An Unmanned Aerial Vehicle Fast Identification System and Method Based on Two-Dimensional Code Mapping

Abstract

The present invention discloses a rapid identification system and method for unmanned aerial vehicles based on two-dimensional code mapping. The system includes an unmanned aerial vehicle, a remote control station, and a monitoring device. Among them: the unmanned aerial vehicle transmits a video transmission signal to the ground remote control station, the remote control station transmits a remote control signal to the unmanned aerial vehicle, and the monitoring device collects the remote control signal and the video transmission signal; the method includes the following steps: Step 1, data acquisition and merging; Step 2, obtaining a composite time-frequency of the merged time-domain signal; Step 3, removing random noise; Step 4, extracting the unmanned aerial vehicle remote control signal; Step 5, identifying the unmanned aerial vehicle remote control signal, so as to judge the model of the unmanned aerial vehicle, getting rid of the limitations of the traditional single linear time-frequency transformation or non-linear time-frequency transformation to describe signal characteristics.

Description

Technical Field

[0001] The present invention relates to the identification of unmanned aerial vehicles (UAVs), and specifically to a rapid UAV identification system and method based on two-dimensional code mapping. Background Art

[0002] UAVs were first applied in the military field. Later, due to their advantages such as small size, light weight, and low cost, they have been widely used in the civilian field. With the increasing application scope of UAVs and the increasingly serious phenomenon of "unauthorized flight", on the one hand, the government needs to increase the publicity of safe and standardized use of UAVs through social media and public platforms, so as to enhance people's safety and legal concepts regarding UAVs and create a social environment where UAVs are used in compliance with the law. On the other hand, there is an urgent need to strengthen the countermeasure management of illegal UAVs technically. The signal identification of UAVs is an important prerequisite for their countermeasure and management, and is also a current research hotspot, which has great significance and application prospects in the field of UAV countermeasures.

[0003] The literature (Li Guangwei. Application of Spectrum Detection Technology in the Field of UAV Detection and Countermeasures [J]) summarized the application and prospects of electromagnetic spectrum measurement technology in UAV detection and countermeasures, and gave several detection methods, but there was no specific algorithm verification. The literature (Liu Li. Research on the Analysis and Identification Technology of Civil UAV Frequency Hopping Signals [D]) proposed a classification system for frequency hopping signals, which could realize the automatic classification of UAV frequency hopping signals. However, this method requires a large amount of UAV sample data and the algorithm is relatively complex.

[0004] Two-dimensional code technology has been widely used in all aspects of production and life, such as automatic tracking of parts on the production line, product traceability, medical first aid, e-commerce, media tourism, etc. However, the above applications of two-dimensional codes all focus on recording numbers, letters, and text. How to use the fast response characteristics of two-dimensional codes to map and store two-dimensional codes for time-frequency spectra to achieve rapid identification of UAV models has important research significance. Summary of the Invention

[0005] The purpose of the present invention is to provide a rapid UAV identification system and method based on two-dimensional code mapping, which breaks away from the limitations of traditional single linear time-frequency transformation or non-linear time-frequency transformation to describe signal characteristics.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions:

[0007] A rapid UAV identification system based on two-dimensional code mapping includes a UAV, a remote control station, and a monitoring device. Among them: the UAV transmits video transmission signals to the ground remote control station, the remote control station transmits remote control signals to the UAV, and the monitoring device collects remote control signals and video transmission signals.

[0008] A rapid UAV recognition method based on QR code mapping, comprising the following steps:

[0009] Step 1, data acquisition and merging

[0010] Use a data acquisition system to collect the UAV model, perform analog-to-digital conversion on the received analog signal, and convert the received signal into two digital baseband signals of in-phase component I and quadrature component Q through quadrature demodulation and digital down-conversion. Then, merge the I and Q signals to obtain a time-domain signal, expressed as:

[0011] sig = I + j·Q (1)

[0012] In the formula, j represents the imaginary unit,

[0013] Step 2, perform a composite time-frequency transform on the merged time-domain signal to obtain a composite time-frequency

[0014] First, perform a Wigner-Ville distribution (WVD) transform on the merged time-domain signal to obtain a time-frequency spectrum TFR WVD , then perform a Gabor transform on the merged time-domain signal to obtain a time-frequency spectrum TFR G , and then perform binarization processing on the TFR G result to obtain an auto-term time-frequency spectrum, expressed as:

[0015]

