CN112668498B - Method, system, terminal and application for identifying individual intelligent increment of aerial radiation source - Google Patents
Method, system, terminal and application for identifying individual intelligent increment of aerial radiation source Download PDFInfo
- Publication number
- CN112668498B CN112668498B CN202011621625.1A CN202011621625A CN112668498B CN 112668498 B CN112668498 B CN 112668498B CN 202011621625 A CN202011621625 A CN 202011621625A CN 112668498 B CN112668498 B CN 112668498B
- Authority
- CN
- China
- Prior art keywords
- radiation source
- layer
- identification
- convolution
- training
- 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
- 230000005855 radiation Effects 0.000 title claims abstract description 83
- 238000000034 method Methods 0.000 title claims abstract description 62
- 230000006870 function Effects 0.000 claims abstract description 68
- 238000012549 training Methods 0.000 claims abstract description 55
- 230000009466 transformation Effects 0.000 claims abstract description 23
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 21
- 238000010586 diagram Methods 0.000 claims abstract description 12
- 238000001228 spectrum Methods 0.000 claims abstract description 7
- 238000011176 pooling Methods 0.000 claims description 37
- 238000000605 extraction Methods 0.000 claims description 31
- 210000002569 neuron Anatomy 0.000 claims description 15
- 238000000354 decomposition reaction Methods 0.000 claims description 14
- 230000004927 fusion Effects 0.000 claims description 14
- 238000012545 processing Methods 0.000 claims description 12
- 238000013528 artificial neural network Methods 0.000 claims description 10
- 125000004122 cyclic group Chemical group 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 4
- 238000004590 computer program Methods 0.000 claims description 4
- 238000012937 correction Methods 0.000 claims description 4
- 238000004821 distillation Methods 0.000 claims description 4
- 238000011049 filling Methods 0.000 claims description 4
- 238000012935 Averaging Methods 0.000 claims description 3
- 238000009825 accumulation Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 230000001186 cumulative effect Effects 0.000 claims description 3
- 230000001934 delay Effects 0.000 claims description 3
- 230000005284 excitation Effects 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 238000011068 loading method Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000002156 mixing Methods 0.000 claims description 2
- 238000012544 monitoring process Methods 0.000 abstract description 11
- 238000013500 data storage Methods 0.000 abstract description 2
- 230000009286 beneficial effect Effects 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000003491 array Methods 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 230000008092 positive effect Effects 0.000 description 2
- 238000012512 characterization method Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005215 recombination Methods 0.000 description 1
- 230000006798 recombination Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000007480 spreading Effects 0.000 description 1
Abstract
The invention belongs to the technical field of radiation source individual identification, and discloses an intelligent incremental identification method, a system, a terminal and application of an aerial radiation source individual, wherein a fuzzy function, double-spectrum transformation, hilbert yellow transformation and short-time Fourier transformation are respectively extracted from received aerial radiation source ADS-B (broadcast automatic correlation monitoring) signals, and linear characteristics of the characteristics are fused to obtain a new characteristic diagram; classifying and identifying known types of radiation source individuals through a convolutional neural network to obtain a network model; and training the untrained category data in an incremental learning mode to realize intelligent incremental identification of the air radiation source individuals. The invention has good recognition accuracy under lower signal-to-noise ratio, good recognition capability under different channels, and weak dependence on a single characteristic, solves the problem that training data arrive in batches, greatly shortens the time required by training and reduces the space required by data storage.
Description
Technical Field
The invention belongs to the technical field of radiation source individual identification, and particularly relates to an intelligent incremental identification method, system, terminal and application of an aerial radiation source individual.
Background
At present: radiation source individual identification refers to identifying a target individual by extracting one or more modulation characteristics exhibited by the received signal. With the increasing number of aerial target individuals, it is necessary to judge whether the aerial target radiation source works normally or abnormally according to the nature of the friend or foe of the aerial target, and these judgments are all based on the premise that the individual identification of the aerial target radiation source is needed, and how to judge quickly and accurately is an important means for realizing national security.
Radiation source individual identification techniques based on fingerprint information have begun in the last century. SaK Lang D et al propose a method for extracting fingerprint information using a constellation and inputting the method for identification in convolutional neural networks, but the method is more stringent for time synchronization information (SaK, lang D, wang C, et al Specic Emitter Identification Techniques for the Internet of Things [ J ]. IEEE Access,2020, 8:1644-1652.). Song C, xu J et al propose a non-stationary signal analysis method of empirical mode decomposition (empirical mode decomposition), but this method has the problem of mode aliasing (Song C, xu J, zhan Y.A method for specific emitter identification based on empirical mode decomposition [ C ]// IEEE International Conference onWireless communications. IEEE, 2010.). Gok G, alp Y K et al propose a method based on a Variational Modal Decomposition (VMD) that uses the envelope and instantaneous frequency of the received signal as a set of models to identify the different radiation source signals, but the method is relatively complex and disadvantageous for practical operation (Gok G, alp Y K, arikan O.A New Method for Specific Emitter Identification With Results on Real Radar Measurements [ J ]. IEEE Transactions on Information Forensics and Security,2020, PP (99): 1-1.). Shieh C S et al propose statistics based on conventional parameters of the radiation source such as direction of arrival (DOA), pulse Width (PW), pulse Repetition Frequency (PRF), and Radar Frequency (RF) as basis for classification recognition, but the recognition accuracy of the method is poor at low signal-to-noise ratio (Shieh C S, lin C T.A vector neural network for emitter identification [ J ] IEEE Transactions on Antennas and Propagation,2002, 50 (8): 1120-1127.). Xu Dan, yang Bo et al propose a Kernel Principal Component Analysis (KPCA) prediction learning method that solves the processing problem of different data clusters with complex nonlinear distributions, but it is difficult to adapt the method to other communication scenarios under non-gaussian channels (Dan Xu, bo Yang, wenli Jiang,. An improved SVDU-IKPCA algorithm for Specific Emitter Identification [ C ]// International Conference on Information & automation.ieee, 2008.). Chen Yue et al propose individual identification of communication radiation sources based on IQ map features, but the features extracted by this method are not obvious and perform poorly at low signal-to-noise ratios (Chen Yue, lei Yingke, li Xin, she Ling, mei Fan. Individual identification of communication radiation sources based on IQ map features [ J/OL ]. Signal processing). Chen Peng et al propose a method for using Frechet distance to calculate the distance between signals, pulse envelope or instantaneous frequency to achieve individual identification of the radiation source, which method is also poor in identification performance at low signal-to-noise ratios (P.Chen, G.Li, K.Xu and J.Wan, "Applying the Frechet distance to the specific emitter identification,"2016IEEE 13th International Conference on Signal Processing (ICSP), chengdu,2016, pp.1027-1030.). The method has high dependency on sliding window selection and low recognition accuracy (D' Agrocin S.Specific emitter identification based on amplitude features [ C ]// IEEE International Conference on Signal & Image Processing applications IEEE 2015.).
