CN114296067A - Pulse Doppler radar low-slow small target identification method based on LSTM model - Google Patents

Pulse Doppler radar low-slow small target identification method based on LSTM model Download PDF

Info

Publication number
CN114296067A
CN114296067A CN202210002980.3A CN202210002980A CN114296067A CN 114296067 A CN114296067 A CN 114296067A CN 202210002980 A CN202210002980 A CN 202210002980A CN 114296067 A CN114296067 A CN 114296067A
Authority
CN
China
Prior art keywords
coefficient
value
training
track
loss function
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
Application number
CN202210002980.3A
Other languages
Chinese (zh)
Inventor
鲁瑞莲
金敏
费德介
汪宗福
郑婷
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Chengdu Huirong Guoke Microsystem Technology Co ltd
Original Assignee
Chengdu Huirong Guoke Microsystem Technology Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Chengdu Huirong Guoke Microsystem Technology Co ltd filed Critical Chengdu Huirong Guoke Microsystem Technology Co ltd
Priority to CN202210002980.3A priority Critical patent/CN114296067A/en
Publication of CN114296067A publication Critical patent/CN114296067A/en
Pending legal-status Critical Current

Links

Images

Landscapes

  • Radar Systems Or Details Thereof (AREA)

Abstract

The invention provides a method for identifying a low-slow small target of a pulse Doppler radar based on an LSTM model, which comprises the following steps: receiving a set of low-slow subclass target tracks of the collected pulse Doppler radar, and performing splitting and normalization pretreatment on track set information; initializing a data set based on initialization parameters and the number of current training single targets to execute selection of LSTM forward propagation, and calculating a loss function value corresponding to a current coefficient; if the loss function value is larger than a preset threshold value, updating the input gate coefficient, the output gate coefficient and the forgetting gate state coefficient; if the loss function value is smaller than the threshold value, performing next training single target number training based on the current neural network weight coefficient; after finishing the training of all the training single target numbers in the current period, comparing the loss function value in the current period with the preset threshold value of the loss function stopping iteration; and verifying the identification accuracy in the verification set data based on the final state neural network parameters and outputting.