[0016] In the formula, TFR G (t, f) is the Gabor transform time-frequency spectrum, t represents time, f represents frequency, max|TFR G ()| represents finding the maximum value of the elements, || represents finding the absolute value, ε is a threshold, and ε = μ·mean(|TFR G (t, f)|), μ is a threshold factor, taking μ = 0.15; mean() represents finding the mean;

[0017] Multiply the auto-term time-frequency spectrum with the time-frequency spectrum TFR WVD to obtain a composite time-frequency spectrum, expressed as:

[0018] TFR 复合 (t, f) = TFR 自项 (t, f) ◎ TFR WVD (t, f) (3)

[0019] In the formula, ◎ is the basic product operation;

[0020] Step 3, eliminate random noise

[0021] According to the characteristic that the amplitude of random noise is usually weaker than that of the UAV signal, the influence of random noise on the detection of the remote control signal is eliminated by setting an energy threshold;

[0022] Step 4: UAV remote control signal extraction

[0023] According to the differences in the correlation between the remote control signal and the video transmission signal in the time-frequency domain, the method of low-rank matrix recovery is used to allocate the part with stronger correlation to the low-rank matrix, while the part with weak correlation is stored in the sparse matrix. Since the video transmission signal exists continuously within the bandwidth range, and the remote control signal is a stepped diagonal line with frequency varying with time in the time-frequency domain, most of the video transmission signals are allocated to the low-rank matrix, and the remote control signal is allocated to the sparse matrix, thus realizing the separation of the remote control signal and the video transmission signal;

[0024] Step 5: UAV remote control signal recognition;

[0025] A clear time-frequency representation of the remote control signal is obtained from the sparse matrix, thereby extracting the time-frequency spectrum of the remote control signal. The time-frequency spectrum is mapped into a two-dimensional code, and the mapped two-dimensional code is compared with the two-dimensional code words in the UAV remote control signal two-dimensional code library. Using the time-frequency spectrum similarity search method of the two-dimensional code, the characteristics of the extracted time-frequency spectrum diagram are identified to determine the model of the UAV.

[0026] Furthermore, the data acquisition system in step 1 is a data acquisition system based on the AD9361 zero-IF receiver architecture.

[0027] Furthermore, step 2 also includes using a generalized rectangular transform to analyze the merged time-domain signal with time as the horizontal axis and frequency as the vertical axis. By grouping the Num 采 data points collected in step 1, each group of data has Num 组 points, and a total of Num 采 / Num 组 times of analysis are performed. Each time represents a time point. The Num 组 points correspond to the signal frequencies within the Bw bandwidth, and the frequency of each point is Bw / Num 组 .

[0028] Furthermore, step 2 undergoes a total of Num 采 / Num 组 times of complex time-frequency transform processing. Each output can obtain a TFR 复合 time-frequency spectrum. Then, calculate the maximum value of the TFR 复合 time-frequency spectrum of each group of data along the time axis, and splice the maximum values of each group of data to obtain the TFR 复合 time-frequency spectrum diagram of all sample data.

[0029] Further, step 3 includes the following steps:

[0030] Step 3.1: Set the energy threshold to TFR 阈值 (t, f), then:

[0031]

[0032] In the formula: TFR(t, f) is the time-frequency spectrum, t is time, f is frequency, and the threshold adopts the improved Bernsen method, taking the time-frequency spectrum values within a square region centered at the current position (t 0 , f 0 ) with a side length of r, and estimating to obtain:

[0033]

[0034] In the formula, γ is the noise reduction coefficient, N t is the number of time-domain samples, and N f is the number of frequency-domain samples;

[0035] Step 3.2: Set the noise reduction coefficient γ ∈ [0, 1], substitute it into formula (5) with a step size of 0.1 to obtain the thresholds corresponding to different noise reduction coefficients; perform binary processing on TFR(t, f), and count the number of points with an energy value greater than 0 after noise reduction for each threshold according to formula (6):

[0036]

[0037] In the formula, tot(TFR(t, f)) represents the total number of points TFR(t, f) with an energy greater than 0 in the time-frequency spectrum diagram after noise reduction processing;

[0038] Step 3.3: Use the second-order difference to compare the values of two adjacent spectral matrix units, and obtain their first-order difference result, denoted as dif(TFR(t, f)), then:

[0039] dif(TFR(t, f)) = TFR(t, f) - TFR(t - 1, f - 1) (7)

[0040] Perform another difference processing on formula (7), denoted as dif(dif(TFR(t, f))), then:

[0041] dif(dif(TFR(t, f))) = dif(TFR(t, f)) - dif(TFR(t - 1, f - 1)) (8)

[0042] The optimal noise reduction coefficient is the value corresponding to the maximum peak of the sequence dif(dif(TFR(t, f))), and substituting it into formula (5) can calculate the optimal noise reduction threshold.