The method solves the problem of individual identification of the radiation source to a certain extent, but has poor identification performance in a low signal-to-noise ratio environment and insufficient generalization capability in different channel environments, and the individual identification method has strong dependence on the characteristic extraction method and has larger identification accuracy difference in different characteristic extraction methods. In addition, conventional training strategies must require all data categories as a priori information and perform a uniform training. However, in reality, the data is often not a whole block, but rather is streaming data, and the traditional combined training mode is not suitable for the current environment.
Through the above analysis, the problems and defects existing in the prior art are as follows:
(1) The prior art has the problems that the recognition performance is poor in a low signal-to-noise ratio environment, the generalization capability is insufficient in different channel environments, the individual recognition method has strong dependence on the feature extraction method, the recognition accuracy of different feature extraction methods is large, and the like.
(2) Conventional training strategies must require that all data categories be known in advance and uniformly trained. However, in reality, the data is often not a whole block, but rather is streaming data, and the traditional combined training mode is not suitable for the current environment.
The difficulty of solving the problems and the defects is as follows:
(1) Because of the influence of the current complex electromagnetic environment and channels, the signal of an individual common air radiation source is difficult to achieve a higher signal-to-noise ratio, the signal is lower in signal-to-noise ratio and stronger in interference, and the method for extracting the characteristics is difficult to find a characteristic with strong anti-interference capability and good identification capability on a plurality of channels;
(2) The training of the streaming data needs to save the previous data, while the storage of the old data has larger requirement on the memory space, if all the data are saved and retrained, the previous training result is wasted, a large amount of memory space is wasted, and the training speed is improved and the resource is utilized maximally.
The meaning of solving the problems and the defects is as follows:
the types of the radiation source individuals in the air are more and more, the mutually staggered signals are more and more dense, the effect of interfering signals of hostile countries is stronger and stronger, the battlefield form is changeable instantaneously, and the radiation source individuals can be identified more quickly and accurately, so that the trend of the whole battlefield form can be influenced. Therefore, how to accurately identify the air radiation source individuals in a complex electromagnetic environment with low signal-to-noise ratio and strong interference, and quickly update the individual identification system by utilizing the previous training results and new data, thereby improving the capability of electronic warfare has become a key problem to be solved in modern war.
Disclosure of Invention
Aiming at the problems existing in the prior art, the invention provides an intelligent incremental identification method, system, terminal and application of an air radiation source individual.
The invention is realized in such a way that an intelligent increment identification method for an air radiation source individual is provided, and the intelligent increment identification method for the air radiation source individual comprises the following steps:
extracting four features of fuzzy function, bispectrum transformation, hilbert yellow transformation and short-time Fourier transformation from a received ADS-B (broadcast automatic correlation monitoring) signal of an aerial radiation source respectively, and carrying out linear feature fusion on the features to obtain a new feature map;
classifying and identifying known types of radiation source individuals through a convolutional neural network to obtain a network model;
and training the untrained category data in an incremental learning mode, namely training the network by taking a small amount of old data and new data together on the basis of the original training result, and modifying part of parameters and loss functions in the network to realize intelligent incremental identification of the air radiation source individuals.
Further, the four feature extraction methods are specifically implemented as:
(1) Short-time fourier transform STFT: the short-time fourier transform expression is:
wherein the subscript n is different from the standard Fourier transform, w (n-m) is a window function sequence, x (m) is an input sequence, and different window function sequences can obtain different Fourier transform results; the short-time fourier transform has two independent variables n and w, both a discrete function with respect to time n and a continuous function with respect to angular frequency w;
(2) Blur functions are used to analyze and design various signals, identifying different types of signals. Assuming that the delay difference is tau, the frequency shift is mu, and the input signal is X 1 (t)、X 2 (t) the ambiguity function is defined as:
(3) The double spectrum conversion is carried out, the double spectrum of the received signal r (n) is estimated by a non-parametric method, r (n) is firstly divided into gamma segments, each segment comprises delta samples, and the third-order cyclic cumulative quantity of the received signal r (n) is written as:
wherein χ is γ (τ 1 ,τ 2 ) Is the third-order cyclic accumulation of each segment of signal, τ 1 ,τ 2 For different time delays, w (τ 1 ,τ 2 ) Is a hexagonal window function, and the bispectral estimate of the signal r (n) is expressed as:
(4) The Hilbert-Huang transform is a method of combining empirical mode decomposition with Hilbert transform. The basic principle of the empirical mode decomposition algorithm is as follows: first, find the upper envelope X of the original signal X (t) max (t) and lower envelope X min (t) and averaging the upper and lower envelopes:
second, for the original signal X (t) and the average envelope m 1 (t) subtracting to obtain the remaining signal d 1 (t); for the remaining signal d 1 (t) repeating the above operation until the screening threshold (SD) is smaller, at which point the final suitable first order mode is obtainedComponent c 1 (t), i.e., a first IMF, wherein the SD solution is as follows:
again, for signals X (t) and c 1 (t) obtaining a first-order residual quantity r by taking a difference 1 (t) r is to 1 (t) substituting the original signal X (t) to perform the above operation, repeating n times to obtain an n-order modal function c n (t) and the final standard-compliant residual amount r n (t) the expression of the original signal X (t) decomposed by empirical mode decomposition is:
finally, the Hilbert transform is used for carrying out time-frequency processing on the signals subjected to the empirical mode decomposition processing and converting the signals into Hilbert spectrograms.
