A method for identifying a remote control signal of a drone based on time-frequency distribution

Abstract

The application belongs to the technical field of unmanned aerial vehicle remote control signal recognition system, and proposes an unmanned aerial vehicle remote control signal recognition system and method based on time-frequency distribution, which comprises a data acquisition and merging module, an analysis module, a signal extraction module, a noise elimination module and a signal recognition module, adopts data acquisition and merging, performs generalized rectangle transformation on the merged time domain signal to obtain GRD time-frequency distribution, eliminates random noise, samples time-frequency sequence, and based on a three-dimensional time-frequency curve similarity search method of curvature characteristics, recognizes the extracted time-frequency spectrum characteristics, and recognizes the unmanned aerial vehicle remote control signal. The scheme of the application breaks away from the limitation of traditional Fourier transformation single frequency domain description signal characteristics.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of unmanned aerial vehicle remote control signal identification system, and particularly relates to an unmanned aerial vehicle remote control signal identification method based on time-frequency distribution. BACKGROUND

[0002] In recent years, the miniaturization and low-altitude of unmanned aerial vehicles have made great progress in the technical field, and the application scenarios are increasingly widespread. Small unmanned aerial vehicles can be seen flying and taking pictures in the airspace near urban parks, exhibitions, festivals and other places. At the same time, the widespread use of unmanned aerial vehicles has also brought problems related to privacy and safety, which has brought security risks to aviation activities and key places, promoting the research on unmanned aerial vehicle detection technology. The analysis and identification of unmanned aerial vehicle remote control signals are the premise and important link of unmanned aerial vehicle detection, such as the detection of remote control or image transmission signals when the unmanned aerial vehicle is flying, and the positioning of the unmanned aerial vehicle to interfere with it and prevent it from flying in unsafe areas to protect the safety of low-altitude airspace. Obviously, it is of great significance and value to research and design an unmanned aerial vehicle remote control signal identification and discovery method in a complex and dense communication and frequency band signal environment involving unmanned aerial vehicles in the field of unmanned aerial vehicle countermeasures.

[0003] The document (Li Guangwei. Application of spectrum detection technology in the field of unmanned aerial vehicle detection and countermeasures [J]) summarizes the application and prospect of electromagnetic spectrum measurement technology in the field of unmanned aerial vehicle detection and countermeasures, and gives several detection methods, but there is no specific algorithm verification. The document (Liu Li. Analysis and identification technology of civilian unmanned aerial vehicle frequency hopping signal [D]) proposes a classification system of frequency hopping signals, which can realize the automatic classification of unmanned aerial vehicle frequency hopping signals, but this method needs a large amount of unmanned aerial vehicle sample data and the algorithm is complex.

[0004] Generally, the remote control signal of the unmanned aerial vehicle is a frequency hopping signal, which belongs to a non-stationary signal. The frequency of the signal presents a nonlinear change with time, and the traditional Fourier transform is no longer suitable for the analysis of the frequency hopping signal. Therefore, it is necessary to jointly analyze from the time domain and the frequency domain, hoping to fully show the frequency change of the remote control signal with time. SUMMARY

[0005] The application aims to provide an unmanned aerial vehicle remote control signal identification method based on time-frequency distribution. The method uses time-frequency domain to characterize and process and analyze signals, and gets rid of the limitations of traditional Fourier transform in single frequency domain description of signal characteristics.

[0006] To achieve the above purpose, the application adopts the following technical scheme: an unmanned aerial vehicle remote control signal identification method based on time-frequency distribution, comprising a data acquisition and merging module, an analysis module, a signal extraction module and a signal identification module.

[0007] Further comprising a noise elimination module.

