchronize the local clock sources of each receiving node, such as using Global Positioning System (GPS) clock synchronization technology. GPS satellites provide highly accurate time signals. Each passive radar receiving node can receive the GPS time signal through a GPS receiver and synchronize its local clock source with GPS time. The specific implementation process is as follows: After receiving the GPS time signal, the GPS receiver of each receiving node compares it with the local clock and calculates the time difference between the two. The receiving node then adjusts its local clock based on this time difference to align it with GPS time. This method effectively eliminates signal phase errors caused by time asynchrony.

[0020] Step S103: Performing local oscillator leakage suppression processing on the original signal collected by each receiving node, and using an adaptive filtering algorithm to eliminate the device's own radiation interference; wherein the execution steps of the adaptive filtering algorithm include: establishing a transfer function model of the hardware characteristics of the receiving node, and offsetting the leakage signal component through a reverse compensation operation.

[0021] Local oscillator leakage occurs when the local oscillator signal within the radar receiver leaks into the receiving channel and mixes with the target's scattered signal, interfering with the acquisition and processing of the target signal. Device-specific radiation interference is electromagnetic radiation generated by the radar receiver during operation, which can affect the accurate acquisition of electromagnetic signals from the target area. Adaptive filtering algorithms automatically adjust filter parameters based on the statistical characteristics of the input signal, effectively suppressing interference components in the signal.

[0022] Establishing a transfer function model of the receiving node's hardware characteristics is a process within the adaptive filtering algorithm. This model describes the transformation relationship between the receiving node's hardware input and output signals. It can be derived by analyzing the receiving node's circuit structure and operating principle. For example, for a passive radar receiving node based on an RF front-end circuit, its transfer function model can be established by measuring the relationship between its input and output signals using system identification methods. Specifically, a series of known test signals can be input to the receiving node, its output signals recorded, and then the parameters of the transfer function model can be fitted using methods such as least squares. The fitting process can refer to existing technologies.

[0023] After obtaining the transfer function model, the leakage signal component is offset by the reverse compensation operation. Specifically, the estimated value of the local oscillator leakage signal is calculated based on the transfer function model, and then the inverse signal of this estimated value is superimposed on the original signal to offset the influence of the leakage signal. For example, if the local oscillator leakage signal calculated by the transfer function model is S leak , then -S leak Superimposed on the original signal S rawThe signal S after local oscillator leakage suppression is obtained. processed =S raw -S leak This adaptive filtering algorithm can effectively eliminate the device's own radiation interference and improve signal quality.

[0024] Step S200: performing multi-dimensional signal decomposition processing on the original scattered signal sequence to obtain a time-frequency distribution feature set and a polarization response feature set corresponding to each signal acquisition period.

[0025] Multidimensional signal decomposition analyzes and decomposes the original scattered signal sequence along multiple dimensions to extract richer signal features. The time-frequency distribution feature set describes the signal's distribution characteristics in both time and frequency. It can reveal information such as the motion and electromagnetic properties of the target scatterer. The polarization response feature set describes the signal's response characteristics in different polarization directions. It can reveal information such as the target scatterer's shape and material.

[0026] There are various methods for performing multidimensional signal decomposition on the original scattered signal sequence. For example, time-domain analysis, frequency-domain analysis, and polarization analysis can be combined. By analyzing the signal's distribution in time and frequency, time-frequency distribution characteristics can be obtained; by analyzing the signal's response in different polarization directions, polarization response characteristics can be obtained. In practical applications, the original scattered signal sequence can first be converted into the time and frequency domains, and then the corresponding characteristic parameters can be extracted. At the same time, the signal can be polarization-channel separated to extract the polarization response characteristic parameters, ultimately forming a time-frequency distribution feature set and a polarization response feature set, respectively.

[0027] As an embodiment, step S200, performing multi-dimensional signal decomposition processing on the original scattered signal sequence to obtain a time-frequency distribution feature set and a polarization response feature set corresponding to each signal acquisition period, may specifically include the following steps S210 to S240: Step S210: performing a time domain alignment operation on the original scattered signal sequence to eliminate the signal timing offset caused by environmental interference and generate a standardized time domain signal sequence.

[0028] Time domain alignment adjusts the original scattered signal sequence on the time axis to ensure temporal consistency. Environmental interference, such as electromagnetic interference and noise around the target area, can cause temporal offsets in the original scattered signal sequence, resulting in inaccurate signal arrival times. This signal timing offset can affect subsequent signal analysis and processing, so time domain alignment is necessary to eliminate this offset. The standardized time domain signal sequence obtained after time domain alignment provides a uniform temporal reference, facilitating subsequent feature extraction and analysis.

[0029] The time domain alignment operation can be performed using the correlation analysis method. The specific steps are as follows: First, select a reference signal, which can be a known standard signal or a signal collected without interference. Then, calculate the correlation between the original scattered signal sequence and the reference signal, and find the position with the largest correlation. The time offset corresponding to this position is the amount that needs to be adjusted. Finally, the original scattered signal sequence is time-shifted according to this time offset, thereby eliminating the signal timing offset and generating a standardized time domain signal sequence. For example, if the reference signal is S ref (t), the original scattered signal sequence is S raw (t), by calculating the correlation function ,turn up The maximum value corresponding to , S raw (t) shift in time , get the standardized time domain signal sequence .

[0030] Step S220: using a multi-resolution orthogonal decomposition algorithm to divide the standardized time-domain signal sequence into frequency bands, extracting signal energy distribution parameters and phase continuity parameters in each frequency band to form an initial time-frequency feature set.

[0031] The multi-resolution orthogonal decomposition algorithm is an algorithm that can decompose a signal at different resolutions. It can decompose the signal into sub-signals with different frequency components. Band division is the division of the signal's frequency range into multiple different frequency bands, each corresponding to a different frequency interval. The signal energy distribution parameter is the energy distribution of the signal within each frequency band, which can reflect the scattering ability of the target scatterer at different frequencies. The phase continuity parameter is the continuous change of the signal phase within each frequency band, which can reflect information such as the motion characteristics of the target scatterer. The initial time-frequency feature set is a feature set composed of the signal energy distribution parameters and phase continuity parameters in each frequency band. It is the basis for subsequent further processing and analysis.

[0032] A possible multi-resolution orthogonal decomposition algorithm is, for example, a wavelet decomposition algorithm. Taking the wavelet decomposition algorithm as an example, the specific steps for frequency band division of the standardized time domain signal sequence are as follows: First, select a suitable wavelet basis function, such as the Daubechies wavelet basis. Then, perform wavelet decomposition on the standardized time domain signal sequence and decompose it into wavelet coefficients of different scales and different positions. The wavelet coefficients corresponding to each scale can be regarded as components of the signal in different frequency bands. Next, calculate the signal energy distribution parameters in each frequency band. For example, the signal energy in the frequency band can be obtained by calculating the sum of the squares of the wavelet coefficients. For the phase continuity parameter, it can be extracted by analyzing the phase change of the wavelet coefficients. For example, the phase difference between adjacent wavelet coefficients can be calculated, and the phase continuity parameter can be obtained by statistically analyzing the distribution of these phase differences. Finally, the signal energy distribution parameters and phase continuity parameters in each frequency band are combined to form an initial time-frequency feature set.

[0033] Step S230: synchronously performing polarization channel separation processing on the original scattered signal sequence to obtain scattering intensity difference parameters and polarization coherence parameters under different polarization directions to form an initial polarization response set.

[0034] Polarization channel separation processing separates the original scattered signal sequence according to different polarization directions to obtain signal components in different polarization directions. The scattering intensity difference parameter is the difference in the scattering intensity of the signal under different polarization directions, which can reflect information such as the shape and material of the target scatterer. The polarization coherence parameter is the degree of coherence between signals in different polarization directions, which can reflect the polarization characteristics of the target scatterer. The initial polarization response set is a feature set composed of the scattering intensity difference parameter under different polarization directions and the polarization coherence parameter, which can be used to describe the polarization response characteristics of the target scatterer.

