CN113962262B - Continuous learning-based intelligent radar signal sorting method - Google Patents
Continuous learning-based intelligent radar signal sorting method Download PDFInfo
- Publication number
- CN113962262B CN113962262B CN202111226205.8A CN202111226205A CN113962262B CN 113962262 B CN113962262 B CN 113962262B CN 202111226205 A CN202111226205 A CN 202111226205A CN 113962262 B CN113962262 B CN 113962262B
- Authority
- CN
- China
- Prior art keywords
- radar signal
- data
- softmax
- sorting
- batch
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000012549 training Methods 0.000 claims abstract description 37
- 238000013136 deep learning model Methods 0.000 claims abstract description 24
- 230000006870 function Effects 0.000 claims abstract description 16
- 230000009467 reduction Effects 0.000 claims abstract description 15
- 238000002715 modification method Methods 0.000 claims abstract description 10
- 238000007477 logistic regression Methods 0.000 claims abstract description 6
- 238000012545 processing Methods 0.000 claims abstract description 5
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 239000013598 vector Substances 0.000 claims description 10
- 239000011159 matrix material Substances 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000005070 sampling Methods 0.000 claims description 5
- 238000012986 modification Methods 0.000 claims description 4
- 230000004048 modification Effects 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 230000001902 propagating effect Effects 0.000 claims description 2
- 230000008859 change Effects 0.000 abstract description 5
- 238000010586 diagram Methods 0.000 description 9
- 238000013135 deep learning Methods 0.000 description 6
- 238000012360 testing method Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 238000000605 extraction Methods 0.000 description 5
- 238000013507 mapping Methods 0.000 description 5
- 238000013145 classification model Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 4
- 230000000007 visual effect Effects 0.000 description 4
- 238000013473 artificial intelligence Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000011176 pooling Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 230000006835 compression Effects 0.000 description 2
- 238000007906 compression Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000010923 batch production Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 210000004205 output neuron Anatomy 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 230000017105 transposition Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
-
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Signal Processing (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention relates to a radar signal intelligent sorting method based on continuous learning, which comprises the following steps: acquiring radar signal data and preprocessing to obtain a training set; combining the full convolution self-encoder with the logistic regression classifier to construct a joint deep learning model, and optimizing by adopting an orthogonal weight modification method to obtain a radar signal sorting model capable of continuously learning; dividing radar signals to be sorted into IQ two paths of data, inputting the IQ two paths of data into a trained radar signal sorting model for sorting, performing data dimension reduction processing on the IQ two paths of data by the radar signal sorting model to obtain dimension reduced sparse representation, and using the dimension reduced sparse representation to a softmax function for radar signal sorting to finally output corresponding labels. The invention effectively improves the learning and training efficiency and the real-time performance of the radar signal sorting network, and can be quickly adapted to the dynamic change of the modulation format of the enemy radar signal, thereby obtaining the electromagnetic space control right in the electronic countermeasure of enemy.
Description
Technical Field
The invention belongs to the technical field of electronic countermeasure technology and artificial intelligence, and particularly relates to an intelligent radar signal sorting method based on continuous learning.
Background
Artificial intelligence is one of the current leading scientific fields, while deep learning is a module of artificial intelligence. What is "intelligent" is that it has a high adaptability to complex, dynamically changing environments. Deep neural networks (Deep NeuralNetworks, hereinafter DNN) are powerful tools for learning complex but fixed mapping rules between inputs and outputs, which remain unchanged but limit their application in more complex and dynamically changing situations. In the field of radar signal sorting and identification in the electronic countermeasure field, only limited types of enemy radar signals can be collected. When the communication modulation format of the enemy radar is continuously changed, after new radar data are continuously acquired, the traditional DNN model with good original data training must be restarted to perform learning training to identify the new radar signal modulation format, but when the data volume is large, a large amount of time is consumed by a conventional deep learning training mode, the learning training efficiency is low, the real-time performance is poor, and the implementation is complex in actual operation.
Disclosure of Invention
In order to solve the problems that in the prior art, when the communication modulation format of the enemy radar is continuously changed, the traditional deep learning model needs to restart learning training, learning training efficiency is low and real-time performance is poor, the invention provides a continuous learning-based radar signal intelligent sorting method. When a new task is learned, the continuous learning method does not need to take data features in the old task to learn again, and can also apply the learned features in the old task to the new task, so that the intelligent radar signal sorting method can show good performance in all tasks and show continuity of the intelligent radar signal sorting method.
