Micro-target identification method, system and device in complex electromagnetic environment and medium

Abstract

The invention discloses a micro-target identification method, system and device in a complex electromagnetic environment and a medium, relates to the technical field of micro-target identification, and aims to solve the technical problem that in the prior art, micro-target identification cannot be carried out on collected signals through multi-type feature decision fusion, in particular to signal optimization. In a complex electromagnetic environment, recognition statistics is carried out according to signals transmitted by a radar, signal optimization and feature recognition are carried out on statistical recognition signals, after signal optimization is completed, feature recognition is carried out according to collected signals, specifically, time-domain feature extraction, frequency-domain feature extraction and decision fusion are carried out, and after time-domain and frequency-domain features are obtained, the time-domain and frequency-domain features are obtained. And performing micro-target identification on the acquired signal through multi-type feature decision fusion.

Description

Technical Field

[0001] This invention relates to the field of micro-target recognition technology, specifically to a method, system, device, and medium for micro-target recognition in complex electromagnetic environments. Background Technology

[0002] Micro-target recognition in complex electromagnetic environments is a challenging interdisciplinary technology that integrates signal processing, target detection, and pattern recognition. The core technical challenge lies in the dual limitations of strong interference in the electromagnetic environment (such as multi-source noise, co-frequency interference, and multipath effects) and weak signal characteristics of micro-targets (such as small RCS, low echo energy, and feature ambiguity).

[0003] However, in the current technology for micro-target identification, it is impossible to optimize the collected radar signals, and the availability of the identified signals cannot be guaranteed. At the same time, it is impossible to perform feature recognition on the radar signals, and it is impossible to evaluate the signal quality based on the identified features. It is difficult to provide data support for micro-target identification based on the identified features. In addition, it is impossible to perform micro-target identification on the collected signals through multi-type feature decision fusion.

[0004] To address the aforementioned technical shortcomings, a solution is proposed. Summary of the Invention

[0005] The purpose of this invention is to solve the problems mentioned above by proposing a method, system, device and medium for micro-target identification in complex electromagnetic environments.

[0006] The objective of this invention can be achieved through the following technical solution: a method for identifying micro-targets in a complex electromagnetic environment, comprising the following steps:

[0007] Signal optimization involves identifying and statistically analyzing the signals transmitted by the radar in complex electromagnetic environments, and then optimizing the statistically identified signals.

[0008] Feature recognition: After signal optimization, feature recognition is performed based on the acquired signal, specifically through time-domain feature extraction and frequency-domain feature extraction.

[0009] Decision fusion, after acquiring time-domain and frequency-domain features, uses multi-type feature decision fusion to identify micro-targets in the acquired signals.

[0010] As a preferred embodiment of the present invention, the signal optimization process is as follows:

[0011] The radar signals are collected and then filtered.

[0012] The real-time acquired signal is input to the filter, i.e., the input signal vector:

[0013] x(n)=[x(n),x(n-1),...,x(n-M+1)] T

[0014] Where M is the order of the filter, and T represents the threshold for pulse detection;

[0015] Determine the desired signal for the current computational scenario based on the filter settings, and assign the label d(n);

[0016] Weighted adjustments are made to optimize filter performance; the filter weight vector is:

[0017] w(n) = [w0(n), w1(n), ..., w M-1 (n)] T

[0018] Where each element W in the weight vector i (n);

[0019] After weighted adjustment, the filter output is expressed as:

[0020] y(n)=W T (n)x(n)

[0021] Among them, W T W(n) is the transpose of the weight vector W(n). The output signal of the filter at the current moment is obtained by performing the inner product operation between the weight vector and the input signal vector.

[0022] In a preferred embodiment of the present invention, the inner product operation of vectors is used to obtain the output signal of the filter at the current moment.

[0023] Compare the desired signal with the filter output signal:

[0024] e(n)=d(n)-y(n)=d(n)-W T (n)x(n)

[0025] If the error is 0, it means that the filter output fully meets the expectations, indicating that the current radar signal meets the identification requirements. If the error is not 0, the larger the error, the greater the gap between the filter output and the expected signal, and further adjustments are needed.

[0026] In a preferred embodiment of the present invention, the feature recognition process is as follows:

[0027] Identify and collect time-domain and frequency-domain features;

[0028] The acquired time-domain signal is digitally sampled to obtain a discrete signal sequence;

[0029] Let the discrete signal sequence be X{n}, m=0,1,…,N-1, and the sampling interval be Δt;

[0030] The pulse width is determined by detecting the time interval during which the signal exceeds a set threshold.