Description

Pulse Doppler radar low-slow small target identification method based on LSTM model
Technical Field
The invention belongs to the technical field of target identification of pulse Doppler radars, and particularly relates to a low-slow small target identification method of a pulse Doppler radar based on an LSTM model.
Background
The radar target identification technology is a technology for detecting a target by using a radar and analyzing acquired echo information to determine the attribute and the type of the target. I.e. from features in the echo, the target type is identified. The nature of this is an electromagnetic backscattering problem that requires inversion of target properties given the incident and scattered waves. The kind of target that the radar needs to identify covers all targets of the earth, sea, air, sky, even including terrain, weather, interference and radiation sources. The degree of target identification also has multi-level definition, except classification, identification and recognition, the method can also be extended to the identification of the target, threat assessment and the like, and the classification and identification of the target by radar, and has important military and civil values.
Artificial intelligence is a science of studying and developing intelligent theories, methods, techniques and applications for simulating, extending and expanding people. The radar target identification technology is an important application of artificial intelligence in the field of equipment, along with the development of the artificial intelligence technology, radar identification is continuously improved, and a plurality of research achievements are available in radar identification from pattern identification and machine learning to the development of rapid neural networks, transfer learning and the like in recent years. Although the radar target identification has a wide application range and is successfully applied in some layers, the radar target identification technology still does not form a complete theoretical system, and some radar target identification systems have certain limitations in function, mainly due to the diversification of target types and radar systems and the extreme complexity of environments.
The traditional radar identification technology usually adopts a statistical pattern identification theory. Pattern recognition is a discipline that mainly uses tools such as statistics, probability theory, computational geometry, machine learning, signal processing, and algorithm design to reason from perceptible data, with the central task of finding essential attributes of something. For radar target identification, firstly, stable and symbolic features of a target are extracted according to information such as motion and echo of the target tracked by a radar, and the extracted features are called as an identification feature template, and then, patterns to be identified are divided into respective pattern classes. The recognition/classification for a given pattern will face two types of tasks: supervised classification, which classifies patterns into existing classes, and unsupervised classification, which classifies patterns into unknown classes.
Feature extraction is an important link of the traditional radar identification technology, radar identification features strongly depend on prior knowledge and professional skills of people, and design of a radar target identification algorithm needs deeper target characteristics and a research background of feature extraction. In the traditional radar target identification, the fixed information of a radar sensor is received, digital signal processing is carried out to extract the characteristics of a target to be identified, the extracted characteristics are classified by utilizing the existing characteristic template, and the target is identified according to the membership degree. The traditional target recognition has the main problems that the traditional target recognition works according to a preset recognition mode, the capacity of automatically changing the recognition mode along with the change of a target and an environment is not provided, when the environment changes, an ideal effect is difficult to obtain only by passive feature extraction and classification, and the adaptability to the target and the environment is insufficient.
In order to meet the current, especially future, combat requirements, the identification technology must be further innovated and developed to continuously improve the identification mode and the identification performance so as to adapt to the increasingly complex combat environment.
Disclosure of Invention
In order to solve the technical problem, the invention provides a method for identifying a low-slow small target of a pulse Doppler radar based on an LSTM model, which comprises the following steps:
step 1, receiving a set of low-speed subclass target tracks of an acquired pulse Doppler radar, and performing splitting and normalization pretreatment on track set information; dividing the normalized flight path into a training set and a verification set according to a preset proportion;
step 2, initializing the number of input nodes, the number of neural network layers, the number of training cycles, the number of single training targets, the single iteration weight adjustment proportion, the threshold value of the stop iteration loss function and the output target type; initializing a starting moment input gate, an output gate, a forgetting gate state coefficient, a cell coefficient and a bias value; initializing a hidden layer cell state value and a hidden layer cell value at the starting moment;
step 3, selecting LSTM forward propagation based on the initialization parameters in the step 2 and the data set of the current training single target number, and calculating a loss function value corresponding to a current coefficient;
