CN115356694B

CN115356694B - High-frequency ground wave radar anti-impact interference method, system, computer equipment and application

Info

Publication number: CN115356694B
Application number: CN202211032369.1A
Authority: CN
Inventors: 于长军; 崔娜; 刘爱军; 王霖玮; 吕哲
Original assignee: Harbin Institute of Technology Weihai
Current assignee: Harbin Institute of Technology Weihai
Priority date: 2022-08-26
Filing date: 2022-08-26
Publication date: 2023-08-22
Anticipated expiration: 2042-08-26
Also published as: CN115356694A

Abstract

The invention belongs to the technical field of high-frequency ground wave radar signal identification, and discloses a high-frequency ground wave radar impact interference resistance method, a high-frequency ground wave radar impact interference resistance system, computer equipment and application. The high-frequency ground wave radar anti-impact interference method comprises the following steps: performing distance dimension FFT processing on the radar echo to obtain information of a distance unit where the target is located; taking out time sequences of different sweep frequency periods corresponding to the distance unit where the target is located, and detecting the interference position of the time sequences; adopting an LSTM neural network to reconstruct and restore the original data; performing speed dimension FFT processing on the reconstructed data; and performing constant false alarm detection on the target in the speed dimension to acquire final target distance speed information. The LSTM method provided by the invention can be quickly converged when being applied to data reconstruction of impact interference, and can meet the real-time processing requirement.

Description

High-frequency ground wave radar anti-impact interference method, system, computer equipment and application

Technical Field

The invention belongs to the technical field of high-frequency ground wave radar signal identification, and particularly relates to a high-frequency ground wave radar impact interference resistance method, a high-frequency ground wave radar impact interference resistance system, computer equipment and application.

Background

High-frequency ground wave radars are usually erected on the shore, and interference occurs near the coast and in offshore environments, and particularly impulse interference represented by lightning interference can have a great influence on the target detection performance of the radar. How to suppress the impact interference is a key to improve the performance of the high-frequency ground wave radar. The suppression of the impact interference is mainly reflected in the detection of the interference position and the reconstruction recovery of the original data.

The interference detection method is mature and mainly comprises a wavelet method, a matrix decomposition method, an empirical mode decomposition method and the like. However, how to reconstruct and recover the original data after the interference is detected is still a challenging invention subject, and is also a scientific problem that the present invention is mainly expected to solve. The existing data reconstruction method mainly comprises a linear prediction method, an interpolation method, a zero setting method, a downsampling method and the like, wherein the AR linear prediction method is simple and convenient to calculate, is suitable for matching signal spectrum peaks, and is most commonly used in prediction problems.

Through the above analysis, the problems and defects existing in the prior art are as follows:

(1) For long-term impact interference, since the AR linear prediction method has limited improvement of noise floor, when the interference intensity exceeds 100dB, the AR method hardly detects the target from the restored data.

(2) For intensive impact interference, since normal data capable of being used for AR modeling is interrupted by interference, enough modeling data cannot be provided for an AR model, the extrapolation capability of linear prediction is seriously affected, and the effective recovery of data after interference rejection cannot be completed by conventional linear prediction.

(3) The impact interference suppression effect of the high-frequency ground wave radar in the prior art needs to be further optimized, and the application effect of the high-frequency ground wave radar in a specific application scene is limited.

Disclosure of Invention

In order to overcome the problems in the related art, the disclosed embodiments of the present invention provide a method, a system, a computer device and an application for high-frequency ground wave radar to resist impact interference. The invention aims to solve the problem of how to effectively recover data under the conditions of long-term interference and dense interference so as to improve the target detection performance, and has important significance.

The technical scheme is as follows: a high-frequency ground wave radar anti-impact interference method based on an LSTM neural network comprises the following steps:

s1, carrying out distance dimension FFT processing on radar echoes to obtain information of a distance unit where a target is located;

s2, taking out time sequences of different sweep frequency periods corresponding to the distance unit where the target is located, and detecting the interference position start and end of the time sequences;

S3, reconstructing and recovering the original data by adopting an LSTM neural network;

s4, performing speed dimension FFT processing on the reconstructed data;

and S5, performing constant false alarm detection on the target in the speed dimension to acquire final target distance speed information.

In one embodiment, in step S1, performing distance-dimensional FFT processing on the radar echo includes the steps of:

1) Constructing a high-frequency ground wave radar FMICW signal model: the high-frequency ground wave radar FMICW transmits reference signals as follows:

wherein T is _r For the sweep period, f ₀ For transmitting the carrier frequency, N is the number of sweep cycles contained in a coherent accumulation period, k is the frequency modulation slope, u ₁ And (t) is a truncated uniform rectangular pulse. The target echo signal may be expressed as:

s _R (t)＝Gs _T (t-τ)；

g is the wave propagation attenuation and receiving antenna influence factor, and tau is the propagation delay of the echo waveform relative to the transmitting waveform; mixing a received signal and a transmitted reference signal, and performing low-pass filtering to obtain a difference frequency signal corresponding to an nth frequency sweep period, wherein the difference frequency signal is:

wherein M (t) is a rectangular pulse of the effective reception time of the receiver, the instantaneous phase p (t) and the frequency f (t) of the difference frequency signal s (t) are respectively:

in the case of a single-base radar,wherein R is the target distance, v is the target speed, and c is the speed of light; f (f) _dv Is a velocity Doppler shift due to the velocity v of the target motion, f _R (n) is the distance R between the radar and the target _n At the time, the frequency difference frequency f of the transmitted wave and the received wave _dRv (n) is f _R (n) range-Doppler shift, f, for target velocity v _dRv (n) and k _v Neglecting; performing distance dimension FFT on the difference frequency signal s (t) to obtain distance information of a target, and performing speed dimension FFT to obtain speed information of the target;

2) Simulation of interference signals:

taking a binary sequence y subject to poisson distribution ₁ (n) and taking the exponential distribution of the sequence y ₂ The product of (n) is used for obtaining a random impulse interference sequence y (n) with controllable power intensity and controllable lightning occurrence density;

poisson distributed binary sequence y ₁ Probability density of (n):

binary sequence y of exponential distribution ₂ Probability density of (n):

q(n,λ)＝λe ^-λn ；

the resulting interference sequence

y(n)＝y ₁ (n)*y ₂ (n)；

After the y (n) construction is completed, it is superimposed to the simulated normal radar echo signal sequence s _R (n) obtaining the interfered radar echo sequence s _r (n)；

s _r (n)＝s _R (n)+y(n)；

In the acquisition of radar echo signal s _r (n) after that, with reference signal s _T (n) mixing to obtain a difference frequency signal s (n); performing distance dimension FFT processing on the difference frequency signal s (n), wherein the distance dimension FFT processing is performed on the difference frequency signal s (n) in the same sweep frequency periodFFT transforming the echo of the target distance Doppler frequency shift f corresponding to the frequency corresponding to the peak value in the FFT transformation result corresponding to each sweep period _R (n) obtaining distance information of the target; and taking out time sequences of different sweep frequency periods corresponding to the distance unit where the target is located, and taking the time sequences as input time sequences for detecting the subsequent interference positions.