[0035] Polarization filtering can be used to separate polarization channels. Specifically, for an original scattered signal sequence containing horizontal polarization and vertical polarization components, it can be filtered using a horizontal polarization filter and a vertical polarization filter respectively to obtain a horizontal polarization component and a vertical polarization component. Then, the scattering intensity difference parameter under different polarization directions is calculated. For example, the scattering intensity difference can be obtained by calculating the energy ratio of the horizontal polarization component and the vertical polarization component. For polarization coherence parameters, they can be obtained by calculating the mutual correlation coefficient of signals in different polarization directions. For example, if the horizontal polarization component is , the vertical polarization component is , then the polarization coherence parameter can be expressed as ,in express Finally, the scattering intensity difference parameters and polarization coherence parameters under different polarization directions are combined to form the initial polarization response set.

[0036] Step S240: Performing nonlinear normalization processing on the initial time-frequency feature set and the initial polarization response set to eliminate dimensional differences and retain the feature distribution pattern, thereby generating the time-frequency distribution feature set and the polarization response feature set. The nonlinear normalization processing includes selecting a corresponding normalization function based on the statistical distribution characteristics of each feature parameter, performing truncation compensation operations on feature values ​​that exceed a preset fluctuation threshold, and retaining the shape of the normalized feature distribution curve.

[0037] Nonlinear normalization is a method of unifying different feature parameters to ensure they have the same dimensions and range. Dimensional differences mean that different feature parameters may have different physical units and numerical ranges, which can affect the performance of subsequent feature analysis and classification algorithms. The feature distribution pattern is the distribution pattern of feature parameters within their range of values. Preserving this distribution pattern ensures that the normalized features still reflect the essential information of the original features.

[0038] Select the corresponding normalization function according to the statistical distribution characteristics of each feature parameter. For example, for feature parameters that obey the normal distribution, you can choose to use z - Score normalization function, its formula is , where x is the original eigenvalue, is the mean of the eigenvalues, Is the standard deviation of the eigenvalues. For characteristic parameters that do not obey the normal distribution, you can choose to use a nonlinear normalization function, such as the logarithmic normalization function .

[0039] Truncation compensation is performed on eigenvalues ​​that exceed the preset fluctuation threshold. The preset fluctuation threshold is a range of eigenvalues ​​set based on actual conditions. If a eigenvalue exceeds this range, it is considered an outlier. Truncation compensation can be used to handle outliers. For example, if the preset fluctuation threshold is [a, b], if a eigenvalue x is greater than b, it is truncated to b; if x is less than a, it is truncated to a. At the same time, to preserve the shape of the normalized feature distribution curve, the eigenvalue can be appropriately adjusted after truncation compensation so that it still conforms to the distribution pattern of the original features. Through these operations, the initial time-frequency feature set and the initial polarization response set are nonlinearly normalized, ultimately generating the time-frequency distribution feature set and the polarization response feature set.

[0040] Step S300: extracting a target electromagnetic feature combination from the time-frequency distribution feature set and the polarization response feature set according to a preset electromagnetic feature association rule, and generating a weight allocation parameter.

[0041] Preset electromagnetic feature association rules are pre-defined rules used to describe the associations between the characteristic parameters in the time-frequency distribution feature set and the polarization response feature set. These rules can be formulated based on the electromagnetic properties of the target scatterer and actual detection requirements. A target electromagnetic feature combination is a set of characteristic parameter combinations closely related to target classification, selected from the time-frequency distribution feature set and the polarization response feature set. Weight assignment parameters are weights assigned to each characteristic parameter in the target electromagnetic feature combination, reflecting the importance of each characteristic parameter in target classification.

[0042] A feature selection algorithm can be used to extract the target electromagnetic feature combination from the time-frequency distribution feature set and the polarization response feature set based on preset electromagnetic feature association rules. For example, a feature selection algorithm based on correlation analysis can be used to calculate the correlation between each feature parameter in the time-frequency distribution feature set and the polarization response feature set and the target classification label, and then select feature parameters with high correlation to form the target electromagnetic feature combination. Weight assignment parameters can be generated based on factors such as the importance and stability of each feature parameter. For example, feature importance assessment methods in machine learning, such as those used in the random forest algorithm, can be used to train a random forest model, calculate the degree of influence of each feature parameter on the model's classification results, and then assign a weight value to each feature parameter based on this influence.

[0043] As an embodiment, step S300, extracting a target electromagnetic feature combination from the time-frequency distribution feature set and the polarization response feature set according to a preset electromagnetic feature association rule, and generating a weight distribution parameter, may specifically include the following steps S310 to S340: Step S310: constructing a feature correlation evaluation matrix, and calculating the mutual information and covariance correlation coefficient between each feature parameter in the time-frequency distribution feature set and each feature parameter in the polarization response feature set.

[0044] The feature correlation evaluation matrix is ​​used to evaluate the degree of correlation between the characteristic parameters in the time-frequency distribution feature set and the polarization response feature set. Mutual information is a metric used to measure the dependency between two random variables and reflects the degree of information sharing between the two characteristic parameters. The covariance correlation coefficient is a metric used to measure the linear relationship between two random variables and reflects the linear correlation between the two characteristic parameters.

[0045] The specific steps for constructing the feature correlation evaluation matrix are as follows: First, all the feature parameters in the time-frequency distribution feature set and the polarization response feature set are numbered. Then, for each pair of feature parameters, the mutual information and covariance correlation coefficient between them are calculated. The mutual information can be calculated using the formula in information theory. If two feature parameters X and Y have a joint probability distribution of p(x, y) and marginal probability distributions of p(x) and p(y) respectively, then the mutual information between them is The covariance correlation coefficient can be calculated using the formula , where Cov(X, Y) is the covariance of X and Y, Var(X) and Var(Y) are the variances of X and Y respectively. The calculated mutual information and covariance correlation coefficient are filled into the matrix to form the feature correlation evaluation matrix.

[0046] Step S320: generating a feature redundancy index according to the mutual information and the covariance correlation coefficient, and performing a redundancy filtering operation on the time-frequency distribution feature set and the polarization response feature set based on the feature redundancy index to obtain a candidate feature subset.

[0047] The feature redundancy index measures the degree of redundancy between feature parameters and can be calculated based on the mutual information and covariance correlation coefficient. Redundancy filtering removes redundant feature parameters from the time-frequency distribution feature set and the polarization response feature set, retaining only those with independent information, thereby obtaining a candidate feature subset.

[0048] The feature redundancy index can be generated by weighted averaging, for example, by weighting the sum of the mutual information and the covariance correlation coefficient to obtain the feature redundancy index. If the mutual information is I and the covariance correlation coefficient is , weights are w1 and w2 respectively, then the feature redundancy index The specific steps for performing redundancy filtering based on the feature redundancy index are as follows: a redundancy threshold T is set. For each pair of feature parameters in the feature correlation evaluation matrix, if their feature redundancy index is greater than T, the two feature parameters are considered redundant. One feature parameter is selected to be retained and the other is removed. In this way, redundancy filtering is performed on the time-frequency distribution feature set and the polarization response feature set, and finally a candidate feature subset is obtained.

[0049] Step S330: performing sensitivity analysis on the candidate feature subset, obtaining the discrimination weight of each candidate feature for the target scatterer category, and generating an initial feature weight set based on the discrimination weight.