[0043] Furthermore, the low-rank matrix recovery in step 4 includes first representing the time-frequency spectrum matrix TFR(t,f) as the sum of a low-rank matrix A and a sparse matrix E, and then recovering the low-rank matrix by solving a norm optimization problem, that is, solving the following optimization problem:

[0044]

[0045] In the formula, min a,E rank() represents finding the minimum value for the low-rank matrix A and the sparse matrix E, s.t. represents the constraint condition, rank() represents finding the rank, || || 0 represents finding the 0-norm, λ is the balance factor, taking N t as the number of time-domain samples, N f as the number of frequency-domain samples, and use the fast matrix decomposition algorithm to solve formula (9) to obtain the sparse matrix E.

[0046] Furthermore, the process of the fast matrix decomposition algorithm in step 4 is as follows:

[0047] Step 4.1, regularize the optimization problem formula (9) to obtain the optimization problem formula (10), expressed as:

[0048]

[0049] In the formula, || || 1 is the 1-norm, || || 1.2 is the (1,2)-norm, is the square of the F-norm, where δ = ||TFR(t.f)|| 2 / 1.5, which is the penalty factor for not satisfying the linear equality constraint, is the background prior knowledge-induced nuclear norm, ρ is the initial pre-estimation value of the rank of the low-rank matrix A, taking ρ = 10, σ l is the singular value of the low-rank matrix A, balance parameter one balance parameter two β = 0.5×||TFR(t.f)|| 2 ;

[0050] Step 4.2, iteratively update the matrices A and E alternately:

[0051] When A = A k , then:

[0052]

[0053] When E = E k , then:

[0054]

[0055] In the formula, k represents the number of iterations;

[0056] When the maximum number of iterations reaches 15 times or the error is reached, the iteration ends and the sparse matrix E is obtained.

[0057] Furthermore, the process of mapping the time-frequency feature spectrum in step 5 into a two-dimensional code is as follows:

[0058] Step 5.1.1, data compression

[0059] Convert the time-frequency spectrum color map into a grayscale map and store the time-frequency spectrum map in the QR code;

[0060] Step 5.1.2, QR code generation

[0061] Select a QR code with a symbol version ≥ 30 to store sufficient data. A symbol refers to the square black and white dots that make up the QR code.

[0062] Furthermore, step 5 uses a time-frequency spectrum similarity search method based on a two-dimensional code to identify the features of the extracted time-frequency spectrum map, and the specific steps for determining the model of the unmanned aerial vehicle are as follows:

[0063] Step 5.2.1, set the initial error correction rate level to the lowest level L;

[0064] Step 5.2.2, search for the corresponding two-dimensional code in the two-dimensional code library mapped by the unmanned aerial vehicle remote control signal. If yes, go to step 5.2.3; if no, go to step 5.2.4;

[0065] Step 5.2.3, the corresponding two-dimensional code found in step 5.2.2 is determined to be the two-dimensional code mapped by the desired unmanned aerial vehicle, that is, the corresponding unmanned aerial vehicle model is identified;

[0066] Step 5.2.4, error correction rate level +1;

[0067] Step 5.2.5, determine whether the error correction rate level is greater than the highest level. If no, go to step 5.2.2; if yes, go to step 5.2.6;

[0068] Step 5.2.6, discard the two-dimensional code, that is, the corresponding unmanned aerial vehicle model is not found.

[0069] The present invention has the following beneficial effects:

[0070] The time-frequency transformation method for UAV-acquired signals based on composite time-frequency transformation in the present invention adopts the composite time-frequency transformation of Garbo linear time-frequency transformation and WVD non-linear time-frequency transformation, that is, the time-frequency spectrum obtained by Garbo transformation and the time-frequency spectrum obtained by WVD transformation are subjected to an elementary product. By using the filtering mutual cancellation effect between Gabor transformation and WVD transformation, while suppressing the generated cross terms, the resolution of the time-frequency spectrum is improved, getting rid of the limitations of a single linear time-frequency transformation or non-linear time-frequency transformation, so as to balance time-frequency resolution and overcome interference.

