Method and device for training drone individual identification model, and computer device

Abstract

The application relates to a training method of a UAV individual identification model, comprising the following steps: acquiring sample set information; training an initial single-head convolutional neural network model based on the sample set information, and extracting model parameters of a preset network layer from model parameters of the trained single-head convolutional neural network; constructing an initial multi-head convolutional neural network model based on the preset network layer and a target full connection layer, wherein the target full connection layer comprises a plurality of sub-connection layers corresponding to UAV types one by one; for each UAV type, the sub-connection layer corresponding to the UAV type is trained through sample information corresponding to the UAV type, so as to obtain the UAV individual identification model. In the above method, the model parameters of the preset network layer are determined first, so that the initial multi-head convolutional network model can be constructed conveniently, the sub-connection layers in the initial multi-head convolutional neural network model correspond to the UAV types one by one, and the identification of different UAV types can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle identification, in particular to a training method and device of an unmanned aerial vehicle individual identification model and a computer device. BACKGROUND

[0002] In recent years, with the rapid expansion of the global unmanned aerial vehicle market, the safety hazards caused by unmanned aerial vehicles are becoming more and more serious, and therefore, the individual identification technology of unmanned aerial vehicles has attracted widespread attention. Through the individual identification technology of unmanned aerial vehicles, the individual information of a to-be-tested unmanned aerial vehicle can be identified, so as to determine the correspondence between the to-be-tested unmanned aerial vehicle and sample unmanned aerial vehicles in a sample unmanned aerial vehicle set.

[0003] However, the existing unmanned aerial vehicle individual identification technology based on radio frequency signals is only applicable to different individuals of the same model, and cannot identify different individuals of different models. In the prior art, if the individual identification of unmanned aerial vehicles of multiple models is to be performed, multiple different types of identification models need to be deployed at the same time, which will increase the calculation and storage overhead of model deployment, and there is a problem of increasing the time consumption of individual identification. SUMMARY

[0004] Therefore, it is necessary to provide a training method, device and computer device of an unmanned aerial vehicle individual identification model in view of the above technical problems.

[0005] In a first aspect, the present application provides a training method of an unmanned aerial vehicle individual identification model. The method comprises:

[0006] obtaining sample set information, the sample set information comprising sample information corresponding to multiple unmanned aerial vehicle types, the sample information comprising individual information and radio frequency signals corresponding to each sample unmanned aerial vehicle;

[0007] training an initial single-head convolutional neural network model based on the sample set information to obtain model parameters of the trained single-head convolutional neural network, and extracting model parameters of a preset network layer from the model parameters of the trained single-head convolutional neural network;

[0008] constructing an initial multi-head convolutional neural network model based on the preset network layer and a target fully connected layer, the target fully connected layer comprising multiple sub-connected layers, and the sub-connected layers corresponding one-to-one to the unmanned aerial vehicle types;

[0009] for each unmanned aerial vehicle type, training the sub-connected layer corresponding to the unmanned aerial vehicle type based on the sample information corresponding to the unmanned aerial vehicle type to obtain an unmanned aerial vehicle individual identification model.

[0010] In one embodiment, the training of the initial single-head convolutional neural network model based on the sample set information to obtain the model parameters of the trained single-head convolutional neural network comprises:

[0011] determine initial input information according to the radio frequency signal and preset radio frequency signal extraction conditions;

[0012] input the initial input information into the initial single-head convolutional neural network model to obtain first prediction information output by the initial single-head convolutional neural network model;

[0013] train the initial single-head convolutional neural network model according to the first prediction information and the individual information to obtain model parameters of the trained single-head convolutional neural network.

[0014] In one of the embodiments, the determining of the initial input information according to the radio frequency signal and preset radio frequency signal extraction conditions comprises:

[0015] performing preset synchronization processing and initial normalization processing on the radio frequency signal to obtain a preprocessed radio frequency signal;

[0016] adding a preset noise signal to the preprocessed radio frequency signal to obtain a noise-added signal, and inputting the noise-added signal into a preset channel model to obtain a target signal output by the preset channel model;

[0017] performing feature extraction on the target signal by using a preset feature extraction method, and determining the initial input information according to a feature extraction result.

[0018] In one of the embodiments, the training of the sub-connection layer corresponding to each type of unmanned aerial vehicle by using the sample information corresponding to the type of unmanned aerial vehicle comprises:

[0019] inputting the initial input information corresponding to each type of unmanned aerial vehicle into the initial multi-head convolutional neural network model to obtain second prediction information output by the sub-connection layer corresponding to the type of unmanned aerial vehicle;

[0020] training the sub-connection layer corresponding to each type of unmanned aerial vehicle according to the second prediction information and the sample information corresponding to the type of unmanned aerial vehicle.