[0050] Sensitivity analysis is a method used to analyze the impact of changes in input variables on output results. In this step, it is used to analyze the impact of each feature parameter in the candidate feature subset on the classification of target scatterers. The discrimination weight is the importance of each candidate feature in distinguishing different target scatterer classes. The initial feature weight set consists of the discrimination weights of each candidate feature.

[0051] The sensitivity analysis of the candidate feature subset can be performed using the experimental design method. For example, a series of experiments can be designed, in each experiment, the value of a candidate feature is changed to observe the changes in the target scatterer classification results. Through multiple experiments, the degree of influence of the value change of each candidate feature on the classification results is calculated, thereby obtaining the discrimination weight of each candidate feature. Specifically, the change in classification accuracy can be used to measure the discrimination weight. If a candidate feature x is changed, the discrimination weight will be increased. i After the value of is taken, the classification accuracy changes from A1 to A2, then the discrimination weight w of the candidate feature i =|A 2- A1|. Combine the discrimination weights of each candidate feature to generate an initial feature weight set.

[0052] Step S340: According to the physical property constraints of the target scatterer, the initial feature weight set is adjusted to eliminate the weight deviation caused by environmental noise and generate the weight allocation parameters; wherein, the adjustment includes: introducing an external calibration signal source to verify the physical property correlation of the candidate feature subset, and when it is detected that the feature weight deviates from the standard weight of the calibration signal source by more than a preset threshold, starting the weight compensation algorithm to correct the deviation.

[0053] Physical property constraints on target scatterers are inherent physical properties of the target scatterer, such as shape, material, and size. These properties affect the value and importance of feature parameters. Environmental noise refers to various interference factors encountered during radar detection, which can cause deviations in feature weights. An external calibration signal source, with known physical properties, can be used to verify the physical property relevance of a candidate feature subset. Standard weights are the feature weight values ​​of a candidate feature subset under ideal conditions. A weight compensation algorithm is used to correct for deviations in feature weights.

[0054] The specific steps for introducing an external calibration signal source to verify the physical property relevance of a candidate feature subset are as follows: The external calibration signal source is transmitted to the target area, the signals received by the radar receiving node are collected, and the characteristic parameters of the candidate feature subset are extracted. Based on the known physical properties of the external calibration signal source, a standard weight is calculated for each characteristic parameter. The calculated standard weight is then compared with the feature weights in the initial feature weight set. When it is detected that the feature weight deviates from the standard weight by more than a preset threshold, a weight compensation algorithm is initiated to correct the deviation. For example, methods such as least squares can be used to adjust the feature weights so that the adjusted feature weights are closer to the standard weights. Through these operations, the initial feature weight set is adjusted, ultimately generating the weight allocation parameters.

[0055] Step S400: performing weighted fusion processing on the target electromagnetic feature combination based on the weight distribution parameters to generate a fused feature vector set, and inputting the fused feature vector set into an adaptive classification network.

[0056] Weighted fusion processing involves weighting and summing each characteristic parameter in the target electromagnetic signature combination according to a weight assignment parameter, fusing multiple characteristic parameters into a comprehensive feature vector. A fused feature vector set, consisting of multiple weighted fused feature vectors, contains more comprehensive information about the target scatterer. An adaptive classification network is a neural network that automatically adjusts classification rules based on input data. It can classify the fused feature vector set and output the target scatterer's category identifier.

[0057] The specific steps of weighted fusion processing of target electromagnetic feature combination based on weight distribution parameters are as follows: If there are n feature parameters in the target electromagnetic feature combination , the corresponding weight distribution parameter is , then the feature vector after weighted fusion is This weighted fusion process is performed on each characteristic parameter combination in the target electromagnetic signature combination to obtain multiple fused feature vectors. These fused feature vectors are combined to generate a fused feature vector set. The fused feature vector set is then input into the adaptive classification network, which classifies the fused feature vectors according to its internal classification rules and outputs the category identification of the target scatterer.

[0058] As an embodiment, the step S400, performing weighted fusion processing on the target electromagnetic feature combination based on the weight distribution parameter to generate a fused feature vector set, may specifically include the following steps S410 to S440: Step S410: constructing a feature fusion equation according to the weight distribution parameters, and performing weighted superposition on each feature parameter in the time-frequency distribution feature set and each feature parameter in the polarization response feature set.

[0059] The feature fusion equation describes how to fuse the characteristic parameters from the time-frequency distribution feature set and the polarization response feature set. It determines the weighting coefficient for each characteristic parameter based on the weight assignment parameter. Weighted superposition multiplies each characteristic parameter by its corresponding weight coefficient and then adds these products to obtain a composite characteristic value.

[0060] The specific steps of constructing the feature fusion equation according to the weight distribution parameters are as follows: If there are m feature parameters in the time-frequency distribution feature set , the corresponding weight distribution parameter is ; There are k characteristic parameters in the polarization response feature set , the corresponding weight distribution parameter is Then the feature fusion equation can be expressed as Each characteristic parameter in the time-frequency distribution feature set and the polarization response feature set is substituted into the feature fusion equation for weighted superposition to obtain a comprehensive characteristic value. This weighted superposition process is performed on each combination of characteristic parameters to obtain multiple comprehensive characteristic values, which can be further combined into a fused characteristic vector.

[0061] Step S420: Perform energy consistency detection on the candidate feature combinations after weighted superposition, eliminate abnormal feature combinations caused by weight distribution mismatch, and retain valid feature combinations that meet the energy conservation criterion.

[0062] Energy consistency testing checks whether candidate feature combinations after weighted superposition satisfy the energy conservation criterion, that is, whether the energy of the feature combination is within a reasonable range. Weight allocation mismatch occurs when the weight allocation parameters are improperly set during the weighted superposition process, resulting in abnormal energies in certain feature combinations. Abnormal feature combinations are those that do not satisfy the energy conservation criterion and may affect subsequent classification results, so they need to be eliminated. Valid feature combinations are those that satisfy the energy conservation criterion and can be used for subsequent processing and analysis.

[0063] Energy consistency detection of the candidate feature combination after weighted superposition can be performed using the energy calculation and threshold judgment method. First, the energy of each candidate feature combination is calculated. For example, the energy value can be obtained by calculating the square sum of each feature parameter in the feature combination. Then, an energy threshold range [E min , E max], if the energy value of a candidate feature combination is not within this range, it is considered an abnormal feature combination and is eliminated; if the energy value is within the range, it is considered a valid feature combination and is retained. This energy consistency test can eliminate abnormal feature combinations caused by weight distribution mismatch and improve the quality of feature combinations.

[0064] Step S430: perform dimensionality compression processing on the effective feature combination, use the principal component preservation algorithm to map the high-dimensional feature space to the low-dimensional feature space, and generate a set of reduced-dimensional feature vectors; wherein, the execution steps of the principal component preservation algorithm include: calculating the information entropy ratio of each effective feature combination before and after dimensionality reduction, and when the information entropy ratio is lower than a preset information loss threshold, enabling an alternative dimensionality reduction channel to compensate for the information loss.

[0065] Dimensionality reduction processing converts high-dimensional valid feature combinations into low-dimensional feature vectors to reduce data complexity and computational complexity. The principal component preserving algorithm is an algorithm that can retain the principal component information of the original data as much as possible during the dimensionality reduction process. The information entropy ratio is the ratio of the information entropy of the valid feature combination before and after dimensionality reduction, which can reflect the degree of information loss during the dimensionality reduction process. The preset information loss threshold is a lower limit of the information entropy ratio set according to actual needs. If the information entropy ratio is lower than this threshold, it is considered that the information loss during the dimensionality reduction process is too great, and an alternative dimensionality reduction channel needs to be activated to compensate. The alternative dimensionality reduction channel is a backup dimensionality reduction method that can be used to reduce information loss when the information loss in the main dimensionality reduction channel is too great.