[0071] The present invention adopts a time-frequency filtering method based on low-rank matrix recovery. The UAV signal is transformed into the time-frequency domain by using a composite transformation, and then according to the difference in the correlation of the energy distribution of the remote control signal and the video transmission signal in the time-frequency domain, the remote control signal and the video transmission signal are separated into a sparse matrix and a low-rank matrix through the low-rank matrix recovery method, thereby separating the remote control signal and the video transmission signal and reducing the interference of the video transmission signal on the remote control signal; based on the adaptive noise reduction threshold estimation of quadratic difference, the optimal noise reduction threshold can be obtained, effectively suppressing noise.

[0072] The time-frequency spectrum similarity search method based on two-dimensional code in the present invention transforms the similarity estimation of the time-frequency spectrum diagram into the similarity search of the two-dimensional code through the mapped two-dimensional code. It has a fast recognition speed, is simple and effective, and the two-dimensional code is convenient for storing large-capacity data, and can achieve accurate recognition. Description of the Drawings

[0073] Figure 1 : Flow chart of the UAV fast recognition method based on two-dimensional code mapping in the present invention;

[0074] Figure 2 : Flow chart of the time-frequency spectrum similarity search method based on two-dimensional code in the present invention;

[0075] Figure 3 : Schematic diagram of the UAV fast recognition system based on two-dimensional code mapping in the present invention;

[0076] Figure 4 : Composite time-frequency spectrum of the UAV-acquired signal in the present invention;

[0077] Figure 5 : Time-frequency spectrum of the sampled data in the present invention and the mapped two-dimensional code;

[0078] Figure 6 : Time-frequency spectrum of the remote control signal of the DJI Mavic Air 2 UAV in the present invention and the mapped two-dimensional code library;

[0079] Figure 7 : Time-frequency spectrum of the remote control signal of the DJI Phantom 3 UAV in the present invention and the mapped two-dimensional code library;

[0080] Figure 8: The time spectrum of the remote control signal of the DJI Phantom 4 Pro drone of the present invention and the mapped QR code library;

[0081] Figure 9 : The adaptive optimal noise reduction threshold estimation curve graph of the present invention. Detailed implementation manners

[0082] The following further elaborates on the specific content of the present invention in conjunction with embodiments, but it is not a limitation to the present invention.

[0083] As Figure 3 shown, a fast UAV recognition system based on QR code mapping includes a UAV, a remote control station, and a monitoring device. The UAV transmits a video transmission signal to the ground remote control station. The remote control station transmits a remote control signal to the UAV. The monitoring device collects the remote control signal and the video transmission signal.

[0084] As Figure 1 shown, a fast UAV recognition method based on QR code mapping includes the following steps:

[0085] Step 1: Data acquisition and merging

[0086] A data acquisition system based on the AD9361 zero-IF receiver architecture is used to acquire the signals of a certain model of DJI UAV. The analog / data devices inside the receiver perform analog-to-digital conversion on the received analog signals, and through quadrature demodulation and digital down-conversion, the received signals are converted into two digital baseband signals of the in-phase component I and the quadrature component Q. Then, the I and Q signals are merged to obtain a time-domain signal, expressed as:

[0087] sig = I + j·Q (1)

[0088] In the formula, j represents the imaginary unit,

[0089] Step 2: Perform a composite time-frequency transform on the merged time-domain signal to obtain a composite time-frequency

[0090] The composite time-frequency transform is used to analyze the merged time-domain signal with time as the horizontal axis and frequency as the vertical axis. By grouping the Num 采 acquired data points, each group of data is Num 组 points, and a total of Num 采 / Num 组 times of analysis are performed. Each time represents a time point. The Num 组 points correspond to the signal frequencies within the Bw bandwidth. The frequency of each point is Bw / Num 组 . The implementation method of the composite time-frequency transform is as follows:

[0091] Step 2.1: Perform the WVD transform on the merged time-domain signal to obtain the time-frequency spectrum TFR WVD ;

[0092] Step 2.2: Perform the Gabor transform on the merged time-domain signal to obtain the time-frequency spectrum TFR G , and then binarize the TFR G result to obtain the auto-term time-frequency spectrum;

[0093]

[0094] where TFR G (t, f) is the Gabor transform time-frequency spectrum, t represents time, f represents frequency, max|TFR G ()| represents finding the maximum value of the elements, | | represents finding the absolute value, ε is the threshold, and ε = μ·mean(|TFR G (t, f)|), μ is the threshold factor, taking μ = 0.15, and mean() represents finding the mean;

[0095] Step 2.3: Perform the basic product of the auto-term time-frequency spectrum and the WVD transform result to obtain the composite time-frequency spectrum, expressed as:

[0096] TFR 复合 (t, f) = TFR 自项 (t, f) ◎ TFR WVD (t, f) (3)

[0097] where ◎ is the basic product operation;