[0021] In one of the embodiments, the constructing of the initial multi-head convolutional neural network model based on the preset network layer and the target full connection layer comprises:

[0022] using model parameters of the preset network layer as model parameters of a target network layer in the initial multi-head convolutional neural network model;

[0023] determining a number of sub-connection layers included in the target full connection layer according to a total number of the types of unmanned aerial vehicles, and establishing a correspondence between the types of unmanned aerial vehicles and the sub-connection layers;

[0024] For each of the UAV types, according to the number of sample UAVs corresponding to the UAV type, determine output data information of a sub-connection layer corresponding to the UAV type;

[0025] Based on the model parameters of the target network layer, the corresponding relationship and the output data information, construct the initial multi-head convolutional neural network model.

[0026] In a second aspect, the present application also provides a UAV individual identification method. The method comprises:

[0027] Obtain type identification information of a to-be-tested UAV and to-be-tested radio frequency signals;

[0028] Based on the to-be-tested radio frequency signals, the type identification information and a UAV individual identification model, obtain initial prediction information;

[0029] According to the initial prediction information and a preset individual identification condition, determine individual prediction information of the to-be-tested UAV;

[0030] The UAV individual identification model is trained by the training method of the UAV individual identification model of the first aspect.

[0031] In one embodiment, based on the to-be-tested radio frequency signals, the type identification information and the UAV individual identification model, obtaining the initial prediction information comprises:

[0032] According to the to-be-tested radio frequency signals and a preset radio frequency signal extraction condition, determine input information;

[0033] Input the input information into the UAV individual identification model, and identify the to-be-tested UAV through a sub-connection layer corresponding to the type identification information in the UAV individual identification model to obtain the initial prediction information.

[0034] In a third aspect, the present application also provides a training device of a UAV individual identification model. The device comprises:

[0035] An initial acquisition module is configured to acquire sample set information, wherein the sample set information comprises sample information corresponding to a plurality of UAV types, and the sample information comprises individual information and radio frequency signals corresponding to each sample UAV;

[0036] A model parameter determination module is configured to train an initial single-head convolutional neural network model based on the sample set information, obtain model parameters of the trained single-head convolutional neural network, and extract model parameters of a preset network layer from the model parameters of the trained single-head convolutional neural network;

[0037] A model construction module is configured to construct an initial multi-head convolutional neural network model based on the preset network layer and a target full connection layer, the target full connection layer including a plurality of sub-connection layers, the sub-connection layers corresponding to the types of UAVs one by one.

[0038] A training module is configured to train, for each type of UAV, the sub-connection layer corresponding to the type of UAV by using the sample information corresponding to the type of UAV, to obtain a UAV individual recognition model.

[0039] In one of the embodiments, the model parameter determination module is specifically configured to:

[0040] determine initial input information according to the radio frequency signal and preset radio frequency signal extraction conditions;

[0041] input the initial input information into the initial single-head convolutional neural network model to obtain first prediction information output by the initial single-head convolutional neural network model;

[0042] train the initial single-head convolutional neural network model according to the first prediction information and the individual information, to obtain model parameters of the trained single-head convolutional neural network.

[0043] In one of the embodiments, the model parameter determination module is further specifically configured to:

[0044] perform preset synchronization processing and initial normalization processing on the radio frequency signal to obtain a preprocessed radio frequency signal;

[0045] add a preset noise signal to the preprocessed radio frequency signal to obtain a noise-added signal, and input the noise-added signal into a preset channel model to obtain a target signal output by the preset channel model;

[0046] extract features from the target signal by using a preset feature extraction method, and determine the initial input information according to a feature extraction result.

[0047] In one of the embodiments, the training module is specifically configured to:

[0048] for each type of UAV, input initial input information corresponding to the type of UAV into the initial multi-head convolutional neural network model to obtain second prediction information output by a sub-connection layer corresponding to the type of UAV;

[0049] train the sub-connection layer corresponding to the type of UAV according to the second prediction information and sample information corresponding to the type of UAV.

[0050] In one of the embodiments, the model construction module is specifically configured to:

[0051] The model parameters of the preset network layer are used as model parameters of a target network layer in the initial multi-head convolutional neural network model.

[0052] According to the total number of the types of UAVs, the number of sub-connection layers included in the target full-connection layer is determined, and a correspondence between the types of UAVs and the sub-connection layers is established.