[0066] The specific steps of using the principal component preservation algorithm to map the high-dimensional feature space to the low-dimensional feature space are as follows: First, the effective feature combinations are standardized so that they have the same mean and variance. Then, the covariance matrix of the effective feature combinations is calculated, and the principal components are obtained by solving the eigenvalues ​​and eigenvectors of the covariance matrix. The first n principal components with larger eigenvalues ​​are selected as the main features, and the original effective feature combinations are projected onto these principal components to obtain the reduced-dimensional feature vectors. During the dimensionality reduction process, the information entropy ratio of each effective feature combination before and after dimensionality reduction is calculated. When the information entropy ratio is lower than the preset information loss threshold, an alternative dimensionality reduction channel is enabled. For example, a local linear embedding algorithm can be used for dimensionality reduction to compensate for the amount of information loss. Finally, the reduced-dimensionality feature vectors are combined to generate a set of reduced-dimensionality feature vectors.

[0067] Step S440: Serializing and reorganizing the dimension-reduced feature vector set to generate the fused feature vector set with spatiotemporal correlation characteristics.

[0068] Serialization and reorganization is the process of recombining the eigenvectors in a reduced-dimensionality eigenvector set in a sequence and according to rules, so that they exhibit spatiotemporal correlation. Spatiotemporal correlation is the relationship between eigenvectors in time and space, which can reflect the motion characteristics and spatial distribution characteristics of the target scatterer.

[0069] Time series analysis and spatial correlation analysis can be used to serialize and reorganize the reduced-dimensionality feature vector set. For example, feature vectors can be sorted according to their acquisition time sequence, and adjacent feature vectors can be combined to form a temporally correlated feature vector sequence. Furthermore, feature vectors collected at different locations can be correlated based on the spatial location information of the radar receiving node, and spatially correlated feature vectors can be combined. Through these operations, the reduced-dimensionality feature vector set is serialized and reorganized, ultimately generating a fused feature vector set with temporal and spatial correlation characteristics.

[0070] Step S500: performing a hierarchical classification operation on the fused feature vector set through the adaptive classification network, outputting a category identifier of the target scatterer and a corresponding classification confidence, and generating a radar detection optimization instruction according to the classification confidence.

[0071] Hierarchical classification involves an adaptive classification network classifying a set of fused feature vectors according to a hierarchical structure, gradually narrowing the classification range and ultimately determining the category of the target scatterer. A category identifier is a label used to indicate the category of the target scatterer, such as aircraft, vehicle, or building. Classification confidence measures the reliability of the classification result and can be expressed as a probability value. Radar detection optimization commands are generated based on the classification confidence and are used to optimize the radar detection process, such as adjusting the radar's transmit power, frequency, and detection angle.

[0072] The specific process of performing hierarchical classification on a set of fused feature vectors using an adaptive classification network is as follows: The adaptive classification network first receives the set of fused feature vectors and then, based on its internal hierarchical classification structure, classifies the fused feature vectors, starting from the top level. At each level, the network evaluates the fused feature vectors according to pre-set classification rules, selecting the most qualified subcategories for the next level of classification, until the lowest level is reached. During the classification process, the network calculates the confidence level of each classification result, for example, using a posterior probability to represent the classification confidence level. Ultimately, the network outputs the target scatterer's category identity and the corresponding classification confidence level. Radar detection optimization instructions are generated based on the classification confidence level. If the classification confidence level is low, indicating that the radar detection result is unreliable, instructions can be generated to adjust radar parameters such as transmit power and frequency to improve detection accuracy.

[0073] As an embodiment, the step S500, performing a hierarchical classification operation on the fused feature vector set by the adaptive classification network, and outputting a category identifier of the target scatterer and a corresponding classification confidence, may specifically include the following steps S510 to S540: Step S510: constructing a multi-level classification decision tree, wherein each non-leaf node is configured with a feature discrimination model based on the electromagnetic scattering mechanism, and each leaf node is associated with a preset category identifier of the target scatterer.

[0074] A multi-level classification decision tree is a tree-structured classification model composed of multiple levels of nodes, each representing a classification decision point. Non-leaf nodes are nodes in the decision tree other than leaf nodes and are used to make classification decisions. A feature discrimination model based on electromagnetic scattering mechanisms is built based on the electromagnetic scattering characteristics of a target scatterer. It can determine which subcategory a target scatterer belongs to based on an input feature vector. Leaf nodes are the lowest level nodes in the decision tree. Each leaf node is associated with a preset category identifier for the target scatterer, representing the final classification result. The specific steps for constructing a multi-level classification decision tree are as follows: First, the hierarchical structure and number of nodes of the decision tree are determined based on the category and feature information of the target scatterer. Then, for each non-leaf node, a feature discrimination model is established based on the electromagnetic scattering mechanism. For example, a support vector machine or decision tree algorithm can be used to establish the feature discrimination model. Each leaf node is associated with a preset category identifier for the target scatterer, such as aircraft or vehicle. Through these steps, a multi-level classification decision tree is constructed.

[0075] Step S520: Input the fused feature vector set into the root node of the multi-level classification decision tree, and calculate the matching degree between the fused feature vector and the feature template of each child node based on the feature discrimination model corresponding to the current node; wherein, the calculation step of the feature discrimination model includes: adopting a template update mechanism to adjust the matching threshold of the feature template according to real-time classification result feedback to ensure the adaptability of the classification decision tree to the time-varying electromagnetic environment.

[0076] The feature template is the feature vector template corresponding to each child node, representing the typical characteristics of the target scatterer class corresponding to that child node. The matching degree is the degree of similarity between the fused feature vector and the feature template, which can be expressed as a numerical value. The template update mechanism is used to adjust the feature template matching threshold based on real-time classification feedback, enabling the classification decision tree to adapt to time-varying electromagnetic environments.

[0077] After the fused feature vector set is input into the root node of the multi-level classification decision tree, the matching degree between the fused feature vector and the feature templates of each child node is calculated based on the feature discrimination model corresponding to the current node. The specific calculation process is as follows: First, for each child node's feature template, the distance between the fused feature vector and the template is calculated. For example, methods such as Euclidean distance and cosine similarity can be used to calculate the distance. Then, the distance value obtained by the feature discrimination model is converted into a matching degree. For example, if Euclidean distance is used, the smaller the distance, the higher the matching degree. During the calculation process, the feature discrimination model uses a template update mechanism to ensure adaptability to time-varying electromagnetic environments. Specifically, after a classification is completed, feedback is provided based on the real-time classification results. If the classification accuracy for a feature template is low, it indicates that the template may not be suitable for the current electromagnetic environment. In this case, the matching threshold for that feature template is adjusted. For example, if the matching threshold is originally set to 0.8 and there are many classification errors for that template, the threshold can be appropriately lowered to 0.7 to expand the matching range. Conversely, if the classification accuracy is high, the threshold can be appropriately increased. In this way, by continuously adjusting the matching threshold according to the real-time classification results, the classification decision tree can better adapt to changes in the electromagnetic environment.

[0078] Step S530: Select the child node with the highest matching degree as the next-level decision node, and iteratively perform matching degree calculation and node selection operations until reaching a leaf node.

[0079] After calculating the matching degree between the fused feature vector and the feature templates of each child node, the child node with the highest matching degree is selected from these child nodes as the decision node for the next level. For example, if the current node has three child nodes, and the matching degrees between the fused feature vector and their feature templates are 0.6, 0.8, and 0.7, respectively, then the child node with a matching degree of 0.8 is selected as the decision node for the next level. Next, the fused feature vector is input into this new decision node, and the matching degree between the fused feature vector and the feature templates of each child node under this node is again calculated based on the feature discrimination model corresponding to that node. The child node with the highest matching degree is then selected as the decision node for the next level, and this matching degree calculation and node selection operation are iterated continuously. As the hierarchy continues, the leaf node of the decision tree is eventually reached, completing a complete classification path.

