Unmanned aerial vehicle identification method and system, electronic equipment and storage medium

Abstract

The invention relates to the technical field of signal identification, in particular to an unmanned aerial vehicle identification method, which comprises the steps of generating a local frequency modulation pulse compression sequence according to a sample sequence signal; performing feature extraction on the local frequency modulation pulse compression sequence to obtain a cross-correlation feature vector; acquiring a radio frequency signal of a to-be-identified unmanned aerial vehicle; performing feature extraction on the radio frequency signal to obtain a time-frequency feature vector; fusing the time-frequency feature vector and the cross-correlation feature vector to obtain a feature fusion vector; and identifying the feature fusion vector, and determining an identifier corresponding to the to-be-identified unmanned aerial vehicle. According to the technical scheme provided by the invention, the related characteristics of the ZC sequences used by different unmanned aerial vehicles are determined by analyzing the frame structures and the modulation parameters. By using the ZC sequence features with strong correlation characteristics and the TFI features containing richer signal information, the method shows stronger robustness in a low signal-to-noise ratio and burst interference environment, and can effectively cope with a complex electromagnetic environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of signal recognition, in particular to a UAV recognition method and system, an electronic device and a storage medium. BACKGROUND

[0002] As a transformative technology, UAVs are rapidly emerging in many fields. From entertainment, aerial photography, to agricultural operations, logistics transportation and military operations, UAVs have shown remarkable multi-functional characteristics and practical value. However, in recent years, with the widespread use of UAVs, the potential risks of unauthorized or malicious use of UAVs have raised concerns about safety, privacy protection, and compliance with regulations. It is urgent to improve the accuracy and reliability of UAV remote identification algorithms.

[0003] In related technologies, UAV remote identification (RID) can be achieved through non-communication protocols or communication protocols to strengthen the regulation of UAVs. For example, in the process of identifying UAVs through non-communication protocols, UAVs can be identified based on radar echoes, vision, acoustic or radio frequency signals, while in the process of identifying UAVs through communication protocols, UAVs can be identified through orthogonal frequency division multiplexing (OFDM) downlink map signaling and frequency hopping spread spectrum uplink control signals. However, the robustness of UAV remote identification algorithms based on non-communication protocols is poor, and the UAV remote identification algorithms based on communication protocols are greatly affected by environmental factors, resulting in low accuracy of UAV identification.

[0004] CN120123749A discloses a detection and identification method and system based on UAV radio frequency signals. The method enhances the feature extraction capability in different channel states and signal-to-noise ratio environments by performing intermediate frequency conversion, short-time Fourier transform and Gaussian white noise simulation on the radio frequency signals, significantly improving the identification effect of UAVs in complex environments. Adaptive time window is used in combination with the elbow method to balance the identification performance and reasoning speed. After moving the sliding window to intercept the original signal, key information can be retained, and the amount of calculation can be effectively controlled and the time delay can be reduced. A UAV identification framework based on MaR Block is proposed, which can not only extract image-level features from the overall time-frequency graph, but also split each row (corresponding to the time series of different frequency points) into a one-dimensional sequence to deeply capture the details and correlations of each frequency point with the time dimension. At the same time, through the linkage of the deep learning model in the training and reasoning stages, data collection and subsequent learning of new types or unknown targets are supported.

[0005] However, the above identification method does not consider that the identification process of the unmanned aerial vehicle is affected by the environment, such as light, weather, background complexity, and occlusion. To truly realize high-precision and high-robustness identification of unmanned aerial vehicles in complex, dynamic and diversified real environments, the environmental challenges faced by the visual perception link must be overcome, and a comprehensive identification method with strong environmental impact resistance must be constructed. SUMMARY

[0006] In view of the problem in the prior art that the identification accuracy of an unmanned aerial vehicle is affected by not considering environmental factors during identification, the present application provides an unmanned aerial vehicle identification method, system, electronic device and storage medium.

[0007] The present application is realized by the following technical solutions: An unmanned aerial vehicle identification method, comprising: generating a local frequency modulation pulse compression sequence according to a sample sequence signal; extracting features from the local frequency modulation pulse compression sequence to obtain a cross-correlation feature vector; obtaining a radio frequency signal of an unmanned aerial vehicle to be identified; extracting features from the radio frequency signal to obtain a time-frequency feature vector; fusing the time-frequency feature vector and the cross-correlation feature vector to obtain a feature fusion vector; identifying the feature fusion vector to determine the identification corresponding to the unmanned aerial vehicle to be identified.

[0008] Preferably, the local frequency modulation pulse compression sequence is generated according to the sample sequence signal, comprising: determining a sample signal bandwidth according to the sample sequence signal, comprising: dividing the sample sequence signal to obtain a plurality of overlapping data segments; obtaining a periodogram of the plurality of overlapping data segments; performing average calculation according to the plurality of periodograms to determine the signal bandwidth; calculating the total number of subcarriers according to the sample signal bandwidth; generating the local frequency modulation pulse compression sequence according to the total number of subcarriers.

