CN117310641A - Sea clutter suppression method combining phase processing mechanism and LSTM network model - Google Patents
Sea clutter suppression method combining phase processing mechanism and LSTM network model Download PDFInfo
- Publication number
- CN117310641A CN117310641A CN202311237415.6A CN202311237415A CN117310641A CN 117310641 A CN117310641 A CN 117310641A CN 202311237415 A CN202311237415 A CN 202311237415A CN 117310641 A CN117310641 A CN 117310641A
- Authority
- CN
- China
- Prior art keywords
- sea clutter
- distance
- time
- azimuth
- sampling point
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 47
- 230000001629 suppression Effects 0.000 title claims abstract description 44
- 238000012545 processing Methods 0.000 title claims abstract description 42
- 230000007246 mechanism Effects 0.000 title claims abstract description 28
- 238000012549 training Methods 0.000 claims abstract description 31
- 238000012795 verification Methods 0.000 claims abstract description 21
- 238000005311 autocorrelation function Methods 0.000 claims abstract description 12
- 230000006835 compression Effects 0.000 claims abstract description 12
- 238000007906 compression Methods 0.000 claims abstract description 12
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 11
- 238000010606 normalization Methods 0.000 claims abstract description 5
- 238000005259 measurement Methods 0.000 claims abstract description 4
- 238000005070 sampling Methods 0.000 claims description 71
- 239000013598 vector Substances 0.000 claims description 23
- 238000001208 nuclear magnetic resonance pulse sequence Methods 0.000 claims description 20
- 238000010276 construction Methods 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 4
- 230000008859 change Effects 0.000 claims description 4
- 238000010200 validation analysis Methods 0.000 claims description 4
- 239000011159 matrix material Substances 0.000 claims description 2
- 238000003672 processing method Methods 0.000 claims description 2
- 238000001514 detection method Methods 0.000 abstract description 5
- 238000013528 artificial neural network Methods 0.000 abstract description 4
- 238000013507 mapping Methods 0.000 abstract description 3
- 238000003384 imaging method Methods 0.000 description 14
- 230000006870 function Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 5
- 238000003062 neural network model Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000000354 decomposition reaction Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000010365 information processing Effects 0.000 description 2
- 230000007787 long-term memory Effects 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 238000011426 transformation method Methods 0.000 description 2
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000739 chaotic effect Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000002592 echocardiography Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000007667 floating Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000010287 polarization Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
- G01S7/414—Discriminating targets with respect to background clutter
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
- G01S7/417—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section involving the use of neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
- G06F18/15—Statistical pre-processing, e.g. techniques for normalisation or restoring missing data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biomedical Technology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Software Systems (AREA)
- Remote Sensing (AREA)
- Evolutionary Biology (AREA)
- Radar, Positioning & Navigation (AREA)
- Computer Networks & Wireless Communication (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Probability & Statistics with Applications (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a sea clutter suppression method combining a phase processing mechanism and an LSTM network model, wherein an airborne multifunctional offshore surveillance radar system acquires actual measurement radar echo data, and performs normalization pretreatment on distance units after distance pulse compression; selecting a distance unit with larger sea clutter intensity and without interference of a target signal, and acquiring a sea clutter one-dimensional time sequence; estimating the delay time of the sea clutter time sequence by using an autocorrelation function method, and estimating the embedding dimension of the sea clutter time sequence by using a Cao algorithm; constructing a training data set by using the one-dimensional sea clutter time sequence according to the estimated delay time and the embedding dimension, and constructing a verification data set by using the time sequence of the rest distance units; designing an LSTM network model based on a phase processing mechanism; inputting the training data set into a network, obtaining a sea clutter prediction model by the training network, and inputting the training data set into a verification data set to obtain sea clutter prediction results in the rest distance units; and (5) canceling the predicted sea clutter and the original echo to realize sea clutter suppression. According to the invention, the chaos characteristic of the sea clutter and the nonlinear mapping characteristic of the neural network are utilized, and the LSTM network model combined with the phase processing mechanism is adopted to predict and inhibit the sea clutter, so that the target detection and identification under the sea clutter background can be improved.
Description
Technical Field
The present invention relates to a method for suppressing sea clutter, and more particularly, to a method for suppressing sea clutter by combining a phase processing mechanism and an LSTM (Long Short-Term Memory) network model. The method can improve the sea surface target detection performance of the radar, improve the target imaging quality and the like.
Background
Radar operates in a marine environment and the received backscatter signals from the sea surface are often referred to as sea clutter. Sea clutter often severely limits the radar's ability to detect sea surface targets. Sea surface targets are complex in type, such as ships, navigation buoys, plane missiles above the sea surface, and the like, and all have different radar scattering characteristics and kinematic characteristics. Meanwhile, under the complex ocean backgrounds such as islands, island reefs and floating ice, sea surface targets can show low observability to different degrees, so that the difficulty of radar on sea surface target detection is increased. In addition, under the middle and high sea conditions of high wind speed and large sea wave, the scattering of sea clutter is enhanced, sea surface target signals are easily submerged under strong sea clutter, and a large number of spike signals similar to sea surface targets exist in sea surface echo, so that modeling of the clutter becomes extremely complex, and the detection and monitoring performance of the radar on the sea surface targets are seriously affected. The radar sea clutter suppression has important research significance because the radar sea clutter is suppressed to improve the signal clutter ratio and the sea surface target detection performance.
