Unmanned aerial vehicle identity recognition method and device, computer equipment and storage medium
By using deep learning technology and feature extraction models, the problem of low accuracy in remote drone identification has been solved, and high-precision drone identification has been achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-31
- Publication Date
- 2026-04-14
AI Technical Summary
Existing drone identification technologies suffer from low accuracy in remote identification without relying on signal receiving equipment.
By employing deep learning technology, wireless signal features are mapped layer by layer through signal preprocessing, feature extraction models (such as the ERDRBM neural network model), and SoftMax regression models to achieve drone identification.
Without adding signal receiving equipment, the accuracy of remote identification of drones has been improved, and the device's essential hardware characteristics can be more robustly represented.
Smart Images
Figure CN121859283A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of unmanned aerial vehicles (UAVs), and more particularly to UAV identification methods, devices, computer equipment, and storage media. Background Technology
[0002] Currently, the most common drone identification technology relies on broadcast protocols, requiring each drone to broadcast a unique identifier. Radio fingerprinting, on the other hand, is an identification technology based on the unique signal characteristics of radio transmission devices. Its core principle is to capture subtle differences in signal waveforms and spectral characteristics caused by variations in device hardware manufacturing and usage environments, thereby creating a unique and stable device identifier. Radio fingerprinting technology is currently widely used in fields such as individual identification.
[0003] Unmanned aerial vehicle (UAV) identification relies on the interaction between the radio signals emitted by the UAV and ground-based detection equipment. High-precision signal receiving equipment is required, but this generally increases the cost of existing equipment. Conversely, using less precise signal receiving equipment reduces the accuracy of UAV identification. Therefore, existing technologies suffer from low accuracy when remotely identifying flying UAVs without relying on dedicated signal receiving equipment. Summary of the Invention
[0004] This application provides a method, apparatus, computer equipment, and storage medium for identifying unmanned aerial vehicles (UAVs), which can solve the technical problem of low identification accuracy when remotely identifying UAVs in flight without relying on signal receiving equipment.
[0005] In a first aspect, embodiments of this application provide a method for identifying unmanned aerial vehicles (UAVs), wherein the UAV identification method is applied to a UAV identification device in a UAV identification system, the UAV identification system further includes a signal receiving device, and the method includes: In response to an identification command for identifying the drone, the wireless signal from the target drone is acquired from the signal receiving device. The wireless signal is preprocessed to obtain preprocessed input information; The input information is input into the trained feature extraction model, and the input information is mapped layer by layer based on the multi-layer feature extraction model to obtain the feature information corresponding to the target UAV. The wireless signal fingerprint type of the target UAV is obtained by classifying and identifying the feature information based on the preset SoftMax regression model, and the identity of the target UAV is determined based on the wireless signal fingerprint type.
[0006] In some embodiments, the feature extraction model is a neural network model. The step of inputting the input information into the trained feature extraction model and mapping the input information layer by layer based on the multi-layer feature extraction of the feature extraction model to obtain the feature information corresponding to the target UAV includes: The input information is input into the trained neural network model, wherein the neural network model is composed of a multi-layer deep neural network consisting of a multi-layer restricted Boltzmann structure; The multi-layer restricted Boltzmann machine based on the neural network model maps the input information layer by layer to obtain the feature information corresponding to the target UAV.
[0007] In some embodiments, the neural network model is an ERDRBM neural network model, which consists of multiple layers of restricted Boltzmann machines. The feature information corresponding to the target UAV is a feature vector. The multiple layers of restricted Boltzmann machines based on the neural network model map the input information layer by layer to obtain the feature information corresponding to the target UAV, including: The state of the first hidden layer is obtained through the first layer encoder; Initial parameters are obtained through bottom-up unsupervised learning by embedding the states layer by layer. The ERDRBM neural network model is subjected to supervised parameter fine-tuning via backpropagation to obtain the globally optimal feature vector corresponding to the target UAV, thereby achieving feature extraction of the target UAV.
[0008] In some embodiments, when performing supervised parameter fine-tuning of the ERDRBM neural network model via backpropagation, supervised parameter fine-tuning using dynamic parameters is employed to prevent data overfitting.