[0008] The unmanned aerial vehicle remote control signal recognition method comprises the following steps:

[0009] S100, based on the acquisition and merging module: data acquisition and merging, using a data acquisition system based on a zero intermediate frequency receiver architecture to acquire unmanned aerial vehicle signals, an analog / digital device in the receiver performing analog-to-digital conversion on the received analog signals, and converting the received signals into two digital baseband signals of in-phase component I and quadrature component Q through quadrature demodulation and digital down-conversion, and then merging the I and Q signals, i.e. time domain signal sig = I + j·Q, j representing an imaginary unit,

[0010] S200, based on the analysis module: performing generalized rectangle transformation on the merged time domain signal to obtain a GRD time-frequency distribution,

[0011] The merged time domain signal is analyzed using generalized rectangle transformation with time as the horizontal axis and frequency as the vertical axis, and the collected Num 采 data points are processed, each group of data being Num 组 points, a total of Num 采 / Num 组 times of analysis, each time representing a time point, Num 组 points corresponding to the signal frequency within the Bw bandwidth, and each point frequency being Bw / Num 组 ,

[0012] The GRD time-frequency distribution is:

[0013]

[0014] In the formula, σ is the kernel factor, α is the asymmetry ratio, τ is the time integration parameter; | | represents the absolute value; sinc is the sinc function; t is the working time, f is the frequency, Sig is the time domain signal, and the subscript # represents the conjugate,

[0015] The above undergoes Num 采 / Num 组 times of generalized rectangle transformation processing, and each time the GRD time-frequency spectrum can be obtained; the maximum value of the GRD time-frequency spectrum along the time axis of each group of data is calculated, and the maximum values of the groups of data are spliced to obtain a GRD time-frequency spectrum maximum value graph of all sample data;

[0016] S300, based on the signal extraction module: extracting the unmanned aerial vehicle remote control signal,

[0017] The low-rank matrix recovery first represents the time-frequency spectrum matrix GRD(t, f) as the sum of a low-rank matrix A and a sparse matrix E, and then solves the norm optimization problem to recover the low-rank matrix, i.e. solving the following optimization problem:

[0018]

[0019] In the formula, represents the minimum value for the low-rank matrix A and the sparse matrix E, s.t. represents the constraint condition, rank() represents the rank, || ||0 represents the 0 norm, λ is the balance factor, and The sparse matrix E is obtained by solving formula (1) using a fast matrix decomposition algorithm;

[0020] S400, based on the signal recognition module, the unmanned aerial vehicle remote control signal is recognized,

[0021] The low-rank matrix recovery method can further suppress the image transmission signal interference, and a clear remote control signal time-frequency representation is obtained in the sparse matrix E, so that the time-frequency characteristics of the remote control signal are extracted, and further compared with the patterns in the unmanned aerial vehicle remote control signal feature library,

[0022] A three-dimensional time-frequency curve similarity search method based on curvature characteristics is used to identify the extracted time-frequency spectrum characteristics, and the determined unmanned aerial vehicle model of the unmanned aerial vehicle is obtained;

[0023] A noise elimination module is used between S300 and S400 to eliminate random noise. Assuming that the threshold is GRD(t, f), then

[0024]

[0025] The threshold adopts the Bernsen method, that is, the average value of the maximum value and the minimum value in a square region with a side length of r=3 and the current position (t0, f0) as the center;

[0026]

[0027] In the formula, N t is the time domain sample number, and N f is the frequency domain sample number;

[0028] In S300, the solution method of the sparse matrix E is as follows:

[0029] a) Regularization of the optimization problem formula (1) is obtained:

[0030]

[0031] In the formula, || ||1 is the 1 norm, and || || 1,2 is the (1, 2) norm, is the square of the F norm, wherein μ=||GRD(t, f)||2 / 1.5 is a penalty factor for not satisfying the linear equation constraint, is the background prior knowledge induced nuclear norm, ρ is the initial estimate of the rank of the low-rank matrix A, ρ = 10 is taken here, and β is the balance parameter The balance parameter two β = 0.5 × ||GRD(t.f)||2;

[0032] b) Iteratively update the matrix A, E:

[0033] When A = A k ,

[0034]

[0035] When E = E k ,

[0036]

[0037] When the maximum number of iterations 10 is reached or the error , the iteration ends, and the sparse matrix E, or the sampling data time-frequency sequence, is obtained;

[0038] In S400, the unmanned aerial vehicle remote control signal recognition includes the following steps:

[0039] S501, according to the remote control signal time-frequency spectrum data, fitting the three-dimensional time-frequency curve of the remote control signal and obtaining the fitting equation,

[0040] For the time-frequency spectrum data in the standard library of the unmanned aerial vehicle remote control signal, N points of time-frequency spectrum data are extracted at equal intervals, and a high-order polynomial function is used to fit the three-dimensional curve, here a second-order polynomial is taken, and the fitting equation is:

[0041] tfr1(t, f) = p 1，00 +p 1，10 ×t+p 1，01 ×f+p 1，20 ×t 2 +p 1，11 ×t×f+p 1，02 ×f 2 (1)

[0042] Where, p 1，00 , p 1,10 , p 1,01 , p 1,20 , p 1,11 , p 1,02 are the coefficients of equation (1), t is the time variable, and f is the frequency variable,

[0043] Similarly, the time-frequency three-dimensional fitting curve of the sampling data is obtained, and the fitting equation is:

[0044] tfr2(t, f) = p 2，00 +p 2，10 ×t+p2，01 ×f + p 2,20 ×t 2 +p 2，11 ×t×f + p 2，02 ×f 2 (2)

[0045] where p 2，00 , p 2，10 , p 2，01 , p 2，20 , p 2，11 , p 2，02 are coefficients of equation (2);

[0046] S502, calculating the curvature of the three-dimensional curve

[0047] a) taking a point set S1 = {s 1，1 , s 1，2 , …, s 1，N} along the curve direction of the three-dimensional curve S1 of the UAV standard library in equal arc length, i (i = 1, 2, … N) is the serial number of data in the point set, and N represents the total number of data; s 1，i represents the i-th data in the point set S1;

[0048] b) calculating the curvature value of each element in the point set S1 on the curve, and then the curvature set of the UAV standard curve S1 is:

[0049] k1 = {k 1，1 , k 1，2 , …, k 1，N}, N = 124 (3)

[0050] where k 1，1 , k 1，2 , …, k 1，N are the curvature values corresponding to the points s 1，1 , s 1，2 , …, s 1，N on the three-dimensional curve S1, and the curvature set can be regarded as a multi-dimensional vector representing the characteristics of a curve;

[0051] S503, estimating the Chebyshev distance of the curvature distribution of the three-dimensional time-frequency curve, and judging the similarity of the three-dimensional curve,

[0052] The Chebyshev distance between the curvature distribution k1 of the three-dimensional time-frequency curve S1 of the UAV and the curvature distribution of the sampling data three-dimensional time-frequency curve S2 is:

[0053]

[0054] When , the threshold value is taken here At this point, the curvature distribution k2 of the three-dimensional time-frequency curve S2 of the sampled data is similar to the curvature distribution k1 of the three-dimensional time-frequency curve S1 of a certain UAV, so it can be determined that the model is a certain UAV.

[0055] Furthermore, the curvature calculation method is as follows:

[0056] Rewrite equation (3) as a parametric equation

[0057] γ1(θ)=(t(θ), f(θ), tfr(θ))

[0058] Where θ is a parameter, r1 is the parametric time-frequency curve of the UAV; tfr(θ) is the parametric time spectrum; t(θ) is the parametric time variable; f(θ) is the parametric frequency variable; then the curvature of the three-dimensional curve S1 of a certain UAV is:

[0059]

[0060] Where γ′1=(t′(θ), f′(θ), tfr′(θ)) is a three-dimensional vector composed of first derivatives, γ″1=(t″(θ), f″(θ), tfr″(θ)) is a three-dimensional vector composed of second derivatives, and || represents the absolute value.

[0061] Similarly, the curvature of the three-dimensional curve S2 of the sampled data is:

[0062] k2={k 2，1 k 2，2 , ..., k 2，N}, N=124 (5)

[0063] Where, k 2，1 k 2，2 , ..., k 2，N These correspond to points s on the three-dimensional curve S2 of the sampled data. 2，1 s 2，2 , ..., s 2，N The curvature value.