Volume 8, 3, month 6, 2019, journal of radar school, discloses a system design and key technical research for on-board multi-functional marine surveillance radar, authors Jiang Qian, wu, wang Yanning. The airborne multifunctional offshore surveillance radar comprises an airborne radar device and display control/information processing software (hereinafter shown as figure 1A), wherein the airborne radar device is installed on an aircraft platform and used as an airborne task device to execute a reconnaissance task under the support of a data transmission subsystem and the like; the display control/processing software is loaded into the task control station to monitor and control the state of the airborne radar, receive the reconnaissance data sent in real time through the data transmission analysis chart, form preliminary situation display, and send the processed reconnaissance information to the rear end for information processing.
The current sea clutter suppression method mainly comprises a cyclic cancellation method, a wavelet transformation method, a subspace decomposition method and a neural network model-based method. The circulation cancellation method is based on actually measured sea clutter, different sinusoidal signals are combined to approach to actually measured sea clutter signals, parameter estimation is carried out on the sinusoidal signals according to actually measured sea clutter, the signals combined by the sinusoidal signals are used as sea clutter background signals, and cancellation processing is carried out on the signals and original echoes to inhibit the sea clutter. The wavelet transformation method adopts corresponding threshold processing according to different characteristics of sea clutter signals and target signals in a wavelet domain to realize the suppression of the sea clutter, and the method does not depend on the statistical characteristics of the clutter, and has good adaptability to complex sea clutter. The subspace decomposition method separates out the sea clutter in the radar echo according to the aggregation characteristic of the sea clutter in the subspace to realize the suppression of the sea clutter. The neural network model-based method is proposed based on the chaotic prediction characteristic of sea clutter. The inherent chaos dynamic change law of the sea clutter can be learned, so that the method can be used for predicting and suppressing the sea clutter. In recent years, with the improvement of computer performance, deep learning technology has been rapidly developed in various industries, and the capability of approaching complex nonlinear relations by utilizing the high fault tolerance capability and any precision of a neural network can be used for predicting sea clutter under different sea conditions.
At present, research on predicting a sea clutter time signal based on a neural network model is mainly focused on amplitude prediction of sea clutter, but predicted sea clutter amplitude is difficult to use for suppression of sea clutter, because the sea clutter time signal is a complex signal, a result obtained by cancellation by using a sea clutter amplitude prediction result is a real signal, and subsequent imaging correction and the like are difficult to carry out. And the real part and the imaginary part of the sea clutter time sequence are separated, the real part and the imaginary part are respectively input into a neural network model to obtain corresponding output prediction results, the prediction results of the real part and the imaginary part are recombined to obtain output complex prediction results, and in this way, the phase information between the real part and the imaginary part is ignored.
Disclosure of Invention
The invention provides a sea clutter suppression method combining a phase processing mechanism and an LSTM network model, which utilizes the chaos characteristic of sea clutter and the nonlinear mapping characteristic of the LSTM network model; grasping a determined chaos generation rule from a seemingly random sea clutter time sequence through an established LSTM network model based on a phase processing mechanism so as to realize sea clutter prediction; then, the phase characteristic of the sea clutter is utilized to improve the sea clutter prediction precision, and the defect that the phase characteristic information is ignored in the existing sea clutter prediction method is overcome; finally, the sea clutter prediction result and the original echo (radar echo data SAR) in ) The cancellation realizes sea clutter suppression. According to the invention, the LSTM network model is adopted, so that the phase information can be effectively reserved, the sea clutter prediction result can be improved, and the suppression effect of the sea clutter can be improved.
As shown in FIG. 1, a sea clutter suppression method combining a phase processing mechanism and an LSTM network model of the present invention is embedded in the data terminal unit of FIG. 1A. Data terminal unit in an airborne multi-functional offshore surveillance radar system for receiving airborne data terminal transmissionsIs denoted as SAR in . In the present invention, a part of radar echo data SAR in For constructing training data sets, denoted MSAR Training Another part of the radar echo data SAR in For constructing a validation dataset, denoted MSAR Verification 。
As shown in FIG. 1, the method for suppressing sea clutter by combining a phase processing mechanism and an LSTM network model specifically comprises the steps of constructing an LSTM sea clutter prediction model and completing sea clutter real-time processing based on the LSTM sea clutter prediction model.
The first part comprises the following steps in the process of constructing the LSTM sea clutter prediction model:
firstly, performing distance pulse compression processing on actually measured radar echo data;
in the data terminal unit, firstly, the measured radar echo data SAR is stored in Then, for said radar echo data SAR in The distance pulse compression is carried out, two-dimensional data formed after the compression is recorded as radar echo-distance pulse compression data D, and the data are expressed as a matrix:
N a indicating the number of azimuth pulses.
N f Representing the number of distance samples.
a 1,1 The 1 st distance at the 1 st azimuth time point is shown to be the sampling point.
a 1,2 The 2 nd distance at the 1 st azimuth time point is shown to be the sampling point.
Represents the N at the 1 st azimuth time point acquired f The distances are towards the sampling points.
a 2,1 Representing the 2 nd acquired azimuthThe 1 st distance at the moment is to the sampling point.
a 2,2 The 2 nd distance at the 2 nd azimuth time point is acquired to the sampling point.
Represents the N < th > at the 2 < nd > azimuth time acquired f The distances are towards the sampling points.