[0009] In some embodiments, the preprocessing of the wireless signal to obtain preprocessed input information includes: Determine the type of conversion required for the wireless signal; If the conversion type is a two-dimensional image, then the wireless signal is converted into a two-dimensional image; If the conversion type is vector, then it is converted into a vector based on the statistical characteristics of the wireless signal; If the conversion type is sequence, the wireless signal is converted into time-series data, wherein the in-phase or quadrature components of the wireless signal are used as one-dimensional time-series data. The preprocessed data is used as input information.
[0010] In some embodiments, if the conversion type is a two-dimensional image, then converting the wireless signal into a two-dimensional image includes: If the conversion type is a two-dimensional image, the wireless signal is converted to obtain a two-dimensional time-frequency image, wherein the two-dimensional time-frequency image displays the frequency change trend over time, and the conversion method includes either short-time Fourier transform or wavelet transform.
[0011] In some embodiments, before determining the type of conversion required for the wireless signal, the method includes: Adjusting various signal target data to achieve feature alignment and feature classification; The wireless signals are randomly divided into groups, with each group containing multiple wireless signal data. Different groups are numbered to obtain the grouped data sets to be processed, thereby reducing computational complexity.
[0012] Secondly, embodiments of this application also provide a drone identification device, a drone identification method applied to a drone identification system, the drone identification system further including a signal receiving device, and the drone identification device comprising: The acquisition unit is used to acquire wireless signals from the target drone from the signal receiving device in response to an identification command for identifying the drone. A preprocessing unit is used to preprocess the wireless signal to obtain preprocessed input information; The feature extraction unit is used to input the input information into the trained feature extraction model, and map the input information layer by layer based on the multi-layer feature extraction of the feature extraction model to obtain the feature information corresponding to the target UAV. The classification and identification unit is used to classify and identify the feature information based on a preset SoftMax regression model to obtain the wireless signal fingerprint type of the target UAV, and to determine the identity of the target UAV based on the wireless signal fingerprint type.
[0013] Thirdly, embodiments of this application also provide a computer device for drone identification, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method.
[0014] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a processor, can implement the above-described method.
[0015] This application provides a method, apparatus, computer device, and storage medium for drone identification. The method includes: in response to an identification command for drone identification, acquiring a wireless signal from a target drone from a signal receiving device; preprocessing the wireless signal to obtain preprocessed input information; inputting the input information into a trained feature extraction model, mapping the input information layer by layer based on the multi-layer feature extraction of the feature extraction model to obtain feature information corresponding to the target drone; classifying and identifying the feature information based on a preset SoftMax regression model to obtain the wireless signal fingerprint type of the target drone, and determining the identity of the target drone based on the wireless signal fingerprint type. In this application, deep learning is used to map the raw signal layer by layer to a higher-level feature representation that is abstracted layer by layer. Higher-level features are more robust to noise and channel variations and better represent the essential hardware characteristics of the device. High-precision identification of flying drones is achieved remotely without adding existing signal receiving equipment, improving the accuracy of remote identification. Attached Figure Description
[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1 A flowchart illustrating the drone identification method provided in this application embodiment; Figure 2 A schematic block diagram of a drone identification device provided in an embodiment of this application; Figure 3 A schematic block diagram of a computer device provided in an embodiment of this application. Detailed Implementation
[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0019] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.
[0020] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.
[0021] It should also be further understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.
[0022] With the development of drones, the detection technology for communication signals of non-cooperative drones, such as those flying illegally, has been widely and deeply applied, leading to various needs for drone detection and identification. Remote identification of drones in flight is a monitoring technology that broadcasts the drone's identity and location information in real time via wireless signals. The core functions of this monitoring technology include identity recognition, real-time positioning, flight control tracking, and flight status monitoring.
[0023] Currently, the most common drone identification technology relies on broadcast protocols, requiring each drone to broadcast a unique identifier. Radio fingerprinting, on the other hand, is an identification technology based on the unique signal characteristics of radio transmission devices. Its core principle is to capture subtle differences in signal waveforms and spectral characteristics caused by variations in device hardware manufacturing and usage environments, thereby creating a unique and stable device identifier. Radio fingerprinting technology is currently widely used in fields such as individual identification.
[0024] Unmanned aerial vehicle (UAV) identification relies on the interaction between the radio signals emitted by the UAV and ground-based detection equipment. High-precision signal receiving equipment is required, but this generally increases the cost of existing equipment. Conversely, using less precise signal receiving equipment reduces the accuracy of UAV identification. Therefore, existing technologies suffer from low accuracy when remotely identifying flying UAVs without relying on dedicated signal receiving equipment.
