A multi-dimensional joint SAR interference parameter estimation method

Abstract

The present invention relates to the technical field of SAR parameter estimation, and in particular to a multi-dimensional domain joint SAR interference parameter estimation method, comprising the following steps: Fourier transforming the SAR echo domain signal to be detected to the frequency domain in each azimuth pulse, and performing kurtosis detection in the frequency domain to determine whether there is interference in each azimuth pulse; when there is interference in a certain azimuth pulse, recording the azimuth pulse sequence number, and performing short-time Fourier transform on the pulse with interference to obtain a time-frequency spectrum; classifying the time-frequency spectrum based on a pre-trained convolutional neural network to determine the interference category; and estimating the interference parameters in different ways for different interference categories. The present invention can realize effective and accurate signal parameter estimation for different categories of SAR interference signals.

Description

Technical Field

[0001] The present invention relates to the technical field of SAR parameter estimation, and more particularly to a multi-dimensional domain joint SAR interference parameter estimation method. Background Art

[0002] As a sensor for all-day and all-weather ground observation, synthetic aperture radar is an important means in modern information electronic warfare. Since SAR is a broadband active remote sensing radar system, it is susceptible to electromagnetic interference within the working frequency band, which seriously affects the observation performance. Therefore, the research on anti-interference technology of SAR system is of great value. With the advancement of SAR technology, SAR systems are developing towards high resolution and wide bandwidth, which puts higher requirements on SAR anti-interference technology. In SAR anti-interference processing, accurately obtaining the parameters of the interference signal has a direct impact on the effectiveness of SAR anti-interference methods and the improvement of combat effectiveness.

[0003] At present, the research on SAR cognitive anti-interference mainly focuses on SAR interference modeling, SAR interference detection, SAR anti-interference technology, etc. There are fewer studies on interference parameter estimation methods for SAR interference cognition, but it is one of the foundations of SAR cognitive anti-interference technology research and is of great significance.

[0004] Therefore, how to extract and estimate SAR interference parameters is a problem that those skilled in the art need to solve urgently. Summary of the invention

[0005] In view of this, the present invention provides a multi-dimensional domain joint SAR interference parameter estimation method, which can achieve effective and accurate signal parameter estimation of different types of SAR interference signals.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A multi-dimensional domain joint SAR interference parameter estimation method comprises the following steps:

[0008] Perform Fourier transform of the SAR echo domain signal to be detected into the frequency domain in each azimuth pulse, and perform kurtosis detection in the frequency domain to determine whether there is interference in the pulses in each azimuth direction;

[0009] When there is interference in a certain azimuth pulse, the pulse number of the azimuth pulse is recorded, and the pulse with interference is subjected to short-time Fourier transform to obtain a time-frequency spectrum diagram;

[0010] Classify the time-frequency spectrum based on a pre-trained convolutional neural network to determine the interference category;

[0011] Different methods are used to estimate interference parameters for different interference categories.

[0012] Furthermore, the judgment method for determining whether there is interference in each azimuth pulse is as follows:

[0013] In the frequency domain, interference detection is performed based on the difference between the kurtosis of the random variable and the kurtosis of the Gaussian random variable, where the kurtosis of the Gaussian random variable is always 3 and the kurtosis of the random variable is K. ur , K ur Related to the mean and variance of the random variable, when there is a pulse in a certain direction |K ur When -3|≥0.5, it is considered that the pulse in this azimuth direction is interfered.

[0014] Furthermore, the interference categories include: dot frequency interference signals, linear frequency modulation interference signals and combined interference signals; the combined interference signal is a combination of dot frequency interference signals and linear frequency modulation interference signals.

[0015] Furthermore, the method for estimating the frequency interference signal parameters is as follows:

[0016] After Fourier transform of the point frequency interference signal, the peak value is searched in the frequency domain. The frequency corresponding to the peak position is the center frequency of the point frequency interference signal.

[0017] Perform median filtering along the time domain, then detect the rising edge and falling edge and calculate the average of the coordinate difference between the rising edge and the falling edge to calculate the time width and pulse repetition frequency of the point frequency interference signal;

[0018] If there are multiple different point frequency interference signals, the point frequency interference signals with different center frequencies are separated by setting a band-limited filter, and then the time width and pulse repetition frequency of each point frequency interference signal are calculated respectively.