In one embodiment, in the step S2 of performing the start-end detection of the interference position, the wavelet-matrix combination method is used to detect the interference position, which specifically includes the following steps:

the method comprises the steps that 1-scale wavelet transformation is adopted to roughly detect whether interference is short-time interference, if so, a matrix decomposition method is used to secondarily detect the interference position, and fine positioning [ start, end ] of the interference position is obtained; if the coarse detection result of the scale 1 wavelet is long-term interference, continuing to perform scale 2 and scale 3 wavelet transformation, merging the results of the 3 wavelet transformations with different scales, and performing missing detection to finally obtain an interference start and end position index [ start, end ].

In one embodiment, in step S3, the reconstruction recovery of the original data using the LSTM neural network includes the steps of:

(1) And solving the complex problem by using a multi-layer perceptron of a Cell unit of the LSTM neural network, setting LSTM training learning parameters according to the characteristics of radar echo time sequence signals, and setting an environment for data reconstruction.

(2) The LSTM neural network reconstructs the interference data: taking out the real part and the imaginary part of the complex signal subjected to distance dimension FFT processing in the step S1 respectively, and performing reconstruction processing respectively; taking normal data in front of an interference section as training data input by an LSTM network, giving an index [ start, end ] of the data to be reconstructed in an original signal, and reconstructing data s (n) corresponding to the [ start, end ] index section by the LSTM network;

(3) If a plurality of interference data segments exist, the LSTM method should follow the time sequence of the signal when reconstructing the data, and reconstruct and restore from the front end to the rear end of the echo signal until the reconstruction of the last segment of interference data is completed; after the real part and the imaginary part signals are respectively reconstructed and restored, the real part and the imaginary part signals are added to obtain a final reconstructed signal. In addition, the reconstructed data can also be used as subsequent training data.

In one embodiment, in step S4, performing a speed-dimensional FFT process on the reconstructed data includes the steps of: after interference suppression processing is completed by utilizing an LSTM network, performing speed-dimensional FFT (fast Fourier transform) on the obtained time sequence to obtain a speed-dimensional pulse compression result and obtain target speed Doppler information f _dv (n); and finishing the two-dimensional FFT processing of the signal.

In one embodiment, in step S5, performing constant false alarm detection on the target in the speed dimension includes:

let k be the power value of the unit to be detected, [ k ] ₁ …k _j ]And [ k ] _j+1 …k _N ]The power of the reference units on the left side and the right side of the unit to be detected is respectively, and N is the total number of the reference units. The noise power estimate y is calculated using the following equation:

the threshold factor T can be expressed as:

wherein P is _f The false alarm probability is represented, and the noise power estimated value y is multiplied by a Threshold factor T to obtain a Threshold signal Threshold;

Threshold＝y*T

taking the power value k of the unit to be detected and a Threshold signal Threshold as inputs of a comparison discriminator, if k is more than Threshold, judging that the target exists, outputting 1, otherwise judging that the target does not exist, outputting 0;

the obtaining final target distance speed information comprises the following steps: after carrying out two-dimensional FFT processing on the echo, taking out a velocity Doppler spectrum corresponding to a distance unit where a target is located; and (3) performing CA-CFAR detection on the velocity Doppler spectrum, and finding the position of the spectrum peak value to obtain the distance and velocity information of the target.

In one embodiment, the high-frequency ground wave radar anti-impact interference method based on the LSTM neural network further comprises:

step 1, mixing the radar echo signal with a transmitted reference signal to obtain a difference frequency signal s (n). Performing distance dimension FFT processing on the difference frequency signal s (n) to obtain distance unit information of the target; taking out the distance unit R where the target is located _n Corresponding time series of each sweep frequency period

Step 2, the time sequence obtained in step 1 is combined by wavelet-matrix methodDetecting the interference position to obtain a vector r= [ start, end ] of an interval where the interference is located]；

Step 3, constructing an LSTM regression network; LSTM neural network parameters include: the Cell unit number of the LSTM network is set to 200; the solver is set as an Adam solver, and the Adam solver can comprehensively consider the mean value and the variance of the gradient to calculate the update step length; the gradient threshold is set to 1 to prevent gradient explosion; the initial learning rate was set to 0.005, and the learning rate was reduced by multiplying by a factor of 0.2 after 150 rounds of training; the training wheel number is set to be 200;

step 4, respectively taking out the sequencesAnd (3) reconstructing the real part and the imaginary part respectively. Taking normal echo data of the front end of interfered real part data and an interference position vector r as the input of an LSTM neural network, wherein the output of the LSTM neural network is the reconstructed data;

step 5, after the real part data and the imaginary part data are processed respectively, adding the real part data and the imaginary part data to obtain a reconstructed complex signal; assigning the complex data to interference position data corresponding to the original input signal to obtain a final LSTM reconstruction signal;

Step 6, if a plurality of interference positions exist, processing in steps 1, 2, 3, 4 and 5 is sequentially performed on each interference data segment; and the data after the front section reconstruction can also be used as input training data during the rear section reconstruction.

Step 7, reconstructing the LSTM reconstruction signal s _LSTM (n) performing a velocity-dimensional FFT processing to obtain a distance unit R where the target is located _n A corresponding velocity doppler spectrum;

step 8, performing CA-CFAR detection on the velocity Doppler spectrum obtained in the step 7, extracting velocity information of the target, and obtaining distance and velocity information of the target;

and step 9, if the radar echo contains information of a plurality of targets, sequentially performing the data processing on sequences of different sweep periods corresponding to the distance units where the targets are located, and obtaining the distance and speed information of the targets.

Another object of the present invention is to provide a high-frequency ground wave radar anti-impact interference system based on an LSTM neural network, including:

the distance dimension FFT processing module is used for mixing the radar echo signal with the emission reference signal to obtain a difference frequency signal s (n); and performing distance dimension FFT processing on the difference frequency signal s (n) to obtain distance unit information of the target. Taking out the distance unit R where the target is located _n Corresponding time series of each sweep frequency period

An interference position detection module for time series obtained by wavelet-matrix combination methodDetecting the interference position to obtain a vector r= [ start, end ] of an interval where the interference is located]；