[0009] Preferably, the total number of subcarriers is calculated according to the sample signal bandwidth, comprising: determining a sample autocorrelation sequence of a sample orthogonal frequency division multiplexing signal in the sample sequence signal; finding a peak value of the sample autocorrelation sequence; calculating the total number of subcarriers according to the index corresponding to the peak value and combining the signal bandwidth.

[0010] Preferably, feature extraction is performed on the local frequency modulation pulse compression sequence to obtain a cross-correlation feature vector, including: In the local frequency modulation pulse compression sequence, a preset number of data segments are randomly selected, and each data segment is non-overlapping; For each data segment, the cross-correlation result of the data segment is reshaped; According to the parameter values corresponding to the reshaped cross-correlation results, a sequence cross-correlation feature is selected; The sequence cross-correlation feature is input into a pre-set convolutional neural network to obtain a cross-correlation feature vector.

[0011] Preferably, feature extraction is performed on the radio frequency signal to obtain a time-frequency feature vector, including: The radio frequency signal is converted by using a short-time Fourier transform to obtain a time-frequency graph; The time-frequency graph is subjected to feature extraction to obtain the time-frequency feature; The time-frequency feature is input into a pre-set lightweight neural network to obtain the time-frequency feature vector.

[0012] Preferably, the time-frequency feature vector and the cross-correlation feature vector are fused to obtain a feature fusion vector, including: The time-frequency feature vector and the cross-correlation feature vector are fused by using a probability weighted addition, a feature vector addition, or a feature vector splicing to obtain the feature fusion vector.

[0013] Preferably, the feature fusion vector is identified to determine the identifier corresponding to the to-be-identified unmanned aerial vehicle, including: The feature fusion vector is input into a pre-trained unmanned aerial vehicle identification model, and the identifier corresponding to the to-be-identified unmanned aerial vehicle is output by the unmanned aerial vehicle identification model.

[0014] An unmanned aerial vehicle identification system, including: A first module for generating a local frequency modulation pulse compression sequence according to a sample sequence signal; A second module for performing feature extraction on the local frequency modulation pulse compression sequence to obtain a cross-correlation feature vector; A third module for acquiring a radio frequency signal of a to-be-identified unmanned aerial vehicle; A fourth module for performing feature extraction on the radio frequency signal to obtain a time-frequency feature vector; A fifth module for fusing the time-frequency feature vector and the cross-correlation feature vector to obtain a feature fusion vector; A sixth module for identifying the feature fusion vector to determine the identifier corresponding to the to-be-identified unmanned aerial vehicle.

[0015] An electronic device includes: a memory and a processor, the memory being used to store a computer program; the processor being used to execute the method as described when the computer program is invoked.

[0016] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method.

[0017] Compared with the prior art, the present invention has the following beneficial effects: This invention discloses a drone identification method that generates a local frequency-modulated pulse compression sequence based on a sample sequence signal, extracts features from the local frequency-modulated pulse compression sequence to obtain a cross-correlation feature vector, acquires the radio frequency signal of the drone to be identified, extracts features from the radio frequency signal to obtain a time-frequency feature vector, and then fuses the time-frequency feature vector and the cross-correlation feature vector to obtain a feature fusion vector. The feature fusion vector can then be used to identify and determine the identifier corresponding to the drone to be identified.

[0018] The method provided by this invention makes full use of communication protocol information. Through a more in-depth analysis of frame structure and modulation parameters, it determines for the first time the relevant features of ZC sequences used by different UAVs. By extracting features from local frequency modulation pulse compression sequences (such as ZC sequences) and combining them with the RF signals of the UAVs, the two extracted feature vectors are then fused. This allows for accurate identification of UAVs based on the fused feature vector, thereby improving the accuracy and efficiency of UAV identification.

[0019] Compared with existing technologies, this invention significantly improves the recognition accuracy by exploring different fusion methods of TFI and ZC sequence features. Compared with existing technologies with a RID delay of 0.3172s, this algorithm improves the average RID evaluation index by at least 2.27%.

[0020] In addition, by utilizing the ZC sequence features with strong correlation characteristics and the TFI features containing richer signal information, the present invention exhibits stronger robustness in low signal-to-noise ratio and sudden interference environments, and can effectively cope with complex electromagnetic environments. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of a drone identification scenario involved in a drone identification method proposed in an embodiment of the present invention; Figure 2 A schematic flowchart illustrating a drone identification method provided in an embodiment of the present invention; Figure 3 A schematic diagram illustrating the bandwidth and bandwidth estimation error for integer multiple sampling with different wavelet carrier numbers, provided as an embodiment of the present invention; Figure 4 A schematic diagram illustrating the bandwidth and bandwidth estimation error for non-integer multiple sampling with different wavelet carrier numbers, provided as an embodiment of the present invention; Figure 5 A schematic diagram of an OFDM signal provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the autocorrelation result of a UAV signal provided in an embodiment of the present invention; Figure 7 A schematic diagram illustrating the estimation error of the number of wavelet carriers when sampling at integer multiples, provided in an embodiment of the present invention; Figure 8 A schematic diagram illustrating the estimation error of the number of wavelet carriers when sampling at non-integer multiples, provided in an embodiment of the present invention; Figure 9 This invention provides an embodiment of different UAV RF signals. A schematic diagram of TFI for each sample; Figure 10 This invention provides an embodiment of the RF signal of different drones at SNR=15dB. ; Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0022] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known drone identification technologies, drone identification algorithms, and electronic devices are omitted so as not to obscure the description of the invention with unnecessary detail.