step 4 compares the loss function value in step 3 with a preset threshold value for stopping the iterative loss function,
if the loss function value is larger than a preset threshold value, updating the input gate coefficient, the output gate coefficient and the forgetting gate state coefficient; if the loss function value is smaller than the threshold value, performing next training single target number training based on the current neural network weight coefficient;
step 5, after finishing the training of the number of the single targets in the current period through the steps 3 and 4, comparing the loss function value of the current period with a preset threshold value of the iteration-stopping loss function;
if the current period loss function value is larger than the preset threshold value, executing the next period training;
if the loss function value of the current period is smaller than the preset threshold value, stopping training and outputting all parameters at the current moment as final neural network parameters;
and 6, verifying the data verification identification accuracy in the verification set based on the neural network parameters in the final state in the step 5 and outputting.
Further, step 1 includes the following substeps:
step 1.1, setting a low-slow subclass target track set of the acquired pulse Doppler radar to be expressed as
Figure BDA0003454141930000031
n＝1,...,N,
ln＝1,...,Ln
Wherein
Figure BDA0003454141930000032
Representing the l in the n track in the track setnThe distance between the point traces is determined,
Figure BDA0003454141930000033
representing the l in the n track in the track setnThe azimuth of the point trace is determined,
Figure BDA0003454141930000034
representing the l in the n track in the track setnThe pitch angle of the point trace is,
Figure BDA0003454141930000041
representing the l in the n track in the track setnThe scattering sectional area RCS of the point trace radar, N represents the number of tracks, LnRepresenting the number of trace points in the nth track;
step 1.2, adding a track label according to track acquisition type prior information aiming at the target track set;
step 1.3, the target track set normalizes the track information according to the following formula:
Figure BDA0003454141930000042
n＝1,...,N,
ln＝1,...,Ln
where Σ · denotes a summation operation;
step 1.4, dividing the normalized flight path into training sets T according to a preset proportionnAnd verification set Vn。
Further, wherein N is 12338; marking the unmanned aerial vehicle track as 1 and marking the non-unmanned aerial vehicle track as 0 in the track label; the training set proportion is 70%, and the validation set proportion is 30%.
Further, in step 2, 256 node numbers are input, the initial value of the training period is 1000 times, the number of the training single targets is 500, the single iteration weight ratio ρ is 1%, and the threshold value Tr of the stop iteration loss function is 10-6。
Further, step 3 comprises the sub-steps of:
step 3.1, dividing the data set into data sets according to the normalized training set obtained in step 1 and the size of the training single target number minipatch in step 2
NbN/minipatch lots, wherein each lot is used as a basic operation unit in the following steps;
and 3.2, taking 1 batch as an operation unit to perform the following operations: calculating the output of the input gate, the forgetting gate and the output gate according to the initialization coefficient in the step 2 and the following formula;
Figure BDA0003454141930000051
Figure BDA0003454141930000052
Figure BDA0003454141930000053
where σ (x) denotes the sigmoid activation function:
Figure BDA0003454141930000054
step 3.3, updating the cell state x and the hidden layer value h according to the result of the step 3.2 by combining the following formula:
Figure BDA0003454141930000055
x＝x0□Fg+Ig□G
h＝Og□tanh(x)
where □ denotes the element dot product, tanh (x) denotes the activation function:
Figure BDA0003454141930000056
step 3.4, calculating a classification output value corresponding to the current weight coefficient according to the result of the step 3.3 by combining the following formula;
Figure BDA0003454141930000057
wherein
Figure BDA0003454141930000058
Denotes the lnThe flight paths respectively belong to the probabilities of the types of 0 and 1, and the classification output value corresponding to the current weight coefficient takes the larger value of the two probabilities to correspond to the type
Figure BDA0003454141930000059
Step 3.5, calculating coefficient corresponding loss function L according to the classification result of the step 3.4s；
Wherein the meaning of each variable is: initial start time input coefficient WIInput hidden layer coefficient WhInputting the offset value B and the gate state coefficient WIgCell coefficient WIcOffset value BIOutput gate state coefficient WOgCell coefficient WOcOffset value BOForgetting the door state coefficient WFgCell coefficient WFcOffset value BF(ii) a Initializing the initial cell state value x0Value of cell envelope h0Hidden layer output coefficient WOOffset value BO(ii) a At the starting time, the gate coefficients, the offset values, the cell states and the hidden layer values are initialized to random values in the interval (0, 1).
Further, step 4 includes updating each coefficient; the method specifically comprises the following substeps:
step 4.1, input coefficients
Figure BDA00034541419300000510
Output coefficient
Figure BDA00034541419300000511
Updating:
Figure BDA0003454141930000061
Figure BDA0003454141930000062
wherein I represents a full 1 vector;
step 4.2, updating the door state coefficient:
Figure BDA0003454141930000063
Figure BDA0003454141930000064
Figure BDA0003454141930000065
wherein,
Figure BDA0003454141930000066