In order to achieve the above purpose, the invention adopts the following technical scheme:
an intelligent radar signal sorting method based on continuous learning, which comprises the following steps:
Step one: acquiring radar signal data, preprocessing the radar signal data to obtain a data set, and randomly sampling from the data set to obtain a training set;
Step two: combining a full convolution self-encoder with a logistic regression classifier to construct a joint deep learning model CAE_SOFTMAX, and optimizing a full convolution self-encoder network in the joint deep learning model CAE_SOFTMAX by adopting an orthogonal weight modification method to perform continuous learning to obtain a radar signal sorting model CAE_SOFTMAX_OWM capable of being continuously learned;
Step three: training the radar signal sorting model CAE_SOFTMAX_OWM by using the training set;
step four: acquiring radar signals to be sorted, and dividing the radar signals to be sorted into IQ two paths of data;
step five: inputting the IQ two paths of data into a trained radar signal sorting model CAE_SOFTMAX_OWM for sorting, performing data dimension reduction processing on the input IQ two paths of data by using a full convolution self-encoder by the radar signal sorting model CAE_SOFTMAX_OWM, obtaining a sparse representation after dimension reduction, using the sparse representation after dimension reduction for a SOFTMAX function for radar signal sorting, and finally outputting a corresponding label.
The invention has the beneficial effects that: when the communication modulation format of the enemy radar is continuously changed, the traditional deep learning model DNN needs to restart learning training, and the radar signal sorting model CAE_SOFTMAX_OWM can adapt to the dynamically changed radar signal modulation format through continuous learning without restarting the learning, and meanwhile, the learning features in the old task can be applied to the new modulation format recognition task, so that good sorting effects can be shown in all tasks. The invention effectively improves the learning and training efficiency and the real-time performance of the radar signal sorting network, and can be quickly adapted to the dynamic change of the modulation format of the enemy radar signal, thereby obtaining the electromagnetic space control right in the electronic countermeasure of enemy.
Drawings
FIG. 1 is a diagram of a CAE network architecture of a full convolutional self-encoder of the present invention;
FIG. 2 is a diagram of the joint deep learning model CAE_SOFTMAX network architecture of the present invention;
FIG. 3 is a diagram of the network architecture of the radar signal sorting model CAE_SOFTMAX_OWM of the present invention;
FIG. 4 is a graph comparing classification accuracy of a conventional deep learning model DNN in an experiment and a radar signal classification model CAE_SOFTMAX_OWM of the present invention;
Fig. 5 is a T-SNE visual comparison diagram of the conventional deep learning model DNN and the radar signal sorting model cae_softmax_owm of the present invention in an experiment, fig. 5 (a) is a T-SNE visual diagram of the conventional deep learning model DNN, and fig. 5 (b) is a T-SNE visual diagram of the radar signal sorting model cae_softmax_owm of the present invention.
Detailed Description
The technical scheme of the present invention will be described in detail with reference to the accompanying drawings and preferred embodiments.
A conventional deep learning method (abbreviated DNN) learns complex and fixed mapping rules between inputs and outputs that can be used to sort radar signals. However, once the learning stage is finished, the mapping rule is learned, so that the mapping rule is solidified, the new mapping is difficult to learn conveniently, and flexible response to situation information such as self state, environment change, task change and the like in the actual environment cannot be made, so that the sorting requirement of complex and changeable radar signals is difficult to meet. A deep learning-based stacked self-encoder (SAE) employs implicit information in the modeling data of a multi-layer neural network. Compared with a shallow model, the deep model can better approximate a nonlinear function, and the stacked self-encoder is already applied to signal feature extraction and dimension reduction, and has the defect that the training network parameters are huge in calculation amount due to the adoption of a fully-connected network. The full convolution self-encoder (called CAE for short) is an SAE stack type self-encoder variant structure, the CAE adopts local connection, weight sharing convolution layer, pooling operation and the like to replace an SAE full connection layer, network weight parameters are greatly reduced, and the scale and the calculated amount of the network parameters are reduced, so that the CAE can be utilized to extract the characteristics of radar modulation signals to achieve the purpose of reducing the dimension.