[0023] The terminology used in the following embodiments is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in the specification and appended claims of this invention, the singular expressions “a,” “the,” “the,” and “the” are intended to also include expressions such as “one or more,” unless the context clearly indicates otherwise.

[0024] Drone remote identification solutions can be broadly categorized into two main types: non-communication protocol identification and communication protocol identification. Non-communication protocol identification can be further subdivided into four methods: radar echo-based, visual, acoustic, and radio frequency (RF) signal-based identification. Radar echo-based drone identification methods are prone to missed detections due to the small size and radar cross-section of drones. Visual identification algorithms are limited by the limited number of cameras and discontinuous visual data, resulting in limited coverage and recognition capabilities. Acoustic identification methods are easily affected by environmental noise, and their detection range is very limited. RF signal-based identification algorithms are unaffected by radar cross-section, low-light environments, and non-line-of-sight scenarios; however, drones of the same brand often use the same or similar signal modulation parameters, leading to low distinguishability of time-frequency characteristics such as signal spectrum, power entropy, higher-order cumulants, and cyclic spectral functions. Furthermore, these algorithms are susceptible to interference signals, causing feature distortion.

[0025] Drone signals mainly include downlink image transmission signals based on OFDM and uplink control signals based on frequency hopping spread spectrum. These two types of signals reflect the modulation parameters and communication protocol information of the drone at different levels, and therefore can be used for remote identification of drones.

[0026] Different UAVs have different OFDM signal frame structures, so the autocorrelation characteristics of the cyclic prefix (CP) can be used for remote identification. However, the frame structure used by UAVs is consistent with the LTE standard, and the modulation parameters such as CP length, number of subcarriers, and signal bandwidth are variable. If the CP autocorrelation characteristics are extracted according to fixed parameters, a certain error will occur.

[0027] Remote identification of drones can also be achieved by identifying the modulation parameters of frequency-hopping spread spectrum signals. However, in scenarios sensitive to detection delays, signal segments with short sampling times and small sampling bandwidths are difficult to fully represent visual statistical features. Furthermore, the wide variety of drones means that traversing and searching the spectrum for frequency-hopping signals incurs significant computational overhead.

[0028] Furthermore, frequency-hopping signals represent regularly distributed small targets in the time-frequency graph. Whether using deep learning-based image denoising algorithms or small target enhancement algorithms based on second-order directional derivatives, the target enhancement effect is unsatisfactory under conditions of low signal-to-noise ratio (SNR) and the presence of interfering signals. This is because pixel-level image enhancement typically cannot reliably and consistently improve semantic-level target detection performance; it may even enhance the quality of interfering targets while compromising the quality of detected targets.

[0029] Therefore, several challenges still exist in remote identification based on communication protocols: 1. Remote drone identification algorithms based on non-communication protocols exhibit poor robustness. Radio frequency (RF)-based remote drone identification algorithms typically extract features such as in-phase / quadrature (I / Q) sequences, received signal strength, higher-order cumulants, spectrum, and time-frequency image (TFI), outperforming other algorithms. However, traditional classification methods or neural networks struggle to consistently and reliably extract effective classifications across multiple transmission frequencies, a communication bandwidth of approximately 100 MHz, and long signal durations. Without further extraction of embedded communication protocol information, their performance significantly degrades in scenarios with strong interference, low signal-to-noise ratio, multiple drones coexisting, and non-line-of-sight (NLS) flight.

[0030] 2. Extracting communication protocol information from UAV frequency-hopping spread spectrum (FHSS) signals exhibits poor performance in the presence of interference and low signal-to-noise ratio. Although uplink FHSS signals have higher energy intensity compared to downlink OFDM signals, they typically have narrower transmission bandwidth and shorter duration. Unknown signals with wider bandwidth, longer duration, and higher energy intensity can interfere with signal detectors, making FHSS signal feature extraction challenging.

[0031] 3. Research on the protocol information contained in the downlink OFDM image transmission signal of UAVs is still insufficient. Few studies have paid attention to the variability of CP length, which can lead to symbol shifts in signal frame structure reconstruction. Existing UAV remote identification algorithms that extract protocol information from video transmission signals mainly focus on transmission frequency, bandwidth, interval, and periodicity, while ignoring the preamble sequence embedded in the signal frame, which commonly uses the ZC (Zadoff-Chu) sequence.

[0032] Therefore, this invention proposes a drone identification method. Based on sample sequence signals, a local frequency-modulated pulse compression sequence is generated, and features are extracted from this sequence to obtain a cross-correlation feature vector. Then, the radio frequency (RF) signal of the drone to be identified is acquired, and features are extracted from this signal to obtain a time-frequency feature vector. The time-frequency feature vector and the cross-correlation feature vector are then fused to obtain a feature fusion vector. This feature fusion vector can then be used to identify the drone and determine its identifier. The technical solution provided by this invention fully utilizes communication protocol information. Through a deeper analysis of frame structure and modulation parameters, it identifies for the first time the ZC sequence-related features used by different drones. By extracting features from the local frequency-modulated pulse compression sequence (such as the ZC sequence) and combining this with feature extraction from the drone's RF signal, and then fusing the two feature vectors, the drone can be accurately identified based on the fused feature vector. This improves the accuracy and efficiency of drone identification.