[0080] Step S540: Calculate the classification confidence according to the historical classification accuracy and current matching degree deviation corresponding to the leaf node, and associate it with the preset category identifier to generate a classification result set.

[0081] The historical classification accuracy rate is the percentage of accurate classifications for a leaf node in past classifications. It reflects the reliability of the category identifier represented by the leaf node in previous classifications. The current matching deviation is the difference between the matching degree between the fused feature vector and the feature template of the leaf node and the average matching degree of the leaf node. It reflects the degree of difference between the current classification and previous cases. The classification confidence rate is a metric that comprehensively considers the historical classification accuracy rate and the current matching deviation rate to measure the reliability of the current classification result.

[0082] The specific method of calculating classification confidence is as follows: First, set a weight coefficient to weight the historical classification accuracy and the current matching deviation. If the weight of the historical classification accuracy is 0.7, the weight of the current matching deviation is 0.3. If the historical classification accuracy of a leaf node is 0.9 and the current matching deviation is 0.1 (indicating that the current matching is 0.1 lower than the average matching), then the classification confidence = 0.7×0.9+0.3×(1 - 0.1) = 0.63 + 0.27 = 0.9. After obtaining the classification confidence, it is combined with the preset category identifier associated with the leaf node to generate a classification result set. For example, if the preset category identifier is "airplane" and the classification confidence is 0.9, then the classification result set will include information such as "airplane, 0.9".

[0083] As an embodiment, the method further includes an optimization process of the classification model, which may specifically include the following steps S550 to S580: Step S550: Collect misclassified samples in the historical classification result set, and extract abnormal fusion feature vectors and environmental interference parameters corresponding to the misclassified samples.

[0084] Misclassified samples are samples that were incorrectly classified during the previous classification process. These samples contain information about potential problems with the classification model. Abnormal fused feature vectors are the fused feature vectors corresponding to misclassified samples. These samples have different characteristic patterns from correctly classified samples, possibly due to environmental interference or model defects. Environmental interference parameters are parameters related to the environment at the time these misclassified samples were collected, such as atmospheric humidity, temperature, and electromagnetic noise intensity. These parameters may have affected the classification results.

[0085] Collecting misclassified samples from historical classification results can be accomplished by comparing the classification results with the actual target categories. For example, the category identifiers output by the classification model can be compared with the actual categories obtained through other reliable means (such as optical observation or manual verification) to identify misclassified samples. For these misclassified samples, their corresponding abnormal fused feature vectors are extracted. Specifically, this can be performed by searching the stored fused feature vector data based on the sample's identification information. Furthermore, the environmental interference parameters at the time these samples were collected are recorded. Environmental monitoring sensors can be installed around the radar detection equipment to obtain parameters such as atmospheric humidity and temperature, while electromagnetic monitoring equipment can be used to obtain parameters such as electromagnetic noise intensity.

[0086] Step S560: constructing an interference feature simulator according to the environmental interference parameters, and generating a simulated interference feature set having similar statistical characteristics to the abnormal fusion feature vector.

[0087] The interference signature simulator is a model that generates simulated interference signatures based on environmental interference parameters. Similar statistical characteristics refer to the similarity between the simulated interference signature set and the anomaly fusion feature vector in terms of statistical indicators such as mean, variance, and distribution shape.

[0088] The specific steps for building an interference signature simulator are as follows: First, analyze the relationship between environmental interference parameters and the anomaly fusion feature vector. Methods such as regression analysis can be used to establish a regression model between the environmental interference parameters and the features of each dimension of the anomaly fusion feature vector. For example, linear regression analysis can be used to derive a linear relationship between atmospheric humidity and a dimension of the anomaly fusion feature vector. Then, based on this relationship model, tools such as a random number generator are used to generate simulated interference signatures with similar statistical characteristics within a given range of environmental interference parameters. For example, based on the current atmospheric humidity value, simulated interference signatures with a distribution consistent with that humidity condition are randomly generated using the regression model. By repeating this process, a large number of simulated interference signatures are generated, forming a simulated interference signature set.

[0089] Step S570: injecting the simulated interference feature set into the training data set of the adaptive classification network to retrain the feature discrimination model in the multi-level classification decision tree.

[0090] Injecting a set of simulated interference features into the training dataset of the adaptive classification network allows the classification model to learn characteristic patterns in the presence of environmental interference, improving the model's robustness. Retraining the feature-based discriminant model within the multi-level classification decision tree adjusts the model's parameters to better adapt to input data containing interference features.

[0091] Specifically, the simulated interference feature set is merged with the original training dataset. For example, if the original training dataset contains 1,000 samples and the simulated interference feature set contains 200 samples, the merged new training dataset will have 1,200 samples. This new training dataset is then used to retrain the feature discrimination model in the multi-level classification decision tree. During the training process, optimization algorithms such as gradient descent are used to continuously adjust the model parameters to minimize the classification error of the training data.

[0092] As an implementation manner, the step S570 of retraining the feature discrimination model in the multi-level classification decision tree may specifically include the following steps S571 to S574: Step S571: performing node vulnerability analysis on the multi-level classification decision tree to identify a set of decision nodes that are susceptible to interference features.

[0093] Node vulnerability analysis evaluates the stability and reliability of each decision node in a multi-level classification decision tree in the presence of noise features. Decision nodes susceptible to noise features are those that are prone to incorrect classification results when fed data containing noise features.

[0094] Node vulnerability analysis can be performed by feeding a series of test data containing varying degrees of interference features into a decision tree and observing how the classification results for each node change. For example, the strength of the interference features in the test data is gradually increased, and the classification accuracy of each node is recorded at each level. If the classification accuracy of a node decreases significantly as the strength of the interference features increases, the node is considered vulnerable to the interference features. By performing this analysis on all nodes, a set of decision nodes susceptible to interference features can be identified.

[0095] Step S572: For each vulnerable decision node, generate an adversarial training sample set, where the adversarial training sample set includes an adversarial feature vector generated by perturbing the original feature vector.

[0096] Adversarial training samples are generated to enhance the robustness of vulnerable decision nodes. Adversarial feature vectors are obtained by perturbing the original feature vectors. These perturbations simulate the effects of environmental interference on the feature vectors.

[0097] The specific method for generating an adversarial training sample set is as follows: For each vulnerable decision node, select the corresponding original feature vector. Then, perform a perturbation operation on these original feature vectors. For example, a random noise value can be added to each dimension of the original feature vector, with the noise value range set according to the actual situation. If the original feature vector is [x1, x2, x3], add a random noise value in the range [-0.1, 0.1] to each dimension to obtain the adversarial feature vector [x1+n1, x2+n2, x3+n3], where n1, n2, and n3 are randomly generated noise values ​​in the range [-0.1, 0.1]. By performing this perturbation operation on multiple original feature vectors, a large number of adversarial feature vectors are generated, forming the adversarial training sample set.

[0098] Step S573: merging the adversarial training sample set and the simulated interference feature set to construct an enhanced training data set.

[0099] The adversarial training sample set is merged with the simulated interference feature set to enable the retrained feature discrimination model to learn the information of adversarial samples and simulated interference features at the same time, further improving the robustness of the model.

[0100] The specific merging process is, for example, directly combining the samples in the adversarial training sample set and the simulated interference feature set.

[0101] Step S574: Use the elastic weight solidification algorithm to locally retrain the feature discrimination model corresponding to the fragile decision node to prevent overfitting and maintain global classification stability; wherein, the execution steps of the elastic weight solidification algorithm include: calculating the importance weights of the model parameters, limiting the adjustment range of important parameters, and allowing secondary parameters to be updated according to new training data.

