CN115546608A - Unmanned aerial vehicle data link electromagnetic interference classification and threat assessment method - Google Patents
Unmanned aerial vehicle data link electromagnetic interference classification and threat assessment method Download PDFInfo
- Publication number
- CN115546608A CN115546608A CN202211238588.5A CN202211238588A CN115546608A CN 115546608 A CN115546608 A CN 115546608A CN 202211238588 A CN202211238588 A CN 202211238588A CN 115546608 A CN115546608 A CN 115546608A
- Authority
- CN
- China
- Prior art keywords
- electromagnetic interference
- unmanned aerial
- aerial vehicle
- data chain
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/001—Measuring interference from external sources to, or emission from, the device under test, e.g. EMC, EMI, EMP or ESD testing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
Abstract
The invention relates to an unmanned aerial vehicle data link electromagnetic interference classification and threat assessment method, in particular to the technical field of unmanned aerial vehicle data link electromagnetic interference assessment. The method comprises the following steps: acquiring state parameters of a data chain of the unmanned aerial vehicle to be predicted and I/Q data of an electromagnetic interference signal of the data chain of the unmanned aerial vehicle to be predicted; obtaining a data chain performance parameter histogram of the unmanned aerial vehicle data chain to be predicted and an atlas of electromagnetic interference signals according to the state parameters of the unmanned aerial vehicle data chain to be predicted and the I/Q data of the electromagnetic interference signals; inputting the atlas and the data link performance parameter histogram into a prediction model to obtain the electromagnetic interference type and the threat degree; the prediction model is obtained by training the MIMT-CNN network comprising a multi-channel image feature extraction module, a first addition layer, a feature fusion processing module, a second addition layer and a multi-task output module. The invention can improve the accuracy of the classification of the electromagnetic interference signals and the threat prediction result, and reduce the time and storage cost.
Description
Technical Field
The invention relates to the technical field of unmanned aerial vehicle data link electromagnetic interference assessment, in particular to an unmanned aerial vehicle data link electromagnetic interference classification and threat assessment method.
Background
Unmanned aerial vehicles are widely recognized as application technologies with great potential due to their characteristics of automation, low cost and multiple functions, and are growing rapidly in many fields of military, civil and commercial sectors. The electromagnetic environment of over-the-air wireless communication networks is increasingly complex due to the presence of illegal users and malicious interference. The unmanned aerial vehicle data link is used as an airborne electronic system, and is easy to cause communication abnormity, interruption and even damage due to electromagnetic interference. The method is limited by the task of the unmanned aerial vehicle platform, and if the electromagnetic situation perception of the data chain depends on the ground station to monitor the electromagnetic situation, on one hand, a large amount of expert field experience is needed, so that personnel bear heavy cognitive burden. On the other hand, the unmanned aerial vehicle is slow in electromagnetic interference treatment speed and cannot cope with the instantaneously changing electromagnetic environment. Therefore, unmanned aerial vehicle data links are required to autonomously perform electromagnetic signal identification and electromagnetic interference threat degree prediction.
Many modeling prediction studies on electromagnetic interference are built on platforms from electronic components, circuits to electronic devices, and the like, by constructing equivalent circuit models and topological networks, statistical probability-based models, and machine learning.
Equivalent circuit model based methods require extensive expertise and detailed knowledge of the structure of the electronic device. The use of statistical probability models requires domain experts to perform feature extraction on the data samples of the tested platform, thereby reducing data dimensionality and facilitating classification or prediction of algorithms. Deep learning is a branch of machine learning, and more complex features can be autonomously learned from data through a plurality of nonlinear transformations including up to billions of weight parameters, so that dependence on professional knowledge and feature extraction rules is reduced, and a Convolutional Neural Network (CNN) is a data-driven deep Neural network structure as a typical application of deep learning in computer vision. The method has good feature extraction capability, and is applied to electromagnetic interference signal classification and threat assessment, however, the conventional method for predicting the electromagnetic interference signal classification and threat by adopting the convolutional neural network always needs to separately model each task, so that the calculation and storage cost required by the model is greatly increased, and the relation between the two tasks is neglected, so that the obtained result is not accurate enough.
Disclosure of Invention
The invention aims to provide an unmanned aerial vehicle data chain electromagnetic interference classification and threat assessment method, which can improve the accuracy of electromagnetic interference signal classification and threat prediction results and reduce the time and storage cost.