Further, the characteristic is subjected to linear characteristic fusion to obtain a new characteristic diagram, and the method is concretely implemented as follows: expanding the four features to the same size by adopting an interpolation zero padding method, and then carrying out linear feature fusion on the four features to finally obtain a new feature map; if the individual differences of the radiation sources reflected by a certain characteristic extraction method are not obvious, the neural network judges which radiation source individual belongs to through other characteristics.
Further, the classifying and identifying the known type of radiation source individuals through the convolutional neural network comprises the following steps:
firstly, the characteristic extraction is carried out on the input characteristic diagram by a convolution layer, the conventional processing of the convolution neural network is usually picture data, the characteristic extraction of four modes is carried out on signals, the characteristic fusion is carried out on the signals, and each signal can obtain a four-channel data characteristic diagram which is also suitable for the processing of the convolution neural network. The convolution layer contains a plurality of convolution kernels, each element constituting the convolution kernels corresponds to a weight coefficient and a deviation amount, each neuron in the convolution layer is connected with a plurality of neurons in a region close to the position in the previous layer, and the size of the region depends on the size of the convolution kernels. The calculation formula is as follows:
wherein b is the deviation, ω is the weight, x and y represent the convolution check to convolve the whole feature map, Z l And Z l+1 Convolved inputs and outputs representing layer l+1, also called feature maps, L l+1 Is Z l+1 Assuming equal feature pattern length and width; z (i, j) corresponds to the pixel of the feature map, K is the number of feature map channels, f is the convolution kernel size, s 0 Is the convolution step size, p is the padding size; the convolutional layer contains excitation functions to help express complex features, and the linear rectification function RELU is used to enable neurons in the neural network to have sparse activation, and the expression form is as follows:
secondly, after the feature extraction is carried out by the convolution layer, the output features are transmitted to the pooling layer for feature selection and information filtering; adding pooling operation after the convolution layer, wherein the pooling layer selects pooling areas which are the same as the steps of the convolution kernel scanning feature map, and the pooling areas are controlled by pooling size, step length and filling, and the expression is as follows:
right middle A k Representing the input feature map, and other parameters are co-convolved with the convolutional layer.
And finally, inputting the output of the pooling layer into the full-connection layer to perform nonlinear combination on the extracted features so as to obtain output.
Further, training the untrained category data in an incremental learning mode; based on the original training result, taking a small amount of old data and new data to train the network together, and modifying partial weight values and loss functions in the network, thereby realizing intelligent incremental identification of the air radiation source individuals; the method specifically comprises the following steps:
firstly, obtaining network model parameter information of a known class sample;
secondly, constructing a new data set by the old sample set and the new unknown class sample; mixing a small amount of old data with new data to form a new data set, and randomly selecting the original data set by adopting a random selection principle on the selection of old samples;
finally, loading an old network model, modifying a network output layer, changing the number of output nodes into the number of radiation source individuals actually trained, reducing the step length of each layer of convolution kernel of a convolution layer and lowering the learning rate when the training recognition rate is low, adding dropout or a regularization term for correction when fitting occurs, and optimizing a loss function as follows:
wherein the first term is the classifier loss function and the second term is the distillation loss function, q i Representing the nodes of the different neurons,representing the classifier.
It is a further object of the present invention to provide a computer device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of:
extracting four features of fuzzy function, bispectrum transformation, hilbert yellow transformation and short-time Fourier transformation from a received ADS-B (broadcast automatic correlation monitoring) signal of an aerial radiation source respectively, and carrying out linear feature fusion on the features to obtain a new feature map;
classifying and identifying known types of radiation source individuals through a convolutional neural network to obtain a network model;
and training the untrained category data in an incremental learning mode, namely training the network by taking a small amount of old data and new data together on the basis of the original training result, and modifying part of parameters and loss functions in the network to realize intelligent incremental identification of the air radiation source individuals.
Another object of the present invention is to provide a computer readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the steps of:
extracting four features of fuzzy function, bispectrum transformation, hilbert yellow transformation and short-time Fourier transformation from a received ADS-B (broadcast automatic correlation monitoring) signal of an aerial radiation source respectively, and carrying out linear feature fusion on the features to obtain a new feature map;
classifying and identifying known types of radiation source individuals through a convolutional neural network to obtain a network model;
and training the untrained category data in an incremental learning mode, namely training the network by taking a small amount of old data and new data together on the basis of the original training result, and modifying part of parameters and loss functions in the network to realize intelligent incremental identification of the air radiation source individuals.
The invention further aims at providing an information data processing terminal which is used for realizing the intelligent incremental identification method of the aerial radiation source individuals.
It is another object of the present invention to provide an intelligent incremental identification system for implementing the method for identifying individual intelligent increments of an aerial radiation source, the intelligent incremental identification system comprising:
the feature combination module is used for carrying out Hilbert yellow transformation, short-time Fourier transformation, fuzzy function and bispectrum transformation feature extraction on the received ADS-B (broadcast automatic correlation monitoring) signals and combining the features into a new feature;
and the radiation source individual identification module is used for realizing radiation source individual identification through a convolutional neural network.
And the incremental training module is used for performing incremental training on the network by adopting an incremental learning method.
The invention further aims at providing a radiation source individual identification terminal which is used for realizing the intelligent incremental identification method.
By combining all the technical schemes, the invention has the advantages and positive effects that: in order to identify the radiation source individual, the invention performs four modes of feature extraction and linear feature fusion on the signals, each signal can obtain a four-channel data feature map, the invention is suitable for the application scene of a convolutional neural network, and after the four modes of feature extraction are fused, each map contains information of all feature extraction, thereby avoiding the dependence on a certain single feature. The invention adds pooling operation after the convolution layer, filters redundant information and reduces the training scale. The pooling layer selects pooling area and the step of the convolution kernel scanning characteristic diagram are the same, and the pooling area, step length and filling are controlled. The invention solves the problem of identifying the radiation source individual, so that the last layer of the full-connection layer is provided with a normalized exponential function (softmax function) to output a classification label, thereby obtaining information such as network model parameters and the like.