[0033] See Figure 1 , Figure 1 This is a schematic diagram of a drone identification scenario involved in a drone identification method proposed in an embodiment of the present invention. The drone identification scenario may include: drone 110 and terminal device 120.

[0034] Among them, the drone 110 can be linked to the terminal device 120.

[0035] During the flight of the drone 110, the terminal device 120 can acquire the radio frequency signal (RF signal) of the drone 110. By analyzing the RF signal, the feature vector corresponding to the RF signal is obtained, and the drone 110 can be identified based on the obtained feature vector.

[0036] Specifically, the terminal device 120 can first estimate the signal bandwidth based on a large amount of pre-set sample data, and then estimate the total number of subcarriers based on the estimated signal bandwidth. After that, based on the estimated total number of subcarriers, a local frequency modulation pulse compression sequence (such as a ZC sequence) is generated, and thus a cross-correlation feature vector can be generated based on the local frequency modulation pulse compression sequence.

[0037] Furthermore, the terminal device 120 can also extract features based on the real-time acquired radio frequency signals to obtain time-frequency feature vectors, and fuse the cross-correlation feature vectors and time-frequency feature vectors to obtain feature fusion vectors. Then, the feature fusion vectors are identified to determine the identifier corresponding to the drone 110, so that the drone 110 can be identified based on the identifier corresponding to the drone 110.

[0038] It should be noted that RF signals may include multiple signals such as OFDM image transmission signals, FHSS signals, downlink broadcast (OFDM type) signals, wireless fidelity (Wi-Fi) signals, and Bluetooth signals. This embodiment of the invention does not specifically limit the RF signals.

[0039] The OFDM signal may include a ZC sequence, which is a frequency-modulated pulse compression sequence. The UAV can be identified by the ZC sequence.

[0040] In addition, in practical applications, the drone identification scenario may also include a server. After the terminal device 120 obtains the RF signal of the drone 110, it can forward the RF signal to the server, so that the server can identify the drone 110 in the above manner.

[0041] The present invention will be further described in detail below with reference to specific embodiments. These descriptions are for explanation purposes only and are not intended to limit the scope of the invention.

[0042] This invention discloses a method for identifying unmanned aerial vehicles (UAVs), referring to... Figure 2 ,include: Step 201: Generate a local frequency-modulated pulse compression sequence based on the sample sequence signal. Specifically: Because the frequency-modulated pulse compression sequence of a drone exhibits significant autocorrelation, and different types of drones possess different frequency-modulated pulse compression sequences, drones can be identified using their frequency-modulated pulse compression sequences.

[0043] Before identifying drones, a local frequency-modulated pulse compression sequence can be generated based on a large number of sample sequence signals. This sequence can then be used to identify drones in subsequent steps.

[0044] For example, you can select the sampling rate. A 100MHz, 100ms I / Q sequence is used as the sample sequence signal, including a typical frame consisting of 15 OFDM symbols, each OFDM symbol containing 2048 subcarriers, lasting approximately 1ms. Therefore, the 100ms sample sequence signal (containing...) (A sample number of points) is sufficient to cover the protocol information required to implement drone RID.

[0045] Optionally, the terminal device can first determine the sample signal bandwidth based on the sample sequence signal, and then calculate the total number of subcarriers based on the sample signal bandwidth. Finally, it can generate a local frequency modulation pulse compression sequence based on the total number of subcarriers.

[0046] Specifically, the terminal device can first acquire a large number of pre-stored sample sequence signals, and then analyze and process the sample sequence signals according to a pre-set algorithm and formula to obtain the sample signal bandwidth. Afterwards, based on the autocorrelation sequence of the OFDM signal in the sample sequence signal, combined with the sample signal bandwidth, the total number of subcarriers of the sample sequence signal can be calculated, thereby generating a local frequency modulation pulse compression sequence based on the total number of subcarriers.

[0047] For example, the formula for generating a local frequency-modulated pulse compression sequence can be shown in formula (1): (1) in, This indicates a local frequency-modulated pulse compression sequence. It is the physical root index. It is the sequence length, which is typically found in OFDM systems. .

[0048] Furthermore, in the process of determining the sample signal bandwidth, the terminal device can first divide the sample sequence signal to obtain multiple overlapping data segments, and obtain the periodogram of multiple overlapping data segments. Then, it can perform an average calculation based on multiple periodograms to determine the signal bandwidth.

[0049] For example, the sample sequence signal can be divided into the following categories according to formula (2): The signal bandwidth is obtained by averaging the periodograms of each overlapping data segment using formula (3).

[0050] (2) (3) in, The duration is A window function is used to control spectral leakage; It is the offset length, and J is a constant representing the signal on the frequency spectrum.