In order to achieve the purpose, the invention provides the following scheme:
an unmanned aerial vehicle data chain electromagnetic interference classification and threat assessment method comprises the following steps:
acquiring state parameters of a data chain of the unmanned aerial vehicle to be predicted and I/Q data of an electromagnetic interference signal of the data chain of the unmanned aerial vehicle to be predicted; the state parameters include: automatic gain control voltage, signal-to-noise ratio and bit error rate;
obtaining a data chain performance parameter histogram of the data chain of the unmanned aerial vehicle to be predicted according to the state parameters of the data chain of the unmanned aerial vehicle to be predicted;
obtaining an atlas of the electromagnetic interference signals of the unmanned aerial vehicle data chain to be predicted according to the I/Q data of the electromagnetic interference signals of the unmanned aerial vehicle data chain to be predicted; the atlas comprises a short-time Fourier transform time-frequency spectrogram and a density constellation map;
inputting an atlas of electromagnetic interference signals of the data chain of the unmanned aerial vehicle to be predicted and a data chain performance parameter histogram of the data chain of the unmanned aerial vehicle to be predicted into a prediction model to obtain the electromagnetic interference type and the electromagnetic interference threat degree of the data chain of the unmanned aerial vehicle to be predicted; the prediction model is obtained by training the MIMT-CNN network; the MIMT-CNN network specifically includes: the multi-channel image fusion processing device comprises a multi-channel image feature extraction module, a first addition layer, a feature fusion processing module, a second addition layer and a multi-task output module which are sequentially connected.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects: inputting an atlas of electromagnetic interference signals of the data chain of the unmanned aerial vehicle to be predicted and a data chain performance parameter histogram of the data chain of the unmanned aerial vehicle to be predicted into a prediction model to obtain the electromagnetic interference type and the electromagnetic interference threat degree of the data chain of the unmanned aerial vehicle to be predicted; the prediction model is obtained by training the MIMT-CNN network; the MIMT-CNN network specifically includes: the multi-channel image feature extraction module, the first addition layer, the feature fusion processing module, the second addition layer and the multi-task output module which are connected in sequence are applied to a multi-input channel MIMT-CNN network to link electromagnetic interference classification and threat assessment, the accuracy of electromagnetic interference signal classification and threat prediction results can be improved, and time and storage cost can be reduced.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
Fig. 1 is a flowchart of a method for classifying data link electromagnetic interference and evaluating threat of an unmanned aerial vehicle according to an embodiment of the present invention;
fig. 2 is a flowchart of training a MIMT-CNN network according to an embodiment of the present invention;
fig. 3 is a structural diagram of a MIMT-CNN network according to an embodiment of the present invention;
fig. 4 is a graph showing the variation of the loss function value with the number of iterations in the MIMT-CNN network training process according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
As shown in fig. 1, a Multi-task CNN with Multi-input, MIMT-CNN network is used for risk perception of an electromagnetic environment of an unmanned aerial vehicle data chain, and based on this embodiment of the present invention, a method for classifying electromagnetic interference of an unmanned aerial vehicle data chain and evaluating threats is provided, including:
step 101: acquiring state parameters of a data chain of the unmanned aerial vehicle to be predicted and I/Q data of an electromagnetic interference signal of the data chain of the unmanned aerial vehicle to be predicted. The state parameters include: automatic Gain Control (AGC), signal-to-noise ratio (SNR), and Bit Error Rate (BER).
Step 102: and obtaining a data chain performance parameter histogram of the data chain of the unmanned aerial vehicle to be predicted according to the state parameters of the data chain of the unmanned aerial vehicle to be predicted.
Step 103: obtaining an atlas of the electromagnetic interference signals of the unmanned aerial vehicle data chain to be predicted according to the I/Q data of the electromagnetic interference signals of the unmanned aerial vehicle data chain to be predicted; the atlas includes short-time Fourier transform time-frequency spectrogram and density constellation.
Step 104: inputting an atlas of electromagnetic interference signals of the data chain of the unmanned aerial vehicle to be predicted and a data chain performance parameter histogram of the data chain of the unmanned aerial vehicle to be predicted into a prediction model to obtain the electromagnetic interference type and the electromagnetic interference threat degree of the data chain of the unmanned aerial vehicle to be predicted; the prediction model is obtained by training a Multi-task CNN with Multi-input, MIMT-CNN network; the MIMT-CNN network specifically includes: the system comprises a multi-channel image feature extraction module, a first addition layer, a feature fusion processing module, a second addition layer and a multi-task output module which are sequentially connected.
As an optional implementation manner, the MIMT-CNN network performs feature extraction on the visualized data chain performance parameters (data chain performance parameter histogram), the electromagnetic interference signal short-time fourier transform time-frequency spectrogram and the density constellation map as input of the multi-channel network. Electromagnetic signal classification and threat assessment are then put into a shared parallel CNN for simultaneous learning based on multitask learning. Then, in order to further improve the performance of the model, the structure and the hyper-parameters of the parallel part of the network are optimized by using a Bayesian optimization method. As shown in fig. 2, the method for determining the prediction model includes:
and respectively carrying out data chain electromagnetic interference injection tests under different electromagnetic interference types to obtain state parameters of the unmanned aerial vehicle data chain and I/Q data of electromagnetic interference signals under each electromagnetic interference type.
And determining the electromagnetic interference threat degree of the unmanned aerial vehicle data chain corresponding to each electromagnetic interference type.
And obtaining a data chain performance parameter histogram of the unmanned aerial vehicle data chain under each electromagnetic interference type according to the state parameters of the unmanned aerial vehicle data chain under each electromagnetic interference type.
Obtaining an atlas of electromagnetic interference signals of the unmanned aerial vehicle data chain under each electromagnetic interference type according to the I/Q data of the electromagnetic interference signals of the unmanned aerial vehicle data chain under each electromagnetic interference type; the atlas includes short-time Fourier transform time-frequency spectrogram and density constellation.
And training the MIMT-CNN network to obtain the prediction model by taking the data chain performance parameter histograms of the unmanned aerial vehicle data chains of all the electromagnetic interference types and the image sets of the electromagnetic interference signals of the unmanned aerial vehicle data chains of all the electromagnetic interference types as sample sets, wherein the CNN is a deep neural network specially used for processing grid-form data such as images. By introducing concepts such as local receptive field, neurons, activation, sparseness and the like, the network can sense the abstract characteristics of data deep level like the human brain, and the method is widely applied to the field of computer vision. Let K be the number of samples of each type, data chain normalized parameter histogramThe matrix is
STFT spectrogram matrix is
The normalized density constellation matrix is
The input image matrix is
The output targets of the model are the type of electromagnetic interference and the interference threat level to the data chain. Let electromagnetic interference type label be v k . Because the electromagnetic interference categories are mutually independent, in order to make the category values more reasonable, the discrete classification categories are subjected to one-hot coding
Data chain interference threat degree label is u k Then the label matrix is
The network establishes a mapping of
And training the MIMT-CNN network by taking the label matrix as output and taking the mapping established by the network as input to obtain a prediction model.
As an optional implementation manner, the method includes the steps of obtaining state parameters of a data chain and In-phase/quadrature (I/Q) data of an electromagnetic space through an electromagnetic interference injection experiment, respectively performing the data chain electromagnetic interference injection experiment under different electromagnetic interference types to obtain the state parameters of the unmanned aerial vehicle data chain and the I/Q data of an electromagnetic interference signal under each electromagnetic interference type, and specifically includes:
the method comprises the steps of carrying out a data chain electromagnetic interference injection experiment under different electromagnetic interference types, interference intensity and interference frequency parameters, collecting data chain state parameters (AGC/SNR/BER) from data chain detection software, and collecting I/Q data of data chain receiving signals under the corresponding electromagnetic interference state by using an electromagnetic spectrum detection receiver.
As an optional implementation manner, determining an electromagnetic interference threat level (electromagnetic interference threat level) of the data chain of the unmanned aerial vehicle corresponding to each electromagnetic interference type specifically includes:
aiming at different electromagnetic interference types, a unified electromagnetic interference threat level division method is adopted for evaluating the interference degree of a data chain. According to the working practice of the unmanned aerial vehicle data chain in an application scene, the situation that the data chain is threatened when the difference value between the interference power and the lock losing power is less than 6dB is defined. And dividing the interference into four levels according to the difference value between the interference power and the data link unlocking threshold, wherein the corresponding difference values between the interference power and the unlocking power are respectively 6dB, 3dB, 1dB and 0dB.
As an optional implementation manner, obtaining a data chain performance parameter histogram of the data chain of the unmanned aerial vehicle in each electromagnetic interference type according to the state parameter of the data chain of the unmanned aerial vehicle in each electromagnetic interference type, and converting the data chain state parameter into a visualized parameter histogram, specifically includes:
normalizing the state parameters of the data chain, and then visualizing to obtain a histogram in each electromagnetic interference state, wherein the histogram is used for representing state information when the data chain is interfered; initial state parameters of unmanned aerial vehicle without electromagnetic interference
State parameter p after electromagnetic interference m ∈{p AGC ,p SNR ,p BER }, maximum value of the State parameter
Using a formula
The three state parameters (AGC, SNR, and BER) are normalized separately, wherein,
is the processed data. The original data is linearly transformed by normalization, mapping the original input to [0, 1]]Thereby changing the performance parameters into dimensionless expressions and then visualizing the normalized data in a histogram.
As an optional implementation manner, obtaining an atlas of electromagnetic interference signals of the data chain of the unmanned aerial vehicle in each electromagnetic interference type according to I/Q data of the electromagnetic interference signals of the data chain of the unmanned aerial vehicle in each electromagnetic interference type specifically includes:
the I/Q data is converted into a time-frequency spectrogram through Short-time Fourier transform (STFT), the change of interference signal power along with time and frequency is reflected, a standardized density constellation diagram is drawn after the I/Q data is standardized, and phase information and noise information of interference are reflected.
Performing time-frequency analysis on the I/Q data to obtain an STFT spectrogram, namely a short-time Fourier transform time-frequency spectrogram, to obtain time-frequency information of an interference signal, sliding an analysis window with the length of M on a received signal r (n), namely the I/Q data, and calculating the short-time Fourier transform (DFT) of a windowed sampling signal. Let L be a non-zero overlap length to compensate for signal attenuation at the window edge. Let R be k (f) Is the DFT of the sampled signal at frequency f centered in the window at time k (M-L),
where n is the time sample point, g (n) is a window function of length M,
N r is the length of the sampled signal r (n).
Squared value of DFT modulus for each windowed segment k (f)| 2 Combining to obtain matrix | R STFT (f)| 2 The matrix contains the amplitude and phase of each time and frequency point. To R STFT (f)| 2 Performing visualization to obtain STFT frequencySpectrogram
And standardizing the I/Q data, and drawing a standardized density constellation diagram to obtain the phase information of the interference signal. The received signals (I/Q data) have different amplitudes under different interferences, so that a proper region needs to be selected to observe a point on the constellation diagram. The selected area is too large, the signal sample is compressed in a small area, and the distribution of sampling points cannot be effectively observed; the selection area is too small and some signal samples may be excluded from the image. Therefore, the amplitude of the sampling point is subjected to scale normalization adjustment to [ -1,1], and a standardized constellation diagram with uniform size is obtained. Due to the superposition of the data chain working signal, the interference signal and the noise signal, the sampling points on the constellation diagram are overlapped, and the distribution characteristics of the sampling points are difficult to judge. However, the sample point distribution density of different regions is different, so the feature of the constellation image can be enhanced by using the point density.
Calculating the ratio of the number d (i, j) of the sampling points to the window area in a circular window with the radius r drawn by taking point coordinates (i, j) as the center to obtain the normalized point density rho (i, j) of a certain point
Then coloring the constellation diagram according to the density, wherein each point in the preprocessed constellation diagram is not independent any more and has equal information content. The processing method carries out time dimension accumulation on points in the constellation diagram, so that the data characteristic dimension is higher, more modulation signal priori knowledge is condensed, and the characteristic enhancement of the constellation diagram is realized.
As an alternative embodiment, the MIMT-CNN is constructed and trained with randomly partitioned training data. The hyper-parameters of the network are optimized on the verification set using bayesian optimization. And testing on the test set according to the network model obtained by training, and checking the accuracy and generalization capability of the model according to the actually obtained interference type classification result and the interference threat degree evaluation result. The training of the MIMT-CNN network to obtain the prediction model by using the data chain performance parameter histogram of the unmanned aerial vehicle data chain under each electromagnetic interference type, each electromagnetic interference type and the atlas of the electromagnetic interference signal of the unmanned aerial vehicle data chain under each electromagnetic interference type as a sample set specifically includes:
and dividing the sample set into a training set, a verification set and a test set according to a set proportion.
And performing initial training on the MIMT-CNN network by adopting the training set to obtain the trained MIMT-CNN network.
And optimizing the trained MIMT-CNN network by adopting a verification set to obtain an optimized MIMT-CNN network.
And testing the optimized MIMT-CNN network by adopting a test set to obtain the prediction model.
As shown in fig. 3, the input of the network is divided into a plurality of channels, and the corresponding data chain performance parameter histogram, short-time fourier transform time-frequency spectrogram and density constellation map in the same interference state are respectively input according to the input types. And (4) extracting the characteristics of each channel, then performing characteristic fusion, and finally obtaining multi-task output. The feature extraction layers of all the channels have the same structure, and are formed by superposing a plurality of convolution layers and activation layers on two feature extraction layers, and are used for image feature learning and dimension reduction. And adding the extracted features, and entering two parallel feature fusion layers, wherein the feature fusion layers consist of fusion modules with different numbers. The feature fusion part (feature fusion processing module) captures the association between different tasks by adopting a parallel network structure, and extracts feature information from different angles, thereby obtaining more comprehensive feature information. The network hyper-parameters of the fusion layer are improved by a Bayesian optimization algorithm, and targeted screening fusion is carried out according to the characteristics of the characteristic parameters input by the network. And the fused parameter matrix enters a multitask output part (multitask output module) and respectively outputs an electromagnetic interference classification result and a data chain interference performance grade predicted value. The input layer of each channel of the model is input with RGB (Red greenblue) image with the size of 100 × 100 × 3 at first, threeThe class feature image data is divided by 255 to normalize each pixel value to [0,1]And (5) interval, and then, enabling each channel to enter a feature extraction layer respectively to extract features. Because the input data volume is very large and is not suitable for loading all data at one time to perform gradient calculation and weight value updating, the input data is divided into small batches, and the network can obtain larger generalization capability. Each input channel passes through the same size and number of convolutional layers, active layers, pooling layers, and BN layers. Wherein each channel contains two consecutive feature extraction layers, consisting of two 3 x 3 convolutional layers, two ReLU active layers and one 2 x 2 max pooling layer, for extracting network features and dimensionality reduction. For inputs with widely distributed image features, large convolutional layers can better realize feature extraction, and for inputs with concentrated image feature distribution, small convolutional layers can be better learned. Therefore, according to the Inceptation structure, the feature fusion processing part can enhance the adaptability of the network to the feature scale by using convolution layers with different sizes in parallel. In order to select a better layer parameter and improve the adaptability of the network to the characteristics, the size and the number of the convolutional layers are obtained by Bayesian optimization. The output layer of the network has two branches of classification and regression, and the regression output is carried out by a full connection layer with the size of 1, and the output is the threat degree u of data chain interference. The classification layer consists of a full connection layer with the size of 4 and a Softmax layer, the output is the electromagnetic interference type v,
wherein w i And v is the output value of the ith node in the layer before Softmax, v is the result of classification, and C is the number of output nodes, namely the number of classification classes. Softmax converts multi-class output values to a range of [0, 1%]And the sum is 1, and the probability is the basis of the category judgment.
As an optional implementation manner, the multi-channel image feature extraction module includes three image feature extraction sub-modules connected in parallel and having the same structure, where the image feature extraction sub-modules include: the device comprises an input layer, a feature extraction unit, a first convolution layer, a first ReLU active layer, a second convolution layer, a first Batchnorm layer and a second ReLU active layer which are connected in sequence.
As an optional implementation manner, the feature fusion processing module includes: the system comprises a first feature extraction submodule and a second feature extraction submodule which are connected in parallel, wherein the first feature extraction submodule comprises a first feature fusion layer, a second feature fusion layer and a first full-connection layer which are connected in sequence; the second feature extraction submodule comprises a third feature fusion layer and a second full connection layer which are connected in sequence.
As an optional implementation, the multitask output module includes: the classification units and the third full connection layer are connected in parallel; the classification unit comprises a fourth full connection layer and a Softmax layer which are connected in sequence.
As an optional implementation manner, the feature extraction unit includes two feature extraction layers connected in sequence; the feature extraction layer includes: the third convolution layer, the third ReLU active layer, the fourth convolution layer, the fourth ReLU active layer and the maximum pooling layer are connected in sequence.
As an alternative embodiment, the first feature fusion layer, the second feature fusion layer, and the third feature fusion layer each include: a fifth convolution layer, a second Batchnorm layer and a fifth ReLU activation layer connected in sequence.
As an optional implementation manner, the preliminary training of the MIMT-CNN network by using the training set to obtain the trained MIMT-CNN network specifically includes:
and initially training the MIMT-CNN network by adopting the training set to obtain the trained MIMT-CNN network by taking the minimum loss function as a target, wherein the loss function is used for evaluating the degree of the difference between an actual value and a predicted value, and the smaller the loss function is, the better the performance of the model is. Because the MIMT-CNN network has a plurality of task outputs, the loss function of the network consists of two parts, namely classification loss L v And regression loss L u . Wherein the classification loss L c With the Focal loss, compared with the common cross entropy loss, the Focal loss has no change for the sample with inaccurate classification, and the loss becomes smaller for the sample with accurate classification. Overall, it is equivalent to increaseThe weights of the classified inaccurate samples in the loss function are weighted, forcing the network to focus on difficult samples. L is v ＝-α(1-p v ) γ log(p v ) Where γ is the focusing parameter, γ ≧ 0, and an increase in γ can increase the sensitivity of the network to error classification data, set herein as the default value γ =2.α is a balance coefficient, α =0.25, and the weight of the large number of negative samples can be reduced by setting the value of α. p is a radical of formula v To predict the probability of being in this category, (1-p) v ) γ The modulation factor is used for reducing the weight of the easily classified sample. Due to the change of the modulation coefficient, the samples which are difficult to train play a leading role, so that the model is more concentrated on the samples which are difficult to classify during training, and the accuracy of model classification is improved.
Regression loss L u Using a half mean square error loss function:
wherein
Is a network response, u k Is a predicted target value, K bs Is network response u in the small batch range k K is u k Total number of observations.
Because as the training progresses, two loss functions L v And L u The reduced speed is not uniform. To unify loss to the same order of magnitude, avoid L with small gradients v Is increased by gradient L u Carry away, thus to L v And L u Respectively adopt different weights o 1 And o- 2 . The final loss function is a weighted sum of the classification loss and the regression loss, enhancing the generalization ability of the model.
L all ＝ο 1 L v +ο 2 L u Adjusting to balance the influence of classification loss and regression loss on the convergence rate of training, and obtaining omicron 1 ＝1，ο 2 ＝0.1。
As an optional implementation manner, the testing the optimized MIMT-CNN network by using a test set to obtain the prediction model specifically includes:
and testing the optimized MIMT-CNN network by adopting a test set to obtain the prediction model by taking maximum classification accuracy and minimum RMSE and MAPE as targets, wherein the classification task and the regression task need to be evaluated respectively because the network can complete the classification task and the regression task at the same time. The classification performance is often evaluated with accuracy. Suppose v T And v F The number of correct and wrong classification respectively, the classification accuracy Acc is
Regression performance the prediction accuracy is usually evaluated by Root Mean Squared Error (RMSE), mean Absolute Percentage Error (MAPE). RMSE can measure how well a predicted value fits an actual curve. Compared to MAPE, is more sensitive to errors between predicted and true values.
the invention provides a method for sampling a sample set according to the following steps of 6:2:2 into training set, validation set, and test set, the training set optimizing the weights by updating the model parameters. The validation set is used to optimize the hyper-parameters of the network to prevent over-fitting and under-fitting, with the goal of obtaining a better model. And finally, the network evaluates the final model on the test set.
The data chain working signal is sent by a ground station data chain, and a Binary Phase Shift Keying (BPSK) communication mode is adopted. Therefore, the interference signal type selects continuous wave and broadband white gaussian noise interference which are common in an actual interference scene, and BPSK interference which is close to data link communication parameters. And training the network through the normalized performance parameters, the STFT spectrogram and the normalized density constellation diagram obtained by the electromagnetic sensitivity test.
In the case of training using the adaptive moment estimation Adam optimizer, the resulting loss function value curve for the training process is shown in fig. 4. It can be seen that the function convergence rate is fast, but the converged loss function value is low, about 0.2.
The model classification accuracy of the MIMT-CNN network was 95.45%, with predicted RMSE and MAPE of 0.49, and 10.83%, respectively. The time from input to output is 14.81ms.
The invention has the following technical effects:
1. the convolutional neural network is an intelligent deep learning method, can automatically sense electromagnetic interference risks, increases the autonomy and intelligence level of an unmanned aerial vehicle data chain, and avoids increasing the cognitive burden of operators.
2. The visualized data chain performance parameters, the electromagnetic interference signal STFT spectrogram and the density constellation map are used as model input, and the model precision is improved. The reason is that the information obtained through the electromagnetic sensitivity test has different data formats and change rules, and the obtained heterogeneous information is visualized to be used as model input, so that the interpretability is enhanced on one hand, and the visual sensory understanding of human is met. On the other hand, the single image has large data volume, more effective information content, strong anti-noise capability and high model accuracy. Different kinds of image information can be fused through the MIMT-CNN, so that the model can more accurately recognize the electromagnetic environment, and the model has higher precision.
3. The MIMT-CNN can simultaneously carry out electromagnetic signal classification and threat assessment, and has high prediction efficiency. The model makes full use of the correlation of the characteristics of the output data set, carries out electromagnetic interference threat level assessment while judging the type of electromagnetic interference, and can avoid the time cost and the calculation cost for training a plurality of models.
In the present specification, the embodiments are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (8)
1. An unmanned aerial vehicle data chain electromagnetic interference classification and threat assessment method is characterized by comprising the following steps:
acquiring state parameters of a data chain of the unmanned aerial vehicle to be predicted and I/Q data of an electromagnetic interference signal of the data chain of the unmanned aerial vehicle to be predicted; the state parameters include: automatic gain control voltage, signal-to-noise ratio and bit error rate;
obtaining a data chain performance parameter histogram of the data chain of the unmanned aerial vehicle to be predicted according to the state parameters of the data chain of the unmanned aerial vehicle to be predicted;
obtaining an atlas of the electromagnetic interference signals of the unmanned aerial vehicle data chain to be predicted according to the I/Q data of the electromagnetic interference signals of the unmanned aerial vehicle data chain to be predicted; the atlas comprises a short-time Fourier transform time-frequency spectrogram and a density constellation map;
inputting an atlas of electromagnetic interference signals of the data chain of the unmanned aerial vehicle to be predicted and a data chain performance parameter histogram of the data chain of the unmanned aerial vehicle to be predicted into a prediction model to obtain the electromagnetic interference type and the electromagnetic interference threat degree of the data chain of the unmanned aerial vehicle to be predicted; the prediction model is obtained by training the MIMT-CNN network; the MIMT-CNN network specifically includes: the multi-channel image fusion processing device comprises a multi-channel image feature extraction module, a first addition layer, a feature fusion processing module, a second addition layer and a multi-task output module which are sequentially connected.
2. The unmanned aerial vehicle data chain electromagnetic interference classification and threat assessment method of claim 1, wherein the multi-channel image feature extraction module comprises three image feature extraction sub-modules which are connected in parallel and have the same structure, and the image feature extraction sub-modules comprise: the device comprises an input layer, a feature extraction unit, a first convolution layer, a first ReLU active layer, a second convolution layer, a first Batchnorm layer and a second ReLU active layer which are connected in sequence.
3. The unmanned aerial vehicle data link electromagnetic interference classification and threat assessment method according to claim 1, wherein the feature fusion processing module comprises: the system comprises a first feature extraction submodule and a second feature extraction submodule which are connected in parallel, wherein the first feature extraction submodule comprises a first feature fusion layer, a second feature fusion layer and a first full-connection layer which are sequentially connected; the second feature extraction submodule comprises a third feature fusion layer and a second full connection layer which are connected in sequence.
4. The unmanned aerial vehicle data link electromagnetic interference classification and threat assessment method of claim 1, wherein the multitask output module comprises: the classification units and the third full-connection layer are connected in parallel; the classification unit comprises a fourth full connection layer and a Softmax layer which are connected in sequence.
5. The unmanned aerial vehicle data link electromagnetic interference classification and threat assessment method according to claim 1, wherein the determination method of the prediction model comprises the following steps:
respectively carrying out data chain electromagnetic interference injection tests under different electromagnetic interference types to obtain state parameters of the unmanned aerial vehicle data chain and I/Q data of electromagnetic interference signals under each electromagnetic interference type;
determining the electromagnetic interference threat degree of the unmanned aerial vehicle data chain corresponding to each electromagnetic interference type;
obtaining a data chain performance parameter histogram of the unmanned aerial vehicle data chain under each electromagnetic interference type according to the state parameters of the unmanned aerial vehicle data chain under each electromagnetic interference type;
obtaining an atlas of the electromagnetic interference signals of the unmanned aerial vehicle data chain under each electromagnetic interference type according to the I/Q data of the electromagnetic interference signals of the unmanned aerial vehicle data chain under each electromagnetic interference type; the atlas comprises a short-time Fourier transform time-frequency spectrogram and a density constellation map;
and training the MIMT-CNN network to obtain the prediction model by taking each electromagnetic interference type, a data chain performance parameter histogram of the unmanned aerial vehicle data chain under each electromagnetic interference type and an image set of electromagnetic interference signals of the unmanned aerial vehicle data chain under each electromagnetic interference type as sample sets.
6. The unmanned aerial vehicle data link electromagnetic interference classification and threat assessment method according to claim 2, wherein the feature extraction unit comprises two feature extraction layers connected in sequence; the feature extraction layer includes: the third convolution layer, the third ReLU active layer, the fourth convolution layer, the fourth ReLU active layer and the maximum pooling layer are connected in sequence.
7. The unmanned aerial vehicle data link electromagnetic interference classification and threat assessment method of claim 3, wherein the first feature fusion layer, the second feature fusion layer and the third feature fusion layer each comprise: a fifth convolutional layer, a second batcnorm layer, and a fifth ReLU active layer connected in sequence.
8. The unmanned aerial vehicle data chain electromagnetic interference classification and threat assessment method according to claim 5, wherein the training of the MIMT-CNN network is performed to obtain the prediction model by taking the data chain performance parameter histogram of the unmanned aerial vehicle data chain under each electromagnetic interference type, each electromagnetic interference type and the atlas of the electromagnetic interference signal of the unmanned aerial vehicle data chain under each electromagnetic interference type as a sample set, and specifically comprises:
dividing the sample set into a training set, a verification set and a test set according to a set proportion;
carrying out initial training on the MIMT-CNN network by adopting the training set to obtain a trained MIMT-CNN network;
optimizing the trained MIMT-CNN network by adopting a verification set to obtain an optimized MIMT-CNN network;
and testing the optimized MIMT-CNN network by adopting a test set to obtain the prediction model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211238588.5A CN115546608A (en) | 2022-10-11 | 2022-10-11 | Unmanned aerial vehicle data link electromagnetic interference classification and threat assessment method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211238588.5A CN115546608A (en) | 2022-10-11 | 2022-10-11 | Unmanned aerial vehicle data link electromagnetic interference classification and threat assessment method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115546608A true CN115546608A (en) | 2022-12-30 |
Family
ID=84733165
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211238588.5A Pending CN115546608A (en) | 2022-10-11 | 2022-10-11 | Unmanned aerial vehicle data link electromagnetic interference classification and threat assessment method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115546608A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115865238A (en) * | 2023-02-27 | 2023-03-28 | 中国人民解放军国防科技大学 | Signal interference detection method and device |
| CN116524322A (en) * | 2023-04-10 | 2023-08-01 | 北京盛安同力科技开发有限公司 | SAR image recognition method based on deep neural network |
| CN117459178A (en) * | 2023-12-22 | 2024-01-26 | 武汉阿内塔科技有限公司 | Unmanned aerial vehicle communication interference method and system based on semantic guidance |
-
2022
- 2022-10-11 CN CN202211238588.5A patent/CN115546608A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115865238A (en) * | 2023-02-27 | 2023-03-28 | 中国人民解放军国防科技大学 | Signal interference detection method and device |
| CN115865238B (en) * | 2023-02-27 | 2023-05-02 | 中国人民解放军国防科技大学 | Signal interference detection method and device |
| CN116524322A (en) * | 2023-04-10 | 2023-08-01 | 北京盛安同力科技开发有限公司 | SAR image recognition method based on deep neural network |
| CN117459178A (en) * | 2023-12-22 | 2024-01-26 | 武汉阿内塔科技有限公司 | Unmanned aerial vehicle communication interference method and system based on semantic guidance |
| CN117459178B (en) * | 2023-12-22 | 2024-03-26 | 武汉阿内塔科技有限公司 | Unmanned aerial vehicle communication interference method and system based on semantic guidance |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115546608A (en) | Unmanned aerial vehicle data link electromagnetic interference classification and threat assessment method | |
| CN113837477A (en) | Data dual-drive power grid fault prediction method, device and equipment under typhoon disaster | |
| CN112434643A (en) | Classification and identification method for low-slow small targets | |
| CN111649817B (en) | Distributed optical fiber vibration sensor system and mode identification method thereof | |
| CN114564982B (en) | Automatic identification method for radar signal modulation type | |
| WO2020209918A2 (en) | Detection of cyber attacks targeting avionics systems | |
| CN106656357B (en) | Power frequency communication channel state evaluation system and method | |
| CN114595732B (en) | Radar radiation source sorting method based on depth clustering | |
| CN115392326B (en) | Modulation identification method based on joint multi-modal information and domain countermeasure neural network | |
| CN111985349B (en) | Classification recognition method and system for radar received signal types | |
| CN116866129A (en) | Wireless communication signal detection method | |
| CN115565019A (en) | Single-channel high-resolution SAR image ground object classification method based on deep self-supervision generation countermeasure | |
| Xu et al. | Prediction on EMS of UAV's Data Link Based on SSA-Optimized Dual-Channel CNN | |
| CN114510960A (en) | Method for recognizing distributed optical fiber sensor system mode | |
| TW202217628A (en) | Rf fingerprint signal processing device and rf fingerprint signal processing method | |
| US11817977B2 (en) | Method of determining single sampling frequency of classification target signal in order to predict modulation type of classification target signal, and method and apparatus for predicting modulation type by using classification target signal sampled with single sampling frequency | |
| CN113534059B (en) | Radar active interference identification method based on deep convolutional network under open set scene | |
| CN112861670B (en) | Transmission line hardware detection method and system | |
| US20220272122A1 (en) | Autonomous vehicle control attack detection and countermeasures | |
| CN115361272A (en) | Communication fault detection method and system of baseband chip and electronic equipment | |
| Ebert-Uphoff et al. | Evaluation, tuning and interpretation of neural networks for meteorological applications | |
| CN114970638A (en) | Radar radiation source individual open set identification method and system | |
| CN111160419B (en) | Deep learning-based electronic transformer data classification prediction method and device | |
| CN108429589B (en) | Optical source of optical network based on spectral analysis and optical path identification method | |
| Marina et al. | Automatic Traffic Light Detection Using AI for VLC |
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 |