[0025] To address the aforementioned issues, embodiments of this application provide a method, apparatus, computer equipment, and storage medium for drone identification.
[0026] The drone identification method is applied to the drone identification device in the drone identification system, which also includes a signal receiving device.
[0027] Figure 1 This is a flowchart illustrating the drone identification method provided in an embodiment of this application. Figure 1 As shown, the method includes the following steps S110-S140: The method includes steps S110-S140: Step S110: In response to the identification command for identifying the drone, obtain the wireless signal from the target drone from the signal receiving device; The identification of a drone relies on the interaction between the radio signals emitted by the drone and ground-based detection equipment. In this embodiment, the ground-based detection equipment is the signal receiving equipment.
[0028] The drone itself emits radio signals, which are then received by a signal receiving device. This device detects and analyzes the specific frequency bands of radio waves transmitted via image transmission or remote control signals, forming a communication link between the drone and the remote controller. The remote control's wireless signals can be detected even before the drone takes off or enters its range.
[0029] In this embodiment, the signal receiving device can be of many types. The signal receiving device can be based on radio spectrum detection, radar detection, photoelectric detection such as visible light or infrared detection, or acoustic detection.
[0030] Radio spectrum detection, due to its long-range early warning and identification capabilities, has become the core and preferred sensor in modern counter-drone detection systems. Radar is a key supplement for dealing with silent flight and achieving all-weather precision surveillance. Electro-optical systems are indispensable "eyes" for final confirmation, identification, and evidence collection. Acoustic detection is typically used as an auxiliary and blind-spot-filling method in specific scenarios.
[0031] The drone identification system includes a drone identification device and a signal receiving device. The signal receiving device is responsible for receiving wireless signals, while the drone identification device is responsible for processing and identifying the wireless signals.
[0032] In response to an identification command to identify the drone, the wireless signal from the target drone is acquired from the signal receiving device.
[0033] Step S120: Preprocess the wireless signal to obtain preprocessed input information; After acquiring the wireless signal, preprocessing is required. The core objective of preprocessing is to transform the raw signal, which contains a large amount of interference, into a clean, standardized data format with prominent features, suitable for machine learning models. Essentially, wireless signal preprocessing is crucial for converting the raw physical layer data acquired during detection into clean information layer data required for analysis. Without high-quality, standardized preprocessing, subsequent performance will be significantly compromised. The effects of preprocessing include resolving signal quality issues by removing artifacts, meeting the input requirements of analysis algorithms through standardized processes, and improving the efficiency and accuracy of subsequent processing by optimizing workflows.
[0034] As an example, digital filters such as bandpass and matched filters are used to filter out out-of-band noise and irrelevant signals to achieve preprocessing of wireless signals.
[0035] As an example, operations such as carrier synchronization, symbol synchronization, and automatic gain control are performed to eliminate linear distortions caused by the channel, align and normalize the signal in the time and frequency domains, and allow subsequent analysis to focus on the inherent nonlinear characteristics of the device.
[0036] Step S120 includes steps S1201-S1205: Step S1201: Determine the conversion type that the wireless signal needs to be converted; Determine the type of conversion required for the wireless signal.
[0037] Before step S1201, steps A1-A3 are included: Step A1: Adjust various signal target data to achieve feature alignment and feature classification; Adjusting various types of signal target data to achieve feature alignment and feature classification improves the decision-making ability of target data.
[0038] The system uses digital filters to remove out-of-band noise for noise reduction and filtering, performs carrier synchronization and symbol synchronization, and performs amplitude normalization to eliminate channel effects. It also locates and extracts signal segments containing complete communication frames, and generates time spectrum diagrams, power spectra, etc., to prepare for feature extraction.
[0039] Step A2: Randomly divide the wireless signals into groups, with each group containing multiple wireless signal data. The signal input data is randomly divided into p groups, with q data points in each group.
[0040] Step A3: Number the different groups to obtain the grouped data to be processed, thereby reducing computational complexity.
[0041] Different groups are numbered to obtain the grouped data sets to be processed, thereby reducing computational complexity.