[0053] For each type of UAV, according to the number of sample UAVs corresponding to the type of UAV, output data information of the sub-connection layer corresponding to the type of UAV is determined.

[0054] Based on the model parameters of the target network layer, the correspondence and the output data information, the initial multi-head convolutional neural network model is constructed.

[0055] In a fourth aspect, the present application further provides a UAV individual identification device. The device comprises:

[0056] A to-be-measured UAV information acquisition module is configured to acquire type identification information of a to-be-measured UAV and a to-be-measured radio frequency signal.

[0057] An initial prediction information determination module is configured to obtain initial prediction information based on the to-be-measured radio frequency signal, the type identification information and a UAV individual identification model.

[0058] An individual identification module is configured to determine individual prediction information of the to-be-measured UAV according to the initial prediction information and preset individual identification conditions.

[0059] The UAV individual identification model is trained by the training method of the UAV individual identification model of the first aspect.

[0060] In one embodiment, the initial prediction information determination module is specifically configured to:

[0061] Determine input information according to the to-be-measured radio frequency signal and preset radio frequency signal extraction conditions.

[0062] Input the input information into the UAV individual identification model, identify the to-be-measured UAV through a sub-connection layer in the UAV individual identification model corresponding to the type identification information, and obtain initial prediction information.

[0063] In a fifth aspect, the present application further provides a computer device. The computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the steps of the first aspect or the second aspect when executing the computer program.

[0064] In a sixth aspect, the present application also provides a computer readable storage medium. The computer readable storage medium has a computer program stored thereon, and the computer program, when executed by a processor, implements the steps of the first aspect or the second aspect.

[0065] In a seventh aspect, the present application also provides a computer program product. The computer program product comprises a computer program, and the computer program, when executed by a processor, implements the steps of the first aspect or the second aspect.

[0066] In the method for training the unmanned aerial vehicle individual identification model, sample set information is obtained, the sample set information includes sample information corresponding to a plurality of unmanned aerial vehicle types, and the sample information includes individual information and radio frequency signals corresponding to each sample unmanned aerial vehicle; an initial single-head convolutional neural network model is trained based on the sample set information to obtain model parameters of the trained single-head convolutional neural network, and model parameters of a preset network layer are extracted from the model parameters of the trained single-head convolutional neural network; an initial multi-head convolutional neural network model is constructed based on the preset network layer and a target fully connected layer, the target fully connected layer includes a plurality of sub-connected layers, and the sub-connected layers correspond to the unmanned aerial vehicle types one by one; for each unmanned aerial vehicle type, the sub-connected layer corresponding to the unmanned aerial vehicle type is trained through the sample information corresponding to the unmanned aerial vehicle type to obtain the unmanned aerial vehicle individual identification model. In the method for training the unmanned aerial vehicle individual identification model, the initial single-head convolutional neural network model is first trained to determine the model parameters of the preset network layer, which facilitates the construction of the initial multi-head convolutional network model. The initial multi-head convolutional neural network model includes the sub-connected layers corresponding to the unmanned aerial vehicle types one by one, and can realize the identification of different unmanned aerial vehicle types. BRIEF DESCRIPTION OF DRAWINGS

[0067] Figure 1 A flowchart of the method for training the unmanned aerial vehicle individual identification model in one embodiment;

[0068] Figure 2 A flowchart of the method for training the unmanned aerial vehicle individual identification model in another embodiment;

[0069] Figure 3 A flowchart of the method for training the unmanned aerial vehicle individual identification model in another embodiment;

[0070] Figure 4 A flowchart of the method for training the unmanned aerial vehicle individual identification model and individual identification in one embodiment;

[0071] Figure 5 A structure diagram of the neural network model in one embodiment;

[0072] Figure 6 A flowchart of the method for training the unmanned aerial vehicle individual identification model in one embodiment;

[0073] Figure 7 This is a structural diagram of a training device for an individual drone recognition model in one embodiment;

[0074] Figure 8 This is a structural diagram of a drone individual identification device in one embodiment;

[0075] Figure 9 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0076] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0077] This application provides a training method for an individual drone recognition model. This embodiment illustrates the method by applying it to a terminal. It is understood that this method can also be applied to a server, and to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. The terminal can be, but is not limited to, various personal computers and laptops. The server can be a standalone server or a server cluster consisting of multiple servers. It should be noted that the above-mentioned individual drone recognition model can be used not only for individual drone recognition, but also for individual drone remote controller recognition, etc. Figure 1 As shown, the method includes the following steps:

[0078] Step 101: Obtain sample set information. The sample set information includes sample information corresponding to multiple drone types. The sample information includes individual information and radio frequency signals corresponding to each sample drone.