[0051] For example, when hour, ; like ,but .

[0052] See Figure 3 and Figure 4 , Figure 3 A schematic diagram illustrating the bandwidth and bandwidth estimation error for integer multiple sampling with different wavelet carrier numbers. Figure 4 A schematic diagram illustrating the bandwidth and bandwidth estimation error for non-integer multiple sampling with different wavelet carrier numbers, as shown below. Figure 3 As shown,Figure 3 The text shows different numbers of subcarriers (e.g.) The trend of bandwidth estimation error with signal-to-noise ratio (SNR) under the conditions of 256, 512, 1024, 2048, and 4096. From Figure 3 As can be seen, when sampling at integer multiples, the bandwidth estimation error for different numbers of subcarriers generally decreases with the increase of signal-to-noise ratio, and the curve trends for different numbers of subcarriers are different, indicating that the number of subcarriers affects the accuracy of bandwidth estimation.

[0053] like Figure 4 As shown, Figure 4 The diagram illustrates the variation of bandwidth estimation error with signal-to-noise ratio for different numbers of subcarriers. Figure 3 The comparison shows that the trend of bandwidth estimation error changes differently when sampling in non-integer multiples compared to sampling in integer multiples. Moreover, under the same signal-to-noise ratio, the bandwidth estimation error of sampling in non-integer multiples is relatively larger overall.

[0054] Correspondingly, when the OFDM signal bandwidth is 20MHz, comparing the bandwidth estimation errors of integer sampling and non-integer sampling, it was found that non-integer sampling reduces the bandwidth estimation performance. However, the bandwidth estimation error of commonly used OFDM signals (with a total number of 1024 or 2048 subcarriers) does not exceed 120 kHz. After rounding to the nearest megahertz, the signal bandwidth can be accurately estimated.

[0055] Furthermore, in the process of calculating the total number of subcarriers, the terminal device can first determine the sample autocorrelation sequence of the sample orthogonal frequency division multiplexing signal based on the sample orthogonal frequency division multiplexing signal in the sample sequence signal, find the peak value of the sample autocorrelation sequence, and then calculate the total number of subcarriers based on the index corresponding to the peak value and the signal bandwidth.

[0056] For example, see Figure 5 and Figure 6 , Figure 5 This is a schematic diagram of an OFDM signal. Figure 5 Primarily used to illustrate the structure of OFDM signals, it demonstrates the basic components of an OFDM signal, including subcarriers and a cyclic prefix. OFDM signals transmit by dividing a high-speed data stream into multiple low-speed sub-data streams, which are then modulated onto multiple mutually orthogonal subcarriers. This structure helps improve spectral efficiency and resistance to multipath fading.

[0057] Figure 6 This is a schematic diagram of the autocorrelation results of the UAV signal. The trend of the curve shows that, under specific conditions... A significant peak appeared at the value. Based on the principle of autocorrelation in OFDM signals, this peak corresponds to... Value and the total number of subcarriers of the signal and bandwidth A specific relationship exists.

[0058] Accordingly, the autocorrelation method is used to assist bandwidth estimation, and combined with the autocorrelation sequence of the OFDM signal (as shown in Equation (4)), the autocorrelation sequence of the OFDM signal is determined. A peak appears at [location], by finding [location] The index corresponding to the maximum value The total number of subcarriers can be estimated according to formula (5). .

[0059] (4) (5) in, Represents the autocorrelation sequence. Indicates the sampling rate. Indicates the total number of subcarriers. Indicates bandwidth. This represents the index corresponding to the autocorrelation sequence. express The index corresponding to the maximum value, Indicates the sequence length. Variables (integers) representing discrete signals.

[0060] See Figure 7 and Figure 8 , Figure 7 A schematic diagram illustrating the estimation error of the number of wavelet carriers when sampling in integer multiples. Figure 8 This is a schematic diagram illustrating the estimation error of the number of wavelet carriers when sampling is a non-integer multiple. Figure 7 The text shows different numbers of subcarriers (e.g.) The subcarrier number estimation error varies with the signal-to-noise ratio (SNR) for integer multiple sampling (e.g., 256, 512, 1024, 2048, 4096). As the SNR increases, the estimation error for each subcarrier number gradually decreases, indicating that in high SNR environments, specific methods can achieve high accuracy in subcarrier number estimation for signals sampled at integer multiples. Figure 7 similar, Figure 8 This demonstrates how the subcarrier number estimation error varies with the signal-to-noise ratio when sampling at non-integer multiples. (Comparison) Figure 7 It can be seen that, under the same signal-to-noise ratio, the estimation error of the number of subcarriers sampled in non-integer multiples is relatively large, especially at low signal-to-noise ratios.

[0061] Step 202: Extract features from the local frequency-modulated pulse compressed sequence to obtain cross-correlation feature vectors. Specifically: Corresponding to step 201, the terminal device can perform feature extraction based on the generated local frequency modulation pulse compression sequence to obtain a cross-correlation feature vector, so that the UAV can be identified in subsequent steps based on the cross-correlation feature vector.