[0042] Step S1202: If the conversion type is a two-dimensional image, then convert the wireless signal into a two-dimensional image; If the conversion type is a two-dimensional image, then the wireless signal is converted into a two-dimensional image. As an example, if the conversion type is a two-dimensional image, the wireless signal is converted to obtain a two-dimensional time-frequency image, wherein the two-dimensional time-frequency image displays the frequency variation trend over time, and the conversion method includes either short-time Fourier transform or wavelet transform. Besides converting the wireless signal to obtain a two-dimensional time-frequency image, the wireless signal can also be converted to obtain a power spectrum image.
[0043] Here, by capturing subtle differences in signal waveforms and spectral characteristics caused by variations in equipment hardware manufacturing and operating environments, a unique device identifier is formed. Tolerances and nonlinearities inherent in the manufacturing process cannot be completely eliminated at the time of shipment. Each drone device's hardware already carries a unique physical fingerprint. Component tolerances refer to the nominal parameters of core components constituting the RF front-end, such as crystal oscillators, filters, power amplifiers, and mixers. Nominal parameters, such as center frequency, cutoff frequency, and gain, exhibit slight fluctuations. Nonlinearity refers to the unavoidable introduction of nonlinear distortions, such as harmonics, during signal amplification. This nonlinear curve is unique to each amplifier. These microscopic differences in hardware manifest in the final RF signal during signal generation and transmission, appearing as slight carrier frequency shifts, unique distortions in the modulation constellation diagram, specific sidebands or harmonic components in the spectrum, and unique transient shapes. These characteristics are inherent and difficult-to-tamper-with physical properties of the device.
[0044] Therefore, by converting the wireless signal into a two-dimensional time-frequency image or power spectrum image, these two-dimensional images are equivalent to the wireless fingerprint of each drone. Specific drones can be identified by performing feature analysis on the two-dimensional time-frequency image or power spectrum image.
[0045] Two-dimensional time-frequency images, through time-frequency analysis, map a one-dimensional time-domain signal into a two-dimensional function. The horizontal axis represents time, the vertical axis represents frequency, and the brightness or color of a pixel indicates the energy intensity of that frequency component at that moment. The spectrum is typically obtained through short-time Fourier transform or wavelet transform. Two-dimensional time-frequency images preserve the signal's frequency domain information, including its frequency components and distribution, as well as its time domain information, such as how frequency components evolve over time and instantaneous characteristics, such as unique transient patterns at startup, shutdown, or modulation switching moments, which appear as vertical stripes or special energy burst patterns on the time-frequency graph. This is a strong feature of device fingerprints. Finally, two-dimensional time-frequency images also preserve the signal's modulation trajectory; for linear frequency modulation (LFM) or frequency-modulated signals, the frequency change trajectory is clearly visible.
[0046] Step S1203: If the conversion type is vector, then convert it into a vector based on the statistical characteristics of the wireless signal; If the conversion type is vector, then it is converted into a vector based on the statistical characteristics of the wireless signal. Statistical characteristics of the signal, such as amplitude or phase mean, variance, and higher-order cumulants, are extracted and concatenated into a vector.
[0047] Step S1204: If the conversion type is sequence, the wireless signal is converted into time-series data, wherein the in-phase component or quadrature component of the wireless signal is used as one-dimensional time-series data. If the transformation type is a sequence, the normalized in-phase or orthogonal components are directly used as one-dimensional time series data.
[0048] Step S1205: Use the preprocessed data as input information.
[0049] The preprocessed data is used as input information.
[0050] Step S130: Input the input information into the trained feature extraction model, and map the input information layer by layer based on the multi-layer feature extraction of the feature extraction model to obtain the feature information corresponding to the target UAV; The input information is fed into the trained feature extraction model, and the input information is mapped layer by layer based on the multi-layer feature extraction model to obtain the feature information corresponding to the target UAV.
[0051] The feature extraction model is a neural network model, and step S130 includes steps S1301-S1302: Step S1301: Input the input information into the trained neural network model, wherein the neural network model is composed of a multi-layer deep neural network consisting of a multi-layer restricted Boltzmann structure; This paper applies a neural network model to radio fingerprinting. The neural network model consists of multiple Restricted Boltzmann Machines (RBMs). A Restricted Boltzmann Machine is a generative stochastic neural network based on an energy model. The basic structure of an RBM includes two layers: a visible layer and a hidden layer. While there are full connections between layers, there are no connections between neurons within a layer, achieving full connectivity but with limitations. The feature extraction of a radiation source target is performed using a multi-hidden-layer deep neural network composed of multiple RBMs. Preprocessed m n-dimensional samples are used as input data to the multi-layer RBM network.