Constructing an LSTM regression network module, which is used for constructing an LSTM regression network;

real part and imaginary part reconstruction processing module for respectively extracting sequenceIs used for the real and imaginary parts of (a),reconstructing the real part and the imaginary part respectively; taking normal echo data of the front end of interfered real part data (imaginary part data) and an interference position vector r as inputs of an LSTM neural network, and outputting the LSTM neural network as reconstructed data;

the LSTM reconstruction signal acquisition module is used for adding the real part data and the imaginary part data after the real part data and the imaginary part data are respectively processed to obtain a reconstructed complex signal; assigning the complex data to interference position data corresponding to the original input signal to obtain a final LSTM reconstruction signal;

each interference data segment is processed by a processing module, and if a plurality of interference positions exist, the processing module is used for sequentially performing distance dimension FFT processing, interference position detection, LSTM regression network construction and real part and imaginary part reconstruction processing on each interference data segment; the data after the front section reconstruction can also be used as input training data during the rear section reconstruction;

A speed-dimensional FFT processing module for reconstructing the LSTM reconstruction signal s _LSTM (n) performing a velocity-dimensional FFT processing to obtain a distance unit R where the target is located _n A corresponding velocity doppler spectrum;

the CA-CFAR detection module is used for carrying out CA-CFAR detection on the obtained velocity Doppler spectrum, extracting velocity information of the target and obtaining distance and velocity information of the target;

and the target distance and speed information acquisition module is used for sequentially carrying out the data processing on the sequences of different sweep frequency periods corresponding to the distance units where the targets are positioned if the radar echo contains information of a plurality of targets so as to obtain the distance and speed information of the targets.

It is a further object of the present invention to provide a computer device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the LSTM neural network based high frequency ground wave radar anti-jamming method.

The invention further aims to provide an application of the LSTM neural network-based high-frequency ground wave radar impact interference resistance method in lightning interference, meteor trail interference, radio frequency interference and impulse noise impact interference resistance.

By combining all the technical schemes, the invention has the advantages and positive effects that:

first, aiming at the technical problems in the prior art and the difficulty in solving the problems, the technical problems solved by the technical proposal of the invention are analyzed in detail and deeply by tightly combining the technical proposal to be protected, the results and data in the research and development process, and the like, and some technical effects brought after the problems are solved have creative technical effects. The specific description is as follows: the invention provides a high-frequency ground wave radar anti-impact interference method. In view of the commonality of the impulse disturbances, and the typically nature of lightning impulse disturbances, subsequent impulse disturbance analysis is represented by lightning disturbances. Lightning impulse interference can typically span multiple sweep periods, which also results in complete failure of data during a sweep period. But the data will generally not fail completely during the coherent integration time of the radar. The invention uses LSTM neural network to reconstruct and recover the time series signals of different sweep frequency periods corresponding to the same distance unit, so as to realize the impact interference suppression of the cross sweep frequency period.

Second, advantages of the present invention over the prior art further include: after the interference position is detected, the invention uses the LSTM neural network to reconstruct and recover the interference data, thereby reducing the noise base and realizing the interference suppression of the high-frequency ground wave radar, and compared with the prior art, the invention has the advantages that: the LSTM neural network is mature, can learn and store long-term input, is suitable for modeling time series signals, and is simple to realize. When the method is applied to data reconstruction of impact interference, the method can be quickly converged, and is easy to process in real time. The LSTM neural network is utilized to reconstruct and recover the data after interference rejection, so that the problem that the reconstruction performance of the traditional AR linear prediction method is reduced under the conditions of long-term interference and dense interference can be solved, the Doppler spectrum noise base is further reduced, and the target detection is effectively realized under the conditions of high-strength long-term interference and dense interference, so that the target detection performance of a radar system is improved.

Thirdly, the technical scheme is regarded as a whole or from the perspective of products, and the technical scheme to be protected has the technical effects and advantages as follows: the invention carries out reconstruction recovery on radar echo data under long-time impact interference and dense impact interference based on the LSTM neural network so as to further reduce a noise base and improve target detection performance, thereby realizing impact interference suppression of the high-frequency ground wave radar. The invention provides a new method for suppressing the impact interference of the high-frequency ground wave radar.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the disclosure and together with the description, serve to explain the principles of the disclosure.

FIG. 1 is a flow chart of a high-frequency ground wave radar anti-impact interference method based on an LSTM neural network provided by an embodiment of the invention;

fig. 2 is an internal schematic view of a Cell unit according to an embodiment of the present invention;

FIG. 3 is a schematic diagram of a CA-CFAR provided by an embodiment of the present invention;

FIG. 4 is a schematic block diagram of a high-frequency ground wave radar anti-impact interference method based on an LSTM neural network provided by an embodiment of the invention;

FIG. 5 shows a range unit R of a target in the case of long-term interference provided by an embodiment of the present invention _n Corresponding time series of each sweep frequency periodA figure;

FIG. 6 (a) shows the original normal time sequence and the LSTM reconstructed sequence s in comparison with the original signal time sequence _LSTM A real waveform plot of (n).

FIG. 6 (b) shows the original normal time sequence and the LSTM reconstructed sequence s in comparison with the original signal time sequence _LSTM An imaginary waveform diagram of (n).

Fig. 7 is a graph of the corresponding interference strength of 50dB, provided by the embodiment of the present invention, compared with the signal reconstructed by the AR method and the signal reconstructed by the AR method under the long-term interference condition;

fig. 8 is a graph of comparison results between a signal obtained by reconstruction and a threshold signal of CFAR detection corresponding to the signal obtained by reconstruction by an AR method and a threshold signal amplitude of CFAR detection corresponding to the signal obtained by reconstruction under the condition of long-term interference provided by the embodiment of the present invention;

fig. 9 is a graph of the corresponding interference intensity between 35dB and 50dB, comparing the reconstructed signal with the signal reconstructed by the AR method provided by the embodiment of the present invention under the condition of dense interference;

fig. 10 is a graph of a comparison result between a signal obtained by reconstruction and a threshold signal of CFAR detection corresponding to the signal obtained by reconstruction under the condition of dense interference provided by the embodiment of the present invention and a signal obtained by reconstruction by an AR method and a threshold signal amplitude of CFAR detection;

FIG. 11 is a schematic diagram of an LSTM neural network-based high-frequency ground wave radar anti-impact interference system according to an embodiment of the present invention;

in the figure: 1. a distance dimension FFT processing module; 2. an interference position detection module; 3. constructing an LSTM regression network module; 4. a real part and an imaginary part reconstruction processing module; 5. an LSTM reconstruction signal acquisition module; 6. each interference data segment is processed by a processing module; 7. a speed dimension FFT processing module; 8. a CA-CFAR detection module; 9. and the target distance and speed information acquisition module.

Detailed Description

In order that the above objects, features and advantages of the invention will be readily understood, a more particular description of the invention will be rendered by reference to the appended drawings. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. The invention may be embodied in many other forms than described herein and similarly modified by those skilled in the art without departing from the spirit or scope of the invention, which is therefore not limited to the specific embodiments disclosed below.

1. Explanation of the examples:

the high-frequency ground wave radar impact interference resistance method based on the LSTM neural network provided by the embodiment of the invention is characterized in that the LSTM neural network is utilized to reconstruct and recover the interfered data, so that the problem that the reconstruction and recovery are difficult due to excessive data loss under the conditions of dense interference and long-time interference can be solved. The sensitivity of the LSTM neural network to the time sequence can be utilized to effectively reconstruct data under the condition of strong interference, so that the extraction of the target signal is realized.