The convolution layer used in the invention adopts a larger convolution kernel for processing. Since AlexNet mainly operates on image classification, the pooling layer is added after each layer to reduce redundant information, and one extracted feature of the present invention is effective information, if the pooling layer is added after each layer of convolution layer, part of the information may be lost. Thus, there is no need to add a pooling layer after each convolution layer, but instead add a global pooling layer after all convolution layers, because the network's mining of the radiation source information is sufficient at this time, and the goal of adding global pooling is simply to reduce the data size and speed up the network convergence. Experiments prove that the network structure is truly beneficial to the identification of the radiation source individuals.
The invention obtains the information such as network model parameters of the known class sample, so as to facilitate incremental learning; second, a new data set is constructed from the old sample set and the new unknown class samples. In order to prevent the network from forgetting too much for the old sample information and prevent the network from being fitted with new type data, the invention uses less part of old data and new data to be mixed to form a new data set, adopts the principle of random selection on the selection of the old sample, and randomizes and randomly selects the original data set; finally, the old network model is loaded, the network output layer is modified, the number of output nodes is changed into the number of radiation source individuals actually trained, when the training recognition rate is low, the step length of each layer of convolution kernel of the convolution layer is reduced, the learning rate is reduced, dropout or a regular term is added for correction when fitting occurs, in order to prevent the catastrophic forgetting problem in incremental learning, the loss function is optimized, and the result caused by training errors is aggravated by setting a T value larger than 1, which is equivalent to 'weight training', so that the training accuracy is higher. The new loss function obtained by combining these two terms will be much better than the loss function obtained by using only the classifier.
The invention can effectively realize the classification of ADS-B (broadcast automatic correlation monitoring) signals, has good recognition performance in a low signal-to-noise ratio environment, reduces the dependence on a feature extraction method, adopts an incremental learning method, and greatly reduces the training time and the space required by storage. The invention provides an incremental intelligent identification method of an aerial radiation source individual, which can have good identification accuracy under a lower signal-to-noise ratio, has good identification capability under different channels, has low dependence on a single characteristic, solves the problem that training data arrives in batches, greatly shortens the time required by training and reduces the space required by data storage.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following description will briefly explain the drawings needed in the embodiments of the present application, and it is obvious that the drawings described below are only some embodiments of the present application, and that other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of an intelligent incremental identification method for an aerial radiation source individual provided by an embodiment of the invention.
FIG. 2 is a schematic diagram of an aerial radiation source individual intelligent incremental identification system provided by an embodiment of the present invention;
in fig. 2: 1. a feature combination module; 2. a radiation source individual identification module; 3. and an incremental training module.
FIG. 3 is a graph of recognition accuracy for training by incremental learning after feature recombination and input to a convolutional neural network, provided by an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the following examples in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
Aiming at the problems existing in the prior art, the invention provides an intelligent incremental identification method, an intelligent incremental identification system, an intelligent incremental identification terminal and an intelligent incremental identification application for an individual air radiation source, and the intelligent incremental identification method, the intelligent incremental identification system, the intelligent incremental identification terminal and the intelligent incremental identification application are described in detail below with reference to the accompanying drawings.
As shown in fig. 1, the method for identifying the intelligent increment of the aerial radiation source individual provided by the invention comprises the following steps:
s101: performing Hilbert-Huang transform, short-time Fourier transform, fuzzy function and bispectrum transform feature extraction on the received ADS-B (broadcast automatic correlation monitoring) signals, and combining the features into a new feature;
s102: realizing radiation source individual identification through a convolutional neural network;
s103: incremental training is carried out on the network by adopting an incremental learning method.
Other steps may be performed by those skilled in the art of the intelligent incremental identification method provided by the present invention, and the intelligent incremental identification method provided by the present invention of fig. 1 is merely one specific embodiment.
As shown in fig. 2, the intelligent incremental identifying system provided by the present invention includes:
the feature combination module 1 is used for performing Hilbert yellow transform, short-time Fourier transform, fuzzy function and bispectrum transform feature extraction on the received ADS-B (broadcast automatic correlation monitoring) signals, and combining the features into a new feature;
and the radiation source individual identification module 2 is used for realizing radiation source individual identification through a convolutional neural network.
And the incremental training module 3 is used for performing incremental training on the network by adopting an incremental learning method.
The technical scheme of the invention is further described below with reference to specific embodiments.
The intelligent increment identification method provided by the invention comprises the following steps:
firstly, performing feature extraction and linear feature fusion on the received ADS-B (broadcast automatic correlation monitoring) signals by the four methods to obtain a new feature map;
(1) Short Time Fourier Transform (STFT)
The short-time fourier transform expression is:
the subscript n is different from the standard Fourier transform, w (n-m) is a window function sequence, x (m) is an input sequence, and different window function sequences can obtain different Fourier transform results; the short-time fourier transform has two independent variables n and w, which are both discrete functions with respect to time n and continuous functions with respect to angular frequency w.
(2) Fuzzy function (ambiguity function)
The blurring function may be used to analyze and design various signals, and may also be used to identify different types of signals. Assuming that the delay difference is tau, the frequency shift is mu, and the input signal is X 1 (t)、X 2 (t) the ambiguity function is defined as:
(3) Double spectrum conversion (Bi-spectrum transform)
The bispectrum of the received signal r (n) is estimated by a non-parametric method. r (n) is first divided into Γ segments, each segment containing Δ samples. The third-order cyclic cumulative amount of the received signal r (n) can be written as:
wherein χ is γ (τ 1 ,τ 2 ) Is the third-order cyclic accumulation of each segment of signal, τ 1 ,τ 2 For different time delays, w (τ 1 ,τ 2 ) Is a hexagonal window function, and the bispectral estimate of the signal r (n) is expressed as:
(4) Hilbert yellow transform
The hilbert yellow transform is a method of combining empirical mode decomposition with the hilbert transform. The basic principle of the empirical mode decomposition algorithm is as follows:
first, find the upper envelope X of the original signal X (t) max (t) and lower envelope X min (t) and averaging the upper and lower envelopes:
second, for the original signal X (t) and the average envelope m 1 (t) subtracting to obtain the remaining signal d 1 (t). For the remaining signal d 1 (t) repeating the above operation until the screening threshold (SD) is smaller, at which point the final suitable first-order modal component c is obtained 1 (t), i.e., the first IMF.