[0079] The individual information of the sample drone may include its first serial number among similar sample drones and its second serial number among all sample drones. The radio frequency signal must contain transmitted data. Transmitted data is data sent from the drone (or drone remote controller) to the receiving device (i.e., the terminal's receiving device). The drone and the receiving device are in close proximity, and the transmitted data typically has a high signal-to-noise ratio.

[0080] Specifically, the sample set information includes sample drones of m different drone types, with each drone type having [number of samples]. If there are a certain number of sample drones, then the total number of sample drones is... The terminal needs to collect individual information and radio frequency signals of all sample drone individuals to form sample set information of the sample drones. It should be noted that when the drone individual identification model is trained, the sample set information obtained by the terminal is the individual information and radio frequency signals of the legal drones.

[0081] In step 102, the initial single-head convolutional neural network model is trained based on the sample set information, model parameters of the trained single-head convolutional neural network are obtained, and model parameters of the preset network layer are extracted from the model parameters of the trained single-head convolutional neural network.

[0082] The last layer of the initial single-head convolutional neural network is an output full connection layer, and the data width of the output information of the output full connection layer is consistent with the total number of sample drones. The terminal can train the initial single-head convolutional neural network model according to the output information of the output full connection layer of the initial single-head convolutional neural network model and the second serial number information of all sample drones, optimize the model parameters of the initial single-head convolutional neural network, and extract the model parameters of the preset network layer from the model parameters of the trained single-head convolutional neural network. It should be noted that the optimization algorithm with the model parameter optimization function in the related art can be applied to the embodiments of the present application, and the embodiments of the present application are not limited in this regard.

[0083] In step 103, an initial multi-head convolutional neural network model is constructed based on the preset network layer and a target full connection layer, and the target full connection layer includes a plurality of sub-connection layers, and each sub-connection layer corresponds to one drone type.

[0084] The terminal constructs a target network layer with the same structure and parameters as the preset network layer. The terminal can also construct a target full connection layer as an output full connection layer based on the sample set information. The target full connection layer includes a plurality of sub-connection layers, and each sub-connection layer corresponds to one drone type.

[0085] In step 104, for each drone type, the sub-connection layer corresponding to the drone type is trained through the sample information corresponding to the drone type, and a drone individual identification model is obtained.

[0086] For each drone type, the terminal inputs the radio frequency signals of the sample drones in the drone type into the initial multi-head convolutional neural network model, trains the sub-connection layer corresponding to the drone type according to the individual prediction information output by the sub-connection layer corresponding to the drone type and the first serial number information of the sample drones in the drone type. After the terminal completes the training of each drone type, the drone individual identification model can be obtained, which can predict the corresponding probability of the to-be-tested drone and the sample drone of the same type.

[0087] The training method of the unmanned aerial vehicle individual identification model comprises the following steps: obtaining sample set information, wherein the sample set information comprises sample information corresponding to a plurality of unmanned aerial vehicle types, and the sample information comprises individual information and radio frequency signals corresponding to each sample unmanned aerial vehicle; training an initial single-head convolutional neural network model based on the sample set information to obtain model parameters of the trained single-head convolutional neural network, and extracting model parameters of a preset network layer from the model parameters of the trained single-head convolutional neural network; constructing an initial multi-head convolutional neural network model based on the preset network layer and a target full connection layer, wherein the target full connection layer comprises a plurality of sub-connection layers, and the sub-connection layers correspond to the unmanned aerial vehicle types one by one; and training, for each unmanned aerial vehicle type, the sub-connection layer corresponding to the unmanned aerial vehicle type based on the sample information corresponding to the unmanned aerial vehicle type to obtain the unmanned aerial vehicle individual identification model. In the unmanned aerial vehicle individual identification model training method, the initial single-head convolutional neural network model is first trained to determine the model parameters of the preset network layer, so that the initial multi-head convolutional network model is constructed, and the complexity of training the multi-head convolutional neural network model is reduced. In addition, the initial multi-head convolutional neural network model constructed comprises the sub-connection layers corresponding to the unmanned aerial vehicle types one by one, so that the identification of different unmanned aerial vehicle types can be realized, the calculation and storage overheads during deployment are reduced, the time consumption of unmanned aerial vehicle individual identification is reduced, and the problem that a plurality of models need to be arranged to complete the identification of different unmanned aerial vehicles is solved. The unmanned aerial vehicle individual identification model obtained through the training method can complete the identification of a to-be-detected unmanned aerial vehicle of the same type as the sample unmanned aerial vehicle, and can be applied to the technical field of detecting intruding unmanned aerial vehicles.