Example 1

As shown in fig. 1, the high-frequency ground wave radar anti-impact interference method based on the LSTM neural network provided by the embodiment of the invention includes the following steps:

s101, carrying out distance dimension FFT processing on radar echo to obtain information of a distance unit where a target is located;

s102, taking out time sequences of different sweep frequency periods corresponding to a distance unit where a target is located, and detecting the interference position start and end of the time sequences;

s103, reconstructing and recovering the original data by adopting an LSTM neural network;

s104, performing speed dimension FFT processing on the reconstructed data;

s105, constant false alarm detection is carried out on the target in the speed dimension, and final target distance speed information is obtained.

Example 2

The method for resisting impact interference of high-frequency ground wave radar based on LSTM neural network provided in embodiment 1 further, in step S101, performing distance dimension FFT processing on radar echo includes the following steps:

1) High-frequency ground wave radar FMICW signal model:

the high frequency ground wave radar FMICW transmitted reference signal may be expressed as follows:

wherein T is _r For the sweep period, f ₀ For transmitting the carrier frequency, N is the number of sweep cycles contained in a coherent accumulation period, k is the frequency modulation slope, u ₁ And (t) is a truncated uniform rectangular pulse. Then target returns The wave signal can be expressed as:

s _R (t)＝Gs _T (t-τ)；

g is the wave propagation attenuation and the receive antenna influence factor, τ is the propagation delay of the echo waveform relative to the transmit waveform. Mixing a received signal and a transmitted reference signal, and performing low-pass filtering to obtain a difference frequency signal corresponding to an nth frequency sweep period, wherein the difference frequency signal is:

in the case of a single-base radar,wherein R is the target distance, v is the target speed, and c is the speed of light. f (f) _dv Is a velocity Doppler shift due to the velocity v of the target motion, f _R (n) is the distance R between the radar and the target _n At the time, the frequency difference frequency f of the transmitted wave and the received wave _dRv (n) is f _R (n) distance Doppler shift with respect to target velocity v, typically f _dRv (n) and k _v Is small and negligible. The distance dimension FFT is performed on the difference frequency signal s (t) to obtain the distance information of the target, and then the speed dimension FFT is performed on the distance information to obtain the speed information of the target.

2) Simulation of interference signals:

the interference signals have randomness and different intensities. Here, the interference is approximated by poisson distributionIs distributed exponentially to approximate the power characteristics of the impulse interference. Taking a binary sequence y subject to poisson distribution ₁ (n) and taking the exponential distribution of the sequence y ₂ And (n) obtaining the random impulse interference sequence y (n) with controllable power intensity and controllable lightning occurrence density.

Poisson distributed binary sequence y ₁ Probability density of (n):

binary sequence y of exponential distribution ₂ Probability density of (n):

q(n,λ)＝λe ^-λn ；

the resulting interference sequence

y(n)＝y ₁ (n)*y ₂ (n)；

After the y (n) construction is completed, it is superimposed to the simulated normal radar echo signal sequence s _R (n) obtaining the interfered radar echo sequence s _r (n)。

s _r (n)＝s _R (n)+y(n)；

In the acquisition of radar echo signal s _r After (n), it is combined with a reference signal s _T (n) mixing to obtain a difference frequency signal s (n). Performing distance dimension FFT processing on the difference frequency signal s (n), namely performing FFT conversion on echoes in the same sweep frequency period, wherein the frequency corresponding to the peak value in the FFT conversion result corresponding to each sweep frequency period is corresponding to the target distance Doppler frequency shift f _R (n) obtaining the distance information of the target. And taking out time sequences of different sweep frequency periods corresponding to the distance unit where the target is located, and taking the time sequences as input time sequences for detecting the subsequent interference positions.

Example 3

In the method for resisting impact interference of the high-frequency ground wave radar based on the LSTM neural network provided in embodiment 1, further, in step S102, a time sequence of different sweep periods corresponding to a distance unit where a target is located is taken out, and the detecting of the interference position start and end of the time sequence includes the following steps:

The wavelet method and the matrix method are relatively mature in the field of signal singularity detection, the mean value error of the wavelet method detection is larger, the variance is smaller, the method is more suitable for long-time impact interference detection, and the matrix decomposition method is more beneficial to positioning of short-time impact interference, so that the wavelet-matrix combination method is adopted to detect the interference position.

1) The transformation and inverse transformation of the wavelet function is shown as follows:

psi (t) is a wavelet basis function, and satisfies the conditionψ _a,b (t) is wavelet function family, which is obtained by wavelet function psi (t) through series translation and expansion, satisfying +.>a represents a scale-up factor and b represents a translation factor. The wavelet transformation can give consideration to time resolution and frequency resolution, and after the signal is subjected to wavelet transformation, the high-frequency components of the signal are relatively concentrated in the time domain, and the low-frequency components are relatively concentrated in the frequency domain. In order to prevent false detection of missed detection, the signals after impact interference are respectively subjected to multi-scale wavelet transformation, and the union of detection is taken as the final interference positioning position.

2) Matrix method interference detection:

in the matrix method interference detection, the echo time sequence of the interfered impact is assumed to be K= [ K ] ₁ ,k ₂ ,…k _N ]Where N is the number of sweep cycles, dividing K equally into d segments, each segment having a points, i.e. n=d×a, and constructing a matrix M as follows:

And (3) carrying out singular value decomposition on M, wherein the main singular value corresponding to M is the corresponding impact interference component. And weighting the characteristic items of the singular values, recovering the characteristic items into one-dimensional vectors, and positioning the interference interval through threshold discrimination. And after the interference interval data is set to zero, linear prediction is carried out on the one-dimensional vector of the orthogonal echo component corresponding to the lightning interference, and data recovery of the interference section is completed.

3) Wavelet-matrix combination method: firstly, whether the interference is short-time interference is detected by adopting scale 1 wavelet transformation, and if the interference is short-time interference, the interference position is detected secondarily by using a matrix decomposition method, so that the fine positioning [ start, end ] of the interference position is obtained. If the coarse detection result of the scale 1 wavelet is long-term interference, the scale 2 wavelet transformation and the scale 3 wavelet transformation are continued, and in order to prevent missed detection, the results of the 3 wavelet transformations with different scales are combined to obtain the final interference start and end position indexes [ start, end ].

Example 4

The method for high-frequency ground wave radar anti-impact interference based on LSTM neural network provided in embodiment 1, further, in step S103, the method for reconstructing and recovering the original data by using the LSTM neural network includes the following steps:

(1) LSTM neural network principle

LSTM is a recurrent neural network that can be used to learn and predict time series. The key to LSTM is the state transfer of Cell cells, which transfer from the last Cell to the next Cell, and other parts have little linear interaction. Inside the Cell unit, LSTM controls or adds information through a "gate" to implement forgetting or memorizing functions. The "gate" is a structure that selectively passes information, and is composed of a sigmoid function and a dot product operation. The output value of the sigmoid function is in the [0,1] interval, 0 representing complete discard, and 1 representing complete pass. A Cell unit has three such gates, namely a forget gate, an input gate, and an output gate.