The SD method is as follows:
again, for signals X (t) and c 1 (t) obtaining a first-order residual quantity r by taking a difference 1 (t) r is to 1 (t) substituting the original signal X (t) to perform the above operation, repeating n times to obtain n-order modal function c n (t) and the final standard-compliant residual amount r n (t) the expression of the original signal X (t) decomposed by empirical mode decomposition is:
finally, the Hilbert transform is used for carrying out time-frequency processing on the signals after empirical mode decomposition and converting the signals into Hilbert spectrograms.
After extraction, the four features are subjected to linear feature fusion to obtain a new feature map.
And secondly, inputting the features obtained in the first step into a convolutional neural network for recognition.
Convolutional neural networks are defined as: the convolutional neural network comprises a convolutional layer, a pooling layer, a full-connection layer and an output layer.
The convolutional layer has the function of extracting the characteristics of input data, the convolutional neural network is usually picture data in the past, and the convolutional neural network has very strong characterization and classification discrimination capability. In order to identify the radiation source individual, as described in the first step, the four modes of feature extraction and linear feature fusion are carried out on the signals, so that each signal can obtain a four-channel data feature map, the method is very suitable for the application scene of a convolutional neural network, and after the four modes of feature extraction are fused, each map contains information of all feature extraction, and the dependence on a certain single feature is avoided. First, the input feature map is "feature extracted" (which is distinguished from the feature extraction of the first step, and refers to information acquisition and learning of input data) by a convolution layer, which contains a plurality of convolution kernels, each element constituting the convolution kernels corresponding to a weight coefficient and a bias vector, similar to a neuron of a feedforward neural network. Each neuron in the convolution layer is connected to a plurality of neurons in a region of the preceding layer that is located close to the region, the size of the region being dependent on the size of the convolution kernel. The calculation formula is as follows:
wherein b is the deviation, ω is the weight, x and y represent the convolution check to convolve the whole feature map, Z l And Z l+1 The convolution inputs and outputs representing layer l+1 are also referred to as feature maps (feature maps). L (L) l+1 Is Z l+1 Assuming equal feature pattern length and width. Z (i, j) corresponds to a pixel of the feature map, K is the number of feature map channels, f is the convolution kernel size (Kernel size), s 0 Is the convolution step size (stride), and p is the padding size (padding). The convolution layer contains an excitation function to assist in expressing complex characteristics, and the linear rectification function RELU is used for enabling neurons in a neural network to have sparse activation, and the expression form is as follows:
and secondly, after the convolution layer performs feature extraction, the output features are transferred to a pooling layer for feature selection and information filtering. The characteristic diagram extracted in the first step is generally large in size, and although the characteristic diagram is processed by a convolution layer, the characteristic diagram is still large in size, and a lot of redundant information exists, so that the training speed is not beneficial to improvement. Therefore, the invention adds pooling operation after the convolution layer, filters redundant information and reduces the training scale. The pooling layer selects pooling area and the step of the convolution kernel scanning characteristic diagram are the same, and the pooling area, step length and filling are controlled. The expression is as follows:
right middle A k Representing the input feature map, and other parameters are co-convolved with the convolutional layer.
And finally, the output of the pooling layer is input into the full-connection layer to carry out nonlinear combination on the extracted features so as to obtain output. The invention solves the problem of identifying the radiation source individual, so that the last layer of the full-connection layer is provided with a normalized exponential function (softmax function) to output a classification label, thereby obtaining information such as network model parameters and the like.
The invention mainly uses a convolutional neural network to classify the radiation source individuals, and uses an improved AlexNet network to classify and identify the radiation source individuals due to the strong classifying capability of the AlexNet network on image information. Because of the large size of the data being processed, the convolution layers used in the present invention are processed with a large convolution kernel. Since AlexNet mainly operates on image classification, the pooling layer is added after each layer to reduce redundant information, while the feature extracted in the first step of the present invention is effective information, if the pooling layer is added after each layer of convolution layer, part of the information may be lost. Thus, there is no need to add a pooling layer after each convolution layer, but instead add a global pooling layer after all convolution layers, because the network's mining of the radiation source information is sufficient at this time, and the goal of adding global pooling is simply to reduce the data size and speed up the network convergence. Experiments prove that the network structure is truly beneficial to the identification of the radiation source individuals.
The network provided by the invention contains 6 tens of millions of parameters and 65000 neurons, a five-layer convolution layer and three-layer full-connection layer network, and the final output layer is a softmax layer. A first layer: convolutional layer 1, the number of convolutional kernels is 96, the size of the convolutional kernels is 11 x 11, step size stride=4, and spreading=0; a second layer: the convolution layer 2 inputs a 'feature map' of the previous layer of convolution, the number of the convolutions is 256, and the size of the convolution kernel is as follows: 5×5, padding=2, stride=1; third layer: the convolution layer 3, the input is the output of the second layer, the number of convolution kernels is 384, the convolution kernel size is 3 x 3, and padding=1; fourth layer: the convolution layer 4, the input is the output of the third layer, the number of convolution kernels is 384, the convolution kernel size is 3×3, and padding=1. Fifth layer: and the convolution layer 5 is input as the output of the fourth layer, the number of convolution kernels is 256, the convolution kernel size is 3 x 3, and padding=1. Then, carrying out maximum pooling directly, wherein the pooling size is 4 x 4, and stride=2; the 6 th, 7 th and 8 th layers are all connected layers, the number of neurons of each layer is 4096, and finally softmax is output. RELU activation functions and Dropout operations are used in the fully connected layer.