The invention provides a radar signal intelligent sorting method based on continuous learning, which comprises the following steps:
Step one: firstly, radar signal data for training and testing a model are acquired, for example, the radar signal data can be acquired through a radar signal receiver, and the received radar signal is divided into IQ two paths of data to be stored; then preprocessing the acquired radar signal data to obtain a data set, wherein the data set comprises 24 modulation formats, each modulation format comprises 26 signal-to-noise ratios, each signal-to-noise ratio comprises 3000 pieces of data, each data comprises two paths of IQ signals, and each path of signal comprises 1024 points, so that the data set has the following size: 2555904 ×2×1024, the dataset size is up to 18.01G. And taking a single IQ sample signal in the data set as a picture to reduce the dimension, randomly sampling a training set and a testing set from the data set, wherein the training set is used for training a radar signal sorting model CAE_SOFTMAX_OWM, and carrying out experimental test by adopting the testing set after the model is trained.
Step two: and combining the full convolution self-encoder with the logistic regression classifier to construct a joint deep learning model CAE_SOFTMAX, and optimizing a full convolution self-encoder network in the joint deep learning model CAE_SOFTMAX by adopting an orthogonal weight modification method to perform continuous learning to obtain a radar signal sorting model CAE_SOFTMAX_OWM capable of being continuously learned.
Specifically, the full convolution self-encoder (abbreviated as CAE) is adopted to perform data dimension reduction on 1024×2 IQ modulation signals, the dimension reduction compression network CAE structure is shown in fig. 1, the CAE encoder (Encoder) inputs single signal vectors, then the features are extracted through operations such as downsampling and nonlinear activation of a plurality of convolution layers, and then the dimension reduction representation is output through a full connection layer. The CAE Decoder (Decoder) input is an encoder compressed representation, which is subjected to a full concatenated layer, a plurality of deconvolution upsamples, nonlinear activation, etc., to reconstruct the signal pixel vectors. The full convolution self-encoder adopts root mean square error as a cost function of CAE, and a cost function calculation formula is as follows:
Where m is the size of the batch training mini-batch, x ij represents the j-th element of the i-th input signal pixel x i∈Rn in one mini-batch, Represents the ith reconstructed signal pixel/>, in one mini-batchN is the vector x i and/>Is a length of (c).
The traditional convolution automatic encoder is removed from the pooling layer and the anti-pooling layer, up-sampling and down-sampling are realized by convolution and transposition convolution, so that the full convolution automatic encoder is formed, in addition, the activation function of the coding output layer of the full convolution automatic encoder is replaced by a leakage ReLU, and the output is normalized to the [ -1,1] interval. The full convolution self-encoder sequentially comprises an input layer, five convolution layers, a folding layer, two full connection layers, an unfolding layer and two transposed convolution layers, wherein the activation functions of the two full connection layers are all leakage ReLU. The CAE network parameters are shown in table 1.
Table 1 CAE network parameters
In table 1, in denotes an input layer of the network, conv and Deconv denote a convolution layer and a transposed convolution layer, FC denotes a fully connected layer, and Fold and unfold denote an expanded layer and a collapsed layer between the fully connected layer and the convolution layer or the transposed convolution layer. M represents the number of encoded output neurons from the encoder, using dropout to prevent overfitting, using LeakRelu activation functions to normalize before binarization.
Adding a logistic regression classifier into the dimension reduction compression network CAE, wherein the logistic regression part uses a full connection layer to obtain a joint deep learning model CAE_SOFTMAX shown in figure 2.
Training a joint deep learning model CAE_SOFTMAX by adopting a batch update strategy, wherein the cost function of the joint deep learning model CAE_SOFTMAX is as follows:
Where m is the size of the batch training mini-batch, y ij represents the j-th element of the true class label y i∈Rk of the i-th input vector x i in one mini-batch, Predictive category label/>, representing the i-th input vector x i in a mini-batchK is the vector y i and/>Is a length of (c).
The cae_softmax network parameters are shown in table 2.
TABLE 2 CAE_SOFTMAX network parameters
In table 2, the compressed representation features extracted directly using the CAE network are given to the softmax function for radar signal sorting, and M represents the number of sorted categories.
The orthogonal weight updating method is combined with a full convolution automatic encoder network, a CAE_SOFTMAX_OWM model is designed and is used as a whole for learning and training, and reconstruction errors and sorting errors are optimized.
When a new task is learned, the orthogonal weight modification method (OWM for short) only modifies the CAE_SOFTMAX network weight parameter in the orthogonal direction of the input space of the old task, and the weight iterative computation hardly acts with the input of the former task, thereby ensuring that the solution searched by the network in the training process of the new task is still in the solution space of the former task. Mathematically, OWM achieves its objective by the weight increment effect obtained by the orthogonal projection operator P and the error back-propagation algorithm.