indicating that the gate state coefficient update value is entered,
Figure BDA0003454141930000067
indicating an update value of the forgetting gate state coefficient,
Figure BDA0003454141930000068
representing the output gate state coefficient update value;
step 4.3, updating cell coefficients:
Figure BDA0003454141930000069
Figure BDA00034541419300000610
Figure BDA00034541419300000611
wherein,
Figure BDA00034541419300000612
indicating the input of the gated cell coefficient update values,
Figure BDA00034541419300000613
indicating an update value of the forgetting gate cell coefficient,
Figure BDA00034541419300000614
representing the output gate cell coefficient update value;
step 4.4, input hidden layer coefficient updating:
Figure BDA00034541419300000615
wherein □ represents the element dot product, and tanh (x) represents the activation function.
Further, the current time parameter set is used as a final neural network parameter; the parameter sets are:
Figure BDA00034541419300000616
further, step 6 comprises the following substeps:
step 6.1, neural network parameters W based on final stateoptClassifying the verification set data and outputting classification results corresponding to the classification output values of the input gate output, the forgetting gate output, the output gate output, the cell state, the hidden layer value and the current weight coefficient
Figure BDA0003454141930000071
Step 6.2, comparing the output classification result with the track label, and counting the recognition rate according to the following formula:
Figure BDA0003454141930000072
by adopting the method, the corresponding network parameters are output based on the long-time memory network LSTM model to the radar low-slow subclass target track characteristic learning training, and the classification function is established based on the network parameters, so that the aim of identifying and classifying the radar system low-slow subclass target in real time is fulfilled. In order to achieve the technical purpose, the invention is realized by adopting the following technical scheme.
Drawings
FIG. 1 is a general flow chart of a low-slow small target identification technique implementation of the present invention;
FIG. 2 is an iteration diagram of LSTM parameter delivery;
FIG. 3 is a diagram of a training process;
fig. 4 is a diagram of real-time recognition effect.
Detailed Description
The invention discloses a pulse Doppler radar low-slow small target identification method based on an LSTM model, which is suitable for a pulse Doppler radar.
A pulse Doppler radar low-slow small target identification method based on an LSTM model comprises the following steps
1) Data preprocessing: collecting a low-speed subclass target track set of the pulse Doppler radar, and performing splitting and normalization pretreatment on track set information; dividing the normalized flight path into a training set and a verification set according to a certain proportion;
2) initializing LSTM model training parameters: initializing the number of input nodes, the number of layers of a neural network, the number of training cycles, the number of targets (minipatch) for training a single time, adjusting the proportion of the weights for a single iteration, stopping iteration loss function threshold and outputting the target types; initializing a starting moment input gate, an output gate, a forgetting gate state coefficient, a cell coefficient and a bias value; initializing a hidden layer cell state value and a hidden layer cell value at the starting moment;
3) based on 2) carrying out LSTM forward propagation on the initialization parameter and the current minipatch data set and calculating a loss function value corresponding to the current coefficient;
4) comparing the loss function value with a threshold value of the loss function for stopping iteration based on 3), and updating each gate coefficient if the loss function value is greater than the threshold value; if the loss function value is smaller than the threshold value, performing the next minipatch training based on the current neural network weight coefficient;
5) based on 3) and 4), after finishing all minimatch training of the current period, comparing the loss function value of the current period with the threshold value of the loss function stopping iteration, and if the loss function value is greater than the threshold value, carrying out training of the next period; if the loss function is smaller than the threshold value, stopping training and outputting the current time parameter as a final neural network parameter;
6) and (5) verifying the identification accuracy in the verification set data based on the final state neural network parameters and outputting.
The following detailed description of embodiments of the invention refers to the accompanying drawings.
Aiming at the problems in the prior art, the invention aims to provide a pulse Doppler radar low-slow small target identification method based on an LSTM model. The method outputs corresponding network parameters for the radar low and slow small class target track characteristic learning training based on the long and short time memory network LSTM model, and establishes a classification function based on the network parameters, thereby achieving the purpose of real-time identification and classification of the radar system low and slow small class target. In order to achieve the technical purpose, the invention is realized by adopting the following technical scheme.
A pulse Doppler radar low-slow small target identification method based on an LSTM model comprises the following steps:
step 1, data preprocessing: collecting a low-speed subclass target track set of the pulse Doppler radar, and performing splitting and normalization pretreatment on track set information; dividing the normalized flight path into a training set and a verification set according to a certain proportion;