[0031] Set threshold V th When X{n1}≥V th And X{n2}≥V th ;

[0032] If n1 is the point where the signal first exceeds the threshold, and n2 is the last point where the signal exceeds the threshold before falling below the threshold again, then the pulse width is T. 宽 = (n2-n1)Δt;

[0033] The rate of change of the signal between adjacent sampling points is calculated using differential methods, and the rise / fall slope is then obtained:

[0034]

[0035]

[0036] Where ks represents the rising edge slope and kj represents the falling edge slope.

[0037] In a preferred embodiment of the present invention, a discrete Fourier transform is performed on the time-domain signal to obtain a frequency-domain signal, and the amplitude spectrum of the frequency-domain signal is analyzed to determine the position of the spectral peak. Specifically:

[0038]

[0039] Where k = 0, 1, ..., N-1, X[g] represents the amplitude spectrum;

[0040] The peak position of the spectrum is And g max It is the index corresponding to the maximum value in the amplitude spectrum;

[0041] The bandwidth is determined by identifying the frequency range below the peak value in the amplitude spectrum. Let the peak amplitude be Ap, and find the amplitude... For two frequency points f1 and f2, the bandwidth B = f2 - f1;

[0042] Analyzing the spectrum at multiple time points and calculating the rate of change of peak frequency over time yields the frequency drift rate. Let the peak positions of the spectrum at different times t1 and t2 be fp1 and fp2, respectively. Then the frequency drift rate is...

[0043] As a preferred embodiment of the present invention, the decision fusion process is as follows:

[0044] The time-domain and frequency-domain features are concatenated into a high-dimensional feature vector. The association between multi-domain features is then directly learned through a classifier. However, the dimensions and distributions of features from different domains vary significantly, so standardization is required first.

[0045] Let any eigenvector be q = {q1, q2, ..., q...} m}, after standardization

[0046] Where, μ m and σ m These are the mean and standard deviation of the feature on the training set, respectively, to ensure that each feature participates in the fusion at the same order of magnitude.

[0047] The time-domain and frequency-domain feature vectors are labeled as q, respectively. s ={q s1 q s2 , ..., q sm} and q p ={q p1 q p2 , ..., q pm Then the concatenated fused feature vector is:

[0048] Support Vector Machines (SVMs) map high-dimensional features to a linearly separable space using kernel functions, making them suitable for handling nonlinear associations of multi-domain features. A decision function is defined as follows:

[0049]

[0050] Where q represents the input feature vector to be classified, sign(·) is the sign function, L represents the number of support vectors, and α m Represented as Lagrange multipliers, y m Let K(q, qm) be the sample label and K(q, qm) be the kernel function.

[0051] The collected feature vectors are input into the function model, and the function input is obtained according to the calculation of the decision function. When the function input is +1, the radar signal to which the current feature vector belongs is marked as a target signal. Conversely, when the function input is -1, the radar signal to which the current feature vector belongs is marked as a non-target signal.

[0052] The present invention also proposes a micro-target recognition system in a complex electromagnetic environment, including a micro-target recognition processor, and the micro-target recognition processor is connected to a signal optimization unit, a feature recognition unit and a decision fusion unit.

[0053] The signal optimization unit identifies and statistically analyzes the signals transmitted by the radar in complex electromagnetic environments, and optimizes the statistically identified signals.

[0054] The feature recognition unit performs feature recognition based on the acquired signal after signal optimization.

[0055] The decision fusion unit, after acquiring time-domain and frequency-domain features, performs micro-target identification on the acquired signals through multi-type feature decision fusion.

[0056] The present invention also proposes a micro-target recognition device in a complex electromagnetic environment, including a micro-target recognition processor, a signal optimization unit, and a decision fusion unit, to execute the aforementioned micro-target recognition method in a complex electromagnetic environment.

[0057] The present invention also proposes a computer-readable storage medium storing a computer program thereon, which, when executed, implements the above-mentioned method for micro-target recognition in a complex electromagnetic environment.

[0058] Compared with the prior art, the beneficial effects of the present invention are:

[0059] 1. In this invention, the statistical identification signal is optimized to solve the signal acquisition deviation caused by multi-source noise in the real-time transmission environment during the signal transmission process, which makes it impossible to identify micro-targets in the acquired signal and reduces the accuracy of signal transmission detection. Moreover, the signal optimization is targeted according to the real-time scene and the image scene to improve the accuracy of micro-target identification in different scenes under complex electromagnetic environment.