[0088] In one embodiment of the present application, as shown in Figure 2 Step 102, training the initial single-head convolutional neural network model based on the sample set information to obtain model parameters of the trained single-head convolutional neural network, specifically comprising:

[0089] Step 201, determining initial input information based on the radio frequency signals and the preset radio frequency signal extraction condition;

[0090] Step 202, inputting the initial input information into the initial single-head convolutional neural network model to obtain first prediction information output by the initial single-head convolutional neural network model;

[0091] Step 203, training the initial single-head convolutional neural network model based on the first prediction information and the individual information to obtain model parameters of the trained single-head convolutional neural network.

[0092] The initial single-head convolutional neural network is a common convolutional neural network model, the last layer of which is an output full connection layer, and the output information of the output full connection layer is the first prediction information. The first prediction information is a one-dimensional matrix and the data width is N CThe data width of the first prediction information is consistent with the total number of the sample drones. The terminal extracts the signal features of the sample drones according to preset radio frequency signal extraction conditions, and takes the signal features as initial input information. The data length of the initial input information is the length of the signal features, and the width of the initial input information changes according to the data format of the signal features. If the data format of the signal features is a real number matrix, the data width of the initial input information is the width of the signal features. If the data format of the signal features is a complex number matrix, the data width of the initial input information is twice the width of the original signal features obtained by splicing the real part and the imaginary part of the complex signal.

[0093] Specifically, when training the initial single-head convolutional neural network model, the initial input information corresponding to the radio frequency signal of only one sample drone is input into the initial single-head convolutional neural network model each time, and the initial neural network model is trained according to the first prediction information output by the output full connection layer and the second serial number information in the individual information of the sample drone. The terminal needs to repeat the above training process until the initial input information corresponding to the radio frequency signals of all sample drones is input into the initial single-head convolutional neural network model, the training of the initial single-head convolutional neural network model is completed, and then the terminal can determine the information of each model parameter in the single-head convolutional neural network model.

[0094] In an embodiment of the present application, as shown in Figure 3 Step 201, determining initial input information according to the radio frequency signal and preset radio frequency signal extraction conditions, specifically including:

[0095] Step 301, performing preset synchronization processing and initial normalization processing on the radio frequency signal to obtain a preprocessed radio frequency signal.

[0096] Specifically, as shown in Figure 4 The terminal needs to first perform data preprocessing on the radio frequency signals of all sample drones, that is, to perform coarse synchronization processing on the radio frequency signals of all sample drones, to preliminarily align the starting positions of the data frames, and to perform initial normalization processing on all radio frequency signals after coarse synchronization to eliminate the interference of the amplitude on the neural network model (which includes the initial single-head convolutional neural network model and the initial multi-head convolutional neural network model).

[0097] Step 302, adding a preset noise signal to the preprocessed radio frequency signal to obtain a noise-added signal, and inputting the noise-added signal into a preset channel model to obtain a target signal output by the preset channel model.

[0098] The signal-to-noise ratio interval of the preset noise signal is -20dB-20dB. The preset channel model can be a channel model randomly extracted from a Rayleigh channel, a Rician channel, an ITU channel model, etc. containing random parameters.

[0099] Specifically, before each training of the neural network model (the neural network model includes an initial single-head convolutional neural network model and an initial multi-head convolutional neural network model), the terminal needs to generate preset noise signals with different signal-to-noise ratios by using a data enhancement method, add the preset noise signals to the preprocessed radio frequency signals of the sample unmanned aerial vehicle, and input the noise-added signals to a randomly extracted preset channel to obtain a target signal output by the preset channel model.

[0100] In step 303, a preset feature extraction method is used to extract features of the target signal, and initial input information is determined according to the feature extraction result.

[0101] Specifically, the preset feature extraction method includes but is not limited to a radio frequency fingerprint extraction method and an ID segment extraction method. The radio frequency fingerprint refers to signal distortion caused by physical defects of a transmitter. The terminal can combine some high-order statistics of random target signal segments to form a radio frequency fingerprint feature, or randomly intercept the target signal segments themselves as radio frequency fingerprint features. The ID segment extraction method is a signal segment obtained by comparing the differences between signals using a correlation method, which remains unchanged for the same unmanned aerial vehicle but is different for different unmanned aerial vehicles. It should be noted that the feature extraction method with signal feature extraction function in the related art can be applied to the embodiments of the present application, and the embodiments of the present application do not limit this.

