Unmanned aerial vehicle detection and identification method

Abstract

The invention discloses an unmanned aerial vehicle detection and identification method, and belongs to the technical field of unmanned aerial vehicle detection. Feature extraction of unmanned aerial vehicle signals; classifying unmanned aerial vehicle signals; and estimating positioning parameters of the unmanned aerial vehicle. The method comprises the following steps: firstly, acquiring an image transmission signal of the unmanned aerial vehicle through software radio equipment, extracting Mel-frequency cepstral coefficient (MFCC) characteristics without preprocessing, then classifying and identifying the extracted characteristics by using a gated circulation unit neural network model (GRU), and finally estimating unmanned aerial vehicle positioning parameters by using a regularization orthogonal matching pursuit algorithm (ROMP) so as to position the unmanned aerial vehicle. According to the method, MFCC is used as radio frequency fingerprint features for classification and identification, compared with a traditional unmanned aerial vehicle identification method, the method is less affected by weather, light, noise and other environments, the precision is high, the identification speed is high, ROMP is used for estimating positioning parameters, and the three-dimensional positioning error of the unmanned aerial vehicle is smaller than 1 meter.

Description

Technical Field

[0001] The invention belongs to the field of unmanned aerial vehicle detection, and specifically proposes a method for detecting and identifying unmanned aerial vehicles. Background Art

[0002] With the continuous development and progress of drone technology, drones have been widely used in various fields. Therefore, efficient unmanned aerial vehicle supervision technology is needed to solve the challenges brought by drones, such as privacy protection and airspace management. Traditional drone identification methods mainly use photoelectric signals, audio signals, radar signals, etc. for detection and identification, which have certain limitations. Whether it is a detection method based on photoelectric signals, drone sound signals or radar echo signals, it may be limited by interference from external factors in practical applications. In contrast, Radio Frequency Fingerprinting Identification (RFFI) is considered to be a more effective unmanned aerial vehicle supervision technology. This technology classifies and identifies drones by analyzing the unique electromagnetic signals emitted by drones. Although many drones may come from the same manufacturer and are the same model, there are still subtle differences at the hardware level. These differences are reflected in the drone's RF signal like a human fingerprint, as the drone's unique identifier. Compared with photoelectric signals and sound signals, RF signals contain more feature information that can reflect the drone hardware and are not easily affected by the environment. Using the features in these signals as the drone's RF fingerprint can more effectively achieve drone detection and identification.

[0003] The Mel Frequency Cepstral Coefficient (MFCC) feature is used as one of the most common features in the frequency domain of audio recognition because of its simple calculation and ability to effectively characterize audio features. The present invention uses it as the radio frequency fingerprint feature of the drone for the classification and identification of the drone, and then uses the Regularized Orthogonal Matching Pursuit (ROMP) algorithm to estimate the azimuth (AOA) and elevation (AOE) positioning parameters of the drone signal for drone positioning. Summary of the invention

[0004] The purpose of the present invention is to provide a method for detecting and identifying unmanned aerial vehicles, which can more effectively supervise unmanned aerial vehicles.

[0005] The present invention provides the following technical solutions:

[0006] A method for detecting and identifying unmanned aerial vehicles, the technical solution includes:

[0007] Firstly, GRU Radio is used to collect and store drone image transmission signals through the software radio product USRP N210. Then, the MFCC features of the drone signals are extracted and input into the GRU network model for classification and recognition. ROMP is used to estimate the AOA and AOE positioning parameters of the drone signals to locate the drone, thereby achieving complete drone detection and recognition.

[0008] A method for detecting and identifying a drone, the specific steps are as follows:

[0009] Step 1: Use USRP N210 to collect and store drone signals.

[0010] Step 2: Extract MFCC features from the signal collected in step 1 as the drone’s RF fingerprint features.

[0011] Step 3: Input the MFCC features into the GRU network model for drone classification and identification.

[0012] Step 4: Use the ROMP algorithm to estimate the positioning parameters of the drone, and then calculate the specific location of the drone by combining the positioning parameters with the drone positioning model. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a schematic diagram of the Mel-frequency cepstral coefficient extraction process in the present invention;

[0014] Figure 2 It is a structural diagram of the GRU network model in the present invention;

[0015] Figure 3 This is a schematic diagram of the UAV positioning model in the present invention;

[0016] Figure 4 This is the entire UAV detection and identification flow chart of the present invention; DETAILED DESCRIPTION

[0017] In order to make the technical problems and technical solutions to be solved by the present invention clearer, the present invention is further described in detail below in conjunction with specific embodiments. The specific embodiments herein are only used to explain the present invention and are not limited to the present invention.

[0018] This embodiment provides a method for detecting and identifying drones, designs a lightweight, fast-recognition, and high-precision GRU neural network, and the system extracts the Mel-frequency cepstral coefficients of the drone image transmission signal as the radio frequency fingerprint feature so that the network model can achieve better performance and better recognition effect. The ROMP algorithm is used in drone positioning to estimate the drone positioning parameters, so that its three-dimensional positioning error is less than 1 meter. The present invention is further described below in conjunction with the accompanying drawings:

[0019] like Figure 1 As shown, a method for detecting and identifying a drone specifically comprises the following steps:

[0020] Step 1: Use GNU Radio software to use USRP N210 to collect and store the image transmission signal data of 5 drones.