[0052] The reason for using RBM is that it can be pre-trained using unlabeled data, and the probabilistic model using RBM may be more robust in reducing the signal-to-noise ratio.
[0053] Step S1302: Based on the neural network model, the multi-layer restricted Boltzmann machine maps the input information layer by layer to obtain the feature information corresponding to the target UAV.
[0054] The multi-layer restricted Boltzmann machine based on the neural network model maps the input information layer by layer to obtain the feature information corresponding to the target UAV.
[0055] The neural network model is an ERDRBM neural network model, where ERDRBM refers to Emitter Recognition based on Deep Restricted Boltzmann Machine (ERDRBM) for radiation source identification. The ERDRBM neural network model consists of multiple layers of restricted Boltzmann machines, and the feature information corresponding to the target UAV is a feature vector. Step S1302 includes steps B1-B3: Step B1: Obtain the state of the first hidden layer through the first layer encoder; The state of the first hidden layer is obtained through the first layer encoder. The deep network, consisting of a single hidden layer, is initialized layer by layer using a greedy algorithm.
[0056] Step B2: Obtain initial parameters through bottom-up unsupervised learning by embedding the states layer by layer; Initial parameters are obtained through unsupervised learning from the bottom up by embedding states. Then, the backpropagation (BP) algorithm is used to perform supervised parameter fine-tuning on the entire ERDRBM deep model to obtain the globally optimal weight vector and achieve feature extraction.
[0057] Stacking Restricted Models (RBMs) and training layer by layer: First, train the first RBM, whose visible layer uses preprocessed data. After training, use the hidden layer activation values of RBM1 (the learned first-layer features) as input to the visible layer of the second RBM. Similarly, train the second RBM to obtain higher-layer features. Repeat this process, stacking multiple RBMs (e.g., 3-5 layers) to form a deep belief network. This process is greedy, training only one layer at a time. The goal of pre-training is to ensure that each RBM layer models the distribution of its input data as well as possible. In this way, the entire network learns the mapping from the raw signal to progressively abstracted feature representations. Higher-level features are more robust to noise and channel variations and better represent the essential hardware characteristics of the device.
[0058] Step B3: Supervised parameter fine-tuning of the ERDRBM neural network model is performed through backpropagation to obtain the globally optimal feature vector corresponding to the target UAV, thereby realizing feature extraction of the target UAV.
[0059] Finally, after extracting features using Softmax for classification and recognition, SoftMax regression in the ERDRBM model is used for classification and recognition.
[0060] The SoftMax regression model is a generalization of the Logistic regression model to multi-class classification problems. The class label can take more than two values, thus obtaining the class label value of the input data, and finally obtaining a non-linear mapping between the input data and the class label value.
[0061] In particular, when performing supervised parameter fine-tuning of the ERDRBM neural network model through backpropagation, dynamic parameters are used for supervised parameter fine-tuning to prevent data overfitting.
[0062] Step S140: Classify and identify the feature information based on the preset SoftMax regression model to obtain the wireless signal fingerprint type of the target UAV, and determine the identity of the target UAV based on the wireless signal fingerprint type.
[0063] The wireless signal fingerprint type of the target UAV is obtained by classifying and identifying the feature information based on the preset SoftMax regression model, and the identity of the target UAV is determined based on the wireless signal fingerprint type.
[0064] This application provides a method, apparatus, computer device, and storage medium for drone identification. The method includes: in response to an identification command for drone identification, acquiring a wireless signal from a target drone from a signal receiving device; preprocessing the wireless signal to obtain preprocessed input information; inputting the input information into a trained feature extraction model, mapping the input information layer by layer based on the multi-layer feature extraction of the feature extraction model to obtain feature information corresponding to the target drone; classifying and identifying the feature information based on a preset SoftMax regression model to obtain the wireless signal fingerprint type of the target drone, and determining the identity of the target drone based on the wireless signal fingerprint type. In this application, deep learning is used to map the raw signal layer by layer to a higher-level feature representation that is abstracted layer by layer. Higher-level features are more robust to noise and channel variations and better represent the essential hardware characteristics of the device. High-precision identification of flying drones is achieved remotely without adding existing signal receiving equipment, improving the accuracy of remote identification.