[0062] Optionally, the terminal device can first randomly select a preset number of data segments from the local frequency-modulated pulse compression sequence. For each data segment, the cross-correlation results of the data segment can be reshaped, and then a cross-correlation feature vector can be generated based on the reshaped cross-correlation results. The data segments do not overlap.

[0063] Furthermore, the terminal device can select sequence cross-correlation features based on the parameter values ​​corresponding to the multiple reshaped cross-correlation results, and then input the sequence cross-correlation features into a pre-set convolutional neural network to obtain cross-correlation feature vectors.

[0064] Specifically, the terminal device can first resample the sample sequence signal to obtain data segments that match the local frequency-modulated pulse compressed sequence, and then perform cross-correlation operations to obtain cross-correlation results. Afterwards, the terminal device can reshape the cross-correlation results corresponding to each data segment, thereby inputting the sequence cross-correlation features into a pre-set convolutional neural network to obtain cross-correlation feature vectors.

[0065] For example, the terminal device can analyze the frame structure (and discover that the local frequency-modulated pulse compression sequence is stably present in the seven OFDM symbols of most UAV signals) to... Randomly selected from the samples Each non-overlapping data segment has a length of [number] segments. Construct a new sample set .

[0066] Cross-correlation is performed according to formula (6) to obtain the cross-correlation result. Then, the cross-correlation result is reshaped according to formula (7), and the maximum value of each column is calculated to obtain the cross-correlation characteristics of each local frequency modulation pulse compression sequence: .

[0067] (6) (7) in, Represents the autocorrelation sequence. This represents the index corresponding to the autocorrelation sequence. This represents the reconstructed cross-correlation result.

[0068] It should be noted that, based on the parameter values ​​corresponding to the multiple cross-correlation results after reshaping, the sequence cross-correlation features can be selected first, and then the sequence cross-correlation features can be input into a pre-set convolutional neural network to obtain the cross-correlation feature vector.

[0069] For example, the calculated TFI features are input into MobileNetV3 (a lightweight convolutional neural network architecture), and the ZC sequence cross-correlation features are input into a convolutional neural network (CNN) to obtain feature vectors respectively. and ,in, and These are the operations performed by MobileNetV3 and CNN, respectively. and It is network weight.

[0070] Step 203: Acquire the radio frequency signal of the drone to be identified.

[0071] During the flight of a drone, the terminal device can acquire the drone's radio frequency signal (RF signal) in real time, and then process the RF signal so that the drone's corresponding identifier can be determined in subsequent steps, thus completing the identification of the drone.

[0072] It should be noted that the embodiments of the present invention are illustrated using the example of a terminal device acquiring RF signals in real time. However, in practical applications, terminal devices can acquire RF signals in various ways. The embodiments of the present invention do not specifically limit the method of acquiring RF signals.

[0073] Step 204: Extract features from the radio frequency signal to obtain the time-frequency feature vector.

[0074] After obtaining the RF signal, feature extraction can be performed on the RF signal to obtain the feature vectors in the RF signal. In subsequent steps, each feature vector can be identified by a pre-set UAV identification model, thereby completing the identification of the UAV.

[0075] Optionally, in the process of feature extraction of radio frequency signals, the radio frequency signals can first be converted by short-time Fourier transform to obtain a time-frequency graph (TFI), then feature extraction can be performed on the time-frequency graph to obtain time-frequency features, and finally the time-frequency features can be input into a pre-set lightweight neural network to obtain a time-frequency feature vector.

[0076] Specifically, based on pre-set parameters and combined with short-time Fourier transform, RF signals can be converted to obtain new information such as the modulation method and transmission pattern of UAV signals.

[0077] After obtaining the TFI, it can be further processed. By identifying and analyzing the TFI, the time-frequency characteristics of the UAV can be obtained. Then, the time-frequency characteristics can be input into a lightweight neural network (such as MobileNetV3) to process the time-frequency characteristics and obtain the time-frequency feature vector.

[0078] For example, such as Figure 9 As shown, Figure 9 For different drone RF signals The TFI diagram of each sample clearly shows the OFDM and frequency hopping spread spectrum protocol information it contains, such as the frequency distribution of the signal changing over time. This information provides an important basis for subsequent identification.

[0079] Furthermore, the TFI of the RF signal can be calculated using the short-time Fourier transform of formula (8), and for For each original sample, the computational complexity is approximately... Within acceptable limits.

[0080] (8) In formula (8) For window functions, the window length is... This is equal to the size of the Fast Fourier Transform, where L represents the number of original samples. is the variable for Fourier transform.

[0081] Step 205: Fuse the time-frequency feature vector and the cross-correlation feature vector to obtain the feature fusion vector.

[0082] After obtaining the time-frequency feature vector and the cross-correlation feature vector respectively, the time-frequency feature vector and the cross-correlation feature vector can be fused to obtain the feature fusion vector. Thus, by utilizing the ZC sequence feature with strong correlation characteristics and the TFI feature containing richer signal information, it can exhibit stronger robustness in low signal-to-noise ratio and sudden interference environments, and can effectively cope with complex electromagnetic environments.