[0019] Furthermore, the method for estimating the linear frequency modulation interference signal parameters is as follows:

[0020] Search for interference echoes containing linear frequency modulation interference signals and having continuous pulse numbers in azimuth, record the pulse numbers in azimuth of the interference echoes, and determine the number of pulses N across azimuths;

[0021] The Radon-STFT method is used to perform Radon transformation on the time-frequency spectrum of N azimuth pulses, and the slope of the straight line in the spectrum is obtained. The slope of the straight line is related to the frequency modulation rate of the linear frequency modulation signal. If the slopes of the straight lines obtained for these N azimuth pulses can match, it is considered that the interference in these N azimuth pulses is generated by the same linear frequency modulation signal.

[0022] Determine the start and end of the cross-azimuth pulse time domain, record the start time and end time corresponding to the start and end ends, and calculate the time width and pulse repetition frequency of the linear frequency modulation interference signal;

[0023] Select any one of the N azimuth pulse signals, perform fractional Fourier transform, find the peak coordinates through peak search, calculate the modulation frequency, multiply the obtained time width by the modulation frequency to obtain the bandwidth of the linear frequency modulation interference signal, and determine the multiple M of the linear frequency modulation interference signal bandwidth relative to the SAR bandwidth;

[0024] Perform short-time Fourier transform on N cross-azimuth pulse signals respectively to obtain time-frequency spectrum, and find the ratio of pulse repetition interval to receiving window duration. Use drawing method to set time-frequency coordinate system, arrange time-frequency spectrum along time axis according to the ratio, and repeatedly arrange time-frequency spectrum on frequency axis according to the positive and negative signs of modulation frequency and multiple M of linear frequency modulation interference signal bandwidth relative to SAR bandwidth. Connect the interference signal time-frequency ridge line, calculate the coordinate of the midpoint of the time-frequency ridge line, and the corresponding frequency axis coordinate value is the center frequency of linear frequency modulation interference signal.

[0025] Furthermore, the time domain start and end of the cross-azimuth pulse signal are searched by the echo domain amplitude ratio method. The search method is expressed as:

[0026]

[0027] Among them, N 1 and N 2 Respectively represent the pulse number of the cross-azimuth pulse signal, N r Indicates the number of points in the echo domain, i indicates the Nth echo 1 or Nth 2 The i-th range point of an azimuth pulse, echo() represents the two-dimensional echo matrix.

[0028] Further, assuming that the number of pulses of the cross-azimuth pulse signal is 2, the time width of the linear frequency modulation interference signal is expressed as:

[0029] t d =PRT echo -X 1 +X 2

[0030] Among them, PRT echo represents the pulse repetition time of the SAR echo, X 1 Indicates the interval between the start time of the first interference pulse and the start time of the first receiving window, X 2 Indicates the interval between the end time of the first interference pulse and the start time of the second receiving window;

[0031] Search for the time domain start end of the next azimuth pulse that matches the modulation frequency, and the pulse repetition frequency is expressed as:

[0032] PRF=1 / (N·PRT echo -X 1 +X3 )

[0033] Among them, X 3 It represents the time interval between the start time of the next interference pulse and the start time of the receiving window, and N represents the number of pulses across azimuth directions.

[0034] Furthermore, for the combined interference signal, the combined interference signal is separated into multiple intrinsic mode functions through the complex empirical mode decomposition method. For each intrinsic mode function, each signal component is located on the time axis and frequency axis of the time-frequency plane, and the signals containing similar frequency components in different time periods are decomposed to obtain the time-frequency spectrum of each signal. The obtained time-frequency spectrum is input into the pre-trained convolutional neural network to obtain the interference category of each signal.