[0065] Figure 2 This is a schematic block diagram of a drone identification device provided in an embodiment of this application. Figure 2As shown, corresponding to the above-described drone identification method, this application also provides a drone identification device 600. The drone identification device 600 includes a unit for performing the aforementioned drone identification method, and can be configured in terminals such as desktop computers, tablets, and laptops. Specifically, please refer to... Figure 2 The drone identification device 600 includes an acquisition unit 601, a preprocessing unit 602, a feature extraction unit 603, and a classification and identification unit 604, wherein: The acquisition unit 601 is used to acquire wireless signals from the target drone from the signal receiving device in response to an identification command for identifying the drone. Preprocessing unit 602 is used to preprocess the wireless signal to obtain preprocessed input information; The feature extraction unit 603 is used to input the input information into the trained feature extraction model, and map the input information layer by layer based on the multi-layer feature extraction of the feature extraction model to obtain the feature information corresponding to the target UAV. The classification and recognition unit 604 is used to classify and recognize the feature information based on a preset SoftMax regression model to obtain the wireless signal fingerprint type of the target drone, and to determine the identity of the target drone based on the wireless signal fingerprint type.
[0066] In some embodiments, the feature extraction model is a neural network model. The feature extraction unit 603, when inputting the input information into the trained feature extraction model, maps the input information layer by layer based on the multi-layer feature extraction of the feature extraction model to obtain the feature information corresponding to the target UAV. Specifically, it is used for: The input information is input into the trained neural network model, wherein the neural network model is composed of a multi-layer deep neural network consisting of a multi-layer restricted Boltzmann structure; The multi-layer restricted Boltzmann machine based on the neural network model maps the input information layer by layer to obtain the feature information corresponding to the target UAV.
[0067] In some embodiments, the neural network model is an ERDRBM neural network model, which is composed of multiple restricted Boltzmann machine layers. The feature information corresponding to the target UAV is a feature vector. The multiple restricted Boltzmann machine layers based on the neural network model map the input information layer by layer to obtain the feature information corresponding to the target UAV, specifically used for: The state of the first hidden layer is obtained through the first layer encoder; Initial parameters are obtained through bottom-up unsupervised learning by embedding the states layer by layer. The ERDRBM neural network model is subjected to supervised parameter fine-tuning via backpropagation to obtain the globally optimal feature vector corresponding to the target UAV, thereby achieving feature extraction of the target UAV.
[0068] In some embodiments, when performing supervised parameter fine-tuning of the ERDRBM neural network model via backpropagation, supervised parameter fine-tuning using dynamic parameters is employed to prevent data overfitting.
[0069] In some embodiments, the preprocessing unit 602 performs preprocessing on the wireless signal to obtain preprocessed input information, specifically for: Determine the type of conversion required for the wireless signal; If the conversion type is a two-dimensional image, then the wireless signal is converted into a two-dimensional image; If the conversion type is vector, then it is converted into a vector based on the statistical characteristics of the wireless signal; If the conversion type is sequence, the wireless signal is converted into time-series data, wherein the in-phase or quadrature components of the wireless signal are used as one-dimensional time-series data. The preprocessed data is used as input information.
[0070] In some embodiments, the preprocessing unit 602, when performing the conversion of the wireless signal into a two-dimensional image if the conversion type is a two-dimensional image, specifically uses the following methods: If the conversion type is a two-dimensional image, the wireless signal is converted to obtain a two-dimensional time-frequency image, wherein the two-dimensional time-frequency image displays the frequency change trend over time, and the conversion method includes either short-time Fourier transform or wavelet transform.
[0071] In some embodiments, the preprocessing unit 602, before performing the conversion type required for determining the wireless signal, specifically includes: Adjusting various signal target data to achieve feature alignment and feature classification; The wireless signals are randomly divided into groups, with each group containing multiple wireless signal data. Different groups are numbered to obtain the grouped data sets to be processed, thereby reducing computational complexity.