[0060] 2. In this invention, multi-feature recognition is performed on the acquired signal. Based on the various types of features of the acquired signal, the decision fusion algorithm can improve the accuracy of micro-target recognition, so as to ensure that the micro-target recognition of the acquired signal can be performed accurately and avoid the inability to perform targeted micro-target recognition due to feature ambiguity, thereby reducing the accuracy of micro-target recognition.

[0061] 3. In this invention, after acquiring time-domain and frequency-domain features, micro-target identification is performed on the acquired signals through multi-type feature decision fusion. The core of decision fusion is to integrate complementary information of multi-domain features to offset the uncertainty of a single feature, thereby improving the accuracy and robustness of micro-target identification. It can also determine the target based on multi-type feature decision fusion. Attached Figure Description

[0062] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0063] Figure 1 This is a diagram illustrating the method processing architecture of the present invention;

[0064] Figure 2 This is a flowchart of the method processing of the present invention;

[0065] Figure 3 This is a system principle block diagram of the present invention. Detailed Implementation

[0066] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0067] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0068] Please see Figure 1-2 As shown, a method for identifying micro-targets in a complex electromagnetic environment includes the following steps:

[0069] Signal optimization involves identifying and statistically analyzing the radar signals in complex electromagnetic environments, and then optimizing the statistically identified signals to address signal acquisition deviations caused by multi-source noise in the real-time transmission environment, which prevents the identification of micro-targets and reduces the accuracy of signal transmission detection. Furthermore, the signal optimization is tailored to real-time and image scenarios to improve the accuracy of micro-target identification in different scenarios under complex electromagnetic environments.

[0070] Feature recognition: After signal optimization, feature recognition is performed on the acquired signal. Specifically, multi-feature recognition is performed on the acquired signal through time-domain feature extraction and frequency-domain feature extraction. Based on the various types of features of the acquired signal, the decision fusion algorithm can improve the accuracy of micro-target recognition, ensuring that the micro-target recognition of the acquired signal can be performed accurately and avoiding the inability to perform targeted micro-target recognition due to feature ambiguity, which would reduce the accuracy of micro-target recognition.

[0071] Decision fusion, after acquiring time-domain and frequency-domain features, uses multi-type feature decision fusion to identify micro-targets in the acquired signals. The core of decision fusion is to integrate complementary information from multiple domain features to offset the uncertainty of a single feature, thereby improving the accuracy and robustness of micro-target identification. It can also determine targets based on multi-type feature decision fusion.

[0072] The signal optimization process is as follows:

[0073] The radar signal is acquired and then filtered. It should be explained that the filtering process allows the filter to automatically adapt to changes in the input signal. By continuously adjusting the filter coefficients (i.e., the weight vector), the mean square error between the filter output signal and the signal we want to obtain is minimized. In the presence of multi-source noise, this method allows the filter to gradually "learn" how to remove noise and make the output closer to the real signal.

[0074] The real-time acquired signal is input to the filter, i.e., the input signal vector:

[0075] x(n)=[x(n),x(n-1),...,x(n-M+1)] T

[0076] Where M is the order of the filter, indicating that the filter will consider the input signal values ​​at the current time n and the previous M-1 times, and combine these values ​​into a vector as the input information of the filter. T represents the threshold for pulse detection.

[0077] Determine the desired signal for the current computational scenario based on the filter settings, and assign the label d(n);

[0078] Weighted adjustments are made to optimize filter performance; the filter weight vector is:

[0079] w(n) = [w0(n), w1(n), ..., w M-1 (n)] T

[0080] Where each element W in the weight vector i (n);

[0081] After weighted adjustment, the filter output is expressed as:

[0082] y(n)=W T (n)x(n)

[0083] Among them, W T W(n) is the transpose of the weight vector W(n). The output signal of the filter at the current moment is obtained by the inner product operation between the weight vector and the input signal vector.

[0084] Compare the desired signal with the filter output signal:

[0085] e(n)=d(n)-y(n)=d(n)-W T (n)x(n)

[0086] If the error is 0, it means that the filter output fully meets the expectations, indicating that the current radar signal meets the identification requirements. If the error is not 0, the larger the error, the greater the gap between the filter output and the expected signal, and further adjustments are needed.

[0087] During communication, the signal is subject to various noise interferences from the transmitting end to the receiving end. The LMS algorithm can process the signal at the receiving end, remove noise by continuously adjusting the filter coefficients, improve the signal quality, and enable the receiver to more accurately restore the original signal.