[0035] Furthermore, the combined interference signal is separated into multiple intrinsic mode functions by complex empirical mode decomposition method, including:

[0036] Assume that the received combined interference signal is y(t), which has K intrinsic mode function components, the number of projection directions is L, the number of iterations is I, and the initialization conditions are k=1 and I=1;

[0037] Calculate y k (t) Projection into L directions

[0038] Finding vectors The local maximum of And calculate the average of the envelopes of different projection directions to get the average rotation vector m k (t);

[0039] From y k Subtract the average rotation vector from (t) to obtain the residual r k,i (t);

[0040] Let I=I+1,y k (t) = r k,i (t), repeat the above steps until the remaining amount r k,i (t) is a monotonic function; if the stopping condition is reached, then r k,I+1 (t) is an intrinsic mode function component, and the acquired signal c is recorded. k (t) = r k,I+1 (t), let y k =y(t), k=k+1, repeat the above steps until k=K, and write the combined interference signal y(t) as K intrinsic mode functions and trend d K (t) is in the form of a sum,

[0041]

[0042] The eigenmode function is defined as finite in time domain and the above equation is re-expressed as:

[0043]

[0044] Among them, c q (t) and c k (t) have the same meaning, is a rectangular function, defined as:

[0045]

[0046] Among them, b q is the bandwidth, τ q Is the pulse width.

[0047] Furthermore, the convolutional neural network adopts a VGG-16 network. During training, SAR echo domain data with different interference categories are obtained, wherein the interference categories include point frequency interference, linear frequency modulation interference, and combined interference; the SAR echo domain data is transformed to the time-frequency domain through short-time Fourier transform to obtain an interference time-frequency domain image to form a training data set.

[0048] It can be seen from the above technical solutions that, compared with the prior art, the present invention has the following beneficial effects:

[0049] The present invention takes into account the diversity of SAR interference modes, first identifies the type of interference, and then adopts different methods to carry out parameter estimation of different types of interference signals respectively, so as to achieve effective and accurate signal parameter estimation of SAR interference. The present invention can provide conditions for predicting interference behavior and studying anti-interference strategies by accurately estimating relevant parameters such as the center frequency and pulse repetition frequency of SAR interference signals. At the same time, the estimation of interference signal parameters can also guide the design of corresponding filters to achieve filtering of interference signals and positioning of radiation sources, and has guiding significance for SAR carrier platforms to avoid interference source movement route planning and destroy interference sources. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0051] Figure 1 A flowchart of the multi-dimensional domain joint SAR interference parameter estimation method provided by the present invention;

[0052] Figure 2Time-frequency spectrum images of several instances in the SAR echo domain interference database provided by the present invention;

[0053] Figure 3 A schematic diagram of classifying SAR interference using the VGG-16 network provided by the present invention;

[0054] Figure 4 A schematic diagram of a method for calculating the time width and pulse repetition frequency of cross-pulse SAR interference provided by the present invention;

[0055] Figure 5 A schematic diagram of a method for calculating the center frequency of cross-pulse SAR interference provided by the present invention. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] like Figure 1 As shown, the embodiment of the present invention discloses a multi-dimensional domain joint SAR interference parameter estimation method, comprising the following steps:

[0058] S1. Perform Fourier transform of the SAR echo domain signal to be detected into the frequency domain in each azimuth pulse, and perform kurtosis detection in the frequency domain to determine whether there is interference in each azimuth pulse;

[0059] S2. When a pulse in a certain azimuth direction is interfered with, the pulse number of the azimuth direction is recorded, and the pulse with interference is subjected to short-time Fourier transform to obtain a time-frequency spectrum diagram;

[0060] S3, classifying the time-frequency spectrum based on the pre-trained convolutional neural network to determine the interference category;

[0061] S4. For different interference categories, different methods are used to estimate interference parameters.

[0062] The above steps are further explained below.

[0063] S1. Detect interference pulses in SAR echoes:

[0064] The SAR echo domain signal to be detected is Fourier transformed to the frequency domain in each azimuth pulse, and the kurtosis is detected in the frequency domain. The kurtosis is a statistical parameter related to the shape of the probability density function (PDF) of the random variable. The kurtosis of a Gaussian random variable is always 3 and has nothing to do with the mean and variance. Assume that the random process X, the kurtosis (K ur )as follows:

[0065]

[0066] Where μ is the mean of the random variable and σ is the variance.

