CN112560596B - Radar interference category identification method and system - Google Patents
Radar interference category identification method and system Download PDFInfo
- Publication number
- CN112560596B CN112560596B CN202011381830.5A CN202011381830A CN112560596B CN 112560596 B CN112560596 B CN 112560596B CN 202011381830 A CN202011381830 A CN 202011381830A CN 112560596 B CN112560596 B CN 112560596B
- Authority
- CN
- China
- Prior art keywords
- data
- sample data
- sample
- training set
- feature vectors
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- 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
- G06N3/045—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
-
- 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 Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Signal Processing (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The application discloses a radar interference category identification method and a radar interference category identification system, which belong to the technical field of data processing and are used for solving the problem that the deep learning technology is not widely applied in the interference identification field because the actual accumulated data sample size of the conventional deep learning technology can not meet the requirement of deep learning on training data. The method comprises the following steps: extracting quantitative radar deception jamming data as new feature vectors, and selecting a preset number of feature vectors from the feature vectors as a training set; generating sample data, and training and adjusting the sample data by adopting a generation countermeasure network to obtain candidate sample data; optimizing candidate sample data by adopting an active learning algorithm, and selecting a target sample from the optimized candidate sample data to form a target sample training set; and constructing a robust recognition model, and recognizing interference types based on the target sample set. The method has reliable calculation result and can be particularly applied to radar interference suppression data analysis.
Description
Technical Field
The application belongs to the technical field of data processing, and particularly relates to a radar interference category identification method and system.
Background
Modern war novel electronic interference has serious influence on radar detection and tracking performance, and accurate and efficient identification of various novel electronic interferences is always key and difficult for electronic countermeasure research. The current research on interference category identification mainly comprises a traditional method based on feature design and extraction and a deep learning technology for automatically extracting features. In the traditional method, feature expression is completed through manual design, so that time consumption of data processing is caused, the method is strongly dependent on expertise and the characteristics of data, and it is difficult to fully mine the association between data. At present, the relatively advanced method can automatically learn the characteristics, and the artificial intelligence technology represented by deep learning can effectively extract semantic characteristics with high robustness and high characterization on information, and has excellent characteristic learning capability. However, the deep learning technique is not widely used in the field of interference recognition, because the actually accumulated data sample size is far from meeting the requirement of deep learning on training data. Therefore, there is an urgent need for a radar interference class identification method that is feasible and effective under small sample conditions, and has strong learning and generalization capabilities.
Disclosure of Invention
The application aims to provide a radar interference type recognition method and a radar interference type recognition system, which are used for solving the problem that the deep learning technology is not widely applied in the interference recognition field because the actual accumulated data sample size of the conventional deep learning technology can not meet the requirement of deep learning on training data.
In order to achieve the above object, the present application provides the following technical solutions:
the radar interference category identification method comprises the following steps:
step S10: extracting quantitative radar deception jamming data as new feature vectors, and selecting a preset number of feature vectors from the feature vectors as a training set;
step S20: generating sample data, and training and adjusting the sample data by adopting a generation countermeasure network to obtain candidate sample data;
step S30: optimizing the candidate sample data by adopting an active learning algorithm, and selecting a target sample from the optimized candidate sample data to form a target sample training set;
step S40: and constructing a robust recognition model, and recognizing interference types based on the target sample set.
According to the radar interference category identification method, preprocessing operation can be carried out on radar interference signal data, error data in the acquired radar deception interference data are removed, larger errors caused by different dimensions in the calculation process are avoided, the convergence speed of a model is increased, and the accuracy of the model is improved; the heuristic active learning method of posterior probability and the generation of the countermeasure network are combined into a new network structure, the data set is supplemented in a targeted manner, and the problems of small scale and insufficient diversity of the training sample set are effectively solved; the number of the depth forests based on multi-granularity feature extraction is far less than that required by the deep learning, and the generalization capability of the model is stronger than that of the deep learning method on a small-scale sample, so that the migration capability of the model is improved; in engineering application, once the model parameters are determined, retraining is not needed, and efficiency is effectively improved. The method has the advantages of simple algorithm principle and reliable calculation result, can be particularly applied to radar interference suppression data analysis, and has wide application value and market prospect.
The radar interference category identification method which is feasible and effective under the condition of a small sample and has strong learning ability and generalization ability has high practicability.
The application also provides a radar interference category recognition system, which comprises a feature vector extraction module, a sample data generation module, a sample data optimization module and a recognition module;
the feature vector extraction module is used for extracting quantitative radar deception jamming data as new feature vectors, and selecting a preset number of feature vectors from the feature vectors as a training set;
the sample data generation module is used for generating sample data, and training and adjusting the sample data by adopting a generation countermeasure network to obtain candidate sample data;
the sample data optimizing module is used for optimizing the candidate sample data by adopting an active learning algorithm, and selecting a target sample from the optimized candidate sample data to form a target sample training set;
the identification module is used for constructing a robust identification model and carrying out interference type identification based on the target sample set.
The radar interference category recognition system corresponds to the beneficial effects obtained by the radar interference category recognition method, and discussion is not repeated here.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the application and do not constitute a limitation on the application. In the drawings:
FIG. 1 is a schematic flow chart of a radar interference class identification method provided by the application;
figure 2 is a schematic diagram of small sample data expansion based on generation of an countermeasure network in the radar interference category recognition method provided by the application,
FIG. 3 is a diagram of a generated network in the radar interference category identification method provided by the application;
fig. 4 is a diagram of a discrimination network in the radar interference category identifying method provided by the application;
fig. 5 is a schematic block diagram of a radar interference class identification system provided by the present application.
Detailed Description
In order to make the technical problems, technical schemes and beneficial effects to be solved more clear, the application is further described in detail below with reference to the accompanying drawings and embodiments. 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 application.
It will be understood that when an element is referred to as being "mounted" or "disposed" on another element, it can be directly on the other element or be indirectly on the other element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or be indirectly connected to the other element.
Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. In the description of the present application, the meaning of "a plurality" is two or more, unless explicitly defined otherwise. The meaning of "a number" is one or more than one unless specifically defined otherwise.
In the description of the present application, it should be understood that the directions or positional relationships indicated by the terms "upper", "lower", "front", "rear", "left", "right", etc., are based on the directions or positional relationships shown in the drawings, are merely for convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as limiting the present application.
In the description of the present application, it should be noted that, unless explicitly specified and limited otherwise, the terms "mounted," "connected," and "connected" are to be construed broadly, and may be either fixedly connected, detachably connected, or integrally connected, for example; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium, and can be communicated with the inside of two elements or the interaction relationship of the two elements. The specific meaning of the above terms in the present application can be understood by those of ordinary skill in the art according to the specific circumstances.
The embodiment of the application provides a radar interference category identification method and a radar interference category identification system under a small sample condition, which comprehensively analyze a method for combining an antagonistic network with a deep forest aiming at radar interference signal data processing requirements, have stronger learning ability and generalization ability, meet engineering application requirements, and have wide application value and market prospect.
As shown in fig. 1 to 4, the radar interference category identification method provided by the application comprises the following steps:
step S10: extracting quantitative radar deception jamming data as new feature vectors, and selecting a preset number of feature vectors from the feature vectors as a training set;
step S20: generating sample data, and training and adjusting the sample data by adopting a generation countermeasure network to obtain candidate sample data;
step S30: optimizing candidate sample data by adopting an active learning algorithm, and selecting a target sample from the optimized candidate sample data to form a target sample training set;
step S40: and constructing a robust recognition model, and recognizing interference types based on the target sample set.
The specific implementation method comprises the following steps:
the intelligent recognition method for the radar interference category under the condition of the small sample is applied to radar interference signal data analysis, and can be specifically executed by referring to the following steps:
acquiring radar deception jamming data and generating a feature vector;
generating a sample expansion;
optimizing the generated sample;
and (5) interference identification and classification.
The specific implementation details of each part are as follows:
1. obtaining radar deception jamming data and generating feature vectors
Taking the first 1500 dimensions of the radar echo signal amplitude characteristics to form a new characteristic vector. In all the feature vectors, 80% of the total number is taken as a training set, the number of the feature vectors in the training set is recorded as M, and the remaining 20% is taken as a test set.
2. Generating sample extensions
Aiming at the problem of lack of sample data, and referring to fig. 2 in detail, the application adopts the generation of a countermeasure network to generate a series of high-quality samples for model training. The generation of the countermeasure network is mainly composed of two parts, namely a generator and a decision maker.
The goal of the generator is to generate data samples that are similar to the real samples; the aim of the arbiter is to be able to correctly distinguish whether its input information is generated by the generator or is actually present. The mode is shown in the following figure, the generator generates some sample data, the discriminator learns to distinguish the generated sample data from real sample data, the generator improves the generator according to the result of the discriminator, the generation is realized through training fine tuning, and finally, new sample data is regenerated, so that the discriminator can discriminate. The loop is ended until the generator and the arbiter cannot be lifted, i.e. the arbiter cannot judge whether the data is generated or real, and the generated model becomes a robust model.
The specific flow is as follows:
(1) Generator and network design
The generator receives a noise signal, generates sample data based on the signal, and inputs the sample data to the arbiter. This section uses a network of convolution structures making it more suitable for processing feature inputs. The whole generator is shown in the following formula:
where r is a random noise and g converts the random noise to data type x, D is a discriminant model, and the output to any input x, D (x) is a real number in the range of 0-1. P (P) r And P g Representing the distribution of the real features and the distribution of the generated features, respectively. Specifically, the transposed convolution is used to transform the noise signature into a data source having the same shape as the input by applying a method such as reshaping the noise. And then normalizing the input of the current layer by adopting a group normalization method to ensure that the mean value of the input is 0 and the variance of the input is 1, thereby realizing the acceleration convergence of the convolutional neural network, ensuring that the convolutional neural network added with the group normalization has very little influence of the initialization of the weight, has very good stability and has very good effect on improving the convolutional performance. After the group normalization is completed, the related characteristic information is obtained by using an activation function, and then the regularized characteristic information is added into the regularized removal dataRedundancy. Finally, the set activation function is used for output in the last layer.
The stronger the generator's ability, the better the data distribution that it generates can fit the real data distribution. The generator network design is shown in fig. 3.
(2) Discriminator and network design
The discriminator receives the data and the real data generated by the generator and outputs a discrimination result. At this time, the arbiter may be regarded as a classifier including a convolutional neural network for object recognition. The calculation loss term calculates the calculation loss of the generator and the calculation loss of the discriminator respectively, and the smoothing function is added simultaneously, so that the overfitting can be prevented, and the generalization capability is enhanced. Pooling layer pooling is negligible in the design process, mainly because the features themselves are very small after multi-layer convolution, and group normalization acceleration training is applied, and feature extraction acceleration training by pooling is not needed. The optimization process is realized by maximum and minimum objective functions, and the formula is as follows:
the arbiter network structure is shown in fig. 4.
(3) Alternate training method for generation and countermeasure
The random noise randomly initializes the generating network to generate a series of generating samples, the training discriminator discriminates true and false, and the error is transmitted forward to the training generator. At this time, the network parameters are fixed and judged, and only the network parameters are updated and generated. Training results in the generation of more realistic false samples, and sequentially continues iteration until the model converges.
3. Optimizing the generated sample
The samples generated in the step 2 are required to be optimized, so that samples favorable for improving the classification effect are searched, the size of the classification training set is reduced, and the efficiency of the classification algorithm is improved on the premise of limited time and resources. The application can realize further screening and optimizing of the generated sample through the active learning technology, and the active learning algorithm can be modeled by the following five components:
A＝(C,L,S,Q,U) (3)
wherein C is one or a set of classifiers; l is a group of marked training sample sets; q is a query function used for querying samples with large information quantity in unlabeled samples; u is the whole unlabeled sample set; s is a supervisor, and can label unlabeled samples. The algorithm is mainly divided into two stages, wherein the first stage is an initialization stage, a small part is selected from unlabeled samples at random, labeled by a supervisor and used as a training set to establish an initial classifier model; the second stage is a cyclic query stage, wherein S is a non-labeled sample set U, a certain non-labeled sample is selected for labeling according to a certain query standard Q, and is added into a training sample set L, and the classifier is retrained until reaching a training stop standard.
4. Interference identification and classification
Based on the optimized sample obtained in the step 3, the application provides a method for constructing a robust recognition model by adopting multi-granularity scanning and an improved cascading depth forest method to realize interference category recognition.
The method comprises the following specific steps:
(1) Multi-granularity feature extraction
Multi-granularity scanning forests are designed to extract more object features. Multiple granularity scanning is to slide the values on the data by using windows with different sizes, and each value has the same label for the same type of data. In the process of multi-granularity scanning, the number of forests and the number of layers of forests belong to two super parameters. The depth forest has the advantages of insensitivity to the super parameters, wherein the depth of the super parameters forest is 2, the breadth is 2, and the categories are respectively a completely random forest and a random forest.
(2) Cascade forest-based interference identification
Cascading forests are integrated forest structures designed for interference classification recognition. The first stage of the cascade forest accepts as input the conversion features from the multi-granularity scan results. The input accepted by each stage is the output of the previous stage forest and the splicing of the original features, and the number of forest layers is continuously deepened until the model converges on the verification set. And taking the characteristic average value obtained by the forest output of the final stage as the final output of the model. In cascade forests, the types, the numbers and the depths of forests all belong to super parameters, 2 types of forests are adopted due to the characteristic of insensitive super parameters of the depth forests, the breadth is 4, the depth is 8, and the number of decision trees is 500. After 1806 dimensions of the final output 1806 of multi-granularity scanning pass through the first forest layer with the breadth of 4, 12 features are obtained, and similarly 1206 dimensions and 606 dimensions respectively obtain 12 features. These 12 features are respectively spliced with the multi-granularity scanned (1818D, 1218D, 618D) conversion features to serve as input features of the next hierarchical forest. The stopping condition of the iterative optimization of the cascade forest is determined by the verification set and the depth of the forest. The reasonable verification set can ensure that the model has certain generalization capability and simultaneously maximizes the accuracy of the model. The verification method is to select some or all layers, observe the output results of the selected layers, and verify the output gains of the layers. When the output gain is no longer increasing, training is stopped, and the output of the current layer is used as the last layer of the model output.
(3) Training an identification model overall process
And carrying out feature extraction on the data subjected to the multi-granularity scanning structure. After the data is subjected to 3-5 sliding windows (granularity) with different sizes to select characteristics, the obtained conversion characteristics have enough difference. In the process of selecting the features by the sliding window, if the original features are too long, the pooling layer can be used for sampling the original features. The conversion characteristics obtained by multi-granularity scanning are used for extracting characteristics of cascade forests, the conversion characteristics of the first-stage forests with different dimensionalities are accepted as input, and the output is probability of belonging to different categories; the input of each stage is the splicing of the output of the previous stage and the conversion characteristics obtained by multi-granularity scanning, the steps are circulated in this way, a multi-stage forest is constructed, and the average value of probability results of each category of the output of the last layer of forest is used as the final output of the model.
According to the radar interference category identification method, preprocessing operation can be carried out on radar interference signal data, error data in the acquired radar deception interference data are removed, larger errors caused by different dimensions in the calculation process are avoided, the convergence speed of a model is increased, and the accuracy of the model is improved; the heuristic active learning method of posterior probability and the generation of the countermeasure network are combined into a new network structure, the data set is supplemented in a targeted manner, and the problems of small scale and insufficient diversity of the training sample set are effectively solved; the number of the depth forests based on multi-granularity feature extraction is far less than that required by the deep learning, and the generalization capability of the model is stronger than that of the deep learning method on a small-scale sample, so that the migration capability of the model is improved; in engineering application, once the model parameters are determined, retraining is not needed, and efficiency is effectively improved. The method has the advantages of simple algorithm principle and reliable calculation result, can be particularly applied to radar interference suppression data analysis, and has wide application value and market prospect.
As an embodiment, in step S10: extracting quantitative radar deception jamming data as new feature vectors, and selecting a preset number of feature vectors from the feature vectors as a training set; step S20: generating sample data, training and adjusting the sample data by adopting a generating countermeasure network, and before obtaining candidate sample data, further comprising the following steps:
step S15: and preprocessing the feature vector data in the training set, and eliminating invalid data.
Further, step S15: preprocessing the feature vector data in the training set, and removing invalid data, wherein the method comprises the following steps:
step S151: determining the integrity of the extracted radar deception jamming data and the data in the training set, and carrying out data complementation on missing data by adopting a form of a characteristic quantity mean value corresponding to the missing data;
step S152: and normalizing the complemented feature vector data.
The preprocessing of the feature vector data in the training set can remove invalid data and improve the effectiveness of the data, and the preprocessing can be mainly realized by the following steps:
(1) Complement missing data
In the process of data acquisition and recording, a phenomenon that a certain piece of data is missing can occur, and the missing data can be replaced by the average value of the corresponding characteristic quantity of the data.
(2) Normalization processing
Normalization is a common method of data preprocessing. The normalization processing of the data mainly has two purposes, namely, the dimension of the data is removed, larger errors caused by different dimensions in the calculation process are avoided, and the convergence speed of the model is increased. The application normalizes the data between [0,1], the normalized expression is as follows:
wherein x is ij Is the ith feature of the jth sample,the maximum and minimum values of the ith feature, respectively.
Through the preprocessing of the feature vector data in the training set, the accuracy and the integrity of experimental data are effectively ensured.
As an embodiment, step S20: generating sample data, training and adjusting the sample data by adopting a generating countermeasure network to obtain candidate sample data, and comprising the following steps of:
step S21: generating simulated sample data similar to real sample data in the training set;
step S22: judging the difference between the simulated sample data and the real sample data;
step S23: and carrying out improvement adjustment on the simulated sample data according to the judgment difference until the difference between the simulated sample data and the real sample data cannot be judged, so as to obtain candidate sample data.
The generated simulated sample data and the real sample data are compared and judged, so that the simulated sample data can be effectively improved and adjusted, the accuracy of the final candidate sample data is ensured, and the recognition effect of the radar interference category recognition method is improved.
As an embodiment, step S30: optimizing candidate sample data by adopting an active learning algorithm, selecting a target sample from the optimized candidate sample data to form a target sample training set, and comprising the following steps:
step S31: predicting posterior probability values of candidate sample data according to a heuristic active learning algorithm of posterior probability;
step S32: ranking the candidate sample data according to the size of the posterior probability value;
step S33: and determining an uncertain region by analyzing the change of posterior probability or each type of distribution condition of each candidate sample data, and selecting samples from the uncertain region to form a target sample training set.
Through optimizing candidate sample data, samples favorable for improving the classification effect can be searched, the size of a classification training set is further reduced, and the efficiency of a classification algorithm is improved on the premise of limited time and resources. In specific implementation, a heuristic active learning algorithm based on posterior probability is adopted, and the posterior probability reflects the certainty factor of the sample category. The algorithm sorts the candidate sample sets according to the magnitude of the posterior probability value of the predicted sample. And determining an uncertain region by analyzing the change of posterior probability or each type of distribution condition of each candidate sample, and selecting samples from the uncertain region to form a training set.
The application also provides a radar interference category recognition system, which comprises a feature vector extraction module, a sample data generation module, a sample data optimization module and a recognition module;
the feature vector extraction module is used for extracting quantitative radar deception jamming data as new feature vectors, and selecting a preset number of feature vectors from the feature vectors as a training set;
the sample data generation module is used for generating sample data, training and adjusting the sample data by adopting a generation countermeasure network to obtain candidate sample data;
the sample data optimizing module is used for optimizing candidate sample data by adopting an active learning algorithm, and selecting a target sample from the optimized candidate sample data to form a target sample training set;
and the identification module is used for constructing a robust identification model and carrying out interference type identification based on the target sample set.
The radar interference type recognition system can preprocess radar interference signal data, eliminates error data in the acquired radar deception interference data, avoids larger errors caused by different dimensions in the calculation process, accelerates the convergence speed of the model, and improves the accuracy of the model; the heuristic active learning method of posterior probability and the generation of the countermeasure network are combined into a new network structure, the data set is supplemented in a targeted manner, and the problems of small scale and insufficient diversity of the training sample set are effectively solved; the number of the depth forests based on multi-granularity feature extraction is far less than that required by the deep learning, and the generalization capability of the model is stronger than that of the deep learning method on a small-scale sample, so that the migration capability of the model is improved; in engineering application, once the model parameters are determined, retraining is not needed, and efficiency is effectively improved. The method has the advantages of simple algorithm principle and reliable calculation result, can be particularly applied to radar interference suppression data analysis, and has wide application value and market prospect.
As an implementation manner, the system further comprises a data processing module, which is used for preprocessing the feature vector data in the training set and eliminating invalid data.
Further, the data processing module is configured to perform the following operations:
determining the integrity of the extracted radar deception jamming data and the data in the training set, and carrying out data complementation on missing data in a form of a characteristic quantity mean value corresponding to the missing data;
and normalizing the complemented feature vector data.
Through the preprocessing of the feature vector data in the training set, the accuracy and the integrity of experimental data are effectively ensured.
As an implementation manner, the sample data generating module is used for generating simulated sample data similar to real sample data in the training set;
judging the difference between the simulated sample data and the real sample data;
and carrying out improvement adjustment on the simulated sample data according to the judgment difference until the difference between the simulated sample data and the real sample data cannot be judged, so as to obtain candidate sample data.
As an implementation manner, the sample data optimization module is configured to predict a posterior probability value of candidate sample data according to a heuristic active learning algorithm of posterior probability;
ranking the candidate sample data according to the size of the posterior probability value;
and determining an uncertain region by analyzing the change of posterior probability or each type of distribution condition of each candidate sample data, and selecting samples from the uncertain region to form a target sample training set.
Through optimizing candidate sample data, samples favorable for improving the classification effect can be searched, the size of a classification training set is further reduced, and the efficiency of a classification algorithm is improved on the premise of limited time and resources. In specific implementation, a heuristic active learning algorithm based on posterior probability is adopted, and the posterior probability reflects the certainty factor of the sample category. The algorithm sorts the candidate sample sets according to the magnitude of the posterior probability value of the predicted sample. And determining an uncertain region by analyzing the change of posterior probability or each type of distribution condition of each candidate sample, and selecting samples from the uncertain region to form a training set.
In the description of the above embodiments, particular features, structures, materials, or characteristics may be combined in any suitable manner in any one or more embodiments or examples.
The foregoing is merely illustrative of the present application, and the present application is not limited thereto, and any person skilled in the art will readily recognize that variations or substitutions are within the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (2)
1. The radar interference category identification method is characterized by comprising the following steps:
step S10: extracting quantitative radar deception jamming data as new feature vectors, and selecting a preset number of feature vectors from the feature vectors as a training set;
step S20: generating sample data, and training and adjusting the sample data by adopting a generation countermeasure network to obtain candidate sample data; the method specifically comprises the following steps: step S21: generating simulated sample data similar to the real sample data in the training set; step S22: judging the difference between the simulated sample data and the real sample data; step S23: performing improvement adjustment on the simulated sample data according to the judgment difference until the difference between the simulated sample data and the real sample data cannot be judged, so as to obtain candidate sample data;
step S30: optimizing the candidate sample data by adopting an active learning algorithm, and selecting a target sample from the optimized candidate sample data to form a target sample training set; the method specifically comprises the following steps: step S31: predicting the posterior probability value of the candidate sample data according to a heuristic active learning algorithm of the posterior probability; step S32: ranking the candidate sample data according to the posterior probability value; step S33: determining an uncertain region by analyzing the change of the posterior probability or each type of distribution condition of each candidate sample data, and selecting samples from the uncertain region to form a target sample training set;
step S40: constructing a robust recognition model, and recognizing interference types based on the target sample training set;
in step S10: extracting quantitative radar deception jamming data as new feature vectors, and selecting a preset number of feature vectors from the feature vectors as a training set; step S20: generating sample data, training and adjusting the sample data by adopting a generating countermeasure network, and before obtaining candidate sample data, further comprising the following steps:
step S15: preprocessing the feature vector data in the training set, and removing invalid data; the method specifically comprises the following steps: step S151: determining the integrity of the extracted radar deception jamming data and the data in the training set, and carrying out data complementation on missing data by adopting a form of a characteristic quantity mean value corresponding to the missing data; step S152: and normalizing the complemented feature vector data.
2. The radar interference category recognition system is characterized by comprising a feature vector extraction module, a sample data generation module, a sample data optimization module and a recognition module;
the feature vector extraction module is used for extracting quantitative radar deception jamming data as new feature vectors, and selecting a preset number of feature vectors from the feature vectors as a training set;
the sample data generation module is used for generating sample data, and training and adjusting the sample data by adopting a generation countermeasure network to obtain candidate sample data;
the sample data optimizing module is used for optimizing the candidate sample data by adopting an active learning algorithm, and selecting a target sample from the optimized candidate sample data to form a target sample training set;
the recognition module is used for constructing a robust recognition model and recognizing interference types based on the target sample training set;
the radar interference category recognition system further comprises a data processing module, wherein the data processing module is used for preprocessing the feature vector data in the training set and removing invalid data;
the data processing module is used for determining the integrity of the extracted radar deception jamming data and the data in the training set, and carrying out data complementation by adopting a form of a characteristic quantity mean value corresponding to the missing data aiming at the missing data;
normalizing the complemented feature vector data;
the sample data generating module is used for generating simulated sample data similar to the real sample data in the training set;
judging the difference between the simulated sample data and the real sample data;
performing improvement adjustment on the simulated sample data according to the judgment difference until the difference between the simulated sample data and the real sample data cannot be judged, so as to obtain candidate sample data;
the sample data optimizing module is used for predicting the posterior probability value of the candidate sample data according to a heuristic active learning algorithm of the posterior probability;
ranking the candidate sample data according to the posterior probability value;
and determining an uncertain region by analyzing the change of the posterior probability or each type of distribution condition of each candidate sample data, and selecting samples from the uncertain region to form a target sample training set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011381830.5A CN112560596B (en) | 2020-12-01 | 2020-12-01 | Radar interference category identification method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011381830.5A CN112560596B (en) | 2020-12-01 | 2020-12-01 | Radar interference category identification method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112560596A CN112560596A (en) | 2021-03-26 |
| CN112560596B true CN112560596B (en) | 2023-09-19 |
Family
ID=75045704
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011381830.5A Active CN112560596B (en) | 2020-12-01 | 2020-12-01 | Radar interference category identification method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112560596B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113255688A (en) * | 2021-05-25 | 2021-08-13 | 电子科技大学成都学院 | Building image data processing method based on improved anti-noise robustness learning algorithm |
| CN114429156B (en) * | 2022-01-21 | 2023-06-20 | 西安电子科技大学 | Radar interference multi-domain characteristic countermeasure learning and detection recognition method |
| CN114818777B (en) * | 2022-03-18 | 2023-07-21 | 北京遥感设备研究所 | Training method and device for active angle deception jamming recognition model |
| CN115840875A (en) * | 2022-11-10 | 2023-03-24 | 北京擎天信安科技有限公司 | Millimeter wave radar abnormal signal detection method and system based on analog transducer |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6178261B1 (en) * | 1997-08-05 | 2001-01-23 | The Regents Of The University Of Michigan | Method and system for extracting features in a pattern recognition system |
| CN108520199A (en) * | 2018-03-04 | 2018-09-11 | 天津大学 | Based on radar image and the human action opener recognition methods for generating confrontation model |
| CN110146855A (en) * | 2019-06-11 | 2019-08-20 | 北京无线电测量研究所 | Radar Intermittent AF panel thresholding calculation method and device |
| CN110570353A (en) * | 2019-08-27 | 2019-12-13 | 天津大学 | Dense connection generation countermeasure network single image super-resolution reconstruction method |
| CN110764064A (en) * | 2019-11-08 | 2020-02-07 | 哈尔滨工业大学 | Radar interference signal identification method based on deep convolutional neural network integration |
| CN110826630A (en) * | 2019-11-08 | 2020-02-21 | 哈尔滨工业大学 | Radar interference signal feature level fusion identification method based on deep convolutional neural network |
| CN111077523A (en) * | 2019-12-13 | 2020-04-28 | 南京航空航天大学 | Inverse synthetic aperture radar imaging method based on generation countermeasure network |
-
2020
- 2020-12-01 CN CN202011381830.5A patent/CN112560596B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6178261B1 (en) * | 1997-08-05 | 2001-01-23 | The Regents Of The University Of Michigan | Method and system for extracting features in a pattern recognition system |
| CN108520199A (en) * | 2018-03-04 | 2018-09-11 | 天津大学 | Based on radar image and the human action opener recognition methods for generating confrontation model |
| CN110146855A (en) * | 2019-06-11 | 2019-08-20 | 北京无线电测量研究所 | Radar Intermittent AF panel thresholding calculation method and device |
| CN110570353A (en) * | 2019-08-27 | 2019-12-13 | 天津大学 | Dense connection generation countermeasure network single image super-resolution reconstruction method |
| CN110764064A (en) * | 2019-11-08 | 2020-02-07 | 哈尔滨工业大学 | Radar interference signal identification method based on deep convolutional neural network integration |
| CN110826630A (en) * | 2019-11-08 | 2020-02-21 | 哈尔滨工业大学 | Radar interference signal feature level fusion identification method based on deep convolutional neural network |
| CN111077523A (en) * | 2019-12-13 | 2020-04-28 | 南京航空航天大学 | Inverse synthetic aperture radar imaging method based on generation countermeasure network |
Non-Patent Citations (5)
| Title |
|---|
| Identification of Radar Active Interference Types Based on Three-dimensional Residual Network;Jingyi Wang等;《2021 3rd International Academic Exchange Conference on Science and Technology Innovation (IAECST)》;第1页 * |
| 基于干扰认知的智能化雷达干扰对抗系统;王鑫;《电子信息对抗技术》(第6期);第48-52页 * |
| 认知雷达对抗中的未知雷达状态识别方法;李岩等;《太赫兹科学与电子信息学报》;第16卷(第2期);第233-238页 * |
| 防空反导雷达高作战效能人机系统设计方法;董博等;《现代防御技术》;第50卷(第4期);第107-115页 * |
| 雷达有源干扰分类与识别方法研究;李娜;《中国优秀硕士学位论文全文数据库 信息科技辑》(第4期);第I136-2648页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112560596A (en) | 2021-03-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112560596B (en) | Radar interference category identification method and system | |
| CN112784881B (en) | Network abnormal flow detection method, model and system | |
| CN112165464B (en) | Industrial control hybrid intrusion detection method based on deep learning | |
| CN111353153B (en) | GEP-CNN-based power grid malicious data injection detection method | |
| CN111753881B (en) | Concept sensitivity-based quantitative recognition defending method against attacks | |
| CN109299741B (en) | Network attack type identification method based on multi-layer detection | |
| CN111598179B (en) | Power monitoring system user abnormal behavior analysis method, storage medium and equipment | |
| CN111835707B (en) | Malicious program identification method based on improved support vector machine | |
| CN112734000A (en) | Intrusion detection method, system, equipment and readable storage medium | |
| CN112738092A (en) | Log data enhancement method, classification detection method and system | |
| CN114760098A (en) | CNN-GRU-based power grid false data injection detection method and device | |
| CN114707571A (en) | Credit data anomaly detection method based on enhanced isolation forest | |
| CN114331731A (en) | PCA and RF based block chain abnormity detection method and related device | |
| Farimani et al. | An HMM for online signature verification based on velocity and hand movement directions | |
| Wani | Microarray classification using sub-space grids | |
| CN115622806B (en) | Network intrusion detection method based on BERT-CGAN | |
| CN116070137A (en) | Open set identification device and method for malicious traffic detection | |
| CN113609480B (en) | Multipath learning intrusion detection method based on large-scale network flow | |
| CN112085170A (en) | Method for detecting difference of image confrontation samples | |
| Lee et al. | A Sequential Approach to Network Intrusion Detection | |
| CN114241263B (en) | Radar interference semi-supervised open set recognition system based on generation of countermeasure network | |
| Kumová | Creating Adversarial Examples in Machine Learning | |
| Wen et al. | Interpretability Derived Backdoor Attacks Detection in Deep Neural Networks: Work-in-Progress | |
| Červíčková | Creating Adversarial Examples in Machine Learning | |
| Aitchison et al. | Novelty Detection in Thermal Video |
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 |