[0064] The principle of the present invention and its beneficial effects: (1).

[0065] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0066] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0067] Figure 1 This is a flowchart of the UAV remote control signal recognition method based on time-frequency analysis according to the present invention.

[0068] Figure 2 This is a flowchart of the three-dimensional time-frequency curve similarity search method based on curvature features according to the present invention.

[0069] Figure 3 This is a schematic diagram of the UAV remote control signal recognition system based on time-frequency analysis according to the present invention.

[0070] Figure 4 This is the generalized matrix time spectrum of the signals collected by the UAV in this invention.

[0071] Figure 5 This is a contour plot of the generalized matrix time spectrum of the signal collected by the UAV in this invention on the time-frequency plane.

[0072] Figure 6 This is the time-frequency three-dimensional fitting curve of the DJI Phantom 4 drone from the drone remote control signal standard library of this invention.

[0073] Figure 7 This is the time-frequency three-dimensional fitting curve of the sampling data in this invention. Detailed Implementation

[0074] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0075] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "vertical", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", and "outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0076] In the description of this invention, unless otherwise specified and limited, it should be noted that the terms "installation", "connection" and "linking" should be interpreted broadly. For example, they can refer to mechanical or electrical connections, or internal connections between two components. They can be direct connections or indirect connections through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms according to the specific circumstances.

[0077] This application provides a method for identifying UAV remote control signals based on time-frequency distribution, including a data acquisition module, an analysis module, a signal extraction module, a noise removal module, and a signal recognition module.

[0078] A method for identifying UAV remote control signals based on time-frequency distribution includes the following steps: S100, data acquisition and merging are performed based on a data acquisition and merging module.

[0079] A data acquisition system based on the AD9361 zero-IF receiver architecture was used to collect signals from a DJI drone model. The receiver's internal analog / data components performed analog-to-digital conversion on the received analog signal, and then converted it into two digital baseband signals—in-phase component I and quadrature component Q—through quadrature demodulation and digital down-conversion. The I and Q signals were then combined to obtain the time-domain signal sig = I + j·Q, where j represents the imaginary unit.

[0080] S200. Based on the analysis module, a generalized rectangular transform is performed on the merged time-domain signal to obtain the GRD time-frequency distribution.

[0081] Since the UAV remote control signal is a non-stationary signal, a generalized rectangular transform is used to analyze the merged time-domain signal with time as the horizontal axis and frequency as the vertical axis. This is achieved by processing 8000×2048 data points, with each group containing 2048 points, for a total of 8000 analyses, each representing a single time point. The 2048 points correspond to signal frequencies within a 40MHz bandwidth, with each point having a frequency of 0.019532625MHz.

[0082] The time-frequency distribution of GRD is as follows:

[0083]

[0084] In the formula, σ is the kernel factor, α is the asymmetric ratio, τ is the time integration parameter; || represents the absolute value; sinc is the singer function; t is time, f is frequency, sig is the time-domain signal, and the superscript # indicates the conjugate.

[0085] The data underwent a total of 8000 generalized rectangular transforms, with each output yielding the GRD time spectrum. The maximum GRD time spectrum value along the time axis for each data set was then calculated. These maximum values ​​were then concatenated to obtain the maximum GRD time spectrum value map for all sample data, as shown below. Figure 4 , Figure 5 As shown, t is the time axis, f is the frequency axis, and tfr is the time-frequency axis. Figure 4 Generalized rectangular time-frequency plot for signals collected by UAVs Figure 5 A contour plot of a generalized rectangular time-frequency plot of signals collected by a drone on the time-frequency plane.

[0086] S300, based on a noise removal module, removes random noise.

[0087] There is a lot of random noise in the actual collected drone signals. When performing time-frequency analysis, since the amplitude of random noise is usually weaker than that of drone signals, an energy threshold is set to eliminate the influence of random noise on the detection of remote control signals.