[0067] Considering that the SAR scene echo conforms to the complex Gaussian distribution, interference detection can be performed based on the difference between the kurtosis of the random variable and the kurtosis of the Gaussian random variable in the frequency domain. ur When -3|≥0.5, it is considered that the pulse in this azimuth direction is interfered, and the pulse number in this azimuth direction is recorded.

[0068] S2. When there is interference in a certain azimuth pulse, the pulse number of the azimuth pulse is recorded, and the pulse with interference is subjected to short-time Fourier transform to obtain a time-frequency spectrum diagram.

[0069] S3. Use the trained convolutional neural network to classify the time-frequency spectrum and determine the interference category.

[0070] Among them, the convolutional neural network uses the VGG-16 network, which can be well applied to classification tasks. The VGG-16 network is trained using the interference signal spectrum database to accurately classify different interference signals for subsequent processing.

[0071] When training specifically, Figure 2-Figure 3 As shown in the figure, SAR echo domain data with RF interference is selected from a simulated SAR echo domain interference database with a large number of interference signals with different parameters and categories, which mainly includes three types: point frequency interference, linear frequency modulation interference, and combined interference. Combined interference means an interference signal that is a combination of point frequency interference and linear frequency modulation interference. The above interference signal is transformed to the time-frequency domain through short-time Fourier transform, and the interference time-frequency domain image is saved to form a training data set of the interference signal time-frequency spectrum.

[0072] The trained VGG-16 network can classify the time-frequency spectrum of the pulse to be detected and determine its interference category.

[0073] S4. Different processes are used to handle different types of interference. The details are as follows:

[0074] (1) The method for estimating the parameters of the point frequency interference signal is as follows:

[0075] After Fourier transform of the point frequency interference signal, the peak value is searched in the frequency domain. The frequency corresponding to the peak position is the center frequency f of the point frequency interference signal. 0 ;

[0076] Median filtering is performed along the time domain to prevent signal breakpoints from affecting parameter estimation. Then, the rising and falling edges are detected and the mean of the coordinate difference between the rising and falling edges is calculated to obtain the time width t of the point frequency interference signal. d , and calculate the pulse repetition frequency PRF; the rising edge and falling edge refer to the position where the interference pulse appears and ends, corresponding to the point on the time coordinate axis, and point frequency interference may have multiple pulses. The purpose of averaging is to reduce the possible error in the primary parameter estimation.

[0077] If there are multiple different point frequency interference signals, the point frequency interference signals with different center frequencies are separated by setting a band-limited filter, and then the time width and pulse repetition frequency of each point frequency interference signal are calculated respectively.

[0078] (2) For linear frequency modulation interference, the general parameter estimation method is to find the peak coordinates (p 0 ,u 0 ), the estimated center frequency f is calculated by the formula 0 With modulation frequency K r :

[0079]

[0080] Among them, N r is the number of points in the echo domain, and the time width t d The calculation method of pulse repetition frequency PRF is the same as the point frequency signal processing method, signal bandwidth B = K r ·t d ;

[0081] However, the linear frequency modulation signal is a broadband signal, and may have long duration or large bandwidth linear frequency modulation interference exceeding the SAR system bandwidth. At this time, the complete interference signal cannot be received within a receiving window of the SAR and the signal has frequency aliasing. The correct signal parameters cannot be obtained according to the general parameter estimation method mentioned above.

[0082] In the embodiment of the present invention, the method for estimating the parameters of the linear frequency modulation interference signal with a long duration / large bandwidth is divided into five steps, specifically:

[0083] In the first step, the echo domain and the STFT (Short Time Fourier Transform) time-frequency domain are combined to search for interference echoes containing linear frequency modulation interference signals and with continuous azimuth pulse numbers, record the azimuth pulse numbers of the interference echoes, and determine the number of pulses N across the azimuth. Specifically, the kurtosis detection is first used to determine the azimuth pulse numbers with interference in the echo domain, and multiple groups of continuous azimuth pulse numbers are found. Then, the STFT time-frequency domain images of several consecutive pulse signals with interference in the echo domain are respectively input into the VGG network to determine whether they belong to linear frequency modulation interference signals, so as to obtain the above results.