[0102] In summary, the terminal adds Gaussian white noise with a random signal-to-noise ratio to each preprocessed radio frequency signal, and inputs the target signal obtained by the random channel model to the target signal, and then extracts features of the target signal to determine the initial input information. At this time, the initial input information (i.e., the input information during training of the neural network model) corresponding to each sample unmanned aerial vehicle is subjected to noise addition processing and random channel model processing, which can greatly improve the noise resistance of the neural network model and the adaptability to the wireless channel.

[0103] In an embodiment of the present application, step 104, for each unmanned aerial vehicle type, the sub-connection layer corresponding to the unmanned aerial vehicle type is trained by using the sample information corresponding to the unmanned aerial vehicle type, specifically including:

[0104] For each unmanned aerial vehicle type, the initial input information corresponding to the unmanned aerial vehicle type is input into the initial multi-head convolutional neural network model to obtain second prediction information output by the sub-connection layer corresponding to the unmanned aerial vehicle type.

[0105] The sub-connection layer corresponding to the unmanned aerial vehicle type is trained according to the second prediction information and the sample information corresponding to the unmanned aerial vehicle type.

[0106] The output full connection layer of the initial multi-head convolutional neural network is a target full connection layer, the target full connection layer includes a plurality of sub-connection layers, and the sub-connection layers correspond one-to-one to the unmanned aerial vehicle types.

[0107] Specifically, for each drone type, when training the initial multi-head convolutional neural network model, only the initial input information corresponding to the radio frequency signal of one sample drone of that drone type is input into the initial multi-head convolutional neural network model at a time. The initial multi-head convolutional neural network model is trained based on the second prediction information output by the sub-fully connected layer corresponding to that drone type and the first sequence number information of that sample drone. The terminal needs to repeat the above training process until the initial input information of all sample drones corresponding to that drone type has been input into the initial multi-head convolutional neural network model, completing the training of the sub-fully connected layer corresponding to that drone type. Other sub-connected layers are not trained. The terminal needs to repeat the above training process until the training of all sub-fully connected layers in the initial multi-head convolutional neural network model is completed.

[0108] In one embodiment of this application, step 103, constructing an initial multi-head convolutional neural network model based on a preset network layer and a target fully connected layer, specifically includes:

[0109] Use the model parameters of the preset network layer as the model parameters of the target network layer in the initial multi-head convolutional neural network model;

[0110] Based on the total number of drone types, determine the number of sub-connected layers contained in the target fully connected layer, and establish the correspondence between drone types and sub-connected layers;

[0111] For each drone type, the output data information of the sub-connection layer corresponding to the drone type is determined based on the number of sample drones corresponding to the drone type.

[0112] Based on the model parameters, correspondences, and output data information of the target network layer, an initial multi-head convolutional neural network model is constructed.

[0113] Specifically, such as Figure 5 As shown, the preset network layer in the initial single-head convolutional neural network model has the same structure as the target network layer in the initial multi-head convolutional neural network model. Therefore, using the model parameters of the preset network layer in the trained initial single-head convolutional neural network model as the model parameters of the target network layer in the initial multi-head convolutional neural network model can greatly reduce the training complexity of the initial multi-head convolutional neural network model and improve the training speed of the initial multi-head convolutional neural network model.

[0114] Further, the terminal can determine the number of sub-connection layers included in the target full connection layer in the initial multi-head convolutional neural network model according to the number of types of unmanned aerial vehicles, and establish a correspondence between the types of unmanned aerial vehicles and the sub-connection layers, so that there is a one-to-one correspondence between the types of unmanned aerial vehicles and the sub-connection layers. For each type of unmanned aerial vehicle, the terminal can also determine the output data information of the sub-connection layer corresponding to the type of unmanned aerial vehicle according to the number of sample unmanned aerial vehicles corresponding to the type of unmanned aerial vehicle, so that the output data size of the sub-connection layer corresponding to the type of unmanned aerial vehicle is consistent with the number of sample unmanned aerial vehicles of the type. It should be noted that the terminal outputs the data information, i.e., the second prediction information, which indicates the corresponding probability between the to-be-tested unmanned aerial vehicle and the sample unmanned aerial vehicle of the same type.

[0115] The terminal can construct the initial multi-head convolutional neural network model based on the model parameters of the target network layer, the correspondence, and the output data information. In the process of training the initial multi-head convolutional neural network model, the model parameters of the target network layer remain unchanged, and only the sub-connection layers are trained.