[0088] The feature recognition process is as follows:

[0089] Identify and collect time-domain and frequency-domain features;

[0090] The acquired time-domain signal is digitally sampled to obtain a discrete signal sequence;

[0091] Let the discrete signal sequence be X{n}, m=0,1,…,N-1, and the sampling interval be Δt;

[0092] The pulse width is determined by detecting the time interval during which the signal exceeds a set threshold.

[0093] Set threshold V th When X{n1}≥V th And X{n2}≥V th ;

[0094] If n1 is the point where the signal first exceeds the threshold, and n2 is the last point where the signal exceeds the threshold before falling below the threshold again, then the pulse width is T. 宽 = (n2-n1)Δt;

[0095] The rate of change of the signal between adjacent sampling points is calculated using differential methods, and the rise / fall slope is then obtained:

[0096]

[0097] Where ks represents the rising edge slope and kj represents the falling edge slope;

[0098] It should be explained that in signal preprocessing and anti-interference scenarios, the rising edge slope and falling edge slope are key features describing the temporal abrupt change characteristics of the signal. Their core function is to distinguish between target signals and interference signals. The echo signals of micro-targets (such as the radar echo of micro UAVs) usually have stable rising / falling edge characteristics (the slope changes gently and regularly), while the rising / falling edges of pulse interference (such as electromagnetic pulses and sudden noise) are often steep and irregular (the absolute value of the slope is extremely large).

[0099] Slope analysis can quickly identify and eliminate interference. In complex electromagnetic environments, signals may be distorted due to multipath effects or co-frequency interference (such as a slowed rising edge or a trailing falling edge). By monitoring slope changes, signal quality can be assessed.

[0100] The frequency domain signal is obtained by performing a discrete Fourier transform on the time domain signal. The amplitude spectrum of the frequency domain signal is then analyzed to determine the location of the spectral peaks. Specifically:

[0101]

[0102] Where k = 0, 1, ..., N-1, X[g] represents the amplitude spectrum;

[0103] The peak position of the spectrum is And g max It is the index corresponding to the maximum value in the amplitude spectrum;

[0104] The bandwidth is determined by identifying the frequency range below the peak value in the amplitude spectrum. Let the peak amplitude be Ap, and find the amplitude... For two frequency points f1 and f2, the bandwidth B = f2 - f1;

[0105] Analyzing the spectrum at multiple time points and calculating the rate of change of peak frequency over time yields the frequency drift rate. Let the peak positions of the spectrum at different times t1 and t2 be fp1 and fp2, respectively. Then the frequency drift rate is...

[0106] It should be explained that the peak position of the spectrum and the frequency drift rate are the core parameters for frequency domain feature extraction. In the identification of micro-targets in complex electromagnetic environments, the core role of the acquisition is to distinguish the target signal from the interference signal and to provide key basis for the analysis of target characteristics.

[0107] The echo signal of a micro-target (such as a micro-drone) usually has a stable peak frequency (such as 5.8 GHz), while the peak frequency of co-channel interference (such as a malicious jammer) may deviate from this range. By comparing the measured peak frequency with the target's theoretical frequency, valid signals can be quickly screened. The frequency drift rate (the rate of change of peak frequency over time) is used to describe the dynamic change of signal frequency.

[0108] The decision fusion process is as follows:

[0109] By concatenating time-domain and frequency-domain features into a high-dimensional feature vector, the correlation between multi-domain features can be directly learned through a classifier (such as SVM or deep learning models). However, the dimensions and distributions of features from different domains vary significantly (e.g., the pulse width in the time domain is in μs, while the peak value in the frequency domain is in MHz), requiring standardization beforehand.

[0110] Let any eigenvector be q = {q1, q2, ..., q...} m}, after standardization, is

[0111] Where, μ m and σ m These are the mean and standard deviation of the feature on the training set, respectively, to ensure that each feature participates in the fusion at the same order of magnitude.

[0112] The time-domain and frequency-domain feature vectors are labeled as q, respectively. s ={q s1 q s2 , ..., q sm} and q p ={q p1 q p2 , ..., q pm Then the concatenated fused feature vector is:

[0113] Support Vector Machines (SVMs) map high-dimensional features to a linearly separable space using kernel functions, making them suitable for handling nonlinear associations of multi-domain features. A decision function is defined as follows:

[0114]

[0115] Where q represents the input feature vector to be classified. In the multi-domain feature fusion scenario, it is a high-dimensional vector composed of features from the time domain, frequency domain, etc., containing various types of information for classification decisions.