Representing the N of the collection a The 1 st distance at each azimuth time points to the sampling point.
Representing the N of the collection a The 2 nd distance at each azimuth time points to the sampling point.
Representing the N of the collection a Nth at each azimuth time f The distances are towards the sampling points.
In the invention, the distance unit normalization b is adopted to compress the echo in the distance direction k The processing method comprises the following steps:
k denotes the identification number of the distance to the sampling point, satisfying k=1, 2, …, N f 。
a k Representing a one-dimensional sea clutter pulse sequence within a kth range bin
a 1,k The kth distance at the 1 st azimuth time point is shown to be the sampling point.
a 2,k The kth distance at the 2 nd azimuth time point is acquired to the sampling point.
Representing the N of the collection a The kth distance at each azimuth time points to the sampling point.
abs(a k ) Representation pair a k Taking the modulus value from all sampling points in the system.
max(abs(a k ) A) represents a k The maximum value after the modulus is taken from all the sampling points in the system.
In the present invention, the radar echo-distance pulse compressed data D is normalized so that the radar echo data SAR in The amplitude of the sampled amplitudes is concentrated between-1 and 1.
A second construction step, selecting a distance unit which has larger sea clutter intensity and is not interfered by a target signal as a selected one-dimensional-sea clutter pulse sequence;
in the invention, a distance unit with larger sea clutter intensity and no interference of target signals is selected from the radar echo-distance pulse compressed data D, and if the selected distance unit is the s-th distance unit, the selected distance unit is recorded as a selected one-dimensional-sea clutter pulse sequence SS s The method comprises the steps of carrying out a first treatment on the surface of the The SS is provided with s Is expressed in the form of a sequence:
a 1,s the s-th distance at the 1 st azimuth time point is shown to be the sampling point.
a 2,s The s-th distance at the 2 nd azimuth time point is shown to be the sampling point.
Representing the N of the collection a The s-th distance at each azimuth time point is directed to the sampling point.
Thirdly, estimating the delay time of the sea clutter time sequence by using an autocorrelation function method;
in the present invention, the autocorrelation function R (τ) for estimating the sea clutter time series delay time is defined as:
τ represents the delay time.
N represents the identification number of the azimuth sampling point, satisfying n=1, 2,.. a 。
a n,s And the s-th distance under the acquired n-th azimuth time point is indicated to the sampling point.
a n+τ,s The s-th distance at the time of the n+τ azimuth is shown to be the sampling point.
Representation a n,s Is a complex conjugate of (a) and (b).
In the present invention, the delay time τ is the value of n corresponding to the time when the autocorrelation function R (τ) falls to 1e of the initial value.
In the present invention, the selected one-dimensional-sea clutter pulse sequence SS is adopted in the formula (4) s Calculating delay Time to obtain estimated delay Time Time_SS s 。
A fourth step of constructing, namely estimating the embedding dimension of the sea clutter time sequence by using a Cao algorithm;
in the present invention, the Cao algorithm for estimating the embedding dimension of the sea clutter time sequence is expressed as follows:
1) Defining a time delay vector:
A i,s,m ={a i,s ,a i+τ,s ,…,a i+(m-1)τ,s } (5)
τ represents the delay time.
m represents the embedding dimension that needs to be estimated.
A i,s,m Representing the ith time delay vector constructed from the embedding dimension m.
a i,s And the s-th distance under the acquired i-th azimuth time is indicated to the sampling point.
a i+τ,s The s-th distance at the time of the i+τ azimuth is shown to be the sampling point.
a i+(m-1)τ,s The s-th distance at s azimuth moments is shown to the sampling point at the acquired i+ (m-1) th tau.
2) Function x (i, m) determining embedding dimension m:
A i,s,m representing the ith time delay vector constructed from the embedding dimension m.
A i,s,m+1 Representing the ith time delay vector constructed from the next embedded dimension m + 1.
A o(i,m),s,m Represents the (i, m) th time delay vector constructed from the embedding dimension m.
A o(i,m),s,m+1 Represents the (i, m) th time delay vector constructed from the next embedding dimension m + 1.
Wherein, is a Euclidean distance calculation formula, and the formula is defined as:
A p,s,m representing the p-th time delay vector constructed from the embedding dimension m. According to formula (5), A is p,s,m ={a p,s ,a p+τ,s ,…,a p+(m-1)τ,s },a p,s Represents the s-th distance to the sampling point, a at the p-th azimuth time point p+τ,s The s-th distance direction sampling point, a under the time of the p+τ -th azimuth is acquired p+(m-1)τ,s The s-th distance at the time of the acquired p < + > (m-1) th tau azimuth is indicated to the sampling point.
A q,s,m Representing the q-th time delay built according to the embedding dimension mA delay vector. According to formula (5), A is q,s,m ={a q,s ,a q+τ,s ,…,a q+(m-1)τ,s },a q,s Represents the s-th distance to the sampling point, a at the q-th azimuth time point q+τ,s The s-th distance direction sampling point, a under the q+τ azimuth time point q+(m-1)τ,s The s-th distance at the acquired q+ (m-1) th tau azimuth time is indicated to the sampling point.
a p+wτ,s Representing a time delay vector A p,s,m W is an integer and the value range of w is more than or equal to 0 and less than or equal to m-1.
a q+wτ,s Representing a time delay vector A q,s,m W is an integer and the value range of w is more than or equal to 0 and less than or equal to m-1.