An internal schematic diagram of the Cell unit is shown in fig. 2.

In a preferred embodiment of the present invention, the Cell state update procedure:

input gate i _t Receiving current input x _t And finally hiddenState h of hiding _t-1 As input, and calculate i according to the following formula _t ：

i _t ＝σ(W _ix x _t +W _ih h _t-1 +b _i )；

After calculation, a value of 0 indicates that no information currently entered will enter a unit state, and a value of 1 indicates that all information currently entered will enter a unit state. Then, the following formula will calculate another value, called candidate value. It is used to calculate the current cell state.

The forget gate will perform the following operations: forgetting a threshold of 0 indicates no c _t-1 Any information passed to c _t A value of 1 means all c _t-1 Information of (c) is propagated to c _t

f _t ＝σ(W _fx x _t +W _fh h _t-1 +b _f )；

Current state calculation:

last state h _t And (3) calculating:

o _t ＝σ(W _ox x _t +W _oh h _t-1 +b _o )；

h _t ＝o _t tanh(c _t )；

LSTM solves the problem that RNN networks have difficulty learning and long-term information storage, employing Cell units to store long-term inputs. The Cell unit of LSTM resembles an accumulator and gating neuron: it will have a weight in the next time step and connect itself, copying the true value of itself's state and the accumulated external signal, but this self-connection is controlled by the multiplication gate that another unit learns and decides when to clear the memory; LSTM is an excellent variant of Recurrent Neural Networks (RNNs), inheriting the characteristics of most RNN models: having multiple layers of perceptrons to solve complex problems; global characteristics and local characteristics can be considered; is suitable for modeling the time sequence; in addition, the method solves the problem of gradient disappearance caused by gradual reduction in the RNN gradient back-propagation process.

The length and the intensity of the impact interference sequence are random, and the arrival time of the impact interference sequence is also random, namely, the influence on the radar echo time sequence is also random, so that the recovery of the original radar echo time sequence signal is realized by considering both the local characteristics of the echo and the integral characteristics of the echo, and meanwhile, the complex variability of the interference is considered, and the LSTM network is very suitable for the modeling recovery of the radar echo time sequence signal.

(2) The process of reconstructing interference data by the LSTM neural network comprises the following steps:

since LSTM neural networks cannot process complex data, the signals after the distance FFT processing are complex signals. The real and imaginary parts of the signal are thus taken out separately, and each is subjected to reconstruction processing. And taking normal data in front of the interference section as training data input by an LSTM network, and giving an index [ start, end ] of the data to be reconstructed in the original signal, wherein the LSTM network can reconstruct data s (n) corresponding to the [ start, end ] index section. If there are multiple interference data segments, the LSTM method should follow the signal time sequence when reconstructing the data, and reconstruct and restore from the front end to the back end of the echo signal until the reconstruction of the last segment of interference data is completed. After the real part and the imaginary part signals are respectively reconstructed and restored, the real part and the imaginary part signals are added to obtain a final reconstructed signal.

Example 5

The method for high-frequency ground wave radar anti-impact interference based on LSTM neural network provided in embodiment 1 further, in step S104, performing a velocity-dimensional FFT process on the reconstructed data includes the following steps:

the FFT process may be equivalent to passing through a digital filter bank, where the signal to noise ratio becomes N times before for a narrowband signal, i.e., the FFT may be equivalent to a coherent accumulation process. For echo signal by two-dimension FFT processing is equivalent to coherent accumulation of the same in the distance and velocity dimensions, respectively. Thus, in a two-dimensional FFT spectrum, the energy of the signal is concentrated at the range and velocity doppler locations corresponding to the target. After interference suppression processing is completed by utilizing an LSTM network, performing speed-dimensional FFT conversion on the obtained time sequence, and obtaining a speed-dimensional pulse compression result to obtain target speed Doppler information f _dv (n). Thus, the two-dimensional FFT processing of the signal is completed.

Example 6

The method for high-frequency ground wave radar anti-impact interference based on LSTM neural network provided in embodiment 1 further, in step S105, performing CFAR constant false alarm detection on the target in the speed dimension includes the following steps:

Target detection for high frequency ground wave over-the-horizon radar is accomplished in the Doppler domain. Since sea clutter, atmospheric noise and target information contained in the echo spectrum after two-dimensional FFT conversion can be approximately gaussian distributed, a unit average constant false alarm (Cell averaging constant false alarm rate, CA-CFAR) detection method can be adopted to complete target detection. The CA-CFAR principle is shown in FIG. 3.

the threshold factor T can be expressed as:

Threshold＝y*T

and after carrying out two-dimensional FFT processing on the echo, taking out a velocity Doppler spectrum corresponding to the distance unit where the target is located, wherein the position where the peak is located is velocity Doppler information corresponding to the target. And carrying out CFAR detection on the target, and finally obtaining the distance and speed information of the target.

Example 7

As shown in fig. 4, the high-frequency ground wave radar anti-impact interference method based on the LSTM neural network provided by the embodiment of the invention includes the following steps:

step 1, mixing the radar echo signal with a transmitted reference signal to obtain a difference frequency signal s (n). And performing distance dimension FFT processing on the difference frequency signal s (n) to obtain distance unit information of the target. Taking out the distance unit R where the target is located _n Corresponding time series of each sweep frequency period

Step 2, the time sequence obtained in step 1 is combined by wavelet-matrix methodDetecting the interference position to obtain a vector r= [ start, end ] of an interval where the interference is located]。

And 3, constructing an LSTM regression network. LSTM neural network parameters are as follows: the Cell unit number of the LSTM network is set to 200; the solver is set as an Adam solver, and the Adam solver can comprehensively consider the mean value and the variance of the gradient to calculate the update step length; the gradient threshold is set to 1 to prevent gradient explosion; the initial learning rate was set to 0.005, and the learning rate was reduced by multiplying by a factor of 0.2 after 150 rounds of training; the training wheel number was set to 200.

Step 4, LSTM network cannot process complex signal, so the sequences are respectively fetchedAnd (3) reconstructing the real part and the imaginary part respectively. The normal echo data of the front end of the interfered real part data (imaginary part data) and the interference position vector r are used as the inputs of the LSTM neural network, and the output of the LSTM neural network is the reconstructed data.

And 5, after the real part data and the imaginary part data are processed respectively, adding the real part data and the imaginary part data to obtain a reconstructed complex signal. And assigning the complex data to interference position data corresponding to the original input signal to obtain a final LSTM reconstruction signal.