Considering a feed-forward network of l+1 layers, the layer numbers are l=0, 1. In all hidden layers, the activation function g (·) is used. W l represents the connection between the first and (l-1) th layers W l∈Rs ×m.xl and y l represent the first layer output and input, respectively, where xl＝g(yl),yl＝Wl Txl-1,xl-1∈Rs,yl∈Rm. s and m represent the dimensions of the input and output, respectively. In OWM, orthogonal projection P l defined in the first layer input space is the key to continuous learning to overcome catastrophic forgetting. In practice, P l may be recursively updated for each task, in a manner similar to the calculation of the correlation inverse matrix OWM method in the RLS algorithm, allowing the determination of the last task P l from the current input P l. It also avoids matrix anti-operations in the original definition P l.
The following provides the implementation steps of the specific calculation process of the OWM method:
Step 1: parameter initialization: the regularization constant β and the learning rate are input, and for l=1, 2,..l, the weights W l (0) and the projection matrix P l (0) of each layer are randomly initialized, setting P l(0)＝Il/β.
Step 2: the ith batch input propagated forward at the jth task then propagates the loss and calculates a weight modification value ΔW l BP (i, j) for the weight W l (i-1, j) using the standard BP algorithm.
Step 3: updating the weight matrix of each layer:
Wl(i,j)＝W(i-1,j)+κ(i,j)ΔWl BP(i,j) j＝1 (3)
Wl(i,j)＝W(i-1,j)+κ(i,j)Pl(j-1)ΔWl BP(i,j) j＝2,3,... (4)
Where κ (i, j) is a predefined learning rate.
Step 4: each batch repeats steps 2 through 3.
Step 5: if the j-th task has completed training, then forward propagating the average of each batch input in turn in the j-th task:
Wherein, Is the response of layer 1 to the j-th task, the i-th batch input average, and P l(0,j)＝Pl(j-1),i＝1,2,...,nj.
Step 6: and iteratively calculating the projection P l of the matrix to update.
If the orthographic projection P l for each batch is updated as in equation (2), α employsAttenuation, the algorithm can achieve the same performance. This approach can be understood as considering each batch process as a different task. It avoids the extra memory space and data reload in step 4, thus significantly speeding up the processing. The learning rate is set as follows:
Step 7: steps 2 to 6 are repeated for the newly added task.
The orthogonal weight modification method realizes effective protection of the existing weight value in the network, is mainly applied to weight adjustment in the forward propagation process, can be completely compatible with the existing BP reverse gradient propagation algorithm, and shows good performance in continuous learning test tasks.
The orthogonal weight modification method OWM is combined with the CAE_SOFTMAX network to obtain the radar signal sorting model CAE_SOFTMAX_OWM in the invention, and the network structure is shown in figure 3. In the cae_softmax_owm network structure, examples of the full convolution self-encoder, examples of the feature extraction module of the full convolution self-encoder, and examples of the continuity learning module in the feature extraction module of the full convolution self-encoder are included. x 1 (1) denotes the full convolution self-encoder reconstructed original signal x 1, other x n (1) being similar. h l denotes an L-th layer feature extraction module of the full convolution self-encoder. owm 1 represents an orthogonal weight modification method of the continuity learning module in the L-th layer feature extraction module corresponding to h 1. Essentially fig. 1 is a self-encoder structure, but a continuity learning method with orthogonal weight modifications embedded therein.
In sorting actual radar signal data, each modulation format of data is equivalent to a task. When the real-time requirements are high, the network needs to be retrained every time a signal of one modulation type is added. In order to reduce this limitation, the invention proposes a method of continuity learning of radar signal sorting, which is flexible in the task of radar signal sorting.
In order to further illustrate the technical effects of the invention, experimental analysis is performed on the traditional deep learning model DNN and the radar signal sorting model CAE_SOFTMAX_OWM of the invention, the experimental operation environment is shown in table 3, continuous learning training is performed on 24 modulation format signals by adopting a CAE_SOFTMAX_OWM network in the experiment, and new tasks are sequentially added according to the sequence of modulation type labels during continuous learning of the model.