3) Define the mean value of x (i, m) as:
the value of E (m) is determined by the time delay τ and the embedding dimension m.
4) Define the function used to study the change in E (m) from m to m+1:
in the present invention, when m increases to a certain value m 0 When m is increased again, the function value maintains E σ (m 0 ) Almost no longer changes in the vicinity, m 0 +1 is the estimated optimal embedding dimension.
Constructing a training data set by using the selected one-dimensional-sea clutter pulse sequence according to the estimated delay time and the optimal embedding dimension;
in the present invention, the training data set MSAR Training The construction method of (A) adopts estimated delay Time Time_SS s And the optimal embedding dimension m 0 +1 selected one-dimensional-sea clutter pulse sequence SS s Divided into short sequences of shorter length(B) Short sequence +.>In the input LSTM network model, the sea clutter time sequence length of the input LSTM network meets the condition that L is more than or equal to mτ, and if the sea clutter time sequence length of a single distance unit is N a At most N can be constructed a Mτ short sequences of length mτ. During construction, each short sequence +.>The latter sampling point value of (2) is used as the corresponding label data, so that the training set number which can be constructed by a single distance unit is at most N a -mτ。
In the present invention, the LSTM network model refers to the contents of pages 248-250 of deep learning by the authors of the people's post and telecommunications publishers, ehn. Goldfeilow. The LSTM (convolutional long short-term memory network, long-term memory network) adds a long-term memory unit on the basis of gating RNN, controls the flow of time sequence information through forgetting gates, input gates and output gates, updates the weight of the network through back propagation, and learns long-term dependency. The RNN neural network using the tanh activation function and the Relu activation function has too short a learning step size and is unstable, and the LSTM neural network can stably learn a multi-step timing relationship.
Constructing a step six, namely realizing a phase processing mechanism by using a pair of real LSTM networks to obtain an LSTM sea clutter prediction model;
in the invention, a phase processing mechanism is realized by a pair of real LSTM networks, and a sea clutter prediction model of the LSTM network combined with the phase processing mechanism is obtained:
f represents LSTM network model, f 1 And f 2 LSTM network representing a pair of real numbers, Z representing complexThe number vectors, A and B, are the real and imaginary parts of the complex vector Z, which are input into the two real network models respectively, and then the output result of the LSTM network model combined with the phase processing mechanism is calculated in a mode similar to a complex multiplication algorithm:
the second part, based on LSTM sea clutter prediction model, completes the sea clutter real-time processing process, including the following steps:
step A, performing distance pulse compression processing on actually measured radar echo data;
in the invention, an airborne multifunctional offshore surveillance radar system acquires actual measurement radar echo data SAR in The radar echo data SAR in The two-dimensional data formed after the distance pulse compression is recorded as radar echo-distance pulse compressed data
In the present invention, the radar echo-distance pulse compressed data D is normalized so that the radar echo data SAR in The amplitude of the sampled amplitudes is concentrated between-1 and 1.
Step B, normalizing the distance units one by one;
in the invention, radar echo-distance pulse compressed data D is processed according to a third construction step and a fourth construction step to obtain a MSAR used for constructing a verification data set Verification 。
In the present invention, the training data set MSAR Training MSAR with verification dataset Verification Is distinguished in that: the verification data set MSAR Verification The radar echo data used do not contain the selected one-dimensional-sea clutter pulse sequence SS in construction step two s 。
Step C, operating an LSTM sea clutter prediction model;
in the present inventionIn the clear, the data set MSAR will be validated Verification Input into LSTM sea clutter prediction model viaAfter the processing, the sea clutter predicted value +.>The method comprisesIs characterized by the sequence:
f is LSTM sea clutter prediction model, nonlinear mapping function.
v represents the distance to sampling point identification number, and the range of the value is 1,2, … and N f (v≠s)。
{a j,v ,a j+1,v ,…,a j+(mτ-1),v And represents the partial data volume in the validation dataset in each input model.
a j,v The v-th distance at the j-th azimuth time of the acquisition is represented to the sampling point.
a j+1,v The v-th distance at the j+1th azimuth time point of the acquisition is represented to the sampling point.
a j+(mτ-1),v The v-th distance at the j+ (mτ -1) -th azimuth time point of the acquisition is represented to the sampling point.
In the present invention, the valueIs the output of the LSTM sea clutter prediction model.
Step D, the sea clutter predicted value and the original distance compressed echo are canceled to realize the suppression of the sea clutter;
in the present invention, sea clutter prediction valueCompressed echo +.>The cancellation is performed, and the prediction result of the sea clutter is also complex value, so that complex difference operation is performed in the cancellation process, and if the target signal exists in the original distance compressed echo, the signal remained in the result after the sea clutter signal cancellation is suppressed is the target signal.
The LSTM sea clutter prediction model constructed by the invention can be singly embedded in a data terminal unit of an airborne multifunctional offshore surveillance radar system, and can also be embedded in processing equipment such as a computer, digital processing equipment and the like which need radar echo data simulation.
Compared with the prior art, the sea clutter suppression method combining the phase processing mechanism and the LSTM network model has the advantages that:
(1) The phase characteristic information of the sea clutter pulse sequence is utilized to provide more effective information describing the sea clutter for the LSTM network model, which is beneficial to training of the LSTM sea clutter prediction model and improving the sea clutter prediction precision by applying the LSTM sea clutter prediction model.