step 2, initializing LSTM model training parameters: initializing the number of input nodes, the number of layers of a neural network, the number of training cycles, the number of targets (minipatch) for training a single time, adjusting the proportion of the weights for a single iteration, stopping iteration loss function threshold and outputting the target types; initializing a starting moment input gate, an output gate, a forgetting gate state coefficient, a cell coefficient and a bias value; initializing a hidden layer cell state value and a hidden layer cell value at the starting moment;
step 3, based on the initialization parameters in the step 2 and the current minipatch data set, selecting LSTM forward propagation and calculating a loss function value corresponding to the current coefficient;
step 4, comparing the loss function value with the iteration-stopping loss function threshold value based on the step 3, and updating each gate coefficient if the loss function value is greater than the threshold value; if the loss function value is smaller than the threshold value, performing the next minipatch training based on the current neural network weight coefficient;
step 5, after finishing all minimatch training in the current period in the steps 3 and 4, comparing the loss function value of the current period with the threshold value of the loss function stopping iteration, and if the loss function value is larger than the threshold value, performing the training in the next period; if the loss function value is smaller than the threshold value, stopping training and outputting the current time parameter as the final neural network parameter;
and 6, verifying the data verification identification accuracy in the verification set based on the final state neural network parameters in the step 5 and outputting.
Referring to fig. 1, a general flowchart is implemented for a pulse doppler radar low-slow small target identification method based on an LSTM model according to the present invention. The pulse Doppler radar low-slow small target identification method based on the LSTM model comprises the following steps:
step 1, data preprocessing: collecting a pulse Doppler radar low-slow subclass target track set, and performing labeling (label) and normalization pretreatment on track set information; dividing the normalized flight path into a training set and a verification set according to a certain proportion;
1a) low-slow subclass target track set of acquired pulse Doppler radar
Figure BDA0003454141930000101
Wherein
Figure BDA0003454141930000102
Representing the l in the n track in the track setnThe distance between the point traces is determined,
Figure BDA0003454141930000103
representing the l in the n track in the track setnThe azimuth of the point trace is determined,
Figure BDA0003454141930000104
representing the l in the n track in the track setnThe pitch angle of the point trace is,
Figure BDA0003454141930000105
representing the l in the n track in the track setnThe scattering cross section area (RCS) of the point-trace radar, N represents the flight pathNumber, LnRepresenting the number of trace points in the nth track;
in this example, but not limited to, N12338 is selected.
1b) Adding a track label according to track acquisition type prior information based on the track set in the step 1 a);
in this example, the track labels are selected from but not limited to the unmanned aerial vehicle track (marked as 1) and the non-unmanned aerial vehicle track (marked as 0);
1c) normalizing the flight path information according to the flight path set in the step 1b) and by combining the following formula:
Figure BDA0003454141930000106
where Σ · denotes a summation operation;
1d) dividing the normalized flight path in the step 1c) into a training set T according to a certain proportionnAnd verification set Vn；
In this example, the training set is selected but not limited to 70% and the verification set is selected 30%.
Step 2, initializing LSTM model training parameters: initializing the number of neural network layers, the number of training cycles, the number of targets (minipatch) for training a single time, adjusting the proportion rho of the single iteration weight, stopping iteration loss function threshold Tr and outputting the target type; initializing an input coefficient, a hidden layer coefficient, a bias value, an input gate, an output gate, a forgetting gate state coefficient, a cell coefficient and a bias value at the starting moment; initializing a cell state value and a cell hidden layer value at the initial moment;
in this example, the number of input nodes is selected from, but not limited to, 256 neural network layers, an initial value of a training period is 1000 times, a single training target number minipatch is 500, a single iteration weight ratio ρ is 1%, and a stop iteration loss function threshold Tr is 10-6Outputting a target category 2, wherein 0 represents a non-unmanned aerial vehicle and 1 represents an unmanned aerial vehicle; initial start time input coefficient WIInput hidden layer coefficient WhInputting the offset value B and the gate state coefficient WIgCell coefficient WIcOffset value BIOutput gate state coefficientWOgCell coefficient WOcOffset value BOForgetting the door state coefficient WFgCell coefficient WFcOffset value BF(ii) a Initializing the initial cell state value x0Value of cell envelope h0Hidden layer output coefficient WOOffset value BO(ii) a At the starting time, the gate coefficients, the offset values, the cell states and the hidden layer values are initialized to random values in the interval (0, 1).
Step 3, based on the initialization parameters in the step 2 and the current minipatch data set, selecting LSTM forward propagation and calculating a loss function value corresponding to the current coefficient;