[0021] Step 2: extract the Mel cepstral coefficients of the data collected in step 1 as the UAV RF fingerprint features. The specific steps are as follows.

[0022] 2.1: Assume that the drone image transmission signal is x(n), and perform windowing processing on it, where N is the length of the input wireless signal and μ is the Hamming window coefficient:

[0023]

[0024] 2.2: Perform discrete Fourier transform on the windowed data x'(n) to transform it into frequency domain data:

[0025]

[0026] 2.3.1: Determine the minimum frequency f L , maximum frequency f H And the number of Mel filters M, in this embodiment, M is 50.

[0027] 2.3.2: The center frequency of each filter is calculated as follows:

[0028]

[0029] in N is the length of discrete Fourier transform, f s is the sampling frequency.

[0030] 2.3.3: The function expression of the Mel filter bank is as follows:

[0031]

[0032] 2.4: The frequency domain data obtained after discrete Fourier transform is filtered through the Mel filter bank, and the output of the Mel filter bank is logarithmically operated to obtain the Mel spectrum, which is calculated as follows:

[0033]

[0034] Where, m = 1, 2, ..., M, M is the number of filters; H m (k) is the mth filter.

[0035] 2.5: Perform discrete cosine transform on the desired logarithmic power spectrum to obtain the Mel frequency cepstrum coefficient. DCT can convert the time domain signal into the frequency domain signal and retain the energy information of the signal. The calculation formula is as follows:

[0036]

[0037] Step 3: Build a GRU network model as the classification and recognition module of the system. The model is built based on MATLAB DeepLearning Toolbox. The GRU network structure mainly includes input layer, GRU layer, fully connected layer, and output layer. The ReLU activation function is used to introduce nonlinearity into the neural network, and the Adam algorithm is selected as the optimization function.

[0038] Step 4: Estimation of the UAV positioning parameters. Use Wireless InSite software to simulate the campus environment and use the ROMP algorithm to perform three-dimensional positioning of the UAV in the campus environment. The UAV positioning model is as follows: Figure 3 shown.

[0039] The ROMP algorithm is an improved version of the OMP algorithm. The algorithm first selects the atoms with the largest absolute value of the inner product with the residual from the measurement matrix as the candidate set, then selects the relevant atoms from the candidate set according to the regularization principle, and finally merges them into the final support set, thereby achieving fast and effective selection of atoms.

[0040] Specifically, the present invention collects 100 sets of data for each drone, for a total of 500 sets. After extracting the MFCC features, the data is randomly divided into a training set and a test set at a ratio of 9:1, and then the training set is input into the GRU model for training, and finally the test set data is used for testing to obtain the classification result of the model for the drone, with a recognition accuracy of 98%.

[0041] The following table shows the average error of the three receivers in estimating the AOA and AOE positioning parameters of the drone using the ROMP algorithm.

[0042] Table 1

[0043]

[0044] At different altitudes of the drone, the average AOA error of the three receivers is less than 0.09 degrees, and the average EOA error is less than 0.6 degrees. Finally, the position of the drone is determined by calculating the positioning parameters. When the confidence level is 60%, the two-dimensional error and three-dimensional error are shown in Table 2.

[0045] Table 2

[0046]

[0047] From the above table, we can see that both the two-dimensional and three-dimensional positioning errors are less than 1 meter, and the error changes slightly with the change of height, which can prove the feasibility of the ROMP algorithm in the present invention to estimate the positioning parameters of the drone. Through the above experimental data, it can be concluded that the drone detection and identification method proposed in the present invention is effective and feasible in practical applications.

Claims

1. A method for detecting and identifying a drone, characterized in that: The steps include: Step 1: Use GNU Radio software to collect drone image transmission signal data in a specific frequency band through the software radio device USRP N210 and store it. Step 2: Extract the Mel-frequency cepstral coefficient features of the data collected in step 1 without preprocessing as the drone RF fingerprint features, and add a label after the features. Step 3: Build a GRU neural network model. Divide the Mel-frequency cepstral coefficient (MFCC) features in step 2 into a training set and a test set at a ratio of 9:

1. Then input the training set into the GRU network model for training. Finally, use the test set data for testing to obtain the classification results of the neural network model for the drone. Step 4: Use the ROMP algorithm to estimate the AOA and AOE positioning parameters of the drone, and calculate the specific location of the drone based on the drone positioning model.

2. A method for detecting and identifying a drone according to claim 1, characterized in that: The output result is a decimal number and a one-dimensional vector, which represent the drone category and the drone's three-dimensional position respectively.

3. A method for detecting and identifying a drone according to claim 1, characterized in that: MFCC is used as the drone radio frequency fingerprint feature, and the number of MFCC features is 50, that is, when obtaining MFCC, the number of Mel filters is 50.

4. A method for detecting and identifying a drone according to claim 1, characterized in that: In step 3, the GRU network model is constructed using MATLAB Deep Learning Toolbox, and MFCC is combined with GRU for drone recognition.

5. A human-machine detection and identification method according to patent claim 1, characterized in that: The positioning parameters of the UAV are estimated based on the Regularized Orthogonal Matching Pursuit (ROMP) algorithm, and finally the UAV position is calculated using the estimated positioning parameters AOA and AOE.

Images