(2) The existing majority of sea clutter prediction methods are used for directly predicting sea clutter amplitude values and discarding phase information, or are used for firstly carrying out real-virtual separation on a sea clutter pulse sequence and then inputting the sea clutter pulse sequence into a network model for training to obtain a sea clutter prediction model aiming at a real part and an imaginary part, and outputting a prediction result and recombining the sea clutter prediction result into complex data.
(3) Compared with the sea clutter prediction network model obtained by complex data training, the sea clutter prediction network model obtained by the complex data training method has higher sea clutter prediction precision, and further can improve the suppression effect of the sea clutter in the LSTM sea clutter prediction model.
(4) Because the phase characteristic information of the sea clutter time pulse sequence and the phase characteristic of the target signal have obvious differences, when the target signal and the sea clutter signal exist in the echo data at a distance, the prediction error of the sea clutter prediction model can be increased, and the target can be detected from the increased prediction error through a reasonably designed error threshold.
Drawings
Fig. 1 is a functional block diagram of a sea clutter suppression method combining a phase processing mechanism with an LSTM network model.
FIG. 1A is a schematic diagram of the components of an on-board multi-functional offshore surveillance radar system.
Fig. 2 is an autocorrelation function method estimated time delay (echo 1).
Fig. 3 is an autocorrelation function method estimated time delay (echo 2).
Fig. 4 is an autocorrelation function method estimated time delay (echo 3).
Fig. 5 is a Cao algorithm estimate embedding dimension (echo 1).
Fig. 6 is a Cao algorithm estimating the embedding dimension (echo 2).
Fig. 7 is a Cao algorithm estimating the embedding dimension (echo 3).
Fig. 8 is an LSTM network model built based on a phase processing mechanism.
FIG. 9 is an echo 1 sea clutter suppression result, with subgraphs (a), (b), (c), and (d) representing distance compressed echo 1, distance compressed echo 1 imaging result, sea clutter suppression result, sea clutter suppressed imaging result, respectively; fig. 9A is a cross-sectional view before and after sea clutter suppression (cross-sectional view at 4000 pulses per range bin).
FIG. 10 is an echo 2 sea clutter suppression result, with subgraphs (a), (b), (c), and (d) representing distance compressed echo 2, distance compressed echo 2 imaging result, sea clutter suppression result, sea clutter suppressed imaging result, respectively; fig. 10A is a cross-sectional view before and after sea clutter suppression (cross-sectional view at the 43 rd pulse of each range bin).
FIG. 11 is an echo 3 sea clutter suppression result, with subgraphs (a), (b), (c), and (d) representing distance compressed echo 3, distance compressed echo 3 imaging result, sea clutter suppression result, sea clutter suppressed imaging result, respectively; fig. 11A is a cross-sectional view before and after sea clutter suppression (a cross-sectional view at 69869 pulse per range bin).
Detailed Description
The present embodiment is implemented on the premise of the technical scheme of the present invention, and a detailed implementation manner and a specific operation process are provided, but the protection scope of the present invention is not limited to the following embodiments. Specific embodiments of the present invention are described in detail below with reference to the accompanying drawings:
in this embodiment, three sets of distance compressed echo data are used, the file numbers of echo 1 and echo 2 are 20210106155330_01_start and 20210106172511_02_start, the file number of echo 3 is 1931107_135603_start, the three sets of data are all actual sea measurement data, echo 1 does not contain a target signal, the target signal exists in the distance units of echo 2, 940-953, 1482-1492 and 2140-2810, the target signal exists in the 9 th distance unit, and the adjacent 8-11 distance units are also affected by the target signal.
The relevant parameters for the three sets of radar echo data are shown in the table below.
Echo 1 | Echo 2 | Echo 3 | |
Antenna polarization mode | HH | HH | VV |
Azimuth angle (°) | 42.18 | 8.01 | 170.26 |
Sea condition | 3-4 | 3-4 | 4 |
Radar frequency (GHz) | 9.3-9.5 | 9.3-9.5 | 9.39 |
Pulse repetition frequency (Hz) | 3000 | 3000 | 1000 |
Number of azimuthal pulses | 6940 | 4390 | 131072 |
Number of distance units | 4346 | 4346 | 14 |
The sea clutter suppression method combining the phase processing mechanism and the LSTM network model comprises the following specific operation processes:
1. and carrying out normalization preprocessing on the distance units of the distance-to-compressed echo signals acquired in the data terminal unit.
2. A distance echo data sequence with stronger sea clutter is selected, wherein the sea clutter time pulse sequence of the 1 st distance unit is selected for echo 1, the sea clutter time pulse sequence of the 933 rd distance unit is selected for echo 2, and the sea clutter time pulse sequence of the 1 st distance unit is selected for echo 3.
3. And calculating a corresponding autocorrelation function of the selected sea clutter time pulse signal, wherein the value of n corresponding to the autocorrelation function value which is reduced to 1-1e of the initial value is delay time. For the sea clutter time pulse signal selected by the echo 1, the value of n is 5; so the time delay τ is 5; for echo 2, a sea clutter time pulse signal is selected, and for echo 3, a sea clutter time pulse signal is selected, and the value of n is 5. The corresponding normalized autocorrelation function diagram is shown in fig. 2, 3 and 4.