[0088] Assuming the threshold is GDR 阈值 (t, f), then

[0089]

[0090] The threshold is set using the Bernsen method, which takes the average of the maximum and minimum values ​​within a square region centered at the current position (t0, f0) with a side length of r (=3, here we take 3 time points);

[0091]

[0092] In the formula, N t =8000 is the number of time-domain samples, N f =2048 is the number of frequency domain samples.

[0093] The S400, based on a signal extraction module, extracts remote control signals from drones.

[0094] Based on the difference in correlation between remote control signals and image transmission signals in the time-frequency domain, a low-rank matrix recovery method is used to allocate the more strongly correlated parts to the low-rank matrix, while the less strongly correlated parts are stored in the sparse matrix. The image transmission signal is continuous within its bandwidth, while the remote control signal is a stepped (or broken) line with frequency varying over time in the time-frequency domain. Therefore, when separating the two, most of the image transmission signal is placed in the low-rank matrix, and the remote control signal is placed in the sparse matrix, thus achieving the goal of separating the remote control signal from the image transmission signal.

[0095] The low-rank matrix recovery first expresses the time-frequency spectrum matrix GRD(t,f) as the sum of the low-rank matrix A and the sparse (noise) matrix E, and then recovers the low-rank matrix by solving the norm optimization problem.

[0096] That is, solve the following optimization problems:

[0097]

[0098] In the formula, This expression represents finding the minimum value of a low-rank matrix A and a sparse matrix E, where st represents the constraint, rank() represents finding the rank, || ||0 represents finding the zero norm, and λ is the balance factor.

[0099] The sparse matrix E is obtained by solving equation (1) using the fast matrix factorization algorithm.

[0100] Furthermore, the solution method is as follows:

[0101] a) Regularizing the optimization problem equation (1) yields the optimization problem:

[0102]

[0103] In the formula, |||1 is the 1-norm, ||| 1.2 It is a (1,2) norm. Let μ be the square of the F-norm, where μ = ||GRD(t,f)||² / 1.5 is the penalty factor for not satisfying the linear equality constraint. The background prior knowledge-induced nuclear norm, ρ is the initial pre-estimated value of the rank of the low-rank matrix A, here taken as ρ = 10, and the balance parameter is one. The equilibrium parameter β = 0.5 = ||GRD(t,f)||2;

[0104] b) Iteratively update matrices A and E alternately:

[0105] When A=A k hour,

[0106]

[0107] When E=E k hour,

[0108]

[0109] When the maximum number of iterations (15) is reached or the error is... The iteration ends when the time is reached.

[0110] This yields the sparse matrix E, or the time-frequency curve of the sampled data, as shown below. Figure 7 As shown.

[0111] The S500, based on its identification module, identifies drone remote control signals.

[0112] The low-rank matrix recovery method can further suppress interference in the image transmission signal, obtaining a clear time-frequency representation of the remote control signal in the sparse matrix E, thereby extracting the time-frequency features of the remote control signal. For example... Figure 6 The three-dimensional curve in the time-frequency plot of the sampled data time-frequency sequence E is further compared with the patterns in the standard library of UAV remote control signals.

[0113] A three-dimensional time-frequency curve similarity search method based on curvature features is used to identify the extracted time-frequency features, such as... Figure 2 The flowchart shown is a three-dimensional time-frequency curve similarity search method based on curvature features, which identifies the drone as a DJI Phantom 4.

[0114] The specific implementation method is as follows:

[0115] S501, based on the time-frequency data of the remote control signal, fits the three-dimensional time-frequency curve of the remote control signal and obtains the fitting equation.

[0116] like Figure 6 This is a three-dimensional time-frequency fitting curve for the DJI Phantom 4 drone remote control signal standard library. 214 time-spectrum data points were extracted at equal intervals from the DJI Phantom 4 time-spectrum data in the drone remote control signal standard library, as shown below. Figure 6 The lower right figure shows the fitting of a three-dimensional curve using a higher-order polynomial function. Here, a second-order polynomial is used, and the fitting result is as follows. Figure 6 Top right image, Figure 6 The left image is Figure 6 The top right image is a contour plot on the time-frequency plane.