[0102] The elastic weight fixation algorithm is used to balance model parameter updates and prevent overfitting. The importance weight of model parameters is the contribution of each model parameter to model performance. Important parameters have a greater impact on model performance, while minor parameters have a relatively smaller impact on model performance.

[0103] The specific steps for executing the elastic weight solidification algorithm are as follows: First, calculate the importance weight of each parameter in the feature discrimination model corresponding to the fragile decision node. Gradient-based methods can be used, such as calculating the absolute value of the gradient of the parameter to measure its importance. The larger the absolute value of the gradient, the greater the impact of the parameter on the model performance, and the higher its importance weight. Then, during the local retraining process, limit the adjustment range of important parameters. For example, set an upper limit for the adjustment range. If the update amount of important parameters exceeds this upper limit, the update amount is limited to this upper limit. For minor parameters, they are allowed to update normally based on the new training data. In this way, the model can be adjusted according to the new training data while preventing overfitting caused by excessive adjustment of important parameters, thereby maintaining global classification stability.

[0104] Step S580: Evaluate the optimization effect by comparing the change in classification accuracy before and after retraining. If the change in classification accuracy does not reach the preset optimization threshold, increase the complexity of the simulated interference features and iteratively perform injection and training operations; the retraining step includes: using an incremental learning strategy to retain the original classification knowledge, and adjusting the splitting criteria and feature weight parameters of the decision tree nodes according to the simulated interference features.

[0105] The change in classification accuracy is the difference between the classification accuracy of the retrained model and the accuracy before and after retraining. It reflects the degree to which retraining improves model performance. The preset optimization threshold is a lower limit for the change in classification accuracy, set based on actual needs. If the change in classification accuracy does not reach this threshold, the optimization effect is not significant and further adjustment is required. The complexity of the simulated interference features can be increased by increasing the dimensionality of the interference features or changing the distribution of the interference features.

[0106] To compare the change in classification accuracy before and after retraining, the model can be tested using the same test dataset, recording the classification accuracy before and after retraining, and then calculating the difference. For example, if the classification accuracy of the model before retraining is 0.8 and after retraining is 0.82, the change in classification accuracy is 0.02. If the preset optimization threshold is 0.05 and the change in classification accuracy does not reach the threshold, the complexity of the simulated interference features needs to be increased. For example, if the simulated interference features originally had only three dimensions, they can be increased to five, or their distribution can be changed from a uniform distribution to a normal distribution. Then, the new set of simulated interference features can be injected into the training dataset and retrained using an incremental learning strategy. This incremental learning strategy retains the original classification knowledge while adjusting the splitting criteria and feature weight parameters of the decision tree nodes based on the new simulated interference features. For example, when splitting a decision tree node, the impact of the new interference features on the split is taken into account and the splitting criteria are adjusted. For feature weight parameters, the weights of each feature are reallocated based on their importance in the new data. This injection and training process is repeated iteratively until the change in classification accuracy reaches the preset optimization threshold.

[0107] As an embodiment, the method further includes an environmental calibration process, specifically including the following steps S600 to S900: Step S600: deploying reference scatterers in a target area, wherein the reference scatterers have known electromagnetic scattering characteristics and spatial position information.

[0108] A reference scatterer is an object used for environmental calibration. Its electromagnetic scattering properties (such as scattering coefficient and polarization) and spatial location are known. The purpose of deploying a reference scatterer is to establish a known electromagnetic scattering reference point in the target area for subsequent calibration of radar detection data.

[0109] When deploying reference scatterers within a target area, the number and placement of reference scatterers should be determined based on the target area's size, shape, and radar detection requirements. For example, within a larger target area, reference scatterers can be deployed at the four corners and the center. Reference scatterers can be objects with stable electromagnetic scattering properties, such as metal spheres and corner reflectors. During deployment, to ensure accurate recording of the spatial location of the reference scatterers, devices such as the Global Positioning System (GPS) can be used for positioning.

[0110] Step S700: collecting calibration signal sequences of the reference scatterer at multiple observation angles, extracting calibration feature vectors and constructing an environmental disturbance compensation database.

[0111] The calibration signal sequence is the electromagnetic signal sequence received by the radar from a reference scatterer at different observation angles. The calibration feature vector is a representative feature vector extracted from the calibration signal sequence. It contains the electromagnetic scattering characteristics of the reference scatterer under a preset environment. The environmental disturbance compensation database stores the calibration feature vectors and related environmental information, which is used to subsequently compensate for environmental disturbances in actual detection data.

[0112] Calibration signal sequences are collected from the reference scatterer at multiple observation angles. The observation angle can be changed by rotating the radar antenna or moving the reference scatterer. For example, the radar antenna is rotated at set angular intervals (e.g., 10 degrees), and the calibration signal sequence is collected for a period of time at each angle. Feature extraction is then performed on the collected calibration signal sequence. The extracted features may include signal energy, frequency distribution, polarization characteristics, and so on. These features are combined to form a calibration feature vector. Environmental information during the acquisition of the calibration signal sequence, such as atmospheric conditions and the distribution of surrounding objects, is also recorded. The calibration feature vector and the corresponding environmental information are stored in the environmental disturbance compensation database for subsequent use.

[0113] Step S800: During the actual detection process, the current environmental parameters are matched with the calibration feature vectors in the environmental disturbance compensation database in real time to calculate the environmental disturbance correction coefficient; wherein the real-time matching includes: using the nearest neighbor search algorithm to find the calibration sample closest to the current environmental parameters, and generating a continuous correction coefficient curve through interpolation calculation.

[0114] Current environmental parameters are environmental parameters acquired in real time during the actual detection process, such as atmospheric humidity, temperature, and electromagnetic noise intensity. The environmental disturbance correction coefficient is used to compensate for the effects of environmental disturbances on radar detection data. The nearest neighbor search algorithm is used to find the closest sample to a given data point in a database.

[0115] During the actual detection process, the current environmental parameters are first acquired in real time. Next, a nearest neighbor search algorithm is used to search the environmental disturbance compensation database for the calibration sample that is closest to the current environmental parameters. For example, the distance (e.g., Euclidean distance) between the current environmental parameters and the environmental parameters of each calibration sample in the database is calculated, and the calibration sample with the smallest distance is selected as the nearest neighbor sample. Next, a continuous correction coefficient curve is generated through interpolation calculation based on the calibration feature vectors of the nearest neighbor sample and its adjacent samples. For example, linear interpolation is used to calculate the correction coefficient under the current environmental parameters based on the correction coefficient values ​​of the nearest neighbor sample and adjacent samples. Through this real-time matching and interpolation calculation, the environmental disturbance correction coefficient is obtained.

[0116] Step S900: applying the environmental disturbance correction coefficient to the fused feature vector set to eliminate feature distortion caused by atmospheric attenuation and multipath effects.

[0117] Atmospheric attenuation is the phenomenon in which electromagnetic signals weaken due to absorption, scattering, and other factors as they propagate through the atmosphere. Multipath is the phenomenon in which electromagnetic signals encounter multiple reflectors during propagation, causing the signal to travel through multiple paths to the radar receiver, resulting in signal distortion. Feature distortion occurs when environmental factors such as atmospheric attenuation and multipath cause changes in the eigenvalues ​​of the fused feature vector, affecting the accuracy of classification results.

[0118] The specific method for applying the environmental disturbance correction coefficient to the fused feature vector set is to multiply each feature vector in the fused feature vector set by the corresponding environmental disturbance correction coefficient. For example, if the fused feature vector is [x1, x2, x3] and the environmental disturbance correction coefficient is k, then the corrected fused feature vector is [kx1, kx2, kx3]. In this way, the fused feature vector set is corrected to eliminate feature distortion caused by atmospheric attenuation and multipath effects, thereby improving the accuracy of subsequent classification.