[0116] In an embodiment of the present application, an unmanned aerial vehicle individual identification method is also provided, as shown in Figure 6 The specific process includes the following steps:

[0117] Step 601: Obtain type identification information of a to-be-tested unmanned aerial vehicle and a to-be-tested radio frequency signal.

[0118] The terminal can obtain the type identification information of the to-be-tested unmanned aerial vehicle by using the unmanned aerial vehicle type identification method. It should be noted that the methods for identifying unmanned aerial vehicle types in the related art can be applied to the embodiments of the present application, and the embodiments of the present application are not limited in this regard. In addition, the type identification information of the to-be-tested unmanned aerial vehicle indicates the type of the to-be-tested unmanned aerial vehicle, and the type of the to-be-tested unmanned aerial vehicle belongs to one of the types of unmanned aerial vehicles in the sample set information.

[0119] Step 602: Obtain initial prediction information based on the to-be-tested radio frequency signal, the type identification information, and the unmanned aerial vehicle individual identification model.

[0120] Specifically, the terminal can obtain the corresponding probability between the to-be-tested unmanned aerial vehicle and the sample unmanned aerial vehicle of the same type by using the to-be-tested radio frequency signal and the unmanned aerial vehicle individual identification model. The initial prediction information is a one-dimensional data matrix, and each data in the matrix represents the corresponding probability between the to-be-tested unmanned aerial vehicle and each sample unmanned aerial vehicle of the same type.

[0121] Step 603: Determine individual prediction information of the to-be-tested unmanned aerial vehicle according to the initial prediction information and a preset individual identification condition.

[0122] The unmanned aerial vehicle individual identification model is trained by the training method of the unmanned aerial vehicle individual identification model.

[0123] Specifically, the terminal extracts the maximum value in the initial prediction information matrix, compares the maximum value with a preset threshold, if the maximum value is greater than the preset threshold, the terminal can determine that the to-be-tested unmanned aerial vehicle is a legal unmanned aerial vehicle, and the terminal can determine the first serial number information corresponding to the position of the maximum value in the initial prediction information matrix according to the correspondence between each position in the initial prediction information matrix and the first serial number information of the sample unmanned aerial vehicle (which can be referred to as target serial number information), and then determine that the to-be-tested unmanned aerial vehicle is the sample unmanned aerial vehicle corresponding to the target serial number information. If the maximum value is less than or equal to the preset threshold, the terminal can determine that the to-be-tested unmanned aerial vehicle is an illegal unmanned aerial vehicle.

[0124] Further, it is assumed that the type of the to-be-tested unmanned aerial vehicle has been obtained by the remaining unmanned aerial vehicle type identification method, and is the kth type in the unmanned aerial vehicle types in the sample set information. The to-be-tested radio frequency signal first undergoes the same preprocessing procedure as in the training, and the corresponding features are extracted, and then the features are input into the multi-head convolutional neural network model obtained by training, and the output of the kth full connection layer is taken for further determination. The terminal compares the maximum value v max in the output data with a preset threshold λ k , where v max =max(v k ). If v max >λ k , it is considered that the to-be-tested radio frequency signal belongs to a legal unmanned aerial vehicle, and otherwise, the to-be-tested radio frequency signal belongs to an illegal unmanned aerial vehicle. If the to-be-tested radio frequency signal is determined to belong to a legal unmanned aerial vehicle, the result of individual identification of the to-be-tested unmanned aerial vehicle is that the signal belongs to the jth individual in the kth type, where j=argmax(v k ).

[0125] In the above unmanned aerial vehicle individual identification method, the unmanned aerial vehicle individual identification model is trained by the training method of the unmanned aerial vehicle individual identification model, and based on the type identification information of the to-be-tested unmanned aerial vehicle and the to-be-tested radio frequency signal, the to-be-tested unmanned aerial vehicle is automatically and accurately identified, so that the different types of unmanned aerial vehicle individuals are quickly and efficiently identified.

[0126] In an embodiment of the present application, step 602, based on the to-be-tested radio frequency signal, the type identification information, and the unmanned aerial vehicle individual identification model, obtaining initial prediction information, specifically comprising:

[0127] determining input information according to the to-be-tested radio frequency signal and the preset radio frequency signal extraction condition;

[0128] inputting the input information into the unmanned aerial vehicle individual identification model, identifying the to-be-tested unmanned aerial vehicle through the sub-connection layer corresponding to the type identification information in the unmanned aerial vehicle individual identification model, and obtaining the initial prediction information.