[0117] The fitting equation is:

[0118] tfr1(t,f)=p 1,00 +p 1,10 ×t+p 1,01 ×f+p 1,20 ×t 2 +p 1,11 ×t×f+p 1,02 ×f 2 (1)

[0119] That

[0120] In, p 1,00 =-0.2872,p 1,10 =2.355×10 -7 ,p 1,01 =-5.11,p 1,20 = -1.026 × 10 -14 ,p 1,11 =5.974×10 -8 ,p 1,02 =11.99

[0121] Similarly, we get the following: Figure 7 The time-frequency three-dimensional fitting curve of the sampled data is shown.

[0122] The fitting equation is:

[0123] tfr2(t, f) = p 2，00 +p 2，10 ×t+p 2，01 ×f+p 2,20 ×t 2 +p 2，11 ×t×f+p 2,02 ×f 2 (2)

[0124] Where, p2,00 =-0.2072,p 2,10 =5.566×10 -8 ,p 2，01 =-0.94,p 2,20 = -1.776 × 10 -15 ,p 2,11 = -1.167 × 10 -7 ,p 2,02 =6.736

[0125] S502, Calculate the curvature of the three-dimensional time-frequency curve.

[0126] a) On the DJI Phantom 4 3D curve S1 in the standard library of drones, take a set of points S1 = {s} with equal arc lengths along the curve direction. 1，1 s 1，2 , ..., s 1，N}, i (i = 1, 2, ..., N) are the indexes of the data in the point set, and N represents the total number of data points, here N = 2^14; s 1,i This represents the i-th data point in the time point set S1;

[0127] b) Calculate the curvature value of each element in the point set S1 on the curve. Then the curvature set of the UAV standard curve S1 is:

[0128] k1={k 1，1 k 1，2 , ..., k 1，N}, N=124 (3)

[0129] Where, k 1,1 ,k 1,2 ,…,k 1,N These correspond to points s on the three-dimensional curve S1, respectively. 1,1 ,s 1,2 ,…,s 1,N The curvature value. In fact, the curvature set can be viewed as a multi-dimensional vector representing the characteristics of a curve. Furthermore, the curvature calculation method is as follows:

[0130] Rewrite equation (3) as a parametric equation

[0131] γ1(θ)=(t(θ),f(θ),tfr(θ))

[0132] Where θ is a parameter, r1 is the parametric time-frequency curve of the drone; tfr(θ) is the parametric time spectrum; t(θ) is the parametric time variable; f(θ) is the parametric frequency variable; then the curvature of the DJI Phantom 4 3D curve S1 is:

[0133]

[0134] Where, γ′1＝ (t′(θ), f′(θ), tfr′(θ)) is a three-dimensional vector composed of first-order derivatives, γ″1=(t″(θ), f″(θ), tfr″(θ)) is a three-dimensional vector composed of second-order derivatives, and || is used to find the absolute value.

[0135] Similarly, the curvature of the three-dimensional curve S2 of the sampled data is:

[0136] k2={k 2，1 k 2，2 , ..., k 2，N}, N=124 (5)

[0137] Where, k 2，1 k 2，2 , ..., k 2，N These correspond to points s on the three-dimensional curve S2 of the sampled data. 2，1 s 2，2 , ..., s 2，N The curvature value.

[0138] S503 estimates the Chebyshev distance of the curvature distribution of a three-dimensional time-frequency curve and determines the similarity of the three-dimensional curves.

[0139] Chebyshev distance can be used to evaluate the difference between two multidimensional vectors in the feature space. For any two space curves, the more similar their shapes are, the smaller the Chebyshev distance value and the higher the similarity. Conversely, it indicates low similarity and large difference.

[0140] The Chebyshev distance between the curvature distribution k1 of the DJI Phantom 4 3D time-frequency curve S1 and the curvature distribution of the sampled data 3D time-frequency curve S2 is:

[0141]

[0142] when At that time, the threshold value is taken here. At this point, the curvature distribution k2 of the three-dimensional time-frequency curve S2 of the sampled data is similar to the curvature distribution k1 of the three-dimensional time-frequency curve S1 of the DJI Phantom 4, which indicates that the model is a DJI Phantom 4.