[0119] As an embodiment, the method further includes a process of verifying the classification result, which may specifically include the following steps S1000 to S1200: Step S1000: Establish a multi-source information fusion verification platform and access optical remote sensing data, infrared detection data and geographic information system data.

[0120] The multi-source information fusion verification platform is used to integrate information from multiple data sources. It can fuse and analyze different types of data to verify the accuracy of radar classification results. Optical remote sensing data is image data of the target area acquired by optical remote sensing equipment (such as satellites and drones). It can provide information such as the target's appearance and shape. Infrared detection data is infrared radiation information of the target acquired by infrared detection equipment. It can reflect characteristics such as the target's temperature distribution. Geographic Information System data is data containing geographic information about the target area (such as topography, landforms, and building distribution).

[0121] Establishing a multi-source information fusion verification platform requires the necessary hardware and software environment. Hardware requirements include servers and storage devices. Software development requires data access interfaces and data fusion algorithms. When integrating optical remote sensing data, infrared detection data, and geographic information system data, it is crucial to ensure that the data format and quality meet the platform's requirements. For example, for optical remote sensing data, image preprocessing, such as denoising and correction, is necessary to improve data quality.

[0122] Step S1100: performing spatial position matching and attribute consistency verification on the category identification of the target scatterer and the associated data in the multi-source information fusion verification platform.

[0123] Spatial position matching compares the spatial position information of the target scatterer with that of the associated data in the multi-source information fusion verification platform to ensure spatial correspondence. Attribute consistency verification checks whether the attributes corresponding to the target scatterer's category identifier are consistent with the attributes of the target in the associated data. For example, if the target scatterer's category identifier is "aircraft" in the radar classification results, the target at the corresponding location in the optical remote sensing data and infrared detection data is checked to see whether it possesses aircraft attributes (such as shape and infrared radiation characteristics).

[0124] When performing spatial position matching, Geographic Information System (GIS) technology can be used to unify the spatial position information of the target scatterer and the spatial position information of the associated data into the same coordinate system for comparison. For attribute consistency verification, an attribute feature library can be established to store the attribute characteristics of different target categories. For example, aircraft attributes may include fuselage shape, wing structure, and target infrared radiation pattern. The target scatterer's category identification is compared with the attribute feature library, and the associated data in the multi-source information fusion verification platform is combined to check whether the attributes are consistent.

[0125] Step S1200: When it is detected that there is an attribute conflict between the classification result and the multi-source data, the classification backtracking mechanism is triggered to re-execute the feature extraction and classification decision operations; wherein, the execution steps of the classification backtracking mechanism include: adjusting the weight allocation strategy in the electromagnetic feature association rule according to the attribute conflict type, and performing a credibility-weighted voting decision on the historical classification results.

[0126] Attribute conflicts occur when the attributes corresponding to the target scatterer's category identifier are inconsistent with those reflected in the associated data in the multi-source information fusion verification platform. The classification backtracking mechanism is a mechanism for handling classification errors. It re-executes feature extraction and classification decision operations to improve classification accuracy.

[0127] When an attribute conflict is detected between the classification results and multi-source data, the type of attribute conflict is first analyzed. For example, is it a shape attribute conflict, a temperature attribute conflict, or another attribute conflict? The weight allocation strategy in the electromagnetic feature association rules is then adjusted based on the attribute conflict type. If the conflict is a shape attribute conflict, the weight of shape-related feature parameters may need to be increased. Next, a credibility-weighted voting decision is performed on the historical classification results. For each historical classification result, a credibility weight is assigned based on factors such as its classification confidence and its relevance to the current situation. For example, historical classification results with high classification confidence and similarities to the current environment are assigned a higher credibility weight. A weighted vote is performed on all historical classification results, and the classification result with the highest votes is selected as the new classification result. Feature extraction and classification decision operations are then re-executed, using the adjusted weight allocation strategy to obtain a more accurate classification result.

[0128] It is understandable that the various algorithms involved in the above-mentioned introductions to the embodiments of the present invention, such as adaptive filtering algorithms, wavelet decomposition algorithms, weight compensation algorithms, etc., can all be learned from the relevant content in the prior art. In order to save space, they will not be expanded too much in the embodiments of this application. In addition, when implementing the scheme of this application, those skilled in the art can supplement the details according to the common knowledge in this field. For example, according to the common knowledge in this field, normalization can be used to eliminate dimensional conflicts before feature fusion, interpolation can be used to eliminate dimensional differences, and historical data, experience or business scenario requirements can be combined to reasonably set thresholds, and models can be trained based on a general model training method, etc. This application no longer provides redundant introductions to overly detailed implementation processes.

[0129] See also Figure 2 , Figure 2The following is a schematic diagram of the structure of a computer system provided in an embodiment of the present invention. The computer system includes at least a processor 101, a communication interface 102, and a memory 103. The processor 101, communication interface 102, and memory 103 may be connected via a bus or other means. The processor 101 (also known as the Central Processing Unit (CPU)) is the computing and control core of the computer system, capable of parsing various instructions within the computer system and processing various data within the computer system. The communication interface 102 may optionally include a standard wired interface or a wireless interface (such as Wi-Fi or a mobile communication interface), which can be used to send and receive data under the control of the processor 101. The communication interface 102 may also be used for data transmission and interaction within the computer system. The memory 103 is a storage device within the computer system, used to store programs and data. It is understood that the memory 103 herein may include both the computer system's built-in memory and, of course, the computer system's supported extended memory. The memory 103 provides storage space for the computer system's operating system, which may include, but is not limited to, Android, iOS, Windows Phone, and the like, though this invention is not limited thereto.

[0130] In one embodiment, the processor 101 executes the passive perception radar target classification method based on electromagnetic detection feature extraction provided in the above embodiment of the present invention by running the computer program in the memory 103 .

Claims

1. A passive sensing radar target classification method based on electromagnetic detection feature extraction, characterized in that: include: Acquire an electromagnetic radiation signal set of a target area, wherein the electromagnetic radiation signal set includes a sequence of original scattered signals collected during multiple signal collection cycles; Performing multi-dimensional signal decomposition processing on the original scattered signal sequence to obtain a time-frequency distribution feature set and a polarization response feature set corresponding to each signal acquisition period; Extracting a target electromagnetic feature combination from the time-frequency distribution feature set and the polarization response feature set according to a preset electromagnetic feature association rule, and generating a weight allocation parameter; Performing weighted fusion processing on the target electromagnetic feature combination based on the weight distribution parameters to generate a fused feature vector set, and inputting the fused feature vector set into an adaptive classification network; A hierarchical classification operation is performed on the fused feature vector set through the adaptive classification network, a category identifier of the target scatterer and a corresponding classification confidence are output, and a radar detection optimization instruction is generated according to the classification confidence.

2. The method according to claim 1, characterized in that Performing multi-dimensional signal decomposition processing on the original scattered signal sequence to obtain a time-frequency distribution feature set and a polarization response feature set corresponding to each signal acquisition period, including: Performing a time domain alignment operation on the original scattered signal sequence to eliminate the signal timing offset caused by environmental interference and generate a standardized time domain signal sequence; Using a multi-resolution orthogonal decomposition algorithm to divide the standardized time domain signal sequence into frequency bands, extracting signal energy distribution parameters and phase continuity parameters in each frequency band to form an initial time-frequency feature set; synchronously performing polarization channel separation processing on the original scattered signal sequence to obtain scattering intensity difference parameters and polarization coherence parameters under different polarization directions to form an initial polarization response set; Nonlinear normalization processing is performed on the initial time-frequency feature set and the initial polarization response set to eliminate dimensional differences and retain feature distribution patterns, thereby generating the time-frequency distribution feature set and the polarization response feature set. The nonlinear normalization processing includes: selecting a corresponding normalization function based on the statistical distribution characteristics of each feature parameter, performing a truncation compensation operation on feature values ​​that exceed a preset fluctuation threshold, and retaining the shape of the normalized feature distribution curve.