[0084] In the second step, considering that the linear frequency modulation interference is linear in the time-frequency domain, the Radon-STFT method is used to perform Radon transform on the time-frequency spectrum of the N azimuth pulses to obtain the slope of the straight line in the spectrum. The slope of the straight line is related to the modulation frequency of the linear frequency modulation signal. If the slopes of the straight lines obtained for these N azimuth pulses can match, that is, the modulation frequencies match, then it is considered that the interference in these N azimuth pulses is generated by the same linear frequency modulation signal. Otherwise, return to the first step.

[0085] The third step, due to the limited length of the receiving window, the time domain starting segment and ending end of the cross-pulse interference may not fall within the receiving window. If one of the two cannot be obtained, the time width of the interference cannot be calculated. The present invention searches for the time domain starting end and ending end of the cross-azimuth pulse signal through the echo domain amplitude ratio method to find the cross-pulse interference that meets the conditions of the simultaneous existence of the starting end and the ending end. Each azimuth pulse signal in the echo domain is a vector. The amplitude ratio method compares the average amplitude difference between the first n points and the last n points of the vector. Since the interference is usually of large power, if it is greater than a certain value, it can be considered that the interference signal appears / ends at the azimuth pulse.

[0086] Assume that the pulse numbers of the cross-azimuth pulses are N 1 、N 2 , the number of echo domain range points is N r (SAR complex echo is a two-dimensional matrix, the row / column dimensions are called range and azimuth respectively, and the number of points in the echo domain range refers to the length of the range dimension), then the above conditions can be expressed as follows:

[0087]

[0088] Among them, i represents the i-th distance point of the N1 / N2-th azimuth pulse of the echo, and echo() represents the two-dimensional echo matrix. When searching according to the above formula, if the number of cross-pulse interference is greater than 2, first determine the starting time point of the interference through the starting end condition, and then search the end end of each subsequent row of pulses until the first end end is found (if the number of cross-pulses is 3, search the end end of the 2nd and 3rd row pulses).

[0089] Record the starting and ending ends of the cross-azimuth pulse time domain that meets the above conditions, and record the start time and end time corresponding to the starting and ending ends, and calculate the time width and pulse repetition frequency of the linear frequency modulation interference signal.

[0090] If there are multiple pulses with both start and end conditions, each pulse needs to be calculated because the interference signal is not constant within the synthetic aperture time.

[0091] If there is no cross-pulse interference that satisfies both the start and end conditions, the interference signal duration and pulse repetition frequency cannot be determined. This is because the SAR receiving window is intermittently open, and the start / end of the interference signal is not received by the SAR, so parameter estimation cannot be achieved.

[0092] like Figure 4 As shown, Nr is the number of points in the SAR echo distance dimension, fs is the sampling frequency of the SAR signal, Tp_echo is the pulse width of the SAR transmission signal, PRT_jam is the pulse repetition time of the interference, and Tp_jam is the interference pulse width. Assuming that the number of pulses of the cross-azimuth pulse signal is 2, the time width of the linear frequency modulation interference signal is expressed as:

[0093] t d =PRT echo -X 1 +X 2

[0094] Among them, PRT echo represents the pulse repetition time of the SAR echo, X 1 Indicates the interval between the start time of the first interference pulse and the start time of the SAR (first) receiving window, X 2 Indicates the interval between the end time of the first interference pulse and the start time of the SAR (second) receiving window.

[0095] The pulse width calculation formula of the interference signal across N pulses is:

[0096] t d =(N-1)*PRT echo -X 1 +X 2

[0097] Search for the time domain start end of the next azimuth pulse that matches the modulation frequency, and the pulse repetition frequency is expressed as:

[0098] PRF=1 / (N·PRT echo -X 1 +X 3 )

[0099] Among them, X 3 Indicates the start time of the next interference pulse and the SAR receiving window (such as Figure 4 , the time interval of the start time of the third receiving window), N represents the number of pulses across the azimuth, N*PRT echoIndicates the PRT of N-fold echo. The two interference signals with matching modulation frequency actually come from the same interference. If the start time of the first interference signal is obtained before, another start time can be obtained by searching the start end of the next interference signal. Then the subtraction of the two is the pulse repetition time, and the reciprocal is the pulse repetition frequency.