Step 6, if there are a plurality of interference positions, the processing of step 1, step 2, step 3, step 4 and step 5 is sequentially performed for each interference data segment. And the data after the front section reconstruction can also be used as input training data during the rear section reconstruction.

Step 7, reconstructing the LSTM reconstruction signal s _LSTM (n) performing a velocity-dimensional FFT processing to obtain a distance unit R where the target is located _n A corresponding velocity doppler spectrum.

And 8, performing CA-CFAR detection on the velocity Doppler spectrum obtained in the step 7, and extracting velocity information of the target. Thus, distance and speed information of the target is obtained.

And step 9, if the radar echo contains information of a plurality of targets, the data processing can be sequentially carried out on sequences of different sweep periods corresponding to the distance units where the targets are located, so as to obtain the distance and speed information of the targets.

Example 8

As shown in fig. 11, the high-frequency ground wave radar anti-impact interference system based on LSTM neural network provided by the embodiment of the present invention includes:

The distance dimension FFT processing module 1 is used for mixing the radar echo signal with the emission reference signal to obtain a difference frequency signal s (n). And performing distance dimension FFT processing on the difference frequency signal s (n) to obtain distance unit information of the target. Taking out the distance unit R where the target is located _n Corresponding time series of each sweep frequency period

An interference position detection module 2 for time series obtained by wavelet-matrix combination methodDetecting the interference position to obtain a vector r= [ start, end ] of an interval where the interference is located]。

The LSTM regression network module 3 is used for constructing an LSTM regression network, and the LSTM neural network parameters are as follows: the Cell unit number of the LSTM network is set to 200; the solver is set as an Adam solver, and the Adam solver can comprehensively analyze the mean and variance of the gradient to calculate an update step length; the gradient threshold is set to 1 to prevent gradient explosion; the initial learning rate was set to 0.005, and the learning rate was reduced by multiplying by a factor of 0.2 after 150 rounds of training; the training wheel number was set to 200.

A real part and an imaginary part reconstruction processing module 4 for respectively extracting the sequencesAnd (3) reconstructing the real part and the imaginary part respectively. The normal echo data of the front end of the interfered real part data (imaginary part data) and the interference position vector r are used as the inputs of the LSTM neural network, and the output of the LSTM neural network is the reconstructed data.

And the LSTM reconstruction signal acquisition module 5 is used for adding the real part data and the imaginary part data after the real part data and the imaginary part data are respectively processed, so as to obtain a reconstructed complex signal. And assigning the complex data to interference position data corresponding to the original input signal to obtain a final LSTM reconstruction signal.

And the processing module 6 is used for sequentially performing distance dimension FFT processing, interference position detection, LSTM regression network construction and real part and imaginary part reconstruction processing on each interference data segment if a plurality of interference positions exist. And the data after the front section reconstruction can also be used as input training data during the rear section reconstruction.

A speed-dimensional FFT processing module 7 for reconstructing the LSTM reconstruction signal s _LSTM (n) performing a velocity-dimensional FFT processing to obtain a distance unit R where the target is located _n Corresponding toVelocity doppler spectrum.

And the CA-CFAR detection module 8 is used for carrying out CA-CFAR detection on the obtained velocity Doppler spectrum and extracting the velocity information of the target. Thus, distance and speed information of the target is obtained.

The target distance and speed information obtaining module 9 is configured to, if the radar echo contains information of a plurality of targets, sequentially perform the data processing on sequences of different sweep periods corresponding to distance units where the targets are located, so as to obtain distance and speed information of the targets.

In the foregoing embodiments, the descriptions of the embodiments are emphasized, and in part, not described or illustrated in any particular embodiment, reference is made to the related descriptions of other embodiments.

The content of the information interaction and the execution process between the devices/units and the like is based on the same conception as the method embodiment of the present invention, and specific functions and technical effects brought by the content can be referred to in the method embodiment section, and will not be described herein.

It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-described division of the functional units and modules is illustrated, and in practical application, the above-described functional distribution may be performed by different functional units and modules according to needs, i.e. the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-described functions. The functional units and modules in the embodiment may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit, where the integrated units may be implemented in a form of hardware or a form of a software functional unit. In addition, the specific names of the functional units and modules are only for distinguishing from each other, and are not used for limiting the protection scope of the present invention. The specific working process of the units and modules in the above system may refer to the corresponding process in the foregoing method embodiment, which is not described herein again.

2. Application examples:

application example 1

The invention provides an LSTM neural network-based high-frequency ground wave radar impact interference suppression method aiming at the problem of impact interference suppression of a high-frequency ground wave radar. In addition, since lightning interference, meteor trail interference, radio frequency interference, impulse noise and the like have similarity in time domain and frequency domain, the impact interference resistance method provided by the invention is applicable to the above scenes.

Application example 2

The embodiment of the invention also provides a computer device, which comprises: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, which when executed by the processor performs the steps of any of the various method embodiments described above.

Application example 3

Embodiments of the present invention also provide a computer readable storage medium storing a computer program which, when executed by a processor, performs the steps of the respective method embodiments described above.

Application example 4

The embodiment of the invention also provides an information data processing terminal, which is used for providing a user input interface to implement the steps in the method embodiments when being implemented on an electronic device, and the information data processing terminal is not limited to a mobile phone, a computer and a switch.

Application example 5

The embodiment of the invention also provides a server, which is used for realizing the steps in the method embodiments when being executed on the electronic device and providing a user input interface.

Application example 6

Embodiments of the present invention provide a computer program product which, when run on an electronic device, causes the electronic device to perform the steps of the method embodiments described above.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the present invention may implement all or part of the flow of the method of the above embodiments, and may be implemented by a computer program to instruct related hardware, where the computer program may be stored in a computer readable storage medium, and when the computer program is executed by a processor, the computer program may implement the steps of each of the method embodiments described above. Wherein the computer program comprises computer program code which may be in source code form, object code form, executable file or some intermediate form etc. The computer readable medium may include at least: any entity or device capable of carrying computer program code to a photographing apparatus/terminal device, recording medium, computer memory, read-only memory (ROM), random access memory (RandomAccessMemory, RAM), electrical carrier signal, telecommunication signal, and software distribution medium. Such as a U-disk, removable hard disk, magnetic or optical disk, etc.

3. Evidence of example related effects:

simulation experiments performed by the embodiment of the invention:

according to the requirements of distance resolution and speed resolution, the radar coherence accumulation time is set to 256 sweep periods in the simulation experiment, the sweep period is 30ms, the sampling point number in a single sweep period is 128, the working frequency is 7MHz, the signal bandwidth is 30kHz, and the target in the simulation is located in a distance unit R _n Where it is located. The interference suppression effect of the LSTM method under the conditions of long-term interference and dense interference is respectively simulated in the simulation experiment.