[0116] sign(·) is the sign function, defined as:

[0117] In the decision function, +1 or -1 is output based on the sign of the calculation result inside the function, thereby determining the category to which the input feature vector belongs;

[0118] L represents the number of support vectors;

[0119] α m Represented as Lagrange multipliers, these are parameters introduced during the solution of the SVM optimization problem. Each support vector corresponds to a Lagrange multiplier, which is determined by the optimization algorithm and is used to measure the importance of each support vector in the classification decision.

[0120] y m Represented as sample labels, with values ​​of +1 or -1, representing the m-th support vector q in the training set. m The category to which it belongs;

[0121] K(q, qm) represents the kernel function used to transform vectors q and qm in low-dimensional space. m Map to a higher-dimensional space and compute the inner product in the higher-dimensional space;

[0122] b is the bias term, which translates the classification hyperplane so that the hyperplane can better separate samples of different categories;

[0123] The collected feature vectors are input into the function model, and the function input is obtained according to the calculation of the decision function. When the function input is +1, the radar signal to which the current feature vector belongs is marked as a target signal. Conversely, when the function input is -1, the radar signal to which the current feature vector belongs is marked as a non-target signal.

[0124] Please see Figure 3 As shown, a micro-target recognition system in a complex electromagnetic environment includes a micro-target recognition processor, which is connected to a signal optimization unit, a feature recognition unit, and a decision fusion unit.

[0125] The signal optimization unit identifies and statistically analyzes the signals transmitted by the radar in complex electromagnetic environments, and optimizes the statistically identified signals.

[0126] The feature recognition unit performs feature recognition based on the acquired signal after signal optimization.

[0127] The decision fusion unit, after acquiring time-domain and frequency-domain features, performs micro-target identification on the acquired signals through multi-type feature decision fusion.

[0128] One of the micro-target recognition devices in a complex electromagnetic environment employs the micro-target recognition system in a complex electromagnetic environment as described in claim 7, comprising a micro-target recognition processor, a signal optimization unit, a decision fusion unit, and so on, to execute the aforementioned micro-target recognition method in a complex electromagnetic environment.

[0129] One of the methods is a computer-readable storage medium storing a computer program that, when executed, implements a micro-target recognition method under a complex electromagnetic environment as described above.

[0130] Thresholds, preset values, preset ranges, etc. are set for result comparison and analysis to determine whether they are good or bad. The value of these thresholds is determined by a combination of large-scale model analysis of sample data and human experience. They can also be adjusted appropriately based on seasonal or common-sense influences.

[0131] Furthermore, the settings for weighting ratios, influence factors, etc., are based on the magnitude of each parameter's influence on the results. The specific values ​​are allocated to ultimately reflect the impact on the results. The settings for input and storage are also determined by a combination of large-scale model analysis of sample data and human experience. Appropriate adjustments can also be made based on seasonal or rational influence conditions.

[0132] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for identifying micro-targets in a complex electromagnetic environment, characterized in that, Includes the following steps: Signal optimization involves identifying and statistically analyzing the signals transmitted by the radar in complex electromagnetic environments, and then optimizing the statistically identified signals. Feature recognition: After signal optimization, feature recognition is performed based on the acquired signal, specifically through time-domain feature extraction and frequency-domain feature extraction. Decision fusion, after acquiring time-domain and frequency-domain features, uses multi-type feature decision fusion to identify micro-targets in the acquired signals.

2. The method for micro-target identification in a complex electromagnetic environment according to claim 1, characterized in that, The signal optimization process is as follows: The radar signals are collected and then filtered. The real-time acquired signal is input to the filter, i.e., the input signal vector: x(n)=[x(n),x(n-1)...,x(n-M+1)] T Where M is the order of the filter, and T represents the threshold for pulse detection; Determine the desired signal for the current computational scenario based on the filter settings, and assign the label d(n); Weighted adjustments are made to optimize filter performance; the filter weight vector is: w(n)=[w0(n),w1(n),...,w M-1 (n)] T Where each element W in the weight vector i (n); After weighted adjustment, the filter output is expressed as: y(n)=W T (n)x(n) Among them, W T W(n) is the transpose of the weight vector W(n). The output signal of the filter at the current moment is obtained by performing the inner product operation between the weight vector and the input signal vector.