[0129] Specifically, the terminal needs to perform data preprocessing on the to-be-tested radio frequency signal, that is, to perform preset coarse synchronization processing and initial normalization processing on the to-be-tested radio frequency signal, then add preset noise and input to the preset channel model. The terminal also needs to perform feature extraction to determine the input information of the to-be-tested unmanned aerial vehicle. The terminal needs to determine the corresponding sub-connection layer according to the type identification information of the to-be-tested unmanned aerial vehicle, and then take the sub-connection layer output information as the initial prediction information.

[0130] In the unmanned aerial vehicle individual identification method, the unmanned aerial vehicle individual identification model is trained by the training method of the unmanned aerial vehicle individual identification model, and the to-be-tested unmanned aerial vehicle is automatically and accurately identified based on the type identification information of the to-be-tested unmanned aerial vehicle and the to-be-tested radio frequency signal, so that the different types of unmanned aerial vehicle individuals are quickly and efficiently identified.

[0131] It should be understood that, although each step in the flowchart involved in each of the above embodiments is displayed in sequence according to the arrow, these steps are not necessarily executed in sequence according to the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least part of the steps in the flowchart involved in each of the above embodiments can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or steps or stages in other steps.

[0132] Based on the same inventive concept, the present application also provides a training device for implementing the above-mentioned training method of the unmanned aerial vehicle individual identification model. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more recommended model training device embodiments provided below can refer to the limitations of the training method of the unmanned aerial vehicle individual identification model in the above text, which will not be repeated here.

[0133] In one embodiment of the present application, as shown in Figure 7 An unmanned aerial vehicle individual identification model training device 700 is provided, which includes an initial acquisition module 710, a model parameter determination module 720, a model construction module 730 and a training module 740, wherein:

[0134] The initial acquisition module 710 is configured to acquire sample set information, the sample set information including sample information corresponding to multiple unmanned aerial vehicle types, the sample information including individual information and radio frequency signals corresponding to each sample unmanned aerial vehicle;

[0135] The model parameter determination module 720 is configured to train the initial single-head convolutional neural network model based on the sample set information, obtain model parameters of the trained single-head convolutional neural network, and extract model parameters of the preset network layer from the model parameters of the trained single-head convolutional neural network.

[0136] The model construction module 730 is configured to construct an initial multi-head convolutional neural network model based on the preset network layer and a target full connection layer, the target full connection layer including a plurality of sub-connection layers, and each sub-connection layer corresponding to one type of unmanned aerial vehicle.

[0137] The training module 740 is configured to, for each type of unmanned aerial vehicle, train the sub-connection layer corresponding to the type of unmanned aerial vehicle by using the sample information corresponding to the type of unmanned aerial vehicle, and obtain an unmanned aerial vehicle individual identification model.

[0138] In an embodiment of the present application, the model parameter determination module 720 is specifically configured to:

[0139] determine initial input information according to the radio frequency signal and a preset radio frequency signal extraction condition;

[0140] input the initial input information into the initial single-head convolutional neural network model to obtain first prediction information output by the initial single-head convolutional neural network model;

[0141] train the initial single-head convolutional neural network model according to the first prediction information and the individual information, and obtain model parameters of the trained single-head convolutional neural network.

[0142] In an embodiment of the present application, the model parameter determination module 720 is specifically configured to:

[0143] perform preset synchronization processing and initial normalization processing on the radio frequency signal to obtain a preprocessed radio frequency signal;

[0144] add a preset noise signal to the preprocessed radio frequency signal to obtain a noise-added signal, input the noise-added signal into a preset channel model, and obtain a target signal output by the preset channel model;

[0145] extract features of the target signal by using a preset feature extraction method, and determine the initial input information according to a feature extraction result.

[0146] In an embodiment of the present application, the training module 740 is specifically configured to:

[0147] for each type of unmanned aerial vehicle, input initial input information corresponding to the type of unmanned aerial vehicle into the initial multi-head convolutional neural network model to obtain second prediction information output by the sub-connection layer corresponding to the type of unmanned aerial vehicle;

[0148] According to the second prediction information and sample information corresponding to the type of the unmanned aerial vehicle, the sub-connection layer corresponding to the type of the unmanned aerial vehicle is trained.

[0149] In an embodiment of the present application, the model construction module 730 is specifically configured to:

[0150] use the model parameters of the preset network layer as the model parameters of the target network layer in the initial multi-head convolutional neural network model;

[0151] determine the number of sub-connection layers contained in the target full-connection layer according to the total number of types of unmanned aerial vehicles, and establish a correspondence between the types of unmanned aerial vehicles and the sub-connection layers;

[0152] for each type of unmanned aerial vehicle, determine the output