[0098] Adopt the combination of the linear time-frequency transform Garbo and the non-linear time-frequency transform WVD, that is, perform the basic product on TFR 自项 (t, f) and TFR WVD (t, f), utilize the filtering mutual cancellation effect between the Gabor transform and the WVD transform to suppress the generated cross terms, and improve the resolution of the time-frequency spectrum to balance the time-frequency resolution and overcome interference;

[0099] The above has gone through a total of Num 采 / Num 组 times of composite transform processing. Each output can obtain the TFR 复合 time-frequency spectrum. Then calculate the maximum value of the TFR 复合 time-frequency spectrum along the time axis for each group of data, and splice the maximum values of each group of data to obtain the TFR 复合 time-frequency spectrum diagram of all sample data, as shown in Figure 4 ;

[0100] Step 3: Remove random noise

[0101] There are many random noises in the actually collected UAV signals. When performing time-frequency analysis, according to the characteristic that the amplitude of random noise is usually weaker than that of UAV signals, by setting an energy threshold, the influence of random noise on the detection of remote control signals is eliminated;

[0102] Step 3.1: Assume the threshold is TFR 阈值 (t, f), then:

[0103]

[0104] In the formula: TFR(t, f) is the time-frequency spectrum, t is the time, f is the frequency, and the threshold adopts the improved Bernsen method, that is, taking the time-frequency spectrum values within a square area with side length r centered at the current position (t 0 , f 0 ) to estimate:

[0105]

[0106] In the formula, γ is the noise reduction coefficient, and in this embodiment, γ = 0.7, N t = 5 is the number of time-domain samples, N f = 5 is the number of frequency-domain samples;

[0107] Generally, the number of noise points in the time-frequency spectrum diagram is much more than the number of signal points, and the energy amplitude is low. Therefore, an initial value can be set first, then noise reduction is performed, and then the total number tot(TFR(t, f)) of points with energy greater than 0 in the time-frequency spectrum diagram after noise reduction processing is calculated. When the noise reduction threshold is close to most of the noise energy, tot(TFR(t, f)) will be greatly reduced, so that a more accurate noise reduction coefficient can be obtained, and a noise reduction threshold closer to the background noise can be obtained;

[0108] Step 3.2: Let the noise reduction coefficient γ ∈ [0, 1], substitute it into formula (5) with a step size of 0.1 to obtain the thresholds corresponding to different noise reduction coefficients; perform binary processing on TFR(t, f), and count the number of points with energy greater than 0 after noise reduction for each threshold according to formula (6):

[0109]

[0110] Step 3.3: Adopt second-order difference, that is, compare the values of two adjacent units to obtain the result of its first-order difference, denoted as dif(TFR(t, f)), then:

[0111] dif(TFR(t, f)) = TFR(t, f) - TFR(t - 1, f - 1) (7)

[0112] Perform a second difference operation on formula (7), denoted as dif(dif(TFR(t,f))), then:

[0113] dif(dif(TFR(t,f))) = dif(TFR(t,f)) - dif(TFR(t - 1,f - 1)) (8)

[0114] The optimal noise reduction coefficient is the value corresponding to the maximum peak of the sequence dif(dif(TFR(t,f))). Substitute it into formula (5) to obtain the optimal noise reduction threshold. As Figure 9 shown, when the noise reduction coefficient γ = -0.7, the optimal noise reduction threshold is obtained;

[0115] Step 4: Drone remote control signal extraction

[0116] According to the differences in the correlation between the remote control signal and the video transmission signal in the time - frequency domain, use the method of low - rank matrix recovery to allocate the part with strong correlation to the low - rank matrix, while the sparse matrix stores the part with weak correlation. The video transmission signal exists continuously within the bandwidth range, while the remote control signal is a stepped oblique (or folded) line with frequency varying with time in the time - frequency domain. Therefore, when separating the two, most of the video transmission signals are allocated to the low - rank matrix, and the remote control signal is allocated to the sparse matrix, thus achieving the purpose of separating the remote control signal and the video transmission signal;

[0117] The low - rank matrix recovery first represents the time - frequency spectrum matrix TFR(t,f) as the sum of a low - rank matrix A and a sparse (noise) matrix E, and then recovers the low - rank matrix by solving a norm optimization problem, that is, solving the following optimization problem:

[0118]

[0119] In the formula, min A,E () represents finding the minimum value for the low - rank matrix A and the sparse matrix E, s.t. represents the constraint condition, rank() represents finding the rank, || || 0 represents finding the 0 - norm, λ is the balancing factor, take Use the fast matrix decomposition algorithm to solve formula (9) to obtain the sparse matrix E. The fast matrix decomposition algorithm includes the following process:

[0120] Step 4.1: Regularize the optimization problem formula (9) to obtain the optimization problem formula (10)

[0121]

[0122] In the formula, || || 1 is the 1 - norm, || || 1.2 is the (1,2) - norm, is the square of the F-norm, where δ = ||TFR(t,f)|| 2 / 1.5, which is a penalty factor for non-satisfaction of linear equality constraints is the background prior knowledge-induced nuclear norm, ρ is the initial pre-estimated value of the rank of the low-rank matrix A, take ρ = 10, σ l are the singular values of the low-rank matrix A, and the balance parameter one The balance parameter two β = 0.5×||TFR(t,f)|| 2 ;

[0123] Step 4.2: Iteratively and alternately update the matrices A and E:

[0124] When A = A k at that time

[0125]

[0126] When E = E k at that time

[0127]

[0128] In the formula, k represents the number of iterations;

[0129] When the maximum number of iterations reaches 15 times or the error at that time, the iteration ends, and the sparse matrix E is obtained, or it is called the time-frequency spectrum of the sampled data, as Figure 5 shown;

[0130] Step 5: UAV remote control signal recognition

[0131] Through the low-rank matrix recovery method, the video transmission signal can be further suppressed, and a clear time-frequency representation of the remote control signal can be obtained in the sparse matrix E, so as to extract the time-frequency spectrum of the remote control signal. Figure 5 Let E be the time-frequency spectrum of the sampled data. On the left is the time-frequency spectrum of the collected UAV remote control signal, and on the right is the mapped two-dimensional code. Here, the two-dimensional code is a QR (Quick Response) code. Map the time-frequency spectrum into a two-dimensional code and compare the mapped two-dimensional code with the two-dimensional code words in the UAV remote control signal two-dimensional code library. Among them: The implementation process of mapping the time-frequency feature spectrum into a two-dimensional code is as follows:

[0132] Step 5.1.1: Data compression

[0133] In order to reduce the amount of data, convert the time-frequency spectrum color map into a grayscale map to facilitate storing the time-frequency spectrum map in the QR code;

[0134] Step 5.1.2: QR code generation

[0135] QR codes have more than 40 different symbol versions from 1 to 40. Symbols refer to the square black and white dots that make up the QR code. In this embodiment, a QR code with a version ≥ 30 (137 symbols × 137 symbols) is selected to store sufficient data;

[0136] The time-frequency spectrum similarity search method based on two-dimensional codes is adopted, as Figure 2 shown, to identify the features of the extracted time-frequency spectrum diagram, and it is determined that the drone is a DJI Phantom 4 Pro model. The specific implementation steps are as follows:

[0137] Step 5.2.1: Set the initial error correction rate level, such as setting it to level L, which is the lowest error correction level;

[0138] Step 5.2.2: Search for the corresponding two-dimensional code in the two-dimensional code library mapped by the drone remote control signal. If so, go to Step 5.2.3; if not, go to Step 5.2.4;

[0139] Step 5.2.3: Determine the two-dimensional code searched in Step 5.2.3 as the two-dimensional code mapped by the desired drone, that is, identify the corresponding drone model;

[0140] Step 5.2.4: Error correction rate level +1. Usually, level L can correct about 7% of errors, level M can correct about 15% of errors, level Q can correct about 25% of errors, and level H can correct about 30% of errors;

[0141] Step 5.2.5: Determine whether the error correction rate level is greater than the highest level (such as level H). If not, go to Step 5.2.2; if so, go to Step 5.2.6;

[0142] Step 5.2.6: Discard this two-dimensional code, that is, the corresponding drone model is not found;

[0143] If the two-dimensional code selected in this embodiment is the one mapped by the DJI Phantom 4 Pro model drone, then the corresponding drone is the DJI Phantom 4 Pro model. The time-frequency spectrum of the remote control signals of the DJI series (DJI Mavic Air 2, DJI Phantom 3, and DJI Phantom 4 Pro) drones and the mapped two-dimensional code library are as Figures 6 to 8 shown. The two-dimensional codes in the drone remote control signal two-dimensional code library can be mapped according to the time-frequency spectra of various drone remote control signals tested in advance and form a drone remote control signal two-dimensional code library. Each two-dimensional code word in the two-dimensional code library is in one-to-one correspondence with the two-dimensional code mapped by the drone remote control signal time-frequency spectrum. In practical applications, the fast search feature of two-dimensional codes can be used to quickly retrieve similar two-dimensional code words in the two-dimensional code library, and then identify the corresponding drone model.