[0100] The fourth step is to target intra-pulse interference by selecting any one of the N azimuth pulse signals, performing a fractional Fourier transform, finding the peak coordinates through peak search, calculating the modulation frequency, multiplying the obtained time width by the modulation frequency to obtain the bandwidth of the linear frequency modulation interference signal, and determining the multiple M of the linear frequency modulation interference signal bandwidth relative to the SAR bandwidth.

[0101] The fifth step is to deal with the inter-pulse interference. After executing the first to third steps, perform short-time Fourier transform on the N cross-azimuth pulse signals to obtain the time-frequency spectrum and find the ratio of the pulse repetition interval to the receiving window length, such as Figure 5 As shown, the number of pulses of cross-azimuth pulse interference is 2, and its pulse sequence number is N 1 、N 2 After that, the time-frequency coordinate system is set by the drawing method, and the time-frequency spectrum is arranged along the time axis according to the obtained ratio. The time-frequency spectrum is repeatedly arranged on the frequency axis according to the positive and negative signs of the frequency modulation rate and the multiple M of the linear frequency modulation interference signal bandwidth relative to the SAR bandwidth. The interference signal time-frequency ridge line is connected, and the coordinates of the midpoint of the time-frequency ridge line are calculated. The corresponding frequency axis coordinate value is the center frequency of the linear frequency modulation interference signal.

[0102] Specifically, if the interference signal spans two azimuth pulses and the bandwidth is M times the SAR bandwidth, M+2 time-frequency diagrams need to be reused. When the modulation rate is positive, the (M+2) / 2 (rounded down) time-frequency diagrams are arranged in a column with the lower left corner aligned with the coordinate origin, and the other half of the time-frequency diagrams are also arranged in a column with the lower left corner aligned with the coordinate (PRT_echo, B*(M+2) / 2). If the modulation rate is negative, then Figure 5 As shown, the lower left corner of the first column of the time-frequency graph is located at the coordinate (0, B*(M+2) / 2), and the lower left corner of the second column of the time-frequency graph is located at the coordinate (PRT_echo, 0).

[0103] If there are multiple different inter-pulse linear frequency modulation interferences at the same time, first separate the linear frequency modulation signals in the FRFT domain, and extract the signals in order of the peak values ​​of the interference in the FRFT domain for separate analysis and parameter estimation.

[0104] (3) For the combined interference signal, the combined interference signal is separated into multiple intrinsic mode functions by the complex empirical mode decomposition method. For each intrinsic mode function, each signal component is located on the time axis and frequency axis of the time-frequency plane, and the signals containing similar frequency components in different time periods are decomposed to obtain the time-frequency spectrum of each signal. The obtained time-frequency spectrum is input into the pre-trained convolutional neural network to obtain the interference category of each signal. Specifically:

[0105] Assume that the received combined interference signal is y(t), which has K intrinsic mode function components, the number of projection directions is L, the number of iterations is I, the initialization conditions are k=1 and I=1, where k is the number of signals in the combined interference signal, and the projection direction satisfies:

[0106]

[0107] Calculate y k (t) Projection into L directions

[0108]

[0109] Finding vectors The local maximum of And calculate the average of the envelopes of different projection directions to get the average rotation vector m k (t), expressed as follows:

[0110]

[0111] From y k Subtract the average rotation vector from (t) to obtain the residual r k,i (t), expressed as:

[0112] r k，i (t) = y k (t)-m k (t);

[0113] Let I=I+1,y k (t) = r k,i (t), repeat the above steps until the remaining amount r k,i (t) is a monotonic function; if the stopping condition is reached, then r k,I+1 (t) is an intrinsic mode function component, and the acquired signal c is recorded. k (t) = r k,I+1 (t), let y k =y(t), k=k+1, repeat the above steps until k=K, and write the combined interference signal y(t) as K intrinsic mode functions and trend d K(t) is in the form of a sum,

[0114]

[0115] Since a signal component can be defined as time-limited, the above equation can also be expressed as:

[0116]

[0117] Among them, c q (t) Same as above c k (t), is a rectangular function, defined as:

[0118]

[0119] Among them, b q is the bandwidth, τ q Is the pulse width.