3a) dividing the data set into N according to the normalized training set obtained in the step 1 and the minimatch size in the step 2bN/minipatch batches, wherein a batch is the following basic operation unit;
3b) taking 1 batch in the step 3a) as an operation unit to perform the following operations: calculating the output of the input gate, the forgetting gate and the output gate according to the initialization coefficient in the step 2 and the following formula;
Figure BDA0003454141930000111
where σ (x) denotes the sigmoid activation function:
Figure BDA0003454141930000112
3c) updating the cell state x and the hidden layer value h according to the result of the step 3b) by combining the following formula:
Figure BDA0003454141930000121
where □ denotes the element dot product, tanh (x) denotes the activation function:
Figure BDA0003454141930000122
3d) calculating a classification output value corresponding to the current weight coefficient according to the result of the step 3c) by combining the following formula;
Figure BDA0003454141930000123
wherein
Figure BDA0003454141930000124
Denotes the lnThe flight paths respectively belong to the probabilities of the types of 0 and 1, and the classification output value corresponding to the current weight coefficient takes the larger value of the two probabilities to correspond to the type
Figure BDA0003454141930000125
3e) Calculating a loss function L from the 3d) classification resultss；
The loss function comprises cross entropy, Focal loss and the like, and in the embodiment of the invention, because the sample proportion is seriously unbalanced, the Focal loss is selected to calculate the loss function:
Figure BDA0003454141930000126
wherein log (·) is a logarithm operation, α is a balance factor, γ is a proportion for adjusting coefficient reduction of a simple sample, and α ═ 0.25 and γ ═ 2 in the present example;
step 4, based on the loss function in the step 3, comparing the loss function with a threshold value of the iteration stopping loss function, and if the loss function is larger than the threshold value, updating the coefficients of all the gates; if the loss function is smaller than the threshold value, performing the next minipatch training based on the current neural network coefficient;
4a) updating each coefficient by combining the following formula;
input coefficient
Figure BDA0003454141930000127
Output coefficient
Figure BDA0003454141930000128
Updating:
Figure BDA0003454141930000131
wherein I represents an all 1 vector. Updating the state coefficient of the door:
Figure BDA0003454141930000132
wherein
Figure BDA0003454141930000133
Indicating that the gate state coefficient update value is entered,
Figure BDA0003454141930000134
indicating an update value of the forgetting gate state coefficient,
Figure BDA0003454141930000135
representing the output gate state coefficient update value; cell coefficient updating:
Figure BDA0003454141930000136
wherein
Figure BDA0003454141930000137
Indicating the input of the gated cell coefficient update values,
Figure BDA0003454141930000138
indicating an update value of the forgetting gate cell coefficient,
Figure BDA0003454141930000139
representing the output gate cell coefficient update value; updating the input hidden layer coefficients:
Figure BDA00034541419300001310
step 5, after finishing all minipatch training in the current period in the steps 3 and 4, comparing the loss function value of the current period with the iteration-stopping loss function threshold, and if the loss function value is greater than the threshold, performing next period training; if the loss function value is smaller than the threshold value, stopping training and outputting the parameter set at the current moment as the final neural network parameter;
the parameter sets are:
Figure BDA00034541419300001311
and 6, verifying the data verification identification accuracy in the verification set based on the final state neural network parameters in the step 5 and outputting.
6a) Based on step 5 final state neural network parameters WoptAnd the formulas 3), 5) and 7) classify the verification set data and output the classification result
Figure BDA0003454141930000141
6b) Comparing the output classification result with the flight path label in the step 1b), and counting the recognition rate by combining the following formula:
Figure BDA0003454141930000142
the effect of the invention is further illustrated by the following simulation comparative tests:
1. an experimental scene is as follows:
acquiring total N (12338) tracks of unmanned aerial vehicle and clutter track data by using radar equipment in a ground clutter environment, and performing network training by using the method based on 0.7 multiplied by N training set tracks to obtain an optimal coefficient WoptAnd performing result verification on the track identification accuracy of the residual verification set based on the coefficient.
2. And (3) analyzing an experimental result:
table 1 shows the recognition accuracy for different numbers of neural network layers and different minipatches; in the parameters of the example, the number of the neural network layers is 100, and the recognition accuracy is highest when the minipatch size is 400, and can reach 93.25%.
TABLE 1
Figure BDA0003454141930000143
Figure BDA0003454141930000151
Finally, it should be noted that the above embodiments are only used for illustrating the technical solutions of the embodiments of the present invention and not for limiting, and although the embodiments of the present invention are described in detail with reference to the above preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions can be made on the technical solutions of the embodiments of the present invention without departing from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (8)