TABLE 3 Experimental Environment configuration
| Deep learning frame | Tensorflow Keras |
| Operating system | Ubuntu16.04 |
| CPU | Inter(R)Core(TM)i7-6700K CPU@4.00GHz |
| GPU | Nvidia GeForce GTX 1080Ti(11GB) |
| RAM | 16GB |
Fig. 4 shows a comparison chart of the classification accuracy of the radar signal classification model cae_softmax_owm of the present invention and the conventional deep learning model DNN on the radar signal classification on the test set, and table 4 shows a comparison of the two model learning training times. From fig. 4 and table 4, it can be seen that the classification accuracy of both classification models approaches as the new task modulation class increases, but the accuracy decreases. However, the continuous learning cae_softmax_owm model is significantly lower than the conventional deep learning model DNN in model learning training time.
Table 4 CAE_SOFTMAX_OWM is compared with DNN model training time (in seconds)
Further, the sparse representation of the CAE_SOFTMAX_OWM model and the DNN model after dimension reduction is visualized by adopting a T-SNE method, wherein the T-SNE method is called a T-distribution field embedding algorithm, the two model output classification result feature layers are drawn into a two-dimensional scatter diagram, and the shape and the color of the scatter diagram corresponding to 24 modulation formats are shown in Table 5. Fig. 5 shows a visual comparison of the T-SNE method for the two models, and it can be seen from fig. 5 that both classification models have a better sorting effect for the 24 modulation formats.
Table 5 scatter diagram shape table corresponding to 24 modulation formats
In summary, according to the continuous learning-based radar signal intelligent sorting method provided by the invention, the full convolution automatic encoder network is adopted to perform dimension reduction processing on radar signals with different modulation formats, so that sparse representation of the radar signals is obtained; then constructing a joint deep learning model CAE_SOFTMAX for dimension reduction and sorting, and combining the joint deep learning model CAE_SOFTMAX with an orthogonal weight modification method to design a radar signal intelligent sorting model method CAE_SOFTMAX_OWM capable of continuous learning. Compared with the traditional deep learning method (DNN for short), the method solves the problems that when the communication modulation format of the enemy radar is continuously changed, the traditional deep learning model DNN needs to restart learning training, learning training efficiency is low and real-time performance is poor, and the radar signal intelligent sorting model CAE_SOFTMAX_OWM can adapt to the radar signal modulation format which is dynamically changed through continuous learning without restarting the learning, and can also learn features in an old task to be applied to a new modulation format recognition task, so that good sorting effects can be shown in all tasks. Experimental results show that the learning training efficiency and the real-time performance of the radar signal sorting network are effectively improved, and the dynamic change of the modulation format of the enemy radar signal can be quickly adapted, so that electromagnetic space control rights are obtained in electronic countermeasure of enemy.
The technical features of the above-described embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above-described embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The above examples illustrate only a few embodiments of the invention, which are described in detail and are not to be construed as limiting the scope of the invention. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the invention, which are all within the scope of the invention. Accordingly, the scope of protection of the present invention is to be determined by the appended claims.
Claims (6)
1. The intelligent radar signal sorting method based on continuous learning is characterized by comprising the following steps:
Step one: acquiring radar signal data, preprocessing the radar signal data to obtain a data set, and randomly sampling from the data set to obtain a training set;
Step two: combining a full convolution self-encoder with a logistic regression classifier to construct a joint deep learning model CAE_SOFTMAX, and optimizing a full convolution self-encoder network in the joint deep learning model CAE_SOFTMAX by adopting an orthogonal weight modification method to perform continuous learning to obtain a radar signal sorting model CAE_SOFTMAX_OWM capable of being continuously learned; the calculation process of the orthogonal weight modification method comprises the following steps:
step 1: inputting a regularization constant beta and a learning rate, and randomly initializing a weight W l (0) and a projection matrix P l (0) of each layer, wherein l=0;
Step 2: inputting forward calculation to the ith batch of the jth task, solving by using BP algorithm, and calculating a weight modification value for the weight W l (i-1, j)
Step 3: updating each layer of weight matrix;
Step 4: each batch repeats steps 2 through 3;
step 5: after training of the task j is completed, sequentially and forward propagating the average value of each batch input;
step 6: iteratively calculating the projection P l of the matrix, and updating;
Step three: training the radar signal sorting model CAE_SOFTMAX_OWM by using the training set;
step four: acquiring radar signals to be sorted, and dividing the radar signals to be sorted into IQ two paths of data;
step five: inputting the IQ two paths of data into a trained radar signal sorting model CAE_SOFTMAX_OWM for sorting, performing data dimension reduction processing on the input IQ two paths of data by using a full convolution self-encoder by the radar signal sorting model CAE_SOFTMAX_OWM, obtaining a sparse representation after dimension reduction, using the sparse representation after dimension reduction for a SOFTMAX function for radar signal sorting, and finally outputting a corresponding label.