4. Estimating an embedding dimension m of the selected sea clutter time sequence by using a Cao algorithm, wherein the selected sea clutter time pulse signal tends to be stable after m is more than 9 for the echo 1, so that the minimum embedding dimension value is 10; selecting a sea clutter time pulse signal for echo 2 to be stable after m is more than 13, so that the minimum embedding dimension value is 14; the sea clutter time pulse signal for echo 2 is chosen to stabilize after m is greater than 11, so the minimum embedding dimension value is 12. The corresponding curves are shown in fig. 5, 6 and 7.
5. Constructing a training data set by using the estimated time delay and the embedding dimension, and constructing a training data set by using 6890 short sequences with the length of 50 as long as 6940 of the selected sea clutter time sequence for echo 1, and constructing a corresponding verification data set by using the rest distance units which are not selected in the same way; for echo 2, since the length of the selected sea clutter time sequence is 4390, 4348 short sequences with length of 42 construct a training data set, and the rest distance units which are not selected also construct a corresponding verification data set; for echo 3, since the length of the selected sea clutter time sequence is 131072, 131012 short sequences of length 60 construct a training data set, and the remaining range units that are not selected also construct a corresponding validation data set.
6. The phase processing mechanism is implemented with a pair of real LSTM networks, resulting in a LSTM network model incorporating the phase processing mechanism, the implementation of which is shown in fig. 8.
7. And training the LSTM network model by using the constructed training data set, wherein the learning rate of the network is set to be 0.0002, the size of a batch of data quantity input into the network is 64, the loss function is a mean square error loss function, and the training round number is set to be 25. The LSTM prediction network model is input and output for single training, the input of the network is a constructed short sequence, the output of the network is a predicted value of the short sequence at the next moment, the predicted value of the network output and the label data corresponding to the input are subjected to mean square error calculation, the network is optimized towards the direction of reducing the mean square error until the network training is finished, and the network has learned the change rule of sea clutter at the moment and has sea clutter prediction capability.
8. And verifying the trained LSTM sea clutter prediction network model by using the verification data set, inputting verification data, and outputting predicted sea clutter. The sea clutter prediction result and the original echo are canceled, wherein (a) in fig. 9 is a distance compressed echo 1, (b) in fig. 9 is a distance compressed echo 1 imaging result, (c) in fig. 9 is a sea clutter suppression implementation result, (d) in fig. 9 is a sea clutter suppression post imaging result, and fig. 9A is a cross-section comparison diagram before and after sea clutter suppression; fig. 10 (a) shows a distance compressed echo 2, fig. 10 (b) shows a distance compressed echo 2 imaging result, fig. 10 (c) shows a sea clutter suppression result, fig. 10 (d) shows a sea clutter suppression post imaging result, and fig. 10A shows a cross-sectional contrast diagram before and after sea clutter suppression; fig. 11 (a) shows the distance compressed echo 3, fig. 11 (b) shows the imaging result of the distance compressed echo 3, fig. 11 (c) shows the result of sea clutter suppression, fig. 11 (d) shows the imaging result after sea clutter suppression, and fig. 11A shows a cross-sectional view before and after sea clutter suppression. In addition, through calculation, the sea clutter energy of the echo 1 after sea clutter suppression is reduced by 12.53dB, the signal-to-clutter ratio of the echo 2 is improved by 5.34dB, and the signal-to-clutter ratio of the echo 3 is improved by 12.35dB. Experimental results show that the sea clutter suppression method combining the phase processing mechanism and the LSTM network model has certain practicability.
Claims (3)
1. Sea clutter suppression method combining phase processing mechanism and LSTM network modelMethod for acquiring actual measurement radar echo data SAR by using airborne multifunctional offshore surveillance radar system in The method comprises the steps of carrying out a first treatment on the surface of the Before sea clutter suppression, an LSTM sea clutter prediction model is required to be built in a data terminal unit; the method is characterized in that: the construction of the LSTM sea clutter prediction model comprises the following steps:
firstly, performing distance pulse compression processing on actually measured radar echo data;
in the data terminal unit, firstly, the measured radar echo data SAR is stored in Then to SAR in The radar echo-distance pulse compressed data D is formed after distance pulse compression, and is expressed as a matrix form:
N a indicating the number of azimuth pulse points;
N f representing the number of distance sampling points;
a 1,1 the 1 st distance direction sampling point under the 1 st azimuth time is acquired;
a 1,2 the 2 nd distance direction sampling point under the 1 st azimuth time is acquired;
represents the N at the 1 st azimuth time point acquired f The distance direction sampling points;
a 2,1 the 1 st distance direction sampling point under the 2 nd azimuth time is acquired;
a 2,2 indicating the 2 nd distance to the sampling point at the 2 nd azimuth time;
represents the N < th > at the 2 < nd > azimuth time acquired f The distance direction sampling points;
representing the N of the collection a The 1 st distance under the azimuth time points to the sampling points;
representing the N of the collection a The 2 nd distance under the azimuth time points to the sampling point;
representing the N of the collection a Nth at each azimuth time f The distance direction sampling points;
normalization b of distance units of adopted distance compressed echo k The processing method comprises the following steps:
k denotes the identification number of the distance to the sampling point, satisfying k=1, 2, …, N f ;