[0143] In the description of this specification, references to terms such as "preferred embodiment," "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0144] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for identifying unmanned aerial vehicle (UAV) remote control signals based on time-frequency distribution, characterized in that, It includes a data acquisition and merging module, an analysis module, a signal extraction module, and a signal recognition module; It also includes a noise removal module; The method for identifying drone remote control signals includes the following steps: S100, based on the acquisition and merging module: Data acquisition and merging. It employs a data acquisition system based on a zero-IF receiver architecture to acquire UAV signals. The receiver's internal analog / data devices perform analog-to-digital conversion on the received analog signals, and then convert the received signals into two digital baseband signals—in-phase component I and quadrature component Q—through quadrature demodulation and digital down-conversion. The I and Q signals are then merged, resulting in the time domain signal sig = I + j·Q, where j represents the imaginary unit. S200, based on the analysis module: performs a generalized rectangular transform on the merged time-domain signal to obtain the GRD time-frequency distribution. The generalized rectangular transform is used to analyze the merged time-domain signal with time as the horizontal axis and frequency as the vertical axis. This analysis is performed on the acquired Num... 采 The data points are processed, and each set of data is Num. 组 Points, total analysis of Num 采 / num 组 Each time represents a point in time, Num 组 The points correspond to the signal frequencies within the Bw bandwidth, and the frequency of each point is Bw / Num. 组 , The time-frequency distribution of GRD is as follows: In the formula, σ is the kernel factor, α is the asymmetric ratio, τ is the time integration parameter; || denotes the absolute value; sinc is the sig function; t is the operating time, f is the frequency, and Sig is the time-domain signal. The superscript # indicates the conjugate. The above has passed through Num 采 / Num 组 After the generalized rectangular transform process, the GRD time spectrum can be obtained for each output; then the maximum value of the GRD time spectrum along the time axis of each group of data is calculated, and the maximum values ​​of each group of data are spliced ​​together to obtain the maximum value map of the GRD time spectrum of all sample data. S300, based on the signal extraction module: extracts remote control signals for UAVs. Low-rank matrix recovery first expresses the time-frequency spectrum matrix GRD(t,f) as the sum of a low-rank matrix A and a sparse matrix E, and then recovers the low-rank matrix by solving a norm optimization problem, i.e., solving the following optimization problem: In the formula, This expression represents finding the minimum value of a low-rank matrix A and a sparse matrix E, where st represents the constraint, rank() represents finding the rank, || ||0 represents finding the zero norm, and λ is the balance factor. The sparse matrix E is obtained by solving equation (1) using the fast matrix factorization algorithm; The S400, based on a signal recognition module, identifies drone remote control signals. The low-rank matrix recovery method can further suppress interference in the image transmission signal, obtaining a clear time-frequency representation of the remote control signal in the sparse matrix E. This allows for the extraction of the time-frequency features of the remote control signal, which are then compared with patterns in the UAV remote control signal feature database. A three-dimensional time-frequency curve similarity search method based on curvature features is used to identify the extracted time-frequency features, thereby determining the specific UAV model. A noise removal module is used between S300 and S400 to remove random noise. Assuming the threshold is GRD(t,f), then... The threshold is set using the Bernson method, which takes the average of the maximum and minimum values ​​within a square region with a side length of r = 3 centered at the current position (t0, f0). In the formula, N t N represents the number of time-domain samples. f The number of frequency domain samples; In S300, the