[0083] It should be noted that in the process of fusing time-frequency feature vectors and cross-correlation feature vectors, probability-weighted addition, feature vector addition, or feature vector concatenation can be used to fuse the time-frequency feature vectors and cross-correlation feature vectors to obtain the feature fusion vector. This embodiment of the invention does not make specific limitations on this.

[0084] Probability Weighted Addition (PWA) Formula (9) is the prediction probability formula, where and They are and Predicted probabilities after fully connected layers and the Softmax function As probability weights, in this invention It can be set to 0.5. The probability-weighted addition method combines the prediction results of the two features by weighting them, taking into account the advantages of TFI and ZC sequence features under different conditions, and to a certain extent balancing the influence of different features on the recognition results.

[0085] (9) In the formula, and They are and The predicted probabilities are obtained after passing through a fully connected layer and a softmax function. The final result is the probability obtained using the PWA method.

[0086] Feature Vector Addition (FVA) The new feature vector is obtained by element-wise addition. ,in, Represents the feature fusion vector. Represents the time-frequency eigenvector. This represents the cross-correlation feature vector. The feature fusion vector retains the magnitude of the original feature vectors. This method directly fuses feature vectors, allowing the two features to complement each other in the same dimension and jointly participate in subsequent prediction probability calculations. It can quickly integrate information from different features and improve recognition efficiency.

[0087] Feature Vector Concatenation (FVC) Will and Concatenate into a new vector ,in, Represents the feature fusion vector. Represents the time-frequency eigenvector. This represents the cross-correlation feature vector. The size of the fused feature vector is twice that of the original vector. This method preserves all original feature information, providing richer data for the neural network, which is then processed through fully connected layers and... Softmax When calculating the predicted probability using functions, the signal characteristics can be reflected more comprehensively, which has an advantage in improving the recognition accuracy.

[0088] Step 206: Identify the feature fusion vector to determine the identifier corresponding to the drone to be identified.

[0089] After obtaining the feature fusion vector, the feature fusion vector can be input into a pre-set UAV recognition model. The UAV recognition model analyzes the feature fusion vector to determine the identification information matched by the feature fusion vector, and then outputs the identification corresponding to the UAV to be identified.

[0090] It should be noted that in practical applications, the network can be trained and its parameters updated first using the classification cross-entropy loss function, thereby obtaining the drone recognition model.

[0091] in, and These are the true labels and the predicted labels, representing the true labels and predicted labels during parameter training and updates. NC The number of data samples is denoted as . During training, network parameters can be continuously adjusted to enable the UAV recognition model to better learn the relationship between features and labels, thereby improving recognition accuracy. For example, the classification cross-entropy loss function can be expressed as shown in formula (10): (10) For example, see Figure 10 , Figure 10 For different UAV RF signals at SNR=15dB As can be clearly seen from the figure, there are obvious cross-correlation peaks. These peaks indicate the correlation between the UAV signal and the corresponding local ZC sequence, proving the feasibility of using ZC sequence for remote UAV identification. At the same time, the distribution of peaks can also reflect the characteristics of the UAV signal burst duration, interval, and arrival period.

[0092] In summary, this invention proposes a method for identifying unmanned aerial vehicles (UAVs). Based on sample sequence signals, a local frequency-modulated pulse (FM) compressed sequence is generated. Features are extracted from the FM compressed sequence to obtain a cross-correlation feature vector. Then, the radio frequency (RF) signal of the UAV to be identified is acquired, and features are extracted from the RF signal to obtain a time-frequency feature vector. The time-frequency feature vector and the cross-correlation feature vector are then fused to obtain a feature fusion vector. This feature fusion vector can then be used to identify the UAV and determine its corresponding identifier. The technical solution provided by this invention fully utilizes communication protocol information. Through deeper analysis of frame structure and modulation parameters, it identifies for the first time the ZC sequence-related features used by different UAVs. By extracting features from the local FM compressed sequence (such as the ZC sequence) and combining it with the UAV's RF signal, and then fusing the two extracted feature vectors, the UAV can be accurately identified based on the fused feature vector. This improves the accuracy and efficiency of UAV identification.

[0093] For example, compared with the prior art, the present invention significantly improves the recognition accuracy by exploring different fusion methods of TFI and ZC sequence features. Compared with the prior art with a RID delay of 0.3172s, the algorithm improves the average RID evaluation index by at least 2.27%.

[0094] In addition, by utilizing the ZC sequence features with strong correlation characteristics and the TFI features containing richer signal information, the present invention exhibits stronger robustness in low signal-to-noise ratio and sudden interference environments, and can effectively cope with complex electromagnetic environments.

[0095] Based on the same inventive concept, this invention also provides a drone identification system, comprising: a first module for generating a local frequency-modulated pulse compression sequence based on a sample sequence signal; a second module for extracting features from the local frequency-modulated pulse compression sequence to obtain a cross-correlation feature vector; a third module for acquiring the radio frequency signal of the drone to be identified; a fourth module for extracting features from the radio frequency signal to obtain a time-frequency feature vector; a fifth module for fusing the time-frequency feature vector and the cross-correlation feature vector to obtain a feature fusion vector; and a sixth module for identifying the feature fusion vector to determine the identifier corresponding to the drone to be identified.