3. The method for micro-target identification in a complex electromagnetic environment according to claim 2, characterized in that, Compare the desired signal with the filter output signal: e(n)=d(n)-y(n)=d(n)-W T (n)x(n) If the error is 0, it means that the filter output fully meets the expectations, indicating that the current radar signal meets the identification requirements. If the error is not 0, the larger the error, the greater the gap between the filter output and the expected signal, and further adjustments are needed.

4. The method for micro-target identification in a complex electromagnetic environment according to claim 3, characterized in that, The feature recognition process is as follows: Identify and collect time-domain and frequency-domain features; The acquired time-domain signal is digitally sampled to obtain a discrete signal sequence; Let the discrete signal sequence be X{n}, m=0,1,…,N-1, and the sampling interval be Δt; The pulse width is determined by detecting the time interval during which the signal exceeds a set threshold. Set threshold V th When X{n1}≥V th And X{n2}≥V th ; If n1 is the point where the signal first exceeds the threshold, and n2 is the last point where the signal exceeds the threshold before falling below the threshold again, then the pulse width is T. 宽 = (n2-n1)Δt; The rate of change of the signal between adjacent sampling points is calculated using differential methods, and the rise / fall slope is then obtained: Where ks represents the rising edge slope and kj represents the falling edge slope.

5. The method for micro-target identification in a complex electromagnetic environment according to claim 4, characterized in that, The frequency domain signal is obtained by performing a discrete Fourier transform on the time domain signal. The amplitude spectrum of the frequency domain signal is then analyzed to determine the location of the spectral peaks. Specifically: Where k = 0, 1, ..., N-1, X[g] represents the amplitude spectrum; The peak position of the spectrum is And g max It is the index corresponding to the maximum value in the amplitude spectrum; The bandwidth is determined by identifying the frequency range below the peak value in the amplitude spectrum. Let the peak amplitude be Ap, and find the amplitude... For two frequency points f1 and f2, the bandwidth B = f2 - f1; Analyzing the spectrum at multiple time points and calculating the rate of change of peak frequency over time yields the frequency drift rate. Let the peak positions of the spectrum at different times t1 and t2 be fp1 and fp2, respectively. Then the frequency drift rate is...

6. The method for micro-target identification in a complex electromagnetic environment according to claim 5, characterized in that, The decision fusion process is as follows: The time-domain and frequency-domain features are concatenated into a high-dimensional feature vector. The association between multi-domain features is then directly learned through a classifier. However, the dimensions and distributions of features from different domains vary significantly, so standardization is required first. Let any eigenvector be q = {q1, q2, ..., q...} m }, after standardization Where, μ m and σ m These are the mean and standard deviation of the feature on the training set, respectively, to ensure that each feature participates in the fusion at the same order of magnitude. The time-domain and frequency-domain feature vectors are labeled as q, respectively. s ={q s1 q s2 , ..., q sm } and q p ={q p1 q p2 , ..., q pm Then the concatenated fused feature vector is: Support Vector Machines (SVMs) map high-dimensional features to a linearly separable space using kernel functions, making them suitable for handling nonlinear associations of multi-domain features. A decision function is defined as follows: Where q represents the input feature vector to be classified, sign(·) is the sign function, L represents the number of support vectors, and α m Represented as Lagrange multipliers, y m Let K(q, qm) be the sample label and K(q, qm) be the kernel function. The collected feature vectors are input into the function model, and the function input is obtained according to the calculation of the decision function. When the function input is +1, the radar signal to which the current feature vector belongs is marked as a target signal. Conversely, when the function input is -1, the radar signal to which the current feature vector belongs is marked as a non-target signal.

7. A micro-target recognition system for complex electromagnetic environments, characterized in that, The micro-target recognition method applied to the above-mentioned complex electromagnetic environment includes a micro-target recognition processor, and the micro-target recognition processor is connected to a signal optimization unit, a feature recognition unit and a decision fusion unit; The signal optimization unit identifies and statistically analyzes the signals transmitted by the radar in complex electromagnetic environments, and optimizes the statistically identified signals. The feature recognition unit performs feature recognition based on the acquired signal after signal optimization. The decision fusion unit, after acquiring time-domain and frequency-domain features, performs micro-target identification on the acquired signals through multi-type feature decision fusion.

8. A micro-target identification device for complex electromagnetic environments, characterized in that, The micro-target recognition system in a complex electromagnetic environment as described in claim 7 includes a micro-target recognition processor, a signal optimization unit, and a decision fusion unit to execute the micro-target recognition method in a complex electromagnetic environment as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed, implements a micro-target recognition method in a complex electromagnetic environment as described in any one of claims 1-6.

Images