[0072] In summary, this application provides a method, apparatus, computer device, and storage medium for drone identification. The method includes: in response to an identification command for drone identification, acquiring a wireless signal from a target drone from a signal receiving device; preprocessing the wireless signal to obtain preprocessed input information; inputting the input information into a trained feature extraction model, mapping the input information layer by layer based on the multi-layer feature extraction of the feature extraction model to obtain feature information corresponding to the target drone; classifying and identifying the feature information based on a preset SoftMax regression model to obtain the wireless signal fingerprint type of the target drone, and determining the identity of the target drone based on the wireless signal fingerprint type. In this application, deep learning is used to map the raw signal layer by layer to a higher-level feature representation that is abstracted layer by layer. Higher-level features are more robust to noise and channel variations and better represent the essential hardware characteristics of the device. High-precision identification of flying drones is achieved remotely without adding existing signal receiving equipment, improving the accuracy of remote identification.
[0073] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned drone identification device and each unit can be referred to the corresponding description in the foregoing method embodiments. For the sake of convenience and brevity, it will not be repeated here.
[0074] The aforementioned drone identification device can be implemented as a computer program, which can, for example... Figure 3 It runs on the computer device shown.
[0075] Please see Figure 3 , Figure 3 This is a schematic block diagram of a computer device 700 provided in an embodiment of this application. The computer device 700 can be a terminal or a server. The terminal can be an electronic device with communication functions, such as a smartphone, tablet, laptop, desktop computer, personal digital assistant, or wearable device. The server can be a standalone server or a server cluster composed of multiple servers.
[0076] See Figure 3 The computer device 700 includes a processor 702, a memory, and a network interface 705 connected via a system bus 701. The memory may include a non-volatile storage medium 703 and internal memory 704.
[0077] The non-volatile storage medium 703 may store an operating system 7031 and a computer program 7032. The computer program 7032 includes program instructions that, when executed, cause the processor 702 to perform a drone identification method.
[0078] The processor 702 provides computing and control capabilities to support the operation of the entire computer device 700.
[0079] The internal memory 704 provides an environment for the operation of the computer program 7032 in the non-volatile storage medium 703. When the computer program 7032 is executed by the processor 702, the processor 702 can execute a drone identification method.
[0080] This network interface 705 is used for network communication with other devices. Those skilled in the art will understand that... Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device 700 to which the present application is applied. The specific computer device 700 may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0081] The processor 702 is used to run the computer program 7032 stored in the memory to perform the following steps: In response to an identification command for identifying a drone, the system acquires wireless signals from the target drone from the signal receiving device; preprocesses the wireless signals to obtain preprocessed input information; inputs the input information into a trained feature extraction model, and maps the input information layer by layer based on the multi-layer feature extraction model to obtain feature information corresponding to the target drone; classifies and identifies the feature information based on a preset SoftMax regression model to obtain the wireless signal fingerprint type of the target drone, and determines the identity of the target drone based on the wireless signal fingerprint type.
[0082] It should be understood that in the embodiments of this application, the processor 702 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.
[0083] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.
[0084] Therefore, this application also provides a storage medium. This storage medium can be a computer-readable storage medium. The storage medium stores a computer program, wherein the computer program includes program instructions. When executed by a processor, the program instructions cause the processor to perform the following steps: In response to an identification command for identifying a drone, the system acquires wireless signals from the target drone from the signal receiving device; preprocesses the wireless signals to obtain preprocessed input information; inputs the input information into a trained feature extraction model, and maps the input information layer by layer based on the multi-layer feature extraction model to obtain feature information corresponding to the target drone; classifies and identifies the feature information based on a preset SoftMax regression model to obtain the wireless signal fingerprint type of the target drone, and determines the identity of the target drone based on the wireless signal fingerprint type.
[0085] The storage medium can be any computer-readable storage medium that can store program code, such as a USB flash drive, external hard drive, read-only memory (ROM), magnetic disk, or optical disk.
[0086] 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, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. 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 application.
[0087] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of each unit is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.
[0088] The steps in the methods of this application embodiment can be adjusted, merged, or deleted according to actual needs. The units in the apparatus of this application embodiment can be merged, divided, or deleted according to actual needs. Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.
[0089] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application.