[0096] Based on the same inventive concept, embodiments of the present invention also provide an electronic device. Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention, such as... Figure 11 As shown, the electronic device provided in this embodiment includes: a memory 111 and a processor 112. The memory 111 is used to store a computer program 113; the processor 112 is used to execute the method described in the above method embodiment when the computer program 113 is invoked.

[0097] The electronic device provided in this embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so they will not be described again here.

[0098] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the methods described in the above-described method embodiments.

[0099] This invention also provides a computer program product that, when run on an electronic device, causes the electronic device to implement the method described in the above-described method embodiments.

[0100] If the integrated units described above are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable storage medium can include at least: any entity or device capable of carrying computer program code to a photographing device / terminal device, a recording medium, a computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0101] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0102] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0103] In the embodiments provided by this invention, it should be understood that the disclosed apparatus / device and method can be implemented in other ways. For example, the apparatus / device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0104] It should be understood that, when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0105] It should also be understood that the term “and / or” as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0106] As used in this specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [described condition or event] is detected" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once [described condition or event] is detected," or "in response to detection of [described condition or event]."

[0107] Furthermore, in the description of this invention and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0108] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of the invention include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0109] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying unmanned aerial vehicles (UAVs), characterized in that, include: Generate a local frequency-modulated pulse compression sequence based on the sample sequence signal; Feature extraction is performed on the local frequency-modulated pulse compression sequence to obtain a cross-correlation feature vector; Acquire the radio frequency signal of the drone to be identified; Feature extraction is performed on the radio frequency signal to obtain a time-frequency feature vector; The time-frequency feature vector and the cross-correlation feature vector are fused to obtain a feature fusion vector; The feature fusion vector is identified to determine the identifier corresponding to the drone to be identified.

2. The UAV identification method according to claim 1, characterized in that, The step of generating a local frequency-modulated pulse compression sequence based on the sample sequence signal includes: Based on the sample sequence signal, determine the sample signal bandwidth, including: The sample sequence signal is divided into multiple overlapping data segments; Obtain a periodicity plot of multiple overlapping data segments; The signal bandwidth is determined by averaging multiple periodograms. The total number of subcarriers is calculated based on the sample signal bandwidth. The local frequency modulation pulse compression sequence is generated based on the total number of subcarriers.

3. The UAV identification method according to claim 2, characterized in that, The total number of subcarriers is calculated based on the sample signal bandwidth, including: determining the sample autocorrelation sequence of the sample orthogonal frequency division multiplexing signal based on the sample orthogonal frequency division multiplexing signal in the sample sequence signal; Find the peak value of the autocorrelation sequence of the sample; The total number of subcarriers is calculated based on the index corresponding to the peak value and the signal bandwidth.

4. The UAV identification method according to claim 1, characterized in that, Feature extraction is performed on the locally frequency-modulated pulse compressed sequence to obtain a cross-correlation feature vector, including: In the local frequency modulation pulse compression sequence, a preset number of data segments are randomly selected, and the data segments do not overlap. For each of the data segments, the cross-correlation results of the data segments are reshaped; Based on the parameter values ​​corresponding to the multiple cross-correlation results after reshaping, sequence cross-correlation features are selected; The cross-correlation features of the sequences are input into a pre-defined convolutional neural network to obtain cross-correlation feature vectors.

5. The UAV identification method according to claim 1, characterized in that, Feature extraction is performed on the radio frequency signal to obtain a time-frequency feature vector, including: The radio frequency signal is converted using short-time Fourier transform to obtain a time-frequency diagram; The time-frequency features are obtained by extracting features from the time-frequency graph; The time-frequency features are input into a pre-set lightweight neural network to obtain the time-frequency feature vector.

6. The UAV identification method according to claim 1, characterized in that, The process of fusing the time-frequency feature vector and the cross-correlation feature vector to obtain a feature fusion vector includes: The time-frequency feature vector and the cross-correlation feature vector are fused using probability-weighted addition, feature vector addition, or feature vector concatenation to obtain the feature fusion vector.

7. The method according to claim 1, characterized in that, The step of identifying the feature fusion vector to determine the identifier corresponding to the drone to be identified includes: The feature fusion vector is input into a pre-trained UAV recognition model, and the UAV recognition model outputs the identifier corresponding to the UAV to be identified.

8. A drone identification system, characterized in that, include: The first module is used to generate a local frequency-modulated pulse compression sequence based on the sample sequence signal; The second module is used to extract features from the local frequency-modulated pulse compression sequence to obtain a cross-correlation feature vector. The third module is used to acquire the radio frequency signal of the drone to be identified; The fourth module is used to extract features from the radio frequency signal to obtain a time-frequency feature vector; The fifth module is used to fuse the time-frequency feature vector and the cross-correlation feature vector to obtain a feature fusion vector; The sixth module is used to identify the feature fusion vector and determine the identifier corresponding to the drone to be identified.

9. An electronic device, characterized in that, include: A memory and a processor, the memory being used to store a computer program; the processor being used to execute the method as described in any one of claims 1 to 7 when the computer program is invoked.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.

Images