solution method for the sparse matrix E is as follows: a) Regularizing the optimization problem equation (1) yields the optimization problem: In the formula, |||1 is the 1-norm, ||| 1.2 It is a (1,2) norm. Let μ be the square of the F-norm, where μ = ||GRD(t,f)||² / 1.5 is the penalty factor for not satisfying the linear equality constraint. The background prior knowledge-induced nuclear norm, ρ is the initial pre-estimated value of the rank of the low-rank matrix A, here taken as ρ = 10, and the balance parameter is one. The equilibrium parameter β = 0.5 × ||GRD(tf)||2; b) Iteratively update matrices A and E alternately: When A=A k hour, When E=E k hour, When the maximum number of iterations (10) is reached or the error is... When the iteration ends, the sparse matrix E is obtained, which is also called the time-frequency sequence of the sampled data; In the S400, the identification of drone remote control signals includes the following steps: S501, based on the time-frequency spectrum data of the remote control signal, fits the three-dimensional time-frequency curve of the remote control signal and obtains the fitting equation. For the time-spectrum data in the standard library of UAV remote control signals, N points of time-spectrum data are extracted at equal intervals, and a three-dimensional curve is fitted using a high-order polynomial function. Here, a second-order polynomial is used, and the fitting equation is: tfr1(t,f)=p 1，00 +p 1，10 ×t+p 1,01 ×f+p 1,20 ×t 2 +p 1，11 ×t×f+p 1，02 ×f 2 (1) Where, p 1,00 ,p 1,10 ,p 1,01 ,p 1,20 ,p 1，11 ,p 1，02 Let f be the coefficient of equation (1), t be the time variable, and f be the frequency variable. Similarly, the time-frequency three-dimensional fitting curve of the sampled data is obtained, and the fitting equation is: tfr2(t,f)=p 2，00 +p 2，10 ×t+p 2，01 ×f+p 2,20 ×t 2 +p 2,11 ×t×f+p 2,02 ×f 2 (2) Where, p 2,00 ,p 2,10 ,p 2,01 ,p 2,20 ,p 2，11 ,p 2，02 The coefficients of equation (2) are given. S502, Calculate the curvature of a three-dimensional curve. a) On a 3D curve S1 of a certain UAV standard library, take a set of points S1 = {s} with equal arc lengths along the curve direction. 1，1 s 1，2 , ..., s 1，N }, i (i = 1, 2, ..., N) are the indexes of the data in the point set, and N represents the total number of data; s 1，i This represents the i-th data point in the time point set S1; b) Calculate the curvature value of each element in the point set S1 on the curve. Then the curvature set of the UAV standard curve S1 is: k1={k 1，1 ,k 1，2 ,…,k 1，N },N=124 (3) Where, k 1,1 ,k 1,2 ,…,k 1,N These correspond to points s on the three-dimensional curve S1, respectively. 1，1 s 1，2 , ..., s 1，N The curvature values, the curvature set can be regarded as a multidimensional vector representing the characteristics of a curve; S503 estimates the Chebyshev distance of the curvature distribution of a three-dimensional time-frequency curve and determines the similarity of the three-dimensional curves. The Chebyshev distance between the curvature distribution k1 of the three-dimensional time-frequency curve S1 of a certain UAV and the curvature distribution of the three-dimensional time-frequency curve S2 of the sampled data is: when At that time, the threshold value is taken here. At this point, the curvature distribution k2 of the three-dimensional time-frequency curve S2 of the sampled data is similar to the curvature distribution k1 of the three-dimensional time-frequency curve S1 of a certain UAV, so it can be determined that the model is a certain UAV.

2. The method for identifying UAV remote control signals based on time-frequency distribution as described in claim 1, characterized in that, The curvature calculation method is as follows: Rewrite equation (3) as a parametric equation γ1(θ)=(t(θ),f(θ),tfr(θ)) Where θ is a parameter, the curvature of the three-dimensional curve S1 of a certain UAV is: Where γ′1=(t′(θ), f′(θ), tfr′(θ)) is a three-dimensional vector composed of first derivatives, γ″1=(t″(θ), f″(θ), tfr″(θ)) is a three-dimensional vector composed of second derivatives, and || represents the absolute value. Similarly, the curvature of the three-dimensional curve S2 of the sampled data is: k2={k 2,1 ,k 2,2 ,…,k 2,N },N=124 (5) Where, k 2，1 k 2，2 , ..., k 2，N These correspond to points s on the three-dimensional curve S2 of the sampled data. 2，1 s 2，2 , ..., s 2，N The curvature value.

Images