[0090] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for identifying unmanned aerial vehicles (UAVs), characterized in that, The drone identification method is applied to a drone identification device in a drone identification system, the drone identification system further including a signal receiving device, and the method includes: In response to an identification command for identifying the drone, the wireless signal from the target drone is acquired from the signal receiving device. The wireless signal is preprocessed to obtain preprocessed input information; The input information is input into the trained feature extraction model, and the input information is mapped layer by layer based on the multi-layer feature extraction model to obtain the feature information corresponding to the target UAV. The wireless signal fingerprint type of the target UAV is obtained by classifying and identifying the feature information based on the preset SoftMax regression model, and the identity of the target UAV is determined based on the wireless signal fingerprint type.
2. The method according to claim 1, characterized in that, The feature extraction model is a neural network model. The input information is input into the trained feature extraction model, and the input information is mapped layer by layer based on the multi-layer feature extraction of the model to obtain the feature information corresponding to the target UAV, including: The input information is input into the trained neural network model, wherein the neural network model is composed of a multi-layer deep neural network consisting of a multi-layer restricted Boltzmann structure; The multi-layer restricted Boltzmann machine based on the neural network model maps the input information layer by layer to obtain the feature information corresponding to the target UAV.
3. The method according to claim 2, characterized in that, The neural network model is an ERDRBM neural network model, which consists of multiple layers of restricted Boltzmann machines. The feature information corresponding to the target UAV is a feature vector. The multiple layers of restricted Boltzmann machines based on the neural network model map the input information layer by layer to obtain the feature information corresponding to the target UAV, including: The state of the first hidden layer is obtained through the first layer encoder; Initial parameters are obtained through bottom-up unsupervised learning by embedding the states layer by layer. The ERDRBM neural network model is subjected to supervised parameter fine-tuning via backpropagation to obtain the globally optimal feature vector corresponding to the target UAV, thereby achieving feature extraction of the target UAV.
4. The method according to claim 3, characterized in that, When performing supervised parameter fine-tuning of the ERDRBM neural network model via backpropagation, dynamic parameters are used for supervised parameter fine-tuning to prevent data overfitting.
5. The method according to claim 1, characterized in that, The preprocessing of the wireless signal to obtain preprocessed input information includes: Determine the type of conversion required for the wireless signal; If the conversion type is a two-dimensional image, then the wireless signal is converted into a two-dimensional image; If the conversion type is vector, then it is converted into a vector based on the statistical characteristics of the wireless signal; If the conversion type is sequence, the wireless signal is converted into time-series data, wherein the in-phase or quadrature components of the wireless signal are used as one-dimensional time-series data. The preprocessed data is used as input information.
6. The method according to claim 5, characterized in that, If the conversion type is a two-dimensional image, then converting the wireless signal into a two-dimensional image includes: If the conversion type is a two-dimensional image, the wireless signal is converted to obtain a two-dimensional time-frequency image, wherein the two-dimensional time-frequency image displays the frequency change trend over time, and the conversion method includes either short-time Fourier transform or wavelet transform.
7. The method according to claim 5, characterized in that, Before determining the type of conversion required for the wireless signal, the process includes: Adjusting various signal target data to achieve feature alignment and feature classification; The wireless signals are randomly divided into groups, with each group containing multiple wireless signal data. Different groups are numbered to obtain the grouped data sets to be processed, thereby reducing computational complexity.
8. A drone identification device, characterized in that, The drone identification method is applied to a drone identification device within a drone identification system, which further includes a signal receiving device. The drone identification device comprises: The acquisition unit is used to acquire wireless signals from the target drone from the signal receiving device in response to an identification command for identifying the drone. A preprocessing unit is used to preprocess the wireless signal to obtain preprocessed input information; The feature extraction unit is used to input the input information into the trained feature extraction model, and map the input information layer by layer based on the multi-layer feature extraction of the feature extraction model to obtain the feature information corresponding to the target UAV. The classification and identification unit is used to classify and identify the feature information based on a preset SoftMax regression model to obtain the wireless signal fingerprint type of the target UAV, and to determine the identity of the target UAV based on the wireless signal fingerprint type.
9. A computer device for identifying unmanned aerial vehicles (UAVs), characterized in that, The device includes a memory, a processor, and a drone identification program stored in the memory and executable on the processor, wherein the processor executes the drone identification program to implement the steps of the drone identification method according to any one of claims 1 to 7.
10. A storage medium, characterized in that, The storage medium stores a program for implementing a drone identification method, which is executed by a processor to implement the steps of the drone identification method as described in any one of claims 1 to 7.