Thirdly, the incremental learning step is as follows: firstly, obtaining information such as network model parameters of a known class sample so as to facilitate incremental learning as described in the second step; second, a new data set is constructed from the old sample set and the new unknown class samples. In order to prevent the network from forgetting too much for the old sample information and prevent the network from being fitted with new type data, the invention uses less part of old data and new data to be mixed to form a new data set, adopts the principle of random selection on the selection of the old sample, and randomizes and randomly selects the original data set; finally, loading an old network model, modifying a network output layer, changing the number of output nodes into the number of radiation source individuals actually trained, reducing the step length of each layer of convolution kernel of a convolution layer and lowering the learning rate when the training recognition rate is low, adding dropout or a regular term for correction when fitting occurs, and optimizing a loss function as follows in order to prevent the catastrophic forgetting problem in incremental learning:
wherein the first term is the classifier loss function and the second term is the distillation loss function, q i Representing the nodes of the different neurons,representing the classifier. But->It is found through observation that the distillation coefficient T is added on the basis of the softmax function, and the result caused by training errors can be aggravated by setting the T value larger than 1, which is equivalent to the training under the 'load training', so that the training accuracy is higher. The new loss function obtained by combining these two terms will be much better than the loss function obtained by using only the classifier. The invention has the advantages and positive effects that: the intelligent incremental identification method for the aerial radiation source individuals can effectively realize the classification of ADS-B (broadcast automatic correlation monitoring) signals, has good identification performance in a low signal-to-noise ratio environment, reduces the dependency on a feature extraction method, adopts an incremental learning method, and greatly reduces the training time and the space required by storage.
The invention verifies the identification method of the air target radiation source through a simulation experiment, wherein the simulation signal contains fingerprint information such as frequency offset, phase distortion, harmonic distortion and the like mentioned in the foregoing, the simulation signal uses an ADS-B (broadcast automatic correlation monitoring) signal of an S-mode transponder extension message (1090ES,1090MHz Mode S Extended Squitter), the working frequency is 1090MHz, the data rate is 1Mbps, and the modulation mode is Pulse Position Modulation (PPM) and binary amplitude keying (2 ASK). The period of the signal was 120 μs, the preamble pulse 8 μs, 4 pulses with a duration of 0.5 μs, and the start times were 0.1 μs, 3.5 μs, and 4.5 μs, respectively. The information pulse occupies 112 mu s, 112 bits of data are transmitted, one bit of data represents a message, the information including the position, altitude, speed, heading, identification number and the like of the airplane is contained, and 01 and 10 binary data used after PPM modulation represent each message. The invention intercepts data with the length of 5 mu s of the ADSB signal leading pulse, uses 600MHz frequency for sampling, sets the signal-to-noise ratio range as-5 dB, sets 3 different individuals in total, and each individual generates 5 different signals, wherein the target 1 phase noise modulation frequency is respectively set as 4MHz, 6MHz, 7MHz, 10MHz and 15MHz, the phase modulation coefficient is respectively set as 0.16,0.27,0.32,0.15,0.25, and the first harmonic component to the fifth harmonic component is set as 1,0.5,0.3,0.2,0.1. The target 2 phase noise modulation frequency is respectively 2MHz, 5MHz, 9MHz, 11MHz and 13MHz, the phase modulation coefficient is respectively 0.21, 0.32,0.15, 0.24 and 0.28, and the first harmonic component to the fifth harmonic component are respectively 1, 0.8, 0.6, 0.4 and 0.2. The target 3-phase noise modulation frequency is respectively 3MHz, 5MHz, 6MHz, 8MHz and 12MHz, the phase modulation coefficient is respectively 0.34, 0.3, 0.23, 0.21 and 0.26, and the first harmonic component to the fifth harmonic component are respectively 1, 0.1, 0.08, 0.05 and 0.03.
Fig. 3 shows the accuracy of individual identification of the radiation source by using four feature extraction methods and different combination methods, and it can be seen that the accuracy of identification reaches 95% when the signal-to-noise ratio is greater than 3dB by using the four feature combination methods.
It should be noted that the embodiments of the present invention can be realized in hardware, software, or a combination of software and hardware. The hardware portion may be implemented using dedicated logic; the software portions may be stored in a memory and executed by a suitable instruction execution system, such as a microprocessor or special purpose design hardware. Those of ordinary skill in the art will appreciate that the apparatus and methods described above may be implemented using computer executable instructions and/or embodied in processor control code, such as provided on a carrier medium such as a magnetic disk, CD or DVD-ROM, a programmable memory such as read only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The device of the present invention and its modules may be implemented by hardware circuitry, such as very large scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., as well as software executed by various types of processors, or by a combination of the above hardware circuitry and software, such as firmware.
The foregoing is merely illustrative of specific embodiments of the present invention, and the scope of the invention is not limited thereto, but any modifications, equivalents, improvements and alternatives falling within the spirit and principles of the present invention will be apparent to those skilled in the art within the scope of the present invention.