[0120] This definition method can locate each signal component on the time axis and frequency axis of the time-frequency plane respectively, that is, it can combine the complex empirical mode decomposition method with time-frequency analysis to decompose signals containing similar frequency components in different time periods.

[0121] For the different separated intrinsic mode functions, the STFT time-frequency spectrum of each signal is obtained and substituted into the VGG-16 classification network to determine which of the above interferences it belongs to and adopt the corresponding method.

[0122] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0123] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-dimensional domain joint SAR interference parameter estimation method, characterized in that: The following steps are involved: Perform Fourier transform of the SAR echo domain signal to be detected into the frequency domain in each azimuth pulse, and perform kurtosis detection in the frequency domain to determine whether there is interference in the pulses in each azimuth direction; When there is interference in a certain azimuth pulse, the pulse number of the azimuth pulse is recorded, and the pulse with interference is subjected to short-time Fourier transform to obtain a time-frequency spectrum diagram; Classify the time-frequency spectrum based on a pre-trained convolutional neural network to determine the interference category; Different methods are used to estimate interference parameters for different interference categories; The interference categories include: dot frequency interference signal, linear frequency modulation interference signal and combined interference signal; the combined interference signal is a combination of dot frequency interference signal and linear frequency modulation interference signal; The method for estimating the linear frequency modulation interference signal parameters is: Search for interference echoes containing linear frequency modulation interference signals and having continuous pulse numbers in azimuth, record the pulse numbers in azimuth of the interference echoes, and determine the number of pulses N across azimuths; The Radon-STFT method is used to perform Radon transformation on the time-frequency spectrum of N azimuth pulses, and the slope of the straight line in the spectrum is obtained. The slope of the straight line is related to the frequency modulation rate of the linear frequency modulation signal. If the slopes of the straight lines obtained for these N azimuth pulses can match, it is considered that the interference in these N azimuth pulses is generated by the same linear frequency modulation signal. Determine the start and end of the cross-azimuth pulse time domain, record the start time and end time corresponding to the start and end ends, and calculate the time width and pulse repetition frequency of the linear frequency modulation interference signal; Select any one of the N azimuth pulse signals, perform fractional Fourier transform, find the peak coordinates through peak search, calculate the modulation frequency, multiply the obtained time width by the modulation frequency to obtain the bandwidth of the linear frequency modulation interference signal, and determine the multiple M of the linear frequency modulation interference signal bandwidth relative to the SAR bandwidth; Perform short-time Fourier transform on N cross-azimuth pulse signals respectively to obtain time-frequency spectrum, and find the ratio of pulse repetition interval to receiving window duration. Use drawing method to set time-frequency coordinate system, arrange time-frequency spectrum along time axis according to the ratio, and repeatedly arrange time-frequency spectrum on frequency axis according to the positive and negative signs of modulation frequency and multiple M of linear frequency modulation interference signal bandwidth relative to SAR bandwidth. Connect the interference signal time-frequency ridge line, calculate the coordinate of the midpoint of the time-frequency ridge line, and the corresponding frequency axis coordinate value is the center frequency of linear frequency modulation interference signal.

2. The multi-dimensional domain joint SAR interference parameter estimation method according to claim 1 is characterized in that: The method for determining whether there is interference in each azimuth pulse is as follows: In the frequency domain, interference detection is performed based on the difference between the kurtosis of the random variable and the kurtosis of the Gaussian random variable, where the kurtosis of the Gaussian random variable is always 3 and the kurtosis of the random variable is K. ur , K ur Related to the mean and variance of the random variable, when there is a pulse in a certain direction |K ur When -3|≥0.5, it is considered that the pulse in this azimuth direction is interfered.