1. A method for identifying a low-slow small target of a pulse Doppler radar based on an LSTM model is characterized by comprising the following steps:
step 1, receiving a set of low-speed subclass target tracks of an acquired pulse Doppler radar, and performing splitting and normalization pretreatment on track set information; dividing the normalized flight path into a training set and a verification set according to a preset proportion;
step 2, initializing the number of input nodes, the number of neural network layers, the number of training cycles, the number of single training targets, the single iteration weight adjustment proportion, the threshold value of the stop iteration loss function and the output target type; initializing a starting moment input gate, an output gate, a forgetting gate state coefficient, a cell coefficient and a bias value; initializing a hidden layer cell state value and a hidden layer cell value at the starting moment;
step 3, selecting LSTM forward propagation based on the initialization parameters in the step 2 and the data set of the current training single target number, and calculating a loss function value corresponding to a current coefficient;
step 4 compares the loss function value in step 3 with a preset threshold value for stopping the iterative loss function,
if the loss function value is larger than a preset threshold value, updating the input gate coefficient, the output gate coefficient and the forgetting gate state coefficient; if the loss function value is smaller than the threshold value, performing next training single target number training based on the current neural network weight coefficient;
step 5, after finishing the training of the number of the single targets in the current period through the steps 3 and 4, comparing the loss function value of the current period with a preset threshold value of the iteration-stopping loss function;
if the current period loss function value is larger than the preset threshold value, executing the next period training;
if the loss function value of the current period is smaller than the preset threshold value, stopping training and outputting all parameters at the current moment as final neural network parameters;
and 6, verifying the data verification identification accuracy in the verification set based on the neural network parameters in the final state in the step 5 and outputting.
2. An identification method as claimed in claim 1, characterized in that step 1 comprises the following sub-steps:
step 1.1, setting a low-slow subclass target track set of the acquired pulse Doppler radar to be expressed as
Figure FDA0003454141920000021
n＝1,...,N,
ln＝1,...,Ln
Wherein
Figure FDA0003454141920000022
Representing the l in the n track in the track setnThe distance between the point traces is determined,
Figure FDA0003454141920000023
representing the l in the n track in the track setnTrack of points orientationThe angle of the corner is such that,
Figure FDA0003454141920000024
representing the l in the n track in the track setnThe pitch angle of the point trace is,
Figure FDA0003454141920000025
representing the l in the n track in the track setnThe scattering sectional area RCS of the point trace radar, N represents the number of tracks, LnRepresenting the number of trace points in the nth track;
step 1.2, adding a track label according to track acquisition type prior information aiming at the target track set;
step 1.3, the target track set normalizes the track information according to the following formula:
Figure FDA0003454141920000026
n＝1,...,N,
ln＝1,...,Ln
where Σ · denotes a summation operation;
step 1.4, dividing the normalized flight path into training sets T according to a preset proportionnAnd verification set Vn。
3. The identification method of claim 1, wherein N-12338; marking the unmanned aerial vehicle track as 1 and marking the non-unmanned aerial vehicle track as 0 in the track label; the training set proportion is 70%, and the validation set proportion is 30%.
4. The identification method of claim 1, wherein in step 2, the number of input nodes is 256, the initial value of the training period is 1000 times, the number of single training targets is 500, the weight ratio p of single iteration is 1%, and the threshold value Tr of the stop iteration loss function is 10 ═ 10-6。
5. An identification method as claimed in claim 1, characterized in that step 3 comprises the sub-steps of:
step 3.1, dividing the data set into N according to the normalized training set obtained in step 1 and the size of the training single target number minipatch in step 2bN/minipatch lots, wherein each lot is used as a basic operation unit in the following steps;
and 3.2, taking 1 batch as an operation unit to perform the following operations: calculating the output of the input gate, the forgetting gate and the output gate according to the initialization coefficient in the step 2 and the following formula;
Figure FDA0003454141920000031
Figure FDA0003454141920000032
Figure FDA0003454141920000033
where σ (x) denotes the sigmoid activation function:
Figure FDA0003454141920000034
step 3.3, updating the cell state x and the hidden layer value h according to the result of the step 3.2 by combining the following formula:
Figure FDA0003454141920000035
Figure FDA0003454141920000036
Figure FDA0003454141920000037
wherein
Figure FDA0003454141920000038
Represents the element dot product, tanh (x) represents the activation function:
Figure FDA0003454141920000039
step 3.4, calculating a classification output value corresponding to the current weight coefficient according to the result of the step 3.3 by combining the following formula;
Figure FDA00034541419200000310
wherein
Figure FDA00034541419200000311
Denotes the lnThe flight paths respectively belong to the probabilities of the types of 0 and 1, and the classification output value corresponding to the current weight coefficient takes the larger value of the two probabilities to correspond to the type
Figure FDA00034541419200000312
Step 3.5, calculating coefficient corresponding loss function L according to the classification result of the step 3.4s；
Wherein the meaning of each variable is: initial start time input coefficient WIInput hidden layer coefficient WhInputting the offset value B and the gate state coefficient WIgCell coefficient WIcOffset value BIOutput gate state coefficient WOgCell coefficient WOcOffset value BOForgetting the door state coefficient WFgCell coefficient WFcOffset value BF(ii) a Initializing the initial cell state value x0Value of cell envelope h0Hidden layer output coefficient WOOffset value BO(ii) a At the starting time, the gate coefficients, the offset values, the cell states and the hidden layer values are initialized to random values in the interval (0, 1).
6. An identification method as claimed in claim 1, characterized in that step 4 comprises updating the coefficients; the method specifically comprises the following substeps:
step 4.1, input coefficients
Figure FDA0003454141920000041
Output coefficient
Figure FDA0003454141920000042
Updating:
Figure FDA0003454141920000043
Figure FDA0003454141920000044
wherein I represents a full 1 vector;
step 4.2, updating the door state coefficient:
Figure FDA0003454141920000045
Figure FDA0003454141920000046
Figure FDA0003454141920000047
wherein,
Figure FDA0003454141920000048
indicating that the gate state coefficient update value is entered,
Figure FDA0003454141920000049
indicating an update value of the forgetting gate state coefficient,
Figure FDA00034541419200000410
representing the output gate state coefficient update value;
step 4.3, updating cell coefficients:
Figure FDA00034541419200000411
Figure FDA00034541419200000412
Figure FDA00034541419200000413
wherein,
Figure FDA00034541419200000414
indicating the input of the gated cell coefficient update values,
Figure FDA00034541419200000415
indicating an update value of the forgetting gate cell coefficient,
Figure FDA00034541419200000416
representing the output gate cell coefficient update value;
step 4.4, input hidden layer coefficient updating:
Figure FDA0003454141920000051
wherein,
Figure FDA0003454141920000052
representing element dot product, and tanh (x) representing activation function.
7. The identification method according to claim 6, wherein the current time parameter set is used as a final neural network parameter; the parameter sets are:
Figure FDA0003454141920000053
8. an identification method as claimed in claim 1, characterized in that step 6 comprises the following sub-steps:
step 6.1, neural network parameters W based on final stateoptClassifying the verification set data and outputting classification results corresponding to the classification output values of the input gate output, the forgetting gate output, the output gate output, the cell state, the hidden layer value and the current weight coefficient
Figure FDA0003454141920000054
Step 6.2, comparing the output classification result with the track label, and counting the recognition rate according to the following formula:
Figure FDA0003454141920000055
CN202210002980.3A 2022-01-04 2022-01-04 Pulse Doppler radar low-slow small target identification method based on LSTM model Pending CN114296067A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210002980.3A CN114296067A (en) 2022-01-04 2022-01-04 Pulse Doppler radar low-slow small target identification method based on LSTM model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210002980.3A CN114296067A (en) 2022-01-04 2022-01-04 Pulse Doppler radar low-slow small target identification method based on LSTM model

Publications (1)

Publication Number Publication Date
CN114296067A true CN114296067A (en) 2022-04-08

Family

ID=80974946

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210002980.3A Pending CN114296067A (en) 2022-01-04 2022-01-04 Pulse Doppler radar low-slow small target identification method based on LSTM model

Country Status (1)

Country Link
CN (1) CN114296067A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116520252A (en) * 2023-04-03 2023-08-01 中国人民解放军93209部队 Intelligent recognition method and system for aerial targets

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116520252A (en) * 2023-04-03 2023-08-01 中国人民解放军93209部队 Intelligent recognition method and system for aerial targets
CN116520252B (en) * 2023-04-03 2024-03-15 中国人民解放军93209部队 Intelligent recognition method and system for aerial targets

Similar Documents

Publication Publication Date Title
CN111913156B (en) Radar radiation source individual identification method based on deep learning model and feature combination
CN112132042B (en) SAR image target detection method based on contrast domain adaptation
CN107300698B (en) Radar target track starting method based on support vector machine
CN110133599B (en) Intelligent radar radiation source signal classification method based on long-time and short-time memory model
US10976429B1 (en) System and method for synthetic aperture radar target recognition utilizing spiking neuromorphic networks
CN108256436A (en) A kind of radar HRRP target identification methods based on joint classification
CN103020978A (en) SAR (synthetic aperture radar) image change detection method combining multi-threshold segmentation with fuzzy clustering
CN109753874A (en) A kind of low slow small classification of radar targets method based on machine learning
Xiao et al. Specific emitter identification of radar based on one dimensional convolution neural network
CN112052933B (en) Particle swarm optimization-based safety testing method and repairing method for deep learning model
CN110082738B (en) Radar target identification method based on Gaussian mixture and tensor recurrent neural network
CN105654136A (en) Deep learning based automatic target identification method for large-scale remote sensing images
CN111539422A (en) Flight target cooperative identification method based on fast RCNN
CN111401168A (en) Multi-layer radar feature extraction and selection method for unmanned aerial vehicle
Guo et al. A novel radar signals sorting method-based trajectory features
CN114818853B (en) Intention recognition method based on bidirectional gating circulating unit and conditional random field
CN114492583B (en) LSTM-based radar search mode identification method
CN116432514A (en) Interception intention recognition strategy simulation system and method for unmanned aerial vehicle attack and defense game
CN114296067A (en) Pulse Doppler radar low-slow small target identification method based on LSTM model
CN112990125B (en) Method for judging whether radiation source radar belongs to target platform
CN116340846A (en) Aliasing modulation signal identification method for multi-example multi-label learning under weak supervision
CN114445456B (en) Data-driven intelligent maneuvering target tracking method and device based on partial model
Zhang et al. Air target recognition algorithm based on mixed depth features in the interference environment
CN113724325B (en) Multi-scene monocular camera pose regression method based on graph convolution network
Guo et al. Deinterleaving of radar pulse based on implicit feature

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