Claims

1. A method for rapid identification of unmanned aerial vehicles (UAVs) based on two-dimensional code mapping, characterized in that, it includes the following steps: Step 1, data collection and merging Use a data collection system to collect the UAV model, and perform analog-to-digital conversion on the received analog signal. Through quadrature demodulation and digital down-conversion, the received signal is converted into two digital baseband signals, the in-phase component I and the quadrature component Q. Then, the I and Q signals are merged to obtain a time-domain signal, expressed as: sig = I + j·Q (1) where j represents the imaginary unit, Step 2, perform a composite time-frequency transform on the merged time-domain signal to obtain a composite time-frequency First, perform the Wiener WVD transform on the merged time-domain signal to obtain the time-frequency spectrum TFR WVD , then perform the Gabor transform on the merged time-domain signal to obtain the time-frequency spectrum TFR G , and then binarize the TFR G result to obtain the auto-term time-frequency spectrum, expressed as: where, TFR G (t, f) is the spectrum during Gabor transform, t represents time, f represents frequency, max|TFR G ()| represents finding the maximum value of elements, || represents finding the absolute value, ε is the threshold, and ε = μ·mean(|TFR G (t, f)|), μ is the threshold factor, taking μ = 0.15, mean() represents finding the mean value; Perform a basic product on the self-term time-frequency spectrum and the time-frequency spectrum TFR WVD to obtain a composite time-frequency spectrum, expressed as: TFR 复合 (t,f) = TFR 自项 (t,f) ◎ TFR WVD (t,f) (3) In the formula, ◎ is the basic product operation; Step 3, eliminate random noise By setting an energy threshold, the influence of random noise on the detection of remote control signals is eliminated. The specific process is as follows: Step 3.1: Set the energy threshold to TFR 阈值 (t, f), then: where TFR(t, f) is the time-frequency spectrum, t is time, f is frequency, and the threshold is obtained by using the improved Bernsen method, taking the time-frequency spectrum values within a square region centered at the current position (t 0 , f 0 ) with side length r, and is estimated as follows: where γ is the noise reduction coefficient, N t is the number of time-domain samples, N f is the number of frequency-domain samples; Step 3.2, Let the noise reduction coefficient γ ∈ [0, 1], substitute it into formula (5) with a step size of 0.1 to obtain the thresholds corresponding to different noise reduction coefficients; perform binary processing on TFR(t, f), and according to formula (6), count the number of points with an energy value greater than 0 after noise reduction for each threshold: In the formula, tot(TFR(t, f)) represents the total number of points TFR(t, f) with an energy greater than 0 in the time-frequency spectrogram after noise reduction processing; Step 3.3, Use second-order difference to compare the values of two adjacent spectral matrix elements, and obtain the first-order difference result, denoted as dif(TFR(t, f)), then: dif(TFR(t, f)) = TFR(t, f) - TFR(t - 1, f - 1) (7) Perform another difference processing on formula (7), denoted as dif(dif(TFR(t, f))), then: dif(dif(TFR(t, f))) = dif(TFR(t, f)) - dif(TFR(t - 1, f - 1)) (8) The optimal noise reduction coefficient is the value corresponding to the maximum peak of the sequence dif(dif(TFR(t, f))). Substitute it into formula (5) to obtain the optimal noise reduction threshold; Step 4, extract the UAV remote control signal According to the differences in the correlation between the remote control signal and the video transmission signal in the time-frequency domain, use the method of low-rank matrix recovery to allocate the part with stronger correlation to the low-rank matrix, and the part with weaker correlation is stored in the sparse matrix. Since the video transmission signal exists continuously within the bandwidth range, and the remote control signal is a stepped diagonal line with frequency changing with time in the time-frequency domain, most of the video transmission signals are distributed in the low-rank matrix, and the remote control signal is distributed in the sparse matrix, thus realizing the separation of the remote control signal and the video transmission signal; Step 5, identify the UAV remote control signal; Obtain a clear time-frequency representation of the remote control signal from the sparse matrix, thereby extracting the time-frequency spectrum of the remote control signal, map the time-frequency spectrum into a two-dimensional code, and compare the mapped two-dimensional code with the two-dimensional codes in the UAV remote control signal two-dimensional code library. Use the time-frequency spectrum similarity search method of the two-dimensional code to identify the characteristics of the extracted time-frequency spectrogram to determine the model of the UAV.