1) Long-term interference condition simulation

In order to verify the performance of the LSTM reconstruction method under the condition of long-term interference, the invention respectively simulates the long-term interference condition that the interference duration is 25 sweep periods and the interference intensity is 50dB to 120 dB. First, for radarNormal echo sampling signal s _R (n) overlapping the interference sequences with the data corresponding to the continuous 25 sweep periods to obtain a received signal s _r (n). Respectively s _r (n) and s _R (n) and s _T (n) mixing to obtain an interfered sequence s ₁ (n) and normal reference sequence s ₂ (n)。

For s ₁ (n) and acceptor s ₂ (n) performing first FFT processing in the distance dimension to obtain a distance pulse compression processed echo signal s ₁₁ (n) and acceptor s ₂₁ (n). Respectively at s ₁₁ (n) and s ₂₁ The distance unit R is taken out in (n) _n Corresponding time seriesAndfor->After interference position detection, the AR method and LSTM method are used to detect +.>Reconstructing to obtain a reconstructed time sequence s _AR (n) and s _LSTM (n) for s _AR (n)、s _LSTM (n) and respectively performing a second FFT treatment to obtain the velocity Doppler spectrum of the signals after the reconstruction of the AR method and the LSTM method. And then->A second FFT process is performed as a reference for the reconstruction effect.

Fig. 4 shows a schematic block diagram of a high-frequency ground wave radar anti-impact interference method based on an LSTM neural network according to an embodiment of the present invention.

Fig. 5 shows a distance unit R _n Time series of impacted interference corresponding to different sweep periodsAnd original signal time sequence +.>Is a real part of the characteristic of (a).

The original normal time sequence is given in FIG. 6 (a)And sequence s recovered by LSTM reconstruction _LSTM A real waveform plot of (n).

FIG. 6 (b) shows the original normal time sequenceAnd sequence s recovered by LSTM reconstruction _LSTM An imaginary waveform diagram of (n).

The effect of comparing the reconstructed Doppler spectrum with the LSTM method and AR method with the original Doppler spectrum is shown in FIG. 7, where the interference intensity is 50dB and the interference duration is 25 sweep cycles. The red dashed line represents the echo signal after the impact disturbance, and the object is seen to have been completely annihilated in noise. The waveform of the echo signal after the reconstruction by the LSTM method is substantially consistent with the waveform of the original signal. The echo data after LSTM reconstruction achieves a lower noise floor than the AR method, especially near the target location, which may be about 10dB lower than the AR method.

Fig. 8 shows the LSTM and AR reconstructed signals and their corresponding CFAR detection thresholds for an interference strength of 120dB and an interference duration of 25 sweep cycles. The reference unit number n=8, the constant false alarm rate is set to P _f ＝5×10 ^-8 . As can be seen from the figure, in this case the signal reconstructed by the AR method is smaller than its corresponding threshold value, and the target has not been detected correctly. At this time, the signal obtained by the LSTM reconstruction method is larger than the threshold value, and the target position can still be correctly detected. The LSTM reconstruction method is superior to the AR method in long-term interference suppression, and can effectively reconstruct data under strong interference and realize target detection under the condition of strong interference.

2) Dense impact disturbance situation simulation

In order to verify the performance of the LSTM reconstruction method under the condition of dense interference, the invention simulates the condition of dense interference with the total interference number of 5. The duration of the single interference is 3-5 sweep periods, and the interference intensity is between 30dB and 50dB. First, 5 positions are extracted in 256 sweep periods, and the extracted positions are taken as interference starting positions. Normal data s corresponding to 3-5 continuous sweep periods after each interference starting position _R (n) superimposing the interference sequence to obtain the received signal s _r (n). Respectively s _r (n) and s _R (n) and s _T (n) mixing to obtain an interfered sequence s ₁ (n) and normal reference sequence s ₂ (n)。

The subsequent processing is similar to the simulation process under the condition of long-time interference, and is similar to that of s ₁ (n) and s ₂ (n) performing first FFT processing in the distance dimension to obtain a distance pulse compression processed echo signal s ₁₁ (n) and s ₂₁ (n). Respectively at s ₁₁ (n) and s ₂₁ The distance unit R is taken out in (n) _n Corresponding time seriesAnd->For->After interference position detection, the AR method and LSTM method are used to detect +.>Reconstructing to obtain a reconstructed time sequence s _AR (n) and s _LSTM (n) for s _AR (n)、s _LSTM (n) and respectively performing a second FFT treatment to obtain the velocity Doppler spectrum of the signals after the reconstruction of the AR method and the LSTM method.

The effect of comparing the reconstructed Doppler spectrum with the original reference Doppler spectrum by the LSTM method and the AR method under dense interference is shown in FIG. 9. The improvement of the LSTM method over the AR method in reconstructing the noise floor is not particularly significant in the case of dense interference, but the LSTM method noise floor can be about 10dB more than the AR method noise floor at the front end of the target location, which is very advantageous for target detection.

The false alarm probability P is given in FIG. 10 _f ＝4.5×10 ^-7 When the reference unit number n=8, the LSTM method and the AR method reconstruct the signal and the CFAR detection threshold corresponding thereto. It can be seen that at this time, the amplitude of the AR reconstruction signal is smaller than the threshold value, and the target cannot be detected, whereas the amplitude of the LSTM method is larger than the threshold value, and the target can still be detected. Under the condition of dense interference, the LSTM method has better target detection performance than the AR reconstruction method, and can be used for data reconstruction under the condition of dense interference.

While the invention has been described with respect to what is presently considered to be the most practical and preferred embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, but on the contrary, is intended to cover various modifications, equivalents, and alternatives falling within the spirit and scope of the invention.

Claims

1. A high-frequency ground wave radar anti-impact interference method based on an LSTM neural network is characterized by comprising the following steps:

s4, performing speed dimension FFT processing on the reconstructed data;

s5, constant false alarm detection is carried out on the target in the speed dimension, and final target distance speed information is obtained;

the method further comprises the steps of:

step 1, mixing radar echo signals with emission reference signals to obtain difference frequency signals s (n), and performing distance dimension FFT processing on the difference frequency signals s (n) to obtain Distance unit information to the target; taking out the distance unit R where the target is located _n Corresponding time series of each sweep frequency period

Step 3, constructing an LSTM regression network;

step 4, respectively taking out the sequencesThe real part and the imaginary part are respectively subjected to reconstruction processing; taking normal echo data of the front end of interfered real part data and an interference position vector r as the input of an LSTM neural network, wherein the output of the LSTM neural network is the reconstructed data;

step 6, if a plurality of interference positions exist, processing in steps 1, 2, 3, 4 and 5 is sequentially performed on each interference data segment; the data after the front section reconstruction can also be used as input training data during the rear section reconstruction;

2. The LSTM neural network-based high-frequency ground wave radar anti-collision interference method according to claim 1, wherein in step S1, performing distance-dimensional FFT processing on radar echoes includes the steps of:

1) Constructing a high-frequency ground wave radar FMICW signal model:

the high-frequency ground wave radar FMICW transmits reference signals as follows:

wherein T is _r For the sweep period, f ₀ For transmitting the carrier frequency, N is the number of sweep cycles contained in a coherent accumulation period, k is the frequency modulation slope, u ₁ (t) is a truncated uniform rectangular pulse, the target echo signal can be expressed as:

s _R (t)＝Gs _T (t-τ)；

wherein G is the wave propagation attenuation and the influence factor of a receiving antenna, and tau is the propagation delay of an echo waveform relative to a transmitting waveform; mixing a received signal and a transmitted reference signal, and performing low-pass filtering to obtain a difference frequency signal corresponding to an nth frequency sweep period, wherein the difference frequency signal is:

Wherein, M (t) is a rectangular pulse of the effective receiving time of the receiver, and the instantaneous phase p (t) and the frequency f (t) of the difference frequency signal s (t) are respectively:

2) Simulation of interference signals:

poisson distributed binary sequence y ₁ Probability density of (n):

binary sequence y of exponential distribution ₂ Probability density of (n):

q(n,λ)＝λe ^-λn ；

the resulting interference sequence

y(n)＝y ₁ (n)*y ₂ (n)；

s _r (n)＝s _R (n)+y(n)；

In the acquisition of radar echo signal s _r (n) after that, with reference signal s _T (n) mixing to obtain a difference frequency signal s (n); performing distance dimension FFT processing on the difference frequency signal s (n), wherein the distance dimension FFT processing is to perform FFT conversion on echoes in the same sweep frequency period, and the frequency corresponding to the peak value in the FFT conversion result corresponding to each sweep frequency period is the corresponding target distance Doppler frequency shift f _R (n) obtaining distance information of the target; and taking out time sequences of different sweep frequency periods corresponding to the distance unit where the target is located, and taking the time sequences as input time sequences for detecting the subsequent interference positions.

3. The method for resisting impact and interference of high-frequency ground wave radar based on LSTM neural network as set forth in claim 1, wherein in step S2, the detection of the interference position is performed by wavelet-matrix combination method, comprising the steps of:

4. The LSTM neural network-based high-frequency ground wave radar anti-impact interference method according to claim 1, wherein in step S3, the reconstruction recovery of the original data using the LSTM neural network comprises the steps of:

(1) Solving the complex problem by using a multi-layer perceptron of a Cell unit of the LSTM neural network, setting LSTM training learning parameters according to the characteristics of radar echo time sequence signals, and setting an environment for data reconstruction;

(2) The LSTM neural network reconstructs the interference data: taking out the real part and the imaginary part of the complex signal subjected to distance dimension FFT processing in the step S1 respectively, and performing reconstruction processing respectively; taking normal data in the front of the interference section as training data input by an LSTM network, and giving an index [ start, end ] of the data to be reconstructed in an original signal, wherein the LSTM network can reconstruct data s (n) corresponding to the [ start, end ] index section;

(3) If a plurality of interference data segments exist, the LSTM method should follow the time sequence of the signal when reconstructing the data, and reconstruct and restore from the front end to the rear end of the echo signal until the reconstruction of the last segment of interference data is completed; after the real part and the imaginary part signals are respectively reconstructed and restored, the real part and the imaginary part signals are added to obtain a final reconstructed signal.

5. The LSTM neural network-based high-frequency ground wave radar anti-collision interference method according to claim 1, wherein in step S4, performing a velocity-dimensional FFT process on the reconstructed data includes the steps of:

after interference suppression processing is completed by utilizing an LSTM network, performing speed-dimensional FFT (fast Fourier transform) on the obtained time sequence to obtain a speed-dimensional pulse compression result and obtain target speed Doppler information f _dv (n); and finishing the two-dimensional FFT processing of the signal.

6. The LSTM neural network-based high-frequency ground wave radar anti-collision interference method according to claim 1, wherein in step S5, performing constant false alarm detection on the target in the velocity dimension includes:

let k be the power value of the unit to be detected, [ k ] ₁ …k _j ]And [ k ] _j+1 …k _N ]The power of the reference units on the left side and the right side of the unit to be detected is respectively, and N is the total number of the reference units; the noise power estimate y is calculated using the following equation:

the threshold factor T can be expressed as:

Threshold＝y*T

7. A system for implementing the LSTM neural network-based high frequency ground wave radar anti-impact interference method of any one of claims 1 to 6, wherein the LSTM neural network-based high frequency ground wave radar anti-impact interference system comprises:

the distance dimension FFT processing module (1) is used for mixing the radar echo signal with the emission reference signal to obtain a difference frequency signal s (n); performing distance dimension FFT processing on the difference frequency signal s (n) to obtain distance unit information of the target; taking out the distance unit R where the target is located _n Corresponding time series of each sweep frequency period

An interference position detection module (2) for time series obtained by wavelet-matrix combination methodDetecting the interference position to obtain a vector r= [ start, end ] of an interval where the interference is located]；

The LSTM regression network module (3) is used for constructing an LSTM regression network;

a real part and an imaginary part reconstruction processing module (4) for respectively taking Sequencing outThe real part and the imaginary part are respectively subjected to reconstruction processing; taking normal echo data of the front end of interfered real part data (imaginary part data) and an interference position vector r as inputs of an LSTM neural network, and outputting the LSTM neural network as reconstructed data;

the LSTM reconstruction signal acquisition module (5) is used for adding the real part data and the imaginary part data after the real part data and the imaginary part data are respectively processed to obtain a reconstructed complex signal; assigning the complex data to interference position data corresponding to the original input signal to obtain a final LSTM reconstruction signal;

each interference data segment is processed by a processing module (6) and is used for sequentially performing distance dimension FFT processing, interference position detection, LSTM regression network construction and real part and imaginary part reconstruction processing on each interference data segment if a plurality of interference positions exist; the data after the front section reconstruction can also be used as input training data during the rear section reconstruction;

a speed-dimensional FFT processing module (7) for reconstructing the LSTM reconstruction signal s _LSTM (n) performing a velocity-dimensional FFT processing to obtain a distance unit R where the target is located _n A corresponding velocity doppler spectrum;

the CA-CFAR detection module (8) is used for carrying out CA-CFAR detection on the obtained velocity Doppler spectrum, extracting velocity information of the target and obtaining distance and velocity information of the target;

And the target distance and speed information acquisition module (9) is used for sequentially carrying out the data processing on the sequences of different sweep frequency periods corresponding to the distance units where the targets are positioned if the radar echo contains information of a plurality of targets so as to obtain the distance and speed information of the targets.

8. A computer device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the LSTM neural network-based high frequency ground wave radar anti-jamming method of any one of claims 1 to 6.

9. Use of the LSTM neural network-based high-frequency ground wave radar anti-impact interference method according to any one of claims 1 to 6 in lightning interference, meteor trail interference, radio frequency interference, impulse noise anti-impact interference.