3. The method according to claim 2, characterized in that According to a preset electromagnetic feature association rule, a target electromagnetic feature combination is extracted from the time-frequency distribution feature set and the polarization response feature set, and a weight distribution parameter is generated, including: Constructing a feature correlation evaluation matrix, and calculating the mutual information and covariance correlation coefficient between each feature parameter in the time-frequency distribution feature set and each feature parameter in the polarization response feature set; generating a feature redundancy index according to the mutual information and the covariance correlation coefficient, and performing a redundancy filtering operation on the time-frequency distribution feature set and the polarization response feature set based on the feature redundancy index to obtain a candidate feature subset; Performing sensitivity analysis on the candidate feature subset to obtain a discrimination weight of each candidate feature for the target scatterer category, and generating an initial feature weight set based on the discrimination weight; According to the physical property constraints of the target scatterer, the initial feature weight set is adjusted to eliminate the weight deviation caused by environmental noise and generate the weight allocation parameters; wherein, the adjustment includes: introducing an external calibration signal source to verify the physical property correlation of the candidate feature subset, and when it is detected that the feature weight deviates from the standard weight of the calibration signal source by more than a preset threshold, starting the weight compensation algorithm to correct the deviation.

4. The method according to claim 3, characterized in that Performing weighted fusion processing on the target electromagnetic feature combination based on the weight distribution parameters to generate a fusion feature vector set includes: Constructing a feature fusion equation according to the weight distribution parameter, and performing weighted superposition on each feature parameter in the time-frequency distribution feature set and each feature parameter in the polarization response feature set; Perform energy consistency detection on the candidate feature combinations after weighted superposition, eliminate abnormal feature combinations caused by weight distribution mismatch, and retain valid feature combinations that meet the energy conservation criterion; Performing dimensionality compression processing on the effective feature combination, mapping the high-dimensional feature space to the low-dimensional feature space using the principal component preservation algorithm, and generating a set of reduced-dimensional feature vectors; Serializing and reorganizing the dimension-reduced feature vector set to generate the fused feature vector set with spatiotemporal correlation characteristics; Among them, the execution steps of the principal component preservation algorithm include: calculating the information entropy ratio of each valid feature combination before and after dimensionality reduction, and when the information entropy ratio is lower than a preset information loss threshold, enabling an alternative dimensionality reduction channel to compensate for the information loss.

5. The method according to claim 4, characterized in that Performing a hierarchical classification operation on the fused feature vector set through the adaptive classification network to output a category identifier of the target scatterer and a corresponding classification confidence, including: Construct a multi-level classification decision tree, in which each non-leaf node is configured with a feature discrimination model based on the electromagnetic scattering mechanism, and each leaf node is associated with a preset category identifier of the target scatterer; Input the fused feature vector set into the root node of the multi-level classification decision tree, and calculate the matching degree between the fused feature vector and the feature template of each sub-node according to the feature discrimination model corresponding to the current node; Select the child node with the highest matching degree as the next-level decision node, and iterate the matching degree calculation and node selection operations until reaching the leaf node; Calculating the classification confidence according to the historical classification accuracy and current matching degree deviation corresponding to the leaf node, and generating a classification result set by associating the preset category identifier; The calculation steps of the feature discrimination model include: using a template update mechanism to adjust the matching threshold of the feature template according to real-time classification result feedback to ensure the adaptability of the classification decision tree to the time-varying electromagnetic environment.

6. The method according to claim 5, characterized in that The method also includes an optimization process of the classification model, including: Collect misclassified samples from the historical classification result set, and extract abnormal fusion feature vectors and environmental interference parameters corresponding to the misclassified samples; Constructing an interference feature simulator according to the environmental interference parameters to generate a simulated interference feature set having similar statistical characteristics to the abnormal fusion feature vector; Injecting the simulated interference feature set into the training data set of the adaptive classification network to retrain the feature discrimination model in the multi-level classification decision tree; The optimization effect is evaluated by comparing the change in classification accuracy before and after retraining. If the change in classification accuracy does not reach a preset optimization threshold, the complexity of the simulated interference features is increased and the injection and training operations are iteratively performed; The retraining step includes: using an incremental learning strategy to retain original classification knowledge, and adjusting the splitting criteria and feature weight parameters of the decision tree nodes according to the simulated interference features.

7. The method according to claim 6, characterized in that Retraining the feature discrimination model in the multi-level classification decision tree includes: Performing node vulnerability analysis on the multi-level classification decision tree to identify a set of decision nodes that are susceptible to interference features; For each vulnerable decision node, generating an adversarial training sample set, wherein the adversarial training sample set includes an adversarial feature vector generated by perturbing the original feature vector; Merging the adversarial training sample set with the simulated interference feature set to construct an enhanced training data set; An elastic weight solidification algorithm is used to locally retrain the feature discrimination model corresponding to the fragile decision node to prevent overfitting and maintain global classification stability; wherein, the execution steps of the elastic weight solidification algorithm include: calculating the importance weights of model parameters, limiting the adjustment range of important parameters, and allowing minor parameters to be updated according to new training data.

8. The method according to claim 1, characterized in that Before obtaining the electromagnetic radiation signal set of the target area, the signal preprocessing step is also included: Deploy multiple passive radar receiving nodes, configure the frequency band coverage and polarization receiving mode of each receiving node, and form a spatially distributed detection array; Synchronize the local clock source of each receiving node to eliminate signal phase errors caused by time asynchrony; The original signal collected by each receiving node is subjected to local oscillator leakage suppression processing, and an adaptive filtering algorithm is used to eliminate the device's own radiation interference; wherein, the execution steps of the adaptive filtering algorithm include: establishing a transfer function model of the hardware characteristics of the receiving node, and offsetting the leakage signal component through a reverse compensation operation.

9. The method according to claim 8, characterized in that The method also includes an environmental calibration process, including: Deploy a reference scatterer in the target area, wherein the reference scatterer has known electromagnetic scattering characteristics and spatial position information; Collecting calibration signal sequences of the reference scatterer at multiple observation angles, extracting calibration feature vectors and constructing an environmental disturbance compensation database; During the detection process, the current environmental parameters are matched in real time with the calibration feature vectors in the environmental disturbance compensation database to calculate the environmental disturbance correction coefficient; wherein the real-time matching includes: using a nearest neighbor search algorithm to find the calibration sample closest to the current environmental parameters, and generating a continuous correction coefficient curve through interpolation calculation; The environmental disturbance correction coefficient is applied to the fused feature vector set to eliminate feature distortion caused by atmospheric attenuation and multipath effects.

10. A computer system, characterized in that: include: a memory storing a computer program; A processor, configured to load the computer program to implement the passive perception radar target classification method based on electromagnetic detection feature extraction as described in any one of claims 1 to 9.

Citation Information

Patent Citations

  • Sea surface small target detection method based on multi-domain multi-dimensional feature fusion

    CN111580064A

  • Performance optimization design method and system for multi-layer composite polaroid

    CN119439495A

  • Unmanned aerial vehicle multi-system navigation decoy system based on intelligent algorithm optimization

    CN119906519A

  • Target signal separation device, passive radar device, and target signal separation method

    JP2020193904A

Cited By

  • Electronic equipment detection method and system based on passive detection

    CN121679743A

  • An electronic device detection method and system based on passive passive detection

    CN121679743B

  • Variable transmission ratio rack-gear sector pair machining error detection method and system

    CN122310702A