a k Representing a one-dimensional sea clutter pulse sequence within a kth range bin
a 1,k Indicating the kth distance to the sampling point at the 1 st azimuth time;
a 2,k indicating the kth distance to the sampling point at the acquired 2 nd azimuth time;
representing the N of the collection a The kth distance under the azimuth time points to the sampling points;
abs(a k ) Representation pair a k Taking the modulus value from all sampling points in the model;
max(abs(a k ) A) represents a k Maximum value after taking the modulus value from all sampling points in the model;
a second construction step, selecting a distance unit which has larger sea clutter intensity and is not interfered by a target signal as a selected one-dimensional-sea clutter pulse sequence;
selecting a distance unit with larger sea clutter intensity and without interference of target signals from the D, and if the distance unit is selected as the s-th distance unit, marking the distance unit as a selected one-dimensional sea clutter pulse sequence SS s The method comprises the steps of carrying out a first treatment on the surface of the The SS is provided with s Is expressed in the form of a sequence:
a 1,s the s-th distance direction sampling point under the 1 st azimuth time point is acquired;
a 2,s the s-th distance direction sampling point under the acquired 2 nd azimuth time is represented;
representing the N of the collection a The s-th distance under the azimuth time points to the sampling points;
thirdly, estimating the delay time of the sea clutter time sequence by using an autocorrelation function method;
the autocorrelation function R (τ) for estimating the sea clutter time series delay time is defined as:
τ represents a delay time;
n represents the identification number of the azimuth sampling point, satisfying n=1, 2,.. a ;
a n,s Representing the s-th distance at the time of the acquired n-th azimuthA direction-leaving sampling point;
a n+τ,s the s-th distance direction sampling point under the time of the n+τ azimuth is shown;
representation a n,s Complex conjugate of (2);
a fourth step of constructing, namely estimating the embedding dimension of the sea clutter time sequence by using a Cao algorithm;
the Cao algorithm for estimating the embedding dimension of the sea clutter time sequence is expressed as follows:
1) Defining a time delay vector:
A i,s,m ={a i,s ,a i+τ,s ,…,a i+(m-1)τ,s } (5)
τ represents a delay time;
m represents the embedding dimension to be estimated;
A i,s,m representing an ith time delay vector constructed from the embedded dimension m;
a i,s the s-th distance direction sampling point under the acquired i-th azimuth time is represented;
a i+τ,s the s-th distance direction sampling point under the acquired i+τ azimuth time is represented;
a i+(m-1)τ,s representing the acquired i+ (m-1) tau, and the s-th distance under s azimuth moments to a sampling point;
2) Function x (i, m) determining embedding dimension m:
A i,s,m representing an ith time delay vector constructed from the embedded dimension m;
A i,s,m+1 representing an ith time delay vector constructed from the next embedded dimension m+1;
A o(i,m),s,m representing the (i, m) th time delay built from the embedding dimension mVector;
A o(i,m),s,m+1 representing the (i, m) th time delay vector constructed from the next embedding dimension m+1;
wherein I and II are a Euclidian distance calculation formula, and the formula is defined as:
A p,s,m representing the p-th time delay vector constructed according to the embedding dimension m; according to formula (5), A is p,s,m ={a p,s ,a p+τ,s ,…,a p+ ( m -1) τ,s },a p,s Represents the s-th distance to the sampling point, a at the p-th azimuth time point p+τ,s The s-th distance direction sampling point, a under the time of the p+τ -th azimuth is acquired p+(m-1)τ,s The s-th distance direction sampling point under the p < + > (m-1) th tau azimuth time point is acquired;
A q,s,m representing the q-th time delay vector constructed according to the embedding dimension m; according to formula (5), A is q,s,m ={a q,s ,a q+τ,s ,…,a q+(m-1)τ,s },a q,s Represents the s-th distance to the sampling point, a at the q-th azimuth time point q+τ,s The s-th distance direction sampling point, a under the q+τ azimuth time point q+(m-1)τ,s The s-th distance direction sampling point under the acquired q < + > (m-1) th tau azimuth time is represented;
a p+wτ,s representing a time delay vector A p,s,m W is an integer and the value range of w is more than or equal to 0 and less than or equal to m-1;
a q+wτ,s representing a time delay vector A q,s,m W is an integer and the value range of w is more than or equal to 0 and less than or equal to m-1;
3) Define the mean value of x (i, m) as:
the E (m) value is determined by the time delay τ and the embedding dimension m;
4) Define the function used to study the change in E (m) from m to m+1:
constructing a training data set by using the selected one-dimensional-sea clutter pulse sequence according to the estimated delay time and the optimal embedding dimension;
training data set MSAR Training The construction method of (A) adopts estimated delay Time Time_SS s And the optimal embedding dimension m 0 +1 selected one-dimensional-sea clutter pulse sequence SS s Divided into short sequences of shorter length(B) Short sequence +.>In the input LSTM network model, the sea clutter time sequence length of the input LSTM network meets the condition that L is more than or equal to mτ, and if the sea clutter time sequence length of a single distance unit is N a At most N can be constructed a -mτ short sequences of length mτ; during construction, each short sequence +.>The latter sampling point value of (2) is used as the corresponding label data, so that the training set number which can be constructed by a single distance unit is at most N a -mτ;
Constructing a step six, namely realizing a phase processing mechanism by using a pair of real LSTM networks to obtain an LSTM sea clutter prediction model;
realizing a phase processing mechanism by using a pair of real LSTM networks to obtain a sea clutter prediction model of the LSTM networks combined with the phase processing mechanism:
f represents LSTM network model, f 1 And f 2 An LSTM network representing a pair of real numbers, Z represents a complex vector, A and B are the real part and the imaginary part of the complex vector Z, the real part and the imaginary part are respectively input into the two real network models, and then the output result of the LSTM network model combined with a phase processing mechanism is calculated by using a complex multiplication algorithm mode:
2. the sea clutter suppression method combining the phase processing mechanism and the LSTM network model according to claim 1, wherein the sea clutter suppression method is characterized in that: the real-time processing process of the sea clutter based on the LSTM sea clutter prediction model comprises the following steps:
step A, performing distance pulse compression processing on actually measured radar echo data;
SAR for radar echo data in Performing range pulse compression to obtain radar echo-range pulse compressed data
Step B, normalizing the distance units one by one;
processing the radar echo-distance pulse compressed data D according to the third construction step and the fourth construction step to obtain the MSAR used for constructing the verification data set Verification ;
Training data set MSAR Training MSAR with verification dataset Verification Is distinguished in that: the verification data set MSAR Verification The radar echo data used do not contain the selected one-dimensional-sea clutter pulse sequence SS in construction step two s ;
Step C, operating an LSTM sea clutter prediction model;
will validate the dataset MSAR Verification Input into LSTM sea clutter prediction model viaAfter the processing, the sea clutter predicted value +.>
The saidIs characterized by the sequence:
f is LSTM sea clutter prediction model;
v represents the distance to sampling point identification number, and the range of the value is 1,2, … and N f (v≠s);
{a j,v ,a j+1,v ,…,a j+ ( mτ -1) ,v -representing the partial data volume in the validation dataset in each input model;
a j,v the v-th distance under the j-th azimuth time of the collection is represented to a sampling point;
a j+1,v the v-th distance under the j+1th azimuth time point is represented to the sampling point;
a j+(mτ-1),v the jth distance direction sampling point under the j < + > (mτ -1) th azimuth time point is shown;
step D, the sea clutter predicted value and the original distance compressed echo are canceled to realize the suppression of the sea clutter;
sea clutter prediction valueAnd->The cancellation is performed, and the prediction result of the sea clutter is also complex value, so that complex difference operation is performed in the cancellation process, and if the target signal exists in the original distance compressed echo, the signal remained in the result after the sea clutter signal cancellation is suppressed is the target signal.
3. The sea clutter suppression method combining a phase processing mechanism and an LSTM network model according to claim 1 or 2, wherein: after normalization of the radar echo-range pulse compression data D, the radar echo data SAR is set in The amplitude of the sampled amplitudes is concentrated between-1 and 1.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211183122X | 2022-09-27 | ||
CN202211183122 | 2022-09-27 |
Publications (1)
Publication Number | Publication Date |
---|---|
CN117310641A true CN117310641A (en) | 2023-12-29 |
Family
ID=89259632
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311237415.6A Pending CN117310641A (en) | 2022-09-27 | 2023-09-25 | Sea clutter suppression method combining phase processing mechanism and LSTM network model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117310641A (en) |
-
2023
- 2023-09-25 CN CN202311237415.6A patent/CN117310641A/en active Pending
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Chen et al. | False-alarm-controllable radar detection for marine target based on multi features fusion via CNNs | |
CN108229404B (en) | Radar echo signal target identification method based on deep learning | |
CN106468770B (en) | Nearly optimal radar target detection method under K Distribution Clutter plus noise | |
CN112882009B (en) | Radar micro Doppler target identification method based on amplitude and phase dual-channel network | |
CN106569193B (en) | The small targets detection in sea clutter method filtered based on anterior-posterior to income reference particle | |
Guo et al. | One-dimensional frequency-domain features for aircraft recognition from radar range profiles | |
CN113221631B (en) | Sequence pulse anti-interference target detection method based on convolutional neural network | |
Wang et al. | SAR target recognition based on probabilistic meta-learning | |
Li et al. | Radar signal recognition algorithm based on entropy theory | |
CN109324315A (en) | Space-time adaptive based on double level block sparsity handles radar clutter suppression method | |
CN113534065B (en) | Radar target micro-motion feature extraction and intelligent classification method and system | |
CN113640768B (en) | Low-resolution radar target identification method based on wavelet transformation | |
CN106772275A (en) | A kind of low-rank matrix based on Alpha Stable distritations recovers detection method | |
CN114117912A (en) | Sea clutter modeling and inhibiting method under data model dual drive | |
CN115616503A (en) | Radar interference signal type identification method based on convolutional neural network model | |
CN108387880A (en) | Detection method of small target under a kind of sea clutter background based on multiple dimensioned oriented Lyapunov indexes | |
CN113759362B (en) | Method, device, equipment and storage medium for radar target data association | |
CN112835008B (en) | High-resolution range profile target identification method based on attitude self-adaptive convolutional network | |
Zhu et al. | Radar HRRP group-target recognition based on combined methods in the backgroud of sea clutter | |
Zhang et al. | A novel method for sea clutter suppression and target detection via deep convolutional autoencoder | |
Li et al. | A bi-sru neural network based on soft attention for hrrp target recognition | |
CN117310641A (en) | Sea clutter suppression method combining phase processing mechanism and LSTM network model | |
CN113421281A (en) | Pedestrian micromotion part separation method based on segmentation theory | |
Yuan | A time-frequency feature fusion algorithm based on neural network for HRRP | |
CN111144316B (en) | Sea surface target detection method based on stack self-encoder |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |