CN115270851A - Time-frequency map reconstruction method based on unmanned aerial vehicle communication signal data enhancement - Google Patents
Time-frequency map reconstruction method based on unmanned aerial vehicle communication signal data enhancement Download PDFInfo
- Publication number
- CN115270851A CN115270851A CN202210719338.7A CN202210719338A CN115270851A CN 115270851 A CN115270851 A CN 115270851A CN 202210719338 A CN202210719338 A CN 202210719338A CN 115270851 A CN115270851 A CN 115270851A
- Authority
- CN
- China
- Prior art keywords
- time
- unmanned aerial
- aerial vehicle
- frequency
- vehicle communication
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The invention discloses a time-frequency map reconstruction method based on unmanned aerial vehicle communication signal data enhancement, which comprises the following steps: collecting various types of unmanned aerial vehicle communication signals; determining a variance of additive white Gaussian noise based on a plurality of first short sampling sequences obtained by segmenting the unmanned aerial vehicle communication signal, and reconstructing the first short sampling sequences based on the variance of the additive white Gaussian noise to obtain second short sampling sequences; determining a time-frequency matrix corresponding to the second short sampling sequence; obtaining a time frequency matrix data set after the construction data is enhanced according to all the time frequency matrixes; and inputting the time-frequency matrix in the time-frequency matrix data set into a trained ResNet network, and outputting the type serial number of the unmanned aerial vehicle. The unmanned aerial vehicle time-frequency map reconstructed by the method has the advantages of controllable aliasing interference, higher quality and abundant samples, solves the problem that data enhancement is inconvenient to directly apply to unmanned aerial vehicle communication signals, and widens the application range of the data enhancement.
Description
Technical Field
The invention belongs to the technical field of signal processing, and relates to a time-frequency map reconstruction method based on unmanned aerial vehicle communication signal data enhancement.
Background
The time-frequency spectrum of the downlink image transmission signal of the unmanned aerial vehicle usually contains time-domain information and frequency-domain information of a radio-frequency signal of the unmanned aerial vehicle, and the time-frequency spectrum textures of different types of unmanned aerial vehicles have the characteristic of difference, so that the time-frequency spectrum of the communication signal of the unmanned aerial vehicle is usually used for the target identification and classification task of the unmanned aerial vehicle. At present, due to various kinds of interference in cities, unmanned aerial vehicle time frequency spectrum aliasing is serious, and the recognition performance of a neural network on the unmanned aerial vehicle time frequency spectrum is poor. Common data enhancement methods include flipping, rotating, scaling or adding noise to the image, which can destroy the texture structure of the original image and thus affect the recognition performance of the classification system. However, the time and frequency distribution of the drone communication signal follows a fixed format and transmission rules, so traditional data enhancements are not readily directly applicable to the drone communication signal.
However, there is currently a method for unmanned aerial vehicle pattern recognition based on short-time fourier transform. However, first the method collects drone signals for all operating channels by artificially controlling the opening and closing of the signal collection equipment. Secondly, carrying out short-time Fourier transform on the unmanned aerial vehicle data to generate a time-frequency map. Then, the training data set and the test data set are divided for the unmanned aerial vehicle atlas. And finally, carrying out target recognition by using a regional convolutional neural network R-CNN (Regions with CNN features) to obtain a recognition result. Although deep learning has a significant effect on image classification, the time-frequency map generated by short-time Fourier transform is higher in identification accuracy. However, the method still has the defects that the unmanned aerial vehicle time-frequency spectrum aliasing is serious due to various urban interferences, and the neural network has poor performance of identifying the unmanned aerial vehicle time-frequency spectrum.
At present, unmanned aerial vehicle identification methods based on unmanned aerial vehicle measurement and control signals are also available. The method is realized by the steps of 1) obtaining a measurement and control signal and generating a time-frequency map by adopting short-time Fourier transform; 2) Performing type labeling on the time-frequency map, and dividing the time-frequency map into a training data set and a test data set; 3) And (3) adopting an unmanned aerial vehicle data set, and performing target identification by using a Convolutional Neural Network (CNN) to obtain an identification result. Although the measurement and control signals of the unmanned aerial vehicles of different styles are collected, the unmanned aerial vehicle time-frequency spectrum generated through short-time Fourier transform can be used for target identification of the unmanned aerial vehicle. However, the method still has the defects that the aliasing of the traditional time-frequency map is serious, so that the identification performance of the classification system is poor. The texture structure of the time-frequency map can be destroyed by adopting image data enhancement, however, the unmanned aerial vehicle communication signal follows a fixed format and a transmission rule, and the data enhancement is not convenient to be directly applied to the unmanned aerial vehicle communication signal. Therefore, the problems that the quality of a data set is poor and data enhancement is inconvenient to directly apply to unmanned aerial vehicle communication signals due to frequency spectrum aliasing in the traditional time-frequency spectrum are solved.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a time-frequency map reconstruction method based on unmanned aerial vehicle communication signal data enhancement. The technical problem to be solved by the invention is realized by the following technical scheme:
a time-frequency map reconstruction method based on unmanned aerial vehicle communication signal data enhancement comprises the following steps:
step 1, collecting various types of unmanned aerial vehicle communication signals;
step 3, determining a time-frequency matrix corresponding to the second short sampling sequence;
and 5, inputting the time-frequency matrix in the time-frequency matrix data set into a trained ResNet network, and outputting the type serial number of the unmanned aerial vehicle.
In one embodiment of the present invention, the step 2 comprises:
step 2.1, uniformly dividing each section of unmanned aerial vehicle communication signal at intervals of L to obtain x sections of first short sampling sequences;
step 2.2, obtaining power spectral density according to the first short sampling sequence;
step 2.3, obtaining the variance of additive white Gaussian noise according to the signal-to-noise ratio and the power spectral density;
step 2.4, generating an additive white Gaussian noise signal according to the variance of the additive white Gaussian noise;
and 2.5, superposing the signal of the Gaussian white noise to the first short sampling sequence to obtain a second short sampling sequence.
In one embodiment of the invention, the power spectral density is:
wherein the content of the first and second substances,
representing the power spectral density of the first short sample sequence, m (i) representing the ith sample point in the first short sample sequence, 1 ≦ i ≦ L, | · | representing an absolute value operation.
In one embodiment of the present invention, the ith sample point in the second short sample sequence is:
where y (i) denotes the ith sample point in the second short sample sequence, m (i) denotes the ith sample point in the first short sample sequence, α denotes a weighting coefficient, fcRepresenting randomnessFrequency deviation, fsThe sampling rate of the system is represented, B (i) represents a signal of additive white Gaussian noise, j represents an imaginary unit symbol, and i is more than or equal to 1 and less than or equal to L.
In one embodiment of the present invention, the step 3 comprises:
and performing STFT on the second short sampling sequence to obtain a time-frequency matrix corresponding to the second short sampling sequence.
In an embodiment of the present invention, the time-frequency matrix is:
wherein the content of the first and second substances,
representing a time-frequency matrix of M rows and N columns generated after the f-th second short sampling sequence is subjected to STFT, wherein f is more than or equal to 1 and less than or equal to x, M is more than or equal to 0 and less than or equal to M-1,0 and less than or equal to N-1,M represents the dimension of a time domain in the STFT, N represents the dimension of a frequency domain in the STFT, sigma represents the summation operation, y represents the summation operation, and y represents the time-frequency matrix of M rows and N columns generated after the f-th second short sampling sequence is subjected to STFTf(i) Denotes the ith element in the f-th second short sample sequence, w (-) denotes the Hamming Window function, d denotes the length of the STFT transform sliding, d = N, e(·)Indicating an exponential operation with a natural constant e as the base.
In one embodiment of the present invention, the step 4 comprises:
and obtaining the named time-frequency matrix according to the type, the channel and the position of the unmanned aerial vehicle to which the time-frequency matrix belongs, and constructing a time-frequency matrix data set through all the named time-frequency matrices.
In an embodiment of the present invention, the method for training the ResNet network includes:
s1, acquiring a training data set, wherein the training data set comprises a plurality of time-frequency maps obtained by performing STFT (space time transform) on unmanned aerial vehicle communication signals;
and S2, inputting the time-frequency map into a ResNet network, and iteratively updating parameters of the ResNet network by using the loss value of the loss function until the loss function is converged to obtain the trained ResNet network.
In one embodiment of the invention, the loss function is:
wherein, loss represents the Loss value of the Loss function, nk represents the total number of samples in the training data set, ii and MM both represent the serial numbers of the samples in the training data set, cc represents the serial number of the label of the type of the unmanned aerial vehicle output by the ResNet network, ii is more than or equal to 1 and less than or equal to Nk, MM is more than or equal to 1 and less than or equal to Nk, and yiiccRepresenting a symbolic function, y is the sequence number of the real type label of the sample sequence number ii is equal to the sequence number cc of the unmanned plane type label output by the ResNet networkiiccIs 1, otherwise, yiiccIs 0,piiccIndicates the prediction probability of the sequence number cc belonging to the type label when the sample sequence number is ii.
The invention has the beneficial effects that:
firstly, because the time-frequency spectrum sample of the unmanned aerial vehicle is reconstructed by the method, the reconstructed time-frequency spectrum of the unmanned aerial vehicle comprises at least 5 types of unmanned aerial vehicles, and each type of unmanned aerial vehicle comprises a plurality of samples, the defects that the quality of a data set is poor due to frequency spectrum aliasing in the traditional time-frequency spectrum in the prior art are overcome, so that the reconstructed time-frequency spectrum of the unmanned aerial vehicle has the advantages of controllable aliasing interference, high quality and rich samples, and can be directly applied to a system for detecting and identifying based on the time-frequency spectrum of the unmanned aerial vehicle.
Secondly, the invention enhances the communication signal of the unmanned aerial vehicle through the data, thereby overcoming the problems that the prior image data enhancement can destroy the texture structure of the original data when the time-frequency map quality is improved, and the communication signal of the unmanned aerial vehicle follows the fixed format and the transmission rule, so that the data enhancement is not convenient to be directly applied to the communication signal of the unmanned aerial vehicle. The method and the device have the advantages that the problem that data enhancement is inconvenient to directly apply to unmanned aerial vehicle communication signals is solved, and the application range of the data enhancement is widened.
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Drawings
Fig. 1 is a schematic flowchart of a time-frequency spectrum reconstruction method based on data enhancement of communication signals of an unmanned aerial vehicle according to an embodiment of the present invention;
FIG. 2 is a diagram of a scenario provided by an embodiment of the present invention;
fig. 3 is a simulation diagram of the time-frequency pattern recognition performance of the unmanned aerial vehicle without data enhancement according to the embodiment of the present invention;
fig. 4 is a simulation diagram for enhancing the time-frequency pattern recognition performance of the unmanned aerial vehicle by using data according to the embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to specific examples, but the embodiments of the present invention are not limited thereto.
Example one
Referring to fig. 1, fig. 1 is a schematic flow chart of a time-frequency map reconstruction method based on data enhancement of an unmanned aerial vehicle communication signal according to an embodiment of the present invention. The embodiment provides a time-frequency map reconstruction method based on unmanned aerial vehicle communication signal data enhancement, and the time-frequency map reconstruction method comprises steps 1-5, wherein:
step 1, collecting various unmanned aerial vehicle communication signals.
Specifically, gather unmanned aerial vehicle communication signal of at least 5 types respectively, every type has an unmanned aerial vehicle at least, and every unmanned aerial vehicle gathers a section length sampling sequence that length is K under every kind of channel, unmanned aerial vehicle communication signal promptly.
Preferably, K =3 × 108。
In one particular embodiment, step 1 may include steps 1.1-1.2, wherein:
step 1.1, an unmanned aerial vehicle working environment and signal acquisition equipment are set up to prepare for subsequent signal acquisition work.
For example, 5 different types of 5 drones are selected in this embodiment, and each type corresponds to one drone. Each drone operates on two channels, 153 and 161 at 5.8GHz, as shown in table 1.
Table 1 sample information table in time frequency database of unmanned aerial vehicle
Referring to fig. 2, the signal acquisition equipment and the working principle of the drone in the embodiment are further described. In a microwave darkroom and under the same arrangement in a complex environment, each unmanned aerial vehicle is placed at a position 100 meters away from the height of the signal acquisition equipment. The Universal Software Radio Peripheral USRPX310 (Universal Software Radio Peripheral) in fig. 2 is connected to the computer through an optical fiber, and the GNU Radio platform in the computer controls the USRP X310 and stores the received communication signal of each unmanned aerial vehicle in the memory of the computer.
And 1.2, respectively acquiring long sampling sequences of each unmanned aerial vehicle under 153 channels and 161 channels.
Firstly, using GNU Radio to control USRP, collecting unmanned aerial vehicle communication signals of each unmanned aerial vehicle under 153 channels at a sampling rate of 100Msa/s for 5 minutes to obtain a long sampling sequence with the sampling length of 5 unmanned aerial vehicles being K, and storing the long sampling sequence in a computer memory.
And secondly, using GNU Radio to control USRP, collecting unmanned aerial vehicle communication signals of each unmanned aerial vehicle under a 161 channel at a sampling rate of 100Msa/s for 5 minutes to obtain a long sampling sequence with the sampling length of 5 unmanned aerial vehicles being K, and storing the long sampling sequence into a computer memory.
And 2, determining the variance of the additive white Gaussian noise based on a plurality of first short sampling sequences obtained by segmenting the communication signal of the unmanned aerial vehicle, and reconstructing the first short sampling sequences based on the variance of the additive white Gaussian noise to obtain a second short sampling sequence.
In a particular embodiment, step 2 may include steps 2.1-2.5, wherein:
and 2.1, uniformly dividing each section of unmanned aerial vehicle communication signal at intervals of L to obtain x sections of first short sampling sequences.
Specifically, the unmanned aerial vehicle communication signal of each long sampling sequence is uniformly divided at intervals of L to obtain x first short sampling sequences after each long sampling sequence is divided.
For example, in the embodiment of the present invention, in accordance with
The number of segments, x, for each sequence of sample points is calculated, wherein,
represents rounding down, since K =3 × 1010L =131072 elements, x =228881, so that through calculation, each long sampling point sequence can be divided into 228881 first short sampling sequences.
Step 2.2, obtaining power spectral density according to the first short sampling sequence, wherein the power spectral density is as follows:
wherein, the first and the second end of the pipe are connected with each other,
representing the power spectral density of the first short sample sequence, m (i) representing the ith sample point in the first short sample sequence, 1 ≦ i ≦ L, | · | representing an absolute value operation.
And 2.3, obtaining the variance of the additive white Gaussian noise according to the signal-to-noise ratio and the power spectral density.
Specifically, the SNR (signal-to-noise ratio) is calculated within the reception bandwidth of 100MHz according to the following formula and a priori condition
To obtain
Wherein ^ represents integral operation, BsRepresenting the drone signal bandwidth, log represents the logarithmic operation based on a natural constant of 10,
represents the power spectral density, σ, of AWGN (Additive White Gaussian Noise)2Representing the variance of the AWGN signal.
And 2.4, generating an additive white Gaussian noise signal according to the variance of the additive white Gaussian noise.
In particular, the variance σ in terms of additive white Gaussian noise2The value of (a) can be generated into AWGN using the MATLAB software platform.
And 2.5, superposing the signal of the Gaussian white noise to the first short sampling sequence to obtain a second short sampling sequence. Superimposing the AWGN signal sample point onto the first short sample sequence using the following equation, and obtaining the ith sample point in the second short sample sequence as:
where y (i) denotes the ith sample point in the second short sample sequence, m (i) denotes the ith sample point in the first short sample sequence, α denotes a weighting coefficient, fcDenotes a random frequency deviation, fsThe sampling rate of the system is represented, B (i) represents a signal of additive white Gaussian noise, j represents an imaginary unit symbol, and i is more than or equal to 1 and less than or equal to L.
And 3, determining a time-frequency matrix corresponding to the second short sampling sequence.
Specifically, a short-time Fourier transform (STFT) transform is performed on each second short sample sequence to obtain a time-frequency matrix corresponding to each second short sample sequence.
In an embodiment of the invention, L =131072 elements, x =228881, STFT transform is performed on each short sample sequence according to the following equation:
wherein the content of the first and second substances,
representing the time-frequency matrix of M rows and N columns generated after the f-th second short sampling sequence is subjected to STFT, wherein f is more than or equal to 1 and less than or equal to x, namely f is more than or equal to 1 and less than or equal to 228881,0 and less than or equal to M-1,0 and less than or equal to N and less than or equal to N-1,M represents the dimension of a time domain in the STFT, M =256, N represents the dimension of a frequency domain in the STFT, N =512, sigma represents the summation operation, y represents the sum operation, and y represents the sum operationf(i) Denotes the ith element in the f-th second short sample sequence, w (-) denotes the Hamming Window function, d denotes the length of the STFT transform sliding, d = N, e(·)Denotes an exponential operation with a natural constant e as the base, and j denotes the imaginary unit symbol.
And 4, obtaining a time frequency matrix data set after the construction data is enhanced according to all the time frequency matrixes, wherein the time frequency matrix data set comprises a plurality of uniquely named time frequency matrixes.
Specifically, the named time-frequency matrix is obtained according to the type, the channel and the position of the unmanned aerial vehicle to which the time-frequency matrix belongs, and a time-frequency matrix data set is constructed through all the named time-frequency matrices.
In this embodiment, each time-frequency matrix is named according to the type V of the unmanned aerial vehicle to which the time-frequency matrix belongs, the channel C in which the time-frequency matrix is located, and the position of the time-frequency matrix in which the time-frequency matrix is located (i.e., the O-th time-frequency matrix) according to the format of V _ C _ O, each named time-frequency matrix is stored under the folder with the name V corresponding to the named time-frequency matrix according to the number of V in the name, and the time-frequency matrix address and the identification tag in the folder are stored in the time-frequency matrix data set.
In the embodiment of the invention, each time-frequency map is named according to the format of V _ C _ O according to the type, channel information and the number of time-frequency matrixes of the unmanned aerial vehicles, so as to distinguish each time-frequency map of different unmanned aerial vehicle models under different channels, wherein V represents the identification labels of 5 unmanned aerial vehicles, V is more than or equal to 0 and less than or equal to 4,C represents 153 and 161 channels of the unmanned aerial vehicles, wherein C is 0 to represent 149 channels, C is 1 to represent 153 channels, C is 2 to represent 157 channels,c is 3 for representing 161 channels, C is 4 for representing 165 channels, O represents the position of an unmanned aerial vehicle in 228881 time-frequency matrixes under a certain channel, O is more than or equal to 1 and less than or equal to 228881, each named time-frequency matrix is respectively stored under a folder with V as a name according to the number corresponding to V in the name, and the absolute path address and the folder name of each time-frequency matrix in the folder are stored in a time-frequency matrix data set
Taking "Mavic Mini" in table 1 as an example, the time-frequency spectrum corresponding to the Mavic Mini is named according to the format of V _ C _ O: defining the unmanned plane of 'Mavic Mini' as the 1 st unmanned plane, marking the identification label V as 1, marking C as 3 because the 161 channel where the unmanned plane is located belongs to the 3 rd channel, marking the unmanned plane time frequency matrix as the 8 th time frequency matrix, and marking O as 8, thereby obtaining the name of the 8 th time frequency matrix of the Mavic Mini under the 161 channel as 2 \u3 \u8.
In this embodiment, the method for training the ResNet network includes:
s1, a training data set is obtained, wherein the training data set comprises a plurality of time-frequency maps obtained by performing STFT on unmanned aerial vehicle communication signals.
Specifically, an un-reconstructed unmanned aerial vehicle communication signal (namely, an original unmanned aerial vehicle communication signal) is randomly selected, the un-reconstructed unmanned aerial vehicle communication signal is converted into a time-frequency map by using STFT, and the time-frequency map is made into an initial time-frequency map data set (namely, a training data set), wherein the unmanned aerial vehicle works under a 161 channel, the type labels account for 80% of rows in total from 0 to 4, and the absolute path addresses and folder names of 183104 samples in total form the training data set.
S2, inputting the time-frequency map into a ResNet network, iteratively updating parameters of the ResNet network by using a loss value of a loss function until the loss function is converged to obtain the trained ResNet network, wherein the loss function is as follows:
where Loss represents the Loss value of the Loss function, nk represents the total number of samples in the training dataset, ii and MM both represent the sample numbers in the training dataset, cc represents the serial number of the ResNet network output drone type tag, 1 ≦ ii ≦ Nk,1 ≦ MM ≦ Nk, e.g., 1 ≦ ii ≦ 183104,1 ≦ MM ≦ 183104,0 ≦ cc ≦ 4,yiiccRepresenting a symbolic function, when the sequence number of the real type tag of the sample sequence number ii is equal to the sequence number cc of the unmanned plane type tag output by the ResNet network, yiiccIs 1, otherwise, yiiccIs 0,piiccIndicates the prediction probability of the sequence number cc belonging to the type label when the sample sequence number is ii.
In addition, the absolute path address of the rest 20% of samples in the sample library and the folder name form a test data set, the test data set is input into a trained ResNet network for testing, and the obtained identification accuracy is made into a confusion matrix.
In the embodiment of the invention, a training data set is formed by using a time-frequency map with type labels of 0-4 under a medium 161 channel according to 80% of absolute path addresses and folder names which account for 228881 in total, and a test data set is formed by the absolute path addresses and folder names which account for 20% in total. Inputting all time-frequency matrixes (instant frequency spectrum) in a training type set into a ResNet network of a Python software database, iteratively updating parameters of neurons, finishing training when a maximum training period max _ epoch =100 is reached, inputting a test data set into the trained ResNet network, counting the number of output type labels equal to the number of real type labels of the unmanned aerial vehicle, dividing the counted result by the total number of samples of the test data set to obtain an identification rate, and making each type of identification rate into a confusion matrix
The structure of the confusion matrix is shown in fig. 3, wherein w rows and q columns take values of 1-5 to represent 5 types of unmanned aerial vehicles, and H representswqIndicating the recognition rate of q when the test is of type w.
And 5, inputting the time-frequency matrix in the time-frequency matrix data set into a trained ResNet (Residual Network) Network, and outputting the type serial number of the unmanned aerial vehicle.
Specifically, in a time-frequency matrix data set generated after the communication signals of the unmanned aerial vehicles are enhanced by data, 20% of absolute path addresses of samples and folder names are randomly selected to form a test set, the test set is input into a trained ResNet network, the serial number of type labels of each type of unmanned aerial vehicle is output, and a formula is utilized to utilize
Calculating the recognition rates of various unmanned aerial vehicles, and making the recognition rates of all types of unmanned aerial vehicles into a confusion matrix, wherein P isVThe total number of the unmanned aerial vehicle types belonging to the V type in the sequence numbers of all the unmanned aerial vehicle type labels output by the ResNet network is represented, Q represents the total number of the test set samples, TVIndicating the accuracy of identification as a type V drone.
In the embodiment of the invention, in a time-frequency matrix data set generated after the unmanned aerial vehicle communication signals are enhanced by using data, absolute path addresses accounting for 20% of the total number and folder names are randomly selected to form a test set. Inputting the test set into a trained ResNet network, counting the number of the serial numbers of the output type labels equal to the real serial number of the unmanned aerial vehicle, dividing the result after each type of counting by the total number of the samples of the test set to obtain an identification rate, and making the identification rate into a confusion matrix
The results are shown in FIG. 4.
The unmanned aerial vehicle time-frequency spectrum reconstructed by the method comprises at least 5 types of unmanned aerial vehicle time-frequency spectrums, each type comprises 10000 samples, and when an unmanned aerial vehicle identification task is executed, the problems that the traditional unmanned aerial vehicle time-frequency spectrum aliasing is serious and the unmanned aerial vehicle time-frequency spectrum identification performance of a neural network is poor due to various urban interferences are solved. Secondly, by means of a communication signal data enhancement technology, background noise with controllable noise power is added to unmanned aerial vehicle communication signals, the hidden danger of aliasing interference can be eliminated, the diversity of samples is increased from the signal source of the unmanned aerial vehicle, the time-frequency spectrum identification performance of the neural network on the unmanned aerial vehicle is improved, and the problem that traditional data enhancement is not convenient to directly apply to the unmanned aerial vehicle communication signals due to the fact that the unmanned aerial vehicle communication signals follow the limitation of a fixed format and a transmission rule is solved.
The effect of the present invention is further explained by combining the simulation experiment as follows:
1. simulation conditions are as follows:
the hardware platform of the simulation experiment of the invention is as follows: intel (R) Core (TM) i7-8700k CPU, main frequency of processor 3.20GHz, memory 16GB, one USRPX310 and 1 fiber.
The software platform of the simulation experiment of the invention is as follows: GNUradio software platform, MATLABR2020a software platform and Python 3.7 software platform.
2. Simulation content and result analysis:
in order to verify the quality of the time-frequency spectrum database of the unmanned aerial vehicle constructed by the invention, the type of the unmanned aerial vehicle is classified by using a residual error neural network ResNet of the database in a Python 3.7 software platform as a classifier. And randomly selecting the absolute path address and the folder name of 80% of samples of 6 different types of unmanned aerial vehicles in an initial time-frequency spectrum database to form a training set, and the absolute path address and the folder name of 20% of samples to form a test set, inputting the training set into a ResNet classifier, and obtaining the trained ResNet classifier after 100 times of iterative training. Inputting the test set into a trained ResNet classifier, outputting predictions of the ResNet classifier on the type of each sample in the test set, and calculating the ratio of the number of the test samples, the types of which are predicted by the ResNet classifier on each sample in the test set and are consistent with the types of the samples, to the total number of the test samples, so as to obtain the recognition accuracy of the 6 unmanned aerial vehicles, wherein the recognition accuracy is shown in a confusion matrix of an initial time-frequency database in fig. 3. In a time-frequency spectrum data set generated after the communication signals of the unmanned aerial vehicles are enhanced by data, the absolute path addresses of 20% of samples of 6 different types of unmanned aerial vehicles in a time-frequency spectrum database and folder names are randomly selected to form a test set, the test set is input into a trained ResNet classifier, the ratio of the number of test samples, the type of each sample predicted by the ResNet classifier in the test set is consistent with the type of the sample, to the total number of the test samples is calculated, and the identification accuracy of the 6 types of unmanned aerial vehicles is obtained and is shown in a confusion matrix of an extended time-frequency database in fig. 4. The higher the target identification accuracy, the better the quality of the unmanned aerial vehicle time-frequency spectrum database constructed by the representation.
The abscissa in fig. 3 and 4, from left to right, represents the predicted type of 6 drones and the 7 types predicted to be without drone signals, respectively, and the ordinate, from top to bottom, represents the true type of 6 drones and the 7 types without drone signals, respectively. The values of the elements in the matrices of fig. 3 and 4 respectively indicate the probability that each type is predicted as each type on the abscissa on the ordinate, the values on the diagonal lines marked by the black areas in fig. 3 and 4 respectively indicate the probabilities of 7 types of correct predictions, and the probability of recognition is expressed as the fraction of the ratio.
By comparing the confusion matrix in fig. 3 and fig. 4, the accuracy of the diagonal line of the confusion matrix in fig. 4 is higher than that in fig. 3, so that the classification accuracy of the time-frequency spectrum generated by adopting the data enhancement of the communication signal of the unmanned aerial vehicle is higher than that of the classification accuracy without adopting the data enhancement, and therefore, the method can realize effective estimation and meet the engineering requirement of the communication signal identification of the unmanned aerial vehicle. Therefore, compared with the prior art, the method provided by the invention solves the problems that the traditional time-frequency map has poor data set quality due to frequency spectrum aliasing, the texture structure of original data can be damaged when the time-frequency map quality is improved through image data enhancement, the unmanned aerial vehicle communication signal follows a fixed format and a transmission rule, the data enhancement is not convenient to be directly applied to the unmanned aerial vehicle communication signal, the data set with high quality cannot be generated, and the identification performance of a classification system is poor.
The method can be used for the identification scenes of targets of a micro unmanned aerial vehicle, a light unmanned aerial vehicle and a small unmanned aerial vehicle.
In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implying any number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the present invention, "a plurality" means two or more unless specifically defined otherwise.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic data point described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples described in this specification can be combined and combined by those skilled in the art.
The foregoing is a more detailed description of the invention in connection with specific preferred embodiments and it is not intended that the invention be limited to these specific details. For those skilled in the art to which the invention pertains, numerous simple deductions or substitutions may be made without departing from the spirit of the invention, which shall be deemed to belong to the scope of the invention.
Claims (9)
1. A time-frequency map reconstruction method based on unmanned aerial vehicle communication signal data enhancement is characterized by comprising the following steps:
step 1, collecting various types of unmanned aerial vehicle communication signals;
step 2, determining a variance of additive white Gaussian noise based on a plurality of first short sampling sequences obtained by segmenting the unmanned aerial vehicle communication signal, and reconstructing the first short sampling sequences based on the variance of the additive white Gaussian noise to obtain second short sampling sequences;
step 3, determining a time-frequency matrix corresponding to the second short sampling sequence;
step 4, obtaining a time frequency matrix data set after the construction data is enhanced according to all the time frequency matrixes, wherein the time frequency matrix data set comprises a plurality of uniquely named time frequency matrixes;
and 5, inputting the time-frequency matrix in the time-frequency matrix data set into a trained ResNet network, and outputting the type serial number of the unmanned aerial vehicle.
2. The time-frequency spectrum reconstruction method based on unmanned aerial vehicle communication signal data enhancement according to claim 1, wherein the step 2 comprises:
step 2.1, uniformly dividing each section of unmanned aerial vehicle communication signal at intervals of L to obtain x sections of first short sampling sequences;
step 2.2, obtaining power spectral density according to the first short sampling sequence;
step 2.3, obtaining the variance of additive white Gaussian noise according to the signal-to-noise ratio and the power spectral density;
step 2.4, generating an additive white Gaussian noise signal according to the variance of the additive white Gaussian noise;
and 2.5, superposing the signal of the Gaussian white noise to the first short sampling sequence to obtain a second short sampling sequence.
3. The time-frequency spectrum reconstruction method based on unmanned aerial vehicle communication signal data enhancement of claim 2, wherein the power spectral density is:
wherein the content of the first and second substances,
representing the power spectral density of the first short sample sequence, m (i) representing the ith sample point in the first short sample sequence, 1 ≦ i ≦ L, | · | representing an absolute value operation.
4. The time-frequency map reconstruction method based on unmanned aerial vehicle communication signal data enhancement of claim 2, wherein an ith sampling point in the second short sampling sequence is:
where y (i) denotes the ith sample point in the second short sample sequence, m (i) denotes the ith sample point in the first short sample sequence, α denotes a weighting coefficient, fcRepresenting random frequency deviation, fsThe sampling rate of the system is represented, B (i) represents a signal of additive white Gaussian noise, j represents an imaginary unit symbol, and i is more than or equal to 1 and less than or equal to L.
5. The time-frequency spectrum reconstruction method based on unmanned aerial vehicle communication signal data enhancement according to claim 1, wherein the step 3 comprises:
and performing STFT on the second short sampling sequence to obtain a time-frequency matrix corresponding to the second short sampling sequence.
6. The time-frequency map reconstruction method based on unmanned aerial vehicle communication signal data enhancement of claim 5, wherein the time-frequency matrix is:
wherein, the first and the second end of the pipe are connected with each other,
representing a time-frequency matrix of M rows and N columns generated after the f-th second short sampling sequence is subjected to STFT transformation, wherein f is more than or equal to 1 and less than or equal to x, M is more than or equal to 0 and less than or equal to M-1,0 and less than or equal to N-1,M represents the dimension of a time domain in the STFT transformation, N represents the dimension of a frequency domain in the STFT transformation, sigma represents summation operation, y represents summation operation, and y represents the dimension of a time domain in the STFT transformationf(i) Denotes the ith element in the f-th second short sample sequence, w (-) denotes the Hamming Window function, d denotes the length of the STFT transform sliding, d = N, e(·)Indicating an exponential operation with a natural constant e as the base.
7. The time-frequency spectrum reconstruction method based on unmanned aerial vehicle communication signal data enhancement according to claim 1, wherein the step 4 comprises:
and obtaining the named time-frequency matrix according to the type, the channel and the position of the unmanned aerial vehicle to which the time-frequency matrix belongs, and constructing a time-frequency matrix data set through all the named time-frequency matrices.
8. The time-frequency spectrum reconstruction method based on unmanned aerial vehicle communication signal data enhancement of claim 1, wherein the training method of the ResNet network comprises the following steps:
s1, acquiring a training data set, wherein the training data set comprises a plurality of time-frequency maps obtained by performing STFT (space time transform) on unmanned aerial vehicle communication signals;
and S2, inputting the time-frequency map into a ResNet network, and iteratively updating parameters of the ResNet network by using the loss value of the loss function until the loss function is converged to obtain the trained ResNet network.
9. The time-frequency spectrum reconstruction method based on unmanned aerial vehicle communication signal data enhancement of claim 8, wherein the loss function is:
wherein, loss represents the Loss value of the Loss function, nk represents the total number of samples in the training data set, ii and MM both represent the serial numbers of the samples in the training data set, cc represents the serial number of the label of the type of the unmanned aerial vehicle output by the ResNet network, ii is more than or equal to 1 and less than or equal to Nk, MM is more than or equal to 1 and less than or equal to Nk, and yiiccRepresenting a symbolic function, when the sequence number of the real type tag of the sample sequence number ii is equal to the sequence number cc of the unmanned plane type tag output by the ResNet network, yiiccIs 1, otherwise, yiiccIs 0,piiccIndicates the prediction probability of the sequence number cc belonging to the type label when the sample sequence number is ii.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210719338.7A CN115270851A (en) | 2022-06-23 | 2022-06-23 | Time-frequency map reconstruction method based on unmanned aerial vehicle communication signal data enhancement |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210719338.7A CN115270851A (en) | 2022-06-23 | 2022-06-23 | Time-frequency map reconstruction method based on unmanned aerial vehicle communication signal data enhancement |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115270851A true CN115270851A (en) | 2022-11-01 |
Family
ID=83761389
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210719338.7A Pending CN115270851A (en) | 2022-06-23 | 2022-06-23 | Time-frequency map reconstruction method based on unmanned aerial vehicle communication signal data enhancement |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115270851A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117200932A (en) * | 2023-09-28 | 2023-12-08 | 中国南方电网有限责任公司超高压输电公司电力科研院 | Identification method and system of unmanned aerial vehicle frequency emission signal based on neural network |
-
2022
- 2022-06-23 CN CN202210719338.7A patent/CN115270851A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117200932A (en) * | 2023-09-28 | 2023-12-08 | 中国南方电网有限责任公司超高压输电公司电力科研院 | Identification method and system of unmanned aerial vehicle frequency emission signal based on neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108600135B (en) | Method for identifying signal modulation mode | |
| CN114429156B (en) | Radar interference multi-domain characteristic countermeasure learning and detection recognition method | |
| CN107808138B (en) | Communication signal identification method based on FasterR-CNN | |
| CN111428817A (en) | Defense method for resisting attack by radio signal identification | |
| Ghadimi et al. | Deep learning-based approach for low probability of intercept radar signal detection and classification | |
| CN111310700A (en) | Intermediate frequency sampling sequence processing method for radiation source fingerprint feature identification | |
| CN114726692B (en) | SERESESESENet-LSTM-based radiation source modulation mode identification method | |
| CN115270851A (en) | Time-frequency map reconstruction method based on unmanned aerial vehicle communication signal data enhancement | |
| CN116047427B (en) | Small sample radar active interference identification method | |
| CN114900256B (en) | Communication scene recognition method and device | |
| Shi et al. | One-shot HRRP generation for radar target recognition | |
| CN115499092A (en) | Astronomical radio transient signal searching method, system, device and readable storage medium | |
| CN117633588A (en) | Pipeline leakage positioning method based on spectrum weighting and residual convolution neural network | |
| CN114841211A (en) | Unmanned aerial vehicle time-frequency spectrum database construction method based on frequency domain data enhancement and expansion | |
| Zhao et al. | Attention-Based CNN Ensemble for Soil Organic Carbon Content Estimation with Spectral Data | |
| CN116385957A (en) | X-ray image contraband detection method, device, equipment and medium | |
| Ruiz‐Munoz et al. | Super resolution for root imaging | |
| CN113239075B (en) | Construction data self-checking method and system | |
| CN114580468A (en) | Interference signal identification method based on time-frequency waterfall graph and convolutional neural network | |
| CN114722942A (en) | Equipment fault diagnosis method and device, electronic equipment and storage medium | |
| US20210241074A1 (en) | System for generating synthetic digital data for data multiplication | |
| CN114358056A (en) | Unmanned aerial vehicle signal identification method and system based on deep learning | |
| CN115118365B (en) | Spectrum prediction interpretation method and system based on mask | |
| CN117200932A (en) | Identification method and system of unmanned aerial vehicle frequency emission signal based on neural network | |
| CN117033983B (en) | Unmanned ship self-noise detection and identification method and system |
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 |