3. The multi-dimensional domain joint SAR interference parameter estimation method according to claim 1, characterized in that: The method for estimating the parameters of the point frequency interference signal is: After Fourier transform of the point frequency interference signal, the peak value is searched in the frequency domain. The frequency corresponding to the peak position is the center frequency of the point frequency interference signal. Perform median filtering along the time domain, then detect the rising edge and falling edge and calculate the average of the coordinate difference between the rising edge and the falling edge to calculate the time width and pulse repetition frequency of the point frequency interference signal; If there are multiple different point frequency interference signals, the point frequency interference signals with different center frequencies are separated by setting a band-limited filter, and then the time width and pulse repetition frequency of each point frequency interference signal are calculated respectively.

4. The multi-dimensional domain joint SAR interference parameter estimation method according to claim 1, characterized in that: The time domain start and end of the cross-azimuth pulse signal are searched by the echo domain amplitude ratio method. The search method is expressed as: Among them, N1 and N2 represent the pulse numbers of the cross-azimuth pulse signal respectively, and N r It represents the number of range points in the echo domain, i represents the i-th range point of the N1-th or N2-th azimuth pulse of the echo, and echo() represents a two-dimensional echo matrix.

5. The multi-dimensional domain joint SAR interference parameter estimation method according to claim 1, characterized in that: Assuming that the number of pulses of the cross-azimuth pulse signal is 2, the time width of the linear frequency modulation interference signal is expressed as: t d =PRT echo -X1+X2 Among them, PRT echo represents the pulse repetition time of the SAR echo, X1 represents the interval between the start time of the first interference pulse and the start time of the first receiving window, and X2 represents the interval between the end time of the first interference pulse and the start time of the second receiving window; Search for the time domain start end of the next azimuth pulse that matches the modulation frequency, and the pulse repetition frequency is expressed as: PRF=1 / (N·PRT echo -X1+X3) Wherein, X3 represents the time interval between the start time of the next interference pulse and the start time of the receiving window, and N represents the number of pulses across azimuth directions.

6. The multi-dimensional domain joint SAR interference parameter estimation method according to claim 1, characterized in that: For the combined interference signal, the complex empirical mode decomposition method is used to separate the combined interference signal into multiple intrinsic mode functions. For each intrinsic mode function, each signal component is located on the time axis and frequency axis of the time-frequency plane, and the signals containing similar frequency components in different time periods are decomposed to obtain the time-frequency spectrum of each signal. The obtained time-frequency spectrum is input into the pre-trained convolutional neural network to obtain the interference category of each signal.

7. The multi-dimensional domain joint SAR interference parameter estimation method according to claim 6, characterized in that: The combined interference signal is separated into multiple intrinsic mode functions by complex empirical mode decomposition method, including: Assume that the received combined interference signal is y(t), which has K intrinsic mode function components, the number of projection directions is L, the number of iterations is I, and the initialization conditions are k=1 and I=1; Calculate y k (t) Projection into L directions Finding vectors The local maximum of And calculate the average of the envelopes of different projection directions to get the average rotation vector m k (t); From y k Subtract the average rotation vector from (t) to obtain the residual r k，i (t); Let I=I+1,y k (t) = r k，i (t), repeat the above steps until the remaining amount r k，i (t) is a monotonic function; if the stopping condition is reached, then r k,I+1 (t) is an intrinsic mode function component, and the acquired signal c is recorded. k (t) = r k,I+1 (t), let y k =y(t), k=k+1, repeat the above steps until k=K, and write the combined interference signal y(t) as K intrinsic mode functions and trend d K (t) is in the form of a sum, The eigenmode function is defined as finite in time domain and the above equation is re-expressed as: Among them, c q (t) and c k (t) have the same meaning, is a rectangular function, defined as: Among them, b q is the bandwidth, τ q Is the pulse width.

8. The multi-dimensional domain joint SAR interference parameter estimation method according to claim 1, characterized in that: The convolutional neural network adopts a VGG-16 network. During training, SAR echo domain data with different interference categories are obtained, wherein the interference categories include point frequency interference, linear frequency modulation interference and combined interference; the SAR echo domain data is transformed to the time-frequency domain through short-time Fourier transform to obtain interference time-frequency domain images to form a training data set.

Images