Claims (7)
1. The method for identifying the intelligent increment of the aerial radiation source individual is characterized by comprising the following steps of:
extracting four features of fuzzy function, bispectrum transformation, hilbert yellow transformation and short-time Fourier transformation from the received ADS-B signal of the aerial radiation source respectively, and carrying out linear feature fusion on the features to obtain a new feature map;
classifying and identifying known types of radiation source individuals through a convolutional neural network to obtain a network model;
training the untrained category data in an incremental learning mode, namely training a network by taking a small part of old data and new data together on the basis of the original training result, and modifying part of parameters and loss functions in the network to realize intelligent incremental identification of an air radiation source individual;
the four feature extraction methods are specifically implemented as follows:
(1) Short-time fourier transform STFT: the short-time fourier transform expression is:
wherein the subscript n is different from the standard Fourier transform, w (n-m) is a window function sequence, x (m) is an input sequence, and different window function sequences can obtain different Fourier transform results; the short-time fourier transform has two independent variables n and w, both a discrete function with respect to time n and a continuous function with respect to angular frequency w;
(2) Fuzzy function for analyzing and designing various signals, identifying different signals, delay difference tau, frequency shift mu, input signal X 1 (t)、X 2 (t) the ambiguity function is defined as:
(3) The double spectrum conversion is carried out, the double spectrum of the received signal r (n) is estimated by a non-parametric method, r (n) is firstly divided into gamma segments, each segment comprises delta samples, and the third-order cyclic cumulative quantity of the received signal r (n) is written as:
wherein χ is γ (τ 1 ,τ 2 ) Is the third-order cyclic accumulation of each segment of signal, τ 1 ,τ 2 For different time delays, w (τ 1 ,τ 2 ) Is a hexagonal window function, and the bispectral estimate of the signal r (n) is expressed as:
(4) The method for combining the Hilbert yellow transformation with the empirical mode decomposition comprises the following basic principles: first, find the upper envelope X of the original signal X (t) max (t) and lower envelope X min (t) and averaging the upper and lower envelopes:
second, for the original signal X (t) and the average envelope m 1 (t) subtracting to obtain the remaining signal d 1 (t); for the remaining signal d 1 (t) repeating the above operation until the screening threshold (SD) is smaller, at which point the final suitable first-order modal component c is obtained 1 (t), i.e., a first IMF, wherein the SD solution is as follows:
again, for signals X (t) and c 1 (t) obtaining a first-order residual quantity r by taking a difference 1 (t) r is to 1 (t) substituting the original signal X (t) to perform the above operation, repeating n times to obtain an n-order modal function c n (t) and the final standard-compliant residual amount r n (t) the expression of the original signal X (t) decomposed by empirical mode decomposition is:
finally, performing time-frequency processing on the signals subjected to empirical mode decomposition by using Hilbert transformation, and converting the signals into Hilbert spectrograms;
the method for obtaining the new feature map by linear feature fusion of the features is implemented as follows: expanding the four features to the same size by adopting an interpolation zero padding method, and then carrying out linear feature fusion on the four features to finally obtain a new feature map; if the individual difference of the radiation sources reflected by a certain characteristic extraction method is not obvious, the neural network judges which radiation source individual belongs to through other characteristics;
training the untrained category data in an incremental learning mode; based on the original training result, taking a small amount of old data and new data to train the network together, and modifying partial weight values and loss functions in the network, thereby realizing intelligent incremental identification of the air radiation source individuals; the method specifically comprises the following steps:
firstly, obtaining network model parameter information of a known class sample;
secondly, constructing a new data set by the old sample set and the new unknown class sample; mixing a small amount of old data with new data to form a new data set, and randomly selecting the original data set by adopting a random selection principle on the selection of old samples;
finally, loading an old network model, modifying a network output layer, changing the number of output nodes into the number of radiation source individuals actually trained, reducing the step length of each layer of convolution kernel of a convolution layer and lowering the learning rate when the training recognition rate is low, adding dropout or a regularization term for correction when fitting occurs, and optimizing a loss function as follows:
wherein the first term is the classifier loss function and the second term is the distillation loss function, q i Representing the nodes of the different neurons,representing the classifier.
2. The method for intelligent incremental identification of an individual source of radiation in air of claim 1 wherein said classifying and identifying the individual source of radiation of a known class by convolutional neural network comprises:
firstly, the characteristic extraction is carried out on the input characteristic diagram by a convolution layer, the conventional processing of the convolution neural network is usually picture data, the characteristic extraction in four modes is carried out on signals, the characteristic fusion is carried out on the signals, and each signal can obtain a four-channel data characteristic diagram which is also suitable for the processing of the convolution neural network; the convolution layer comprises a plurality of convolution kernels, each element composing the convolution kernels corresponds to a weight coefficient and a deviation, each neuron in the convolution layer is connected with a plurality of neurons in a region close to the position in the previous layer, the size of the region depends on the size of the convolution kernels, and a calculation formula is as follows:
wherein b is the deviation, ω is the weight, x and y represent the convolution check to convolve the whole feature map, Z l And Z l+1 Convolved inputs and outputs representing layer l+1, also called feature maps, L l+1 Is Z l+1 Assuming equal feature pattern length and width; z (i, j) corresponds to the pixel of the feature map, K is the number of channels of the feature map, fIs the convolution kernel size, s 0 Is the convolution step size, p is the padding size; the convolutional layer contains excitation functions to help express complex features, and the linear rectification function RELU is used to enable neurons in the neural network to have sparse activation, and the expression form is as follows:
secondly, after the feature extraction is carried out by the convolution layer, the output features are transmitted to the pooling layer for feature selection and information filtering; adding pooling operation after the convolution layer, wherein the pooling layer selects pooling areas which are the same as the steps of the convolution kernel scanning feature map, and the pooling areas are controlled by pooling size, step length and filling, and the expression is as follows:
right middle A k Representing an input feature map, and other parameters being the same as a convolution layer;
and finally, inputting the output of the pooling layer into the full-connection layer to perform nonlinear combination on the extracted features so as to obtain output.
3. A computer device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the method of intelligent incremental identification of individual sources of airborne radiation as claimed in any one of claims 1 to 2.
4. A computer readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the method of individual intelligent incremental identification of an airborne radiation source of any one of claims 1-2.
5. An information data processing terminal, characterized in that the information data processing terminal is used for realizing the intelligent incremental identification method of the aerial radiation source individual according to any one of claims 1-2.
6. An intelligent delta identification system for implementing the intelligent delta identification method of any one of claims 1-2, wherein the aerial radiation source individual intelligent delta identification system comprises:
the feature combination module is used for carrying out Hilbert yellow transformation, short-time Fourier transformation, fuzzy function and bispectrum transformation feature extraction on the received ADS-B signals and combining the features into a new feature;
the radiation source individual identification module is used for realizing radiation source individual identification through a convolutional neural network;
and the incremental training module is used for performing incremental training on the network by adopting an incremental learning method.
7. A radiation source individual identification terminal, characterized in that the radiation source individual identification terminal is used for implementing the aerial radiation source individual intelligent incremental identification method according to any one of claims 1-2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011621625.1A CN112668498B (en) | 2020-12-30 | 2020-12-30 | Method, system, terminal and application for identifying individual intelligent increment of aerial radiation source |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011621625.1A CN112668498B (en) | 2020-12-30 | 2020-12-30 | Method, system, terminal and application for identifying individual intelligent increment of aerial radiation source |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112668498A CN112668498A (en) | 2021-04-16 |
| CN112668498B true CN112668498B (en) | 2024-02-06 |
Family
ID=75412118
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011621625.1A Active CN112668498B (en) | 2020-12-30 | 2020-12-30 | Method, system, terminal and application for identifying individual intelligent increment of aerial radiation source |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112668498B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113435245B (en) * | 2021-05-18 | 2023-06-30 | 西安电子科技大学 | Method, system and application for identifying individual aerial radiation source |
| CN113435246B (en) * | 2021-05-18 | 2024-04-05 | 西安电子科技大学 | Intelligent radiation source individual identification method, system and terminal |
| CN113923088B (en) * | 2021-10-21 | 2023-08-08 | 天津光电通信技术有限公司 | HLNN-based automatic 5G signal digital modulation mode identification method |
| CN114925748A (en) * | 2022-04-20 | 2022-08-19 | 北京市商汤科技开发有限公司 | Model training and modal information prediction method, related device, equipment and medium |
| CN114664290B (en) * | 2022-05-17 | 2022-08-19 | 深圳比特微电子科技有限公司 | Sound event detection method and device and readable storage medium |
| CN116559809A (en) * | 2023-03-28 | 2023-08-08 | 南京桂瑞得信息科技有限公司 | Radar radiation source individual identification method based on multi-source fusion network |
| CN117544963B (en) * | 2024-01-04 | 2024-03-26 | 四川大学 | Method and equipment for identifying radiation source of cross-mode communication signal based on FTGan-Yolo |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108090412A (en) * | 2017-11-17 | 2018-05-29 | 西北工业大学 | A kind of radar emission source category recognition methods based on deep learning |
| CN109492765A (en) * | 2018-11-01 | 2019-03-19 | 浙江工业大学 | A kind of image Increment Learning Algorithm based on migration models |
| GB201906560D0 (en) * | 2018-08-24 | 2019-06-26 | Petrochina Co Ltd | Method and apparatus for automatically extracting image features of electrical imaging well logging |
| KR20190110939A (en) * | 2018-03-21 | 2019-10-01 | 한국과학기술원 | Environment sound recognition method based on convolutional neural networks, and system thereof |
| CN110472545A (en) * | 2019-08-06 | 2019-11-19 | 中北大学 | The classification method of the power components image of taking photo by plane of knowledge based transfer learning |
| CN112087774A (en) * | 2020-09-14 | 2020-12-15 | 桂林电子科技大学 | Communication radiation source individual identification method based on residual error neural network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11080558B2 (en) * | 2019-03-21 | 2021-08-03 | International Business Machines Corporation | System and method of incremental learning for object detection |
-
2020
- 2020-12-30 CN CN202011621625.1A patent/CN112668498B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108090412A (en) * | 2017-11-17 | 2018-05-29 | 西北工业大学 | A kind of radar emission source category recognition methods based on deep learning |
| KR20190110939A (en) * | 2018-03-21 | 2019-10-01 | 한국과학기술원 | Environment sound recognition method based on convolutional neural networks, and system thereof |
| GB201906560D0 (en) * | 2018-08-24 | 2019-06-26 | Petrochina Co Ltd | Method and apparatus for automatically extracting image features of electrical imaging well logging |
| CN109492765A (en) * | 2018-11-01 | 2019-03-19 | 浙江工业大学 | A kind of image Increment Learning Algorithm based on migration models |
| CN110472545A (en) * | 2019-08-06 | 2019-11-19 | 中北大学 | The classification method of the power components image of taking photo by plane of knowledge based transfer learning |
| CN112087774A (en) * | 2020-09-14 | 2020-12-15 | 桂林电子科技大学 | Communication radiation source individual identification method based on residual error neural network |
Non-Patent Citations (2)
| Title |
|---|
| 何丽 ; 韩克平 ; 朱泓西 ; 刘颖 ; .双分支迭代的深度增量图像分类方法.模式识别与人工智能.2020,(第02期),全文. * |
| 高欣宇 ; 张文博 ; 姬红兵 ; 欧阳成 ; .新型雷达辐射源识别.中国图象图形学报.2020,(第06期),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112668498A (en) | 2021-04-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112668498B (en) | Method, system, terminal and application for identifying individual intelligent increment of aerial radiation source | |
| CN110865357B (en) | Laser radar echo signal noise reduction method based on parameter optimization VMD | |
| CN108845306B (en) | Laser radar echo signal denoising method based on variational modal decomposition | |
| CN110532932B (en) | Method for identifying multi-component radar signal intra-pulse modulation mode | |
| CN110673109B (en) | Full waveform data decomposition method for satellite-borne large-light-spot laser radar | |
| CN112560803A (en) | Radar signal modulation identification method based on time-frequency analysis and machine learning | |
| Liu | Multi-feature fusion for specific emitter identification via deep ensemble learning | |
| CN114025379A (en) | Broadband multi-signal detection method, device and equipment | |
| CN113298846A (en) | Intelligent interference detection method based on time-frequency semantic perception | |
| CN112949387A (en) | Intelligent anti-interference target detection method based on transfer learning | |
| Nuhoglu et al. | Image segmentation for radar signal deinterleaving using deep learning | |
| CN112859012A (en) | Radar deception jamming identification method based on cascade convolution neural network | |
| CN110632588B (en) | Zero intermediate frequency secondary radar direct current offset compensation algorithm based on FPGA | |
| CN111046697A (en) | Adaptive modulation signal identification method based on fuzzy logic system | |
| CN111951611A (en) | ADS-B weak signal detection device and method based on multi-feature fusion | |
| CN116626631A (en) | Automatic radar model identification method and system combining intra-pulse and inter-pulse characteristics | |
| KR101427149B1 (en) | A method of recognizing complex PRI modulation type of radar signals based on 2-layer SVM using exponential moving average | |
| CN115980689A (en) | Point cloud detection-based radiation source signal sorting method, device, equipment and medium | |
| Guven et al. | Classifying LPI radar waveforms with time-frequency transformations using multi-stage CNN system | |
| CN114710215A (en) | Method for fast blind detection of frequency hopping signal | |
| Wang et al. | Modulation Recognition of Overlapping Radar Signals under Low SNR Based on Se-Incepatnet | |
| Ai et al. | Radar Signal Separation Recognition Method based on Semantic Segmentation | |
| Wang et al. | Strategies for Enhanced Signal Modulation Classifications Under Unknown Symbol Rates and Noise Conditions | |
| CN113435246B (en) | Intelligent radiation source individual identification method, system and terminal | |
| CN116827486B (en) | Blind detection system and method for short-wave communication signals |
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 |