2. The method for rapid identification of UAVs based on two-dimensional code mapping according to claim 1, characterized in that, The data acquisition system in step 1 is a data acquisition system based on the AD9361 zero-IF receiver architecture.

3. The method for rapid identification of unmanned aerial vehicles based on QR code mapping according to claim 1, characterized in that, Step 2 further includes analyzing the merged time-domain signal with time as the horizontal axis and frequency as the vertical axis by using a generalized rectangular transformation. By grouping the Num 采 data points collected in Step 1, with each group having Num 组 points, a total of Num 采 / Num 组 analyses are performed, each representing a time point. The Num 组 points correspond to the signal frequencies within the Bw bandwidth, and the frequency of each point is Bw / Num 组 .

4. The method for rapid identification of unmanned aerial vehicles based on QR code mapping according to claim 3, characterized in that, The said step 2 undergoes a total of Num 采 / Num 组 times of combined time-frequency transformation processing. Each output can obtain a TFR 复合 time-frequency spectrum. Then, calculate the maximum value of the TFR 复合 time-frequency spectrum of each group of data along the time axis. Concatenate the maximum values of each group of data to obtain the TFR 复合 time-frequency spectrum diagram of all sample data.

5. The method for rapid identification of unmanned aerial vehicles based on QR code mapping according to claim 1, characterized in that, the low-rank matrix recovery in step 4 includes first expressing the time-frequency spectrum matrix TFR(t,f) as the sum of a low-rank matrix A and a sparse matrix E, and then recovering the low-rank matrix by solving a norm optimization problem, that is, solving the following optimization problem: where, min A,E rank() represents finding the minimum value for the low-rank matrix A and the sparse matrix E, s.t. represents the constraint condition, rank() represents finding the rank, || || 0 represents finding the 0-norm, λ is the balancing factor, taking N t as the number of time-domain samples, N f as the number of frequency-domain samples, use the fast matrix decomposition algorithm to solve formula (9) to obtain the sparse matrix E.

6. The method for rapid identification of unmanned aerial vehicles based on QR code mapping according to claim 5, characterized in that, the process of the fast matrix decomposition algorithm in step 4 is as follows: Step 4.

1. Regularize the optimization problem formula (9) to obtain the optimization problem formula (10), expressed as: In the formula, || || 1 is the 1-norm, || || 1.2 is the (1,2)-norm, is the square of the F-norm, where δ = ||TFR(t.f)|| 2 / 1.5, which is the penalty factor for not satisfying the linear equality constraint, is the background prior knowledge-induced nuclear norm, ρ is the initial pre-estimated value of the rank of the low-rank matrix A, take ρ = 10, σ l is the singular value of the low-rank matrix A, balance parameter one balance parameter two β = 0.5×||TFR(t.f)|| 2 ; Step 4.

2. Iteratively update the matrices A and E alternately: When A = A k then: When E = E k then: where k represents the number of iterations; When the maximum number of iterations reaches 15 times or the error is reached, the iteration ends and the sparse matrix E is obtained.

7. The method for rapid identification of unmanned aerial vehicles based on QR code mapping according to claim 1, characterized in that, the process of mapping the time-frequency feature spectrum in step 5 into a QR code is as follows: Step 5.1.

1. Data compression Convert the time-frequency spectrum color map into a grayscale map and store the time-frequency spectrum map in the QR code; Step 5.1.

2. QR code generation Select a QR code with a symbol version ≥ 30 to store sufficient data. A symbol refers to the square black and white dots that make up the QR code.

8. The method for rapid identification of unmanned aerial vehicles based on QR code mapping according to claim 1, characterized in that, the method for identifying the features of the extracted time-frequency spectrum map by using the time-frequency spectrum similarity search method based on QR code in step 5, so as to determine the specific model of the unmanned aerial vehicle, the specific steps are as follows: Step 5.2.

1. Set the initial error correction rate level to the lowest level L; Step 5.2.

2. Search for the corresponding QR code in the QR code library mapped by the unmanned aerial vehicle remote control signal. If so, go to step 5.2.3; if not, go to step 5.2.4; Step 5.2.

3. The QR code found in step 5.2.2 is determined to be the QR code mapped by the desired unmanned aerial vehicle, that is, the corresponding unmanned aerial vehicle model is identified; Step 5.2.

4. Error correction rate level +1; Step 5.2.

5. Determine whether the error correction rate level is greater than the highest level. If not, go to step 5.2.2; if so, go to step 5.2.6; Step 5.2.

6. Discard the QR code, that is, the corresponding unmanned aerial vehicle model is not found.

Images