2. The intelligent sorting method for radar signals based on continuous learning according to claim 1, wherein the full convolution self-encoder sequentially comprises an input layer, five convolution layers, one folding layer, two full connection layers, one unfolding layer and two transposed convolution layers, wherein the activation functions of the two full connection layers are all Leaky ReLU.
3. The intelligent sorting method of radar signals based on continuous learning according to claim 1 or 2, wherein the full convolution self-encoder adopts root mean square error as a cost function, and the cost function calculation formula is:
Where m is the size of the batch training mini-batch, x ij represents the j-th element of the i-th input signal pixel x i∈Rn in one mini-batch, Represents the ith reconstructed signal pixel/>, in one mini-batchN is the vector x i and/>Is a length of (c).
4. The continuous learning-based radar signal intelligent sorting method according to claim 1 or 2, wherein the joint deep learning model cae_softmax is trained by adopting a batch update strategy, and a cost function of the joint deep learning model cae_softmax is:
Where m is the size of the batch training mini-batch, y ij represents the j-th element of the true class label y i∈Rk of the i-th input vector x i in one mini-batch, Predictive category label/>, representing the i-th input vector x i in a mini-batchK is the vector y i and/>Is a length of (c).
5. The intelligent sorting method of radar signals based on continuous learning according to claim 1 or 2, wherein the data set comprises 24 modulation formats, each modulation format comprises 26 signal-to-noise ratios, each signal-to-noise ratio comprises 3000 pieces of data, each data comprises IQ two signals, and each signal comprises 1024 points.
6. The intelligent sorting method of radar signals based on continuous learning according to claim 1 or 2, wherein the radar signal data is obtained by a radar signal receiver.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111226205.8A CN113962262B (en) | 2021-10-21 | 2021-10-21 | Continuous learning-based intelligent radar signal sorting method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111226205.8A CN113962262B (en) | 2021-10-21 | 2021-10-21 | Continuous learning-based intelligent radar signal sorting method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113962262A CN113962262A (en) | 2022-01-21 |
| CN113962262B true CN113962262B (en) | 2024-05-14 |
Family
ID=79465829
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111226205.8A Active CN113962262B (en) | 2021-10-21 | 2021-10-21 | Continuous learning-based intelligent radar signal sorting method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113962262B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115508795B (en) * | 2022-10-20 | 2024-04-02 | 中国人民解放军空军航空大学 | Method for dynamically generating detection interference integrated shared signal |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5162805A (en) * | 1975-02-19 | 1992-11-10 | The United States Of America As Represented By The Secretary Of The Navy | Frequency diversity sidelobe canceller |
| CN101587546A (en) * | 2009-06-12 | 2009-11-25 | 浙江大学 | Identification method of lower limb action pattern based on supporting vector multivariate classification |
| WO2014205231A1 (en) * | 2013-06-19 | 2014-12-24 | The Regents Of The University Of Michigan | Deep learning framework for generic object detection |
| CN109116312A (en) * | 2018-07-20 | 2019-01-01 | 电子科技大学 | Radar signal classification method based on QMFB and convolutional neural networks |
| CN110569752A (en) * | 2019-08-20 | 2019-12-13 | 河海大学 | convolutional neural network-based radar signal category determination method |
| CN110991513A (en) * | 2019-11-27 | 2020-04-10 | 佛山科学技术学院 | Image target recognition system and method with human-like continuous learning capability |
| CN111948622A (en) * | 2020-08-07 | 2020-11-17 | 哈尔滨工程大学 | Linear frequency modulation radar signal TOA estimation algorithm based on parallel CNN-LSTM |
| CN112859025A (en) * | 2021-01-05 | 2021-05-28 | 河海大学 | Radar signal modulation type classification method based on hybrid network |
| CN113033473A (en) * | 2021-04-15 | 2021-06-25 | 中国人民解放军空军航空大学 | ST2DCNN + SE-based radar overlapped signal identification method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10437936B2 (en) * | 2018-02-01 | 2019-10-08 | Jungle Disk, L.L.C. | Generative text using a personality model |
-
2021
- 2021-10-21 CN CN202111226205.8A patent/CN113962262B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5162805A (en) * | 1975-02-19 | 1992-11-10 | The United States Of America As Represented By The Secretary Of The Navy | Frequency diversity sidelobe canceller |
| CN101587546A (en) * | 2009-06-12 | 2009-11-25 | 浙江大学 | Identification method of lower limb action pattern based on supporting vector multivariate classification |
| WO2014205231A1 (en) * | 2013-06-19 | 2014-12-24 | The Regents Of The University Of Michigan | Deep learning framework for generic object detection |
| CN109116312A (en) * | 2018-07-20 | 2019-01-01 | 电子科技大学 | Radar signal classification method based on QMFB and convolutional neural networks |
| CN110569752A (en) * | 2019-08-20 | 2019-12-13 | 河海大学 | convolutional neural network-based radar signal category determination method |
| CN110991513A (en) * | 2019-11-27 | 2020-04-10 | 佛山科学技术学院 | Image target recognition system and method with human-like continuous learning capability |
| CN111948622A (en) * | 2020-08-07 | 2020-11-17 | 哈尔滨工程大学 | Linear frequency modulation radar signal TOA estimation algorithm based on parallel CNN-LSTM |
| CN112859025A (en) * | 2021-01-05 | 2021-05-28 | 河海大学 | Radar signal modulation type classification method based on hybrid network |
| CN113033473A (en) * | 2021-04-15 | 2021-06-25 | 中国人民解放军空军航空大学 | ST2DCNN + SE-based radar overlapped signal identification method |
Non-Patent Citations (3)
| Title |
|---|
| A Comprehensive Survey of Machine Learning Applied to Radar Signal Processing;Ping Lang 等;《arXiv:2009.13702v1》;20200929;1-49 * |
| 基于神经网络的复杂雷达信号分类方法研究;文茜;《中国优秀硕士学位论文全文数据库 信息科技辑》;20190515;I136-637 * |
| 认知电子战研究进展;刘松涛 等;《探测与控制学报》;20201031;第第42卷卷(第第5期期);1-15 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113962262A (en) | 2022-01-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110263227B (en) | Group partner discovery method and system based on graph neural network | |
| CN111882040B (en) | Convolutional neural network compression method based on channel number search | |
| CN109711426B (en) | Pathological image classification device and method based on GAN and transfer learning | |
| CN113361664B (en) | Image recognition system and method based on quantum convolution neural network | |
| CN112288011B (en) | Image matching method based on self-attention deep neural network | |
| CN112966114B (en) | Literature classification method and device based on symmetrical graph convolutional neural network | |
| CN113191387A (en) | Cultural relic fragment point cloud classification method combining unsupervised learning and data self-enhancement | |
| CN111723914A (en) | Neural network architecture searching method based on convolution kernel prediction | |
| CN112101364B (en) | Semantic segmentation method based on parameter importance increment learning | |
| CN113240683B (en) | Attention mechanism-based lightweight semantic segmentation model construction method | |
| CN116740527A (en) | Remote sensing image change detection method combining U-shaped network and self-attention mechanism | |
| CN114821050A (en) | Named image segmentation method based on transformer | |
| CN113935489A (en) | Variational quantum model TFQ-VQA based on quantum neural network and two-stage optimization method thereof | |
| CN113962262B (en) | Continuous learning-based intelligent radar signal sorting method | |
| CN115965864A (en) | Lightweight attention mechanism network for crop disease identification | |
| CN116977001A (en) | Geological disaster prevention and treatment engineering cost management system and method thereof | |
| CN115393690A (en) | Light neural network air-to-ground observation multi-target identification method | |
| CN113420593B (en) | Small sample SAR automatic target recognition method based on hybrid inference network | |
| CN112801029B (en) | Attention mechanism-based multitask learning method | |
| CN117611838A (en) | Multi-label image classification method based on self-adaptive hypergraph convolutional network | |
| CN110288002B (en) | Image classification method based on sparse orthogonal neural network | |
| CN113688946B (en) | Multi-label image recognition method based on spatial correlation | |
| Arribas et al. | Neural architectures for parametric estimation of a posteriori probabilities by constrained conditional density functions | |
| CN114595641A (en) | Method and system for solving combined optimization problem | |
| Ko et al. | Deep model compression and inference speedup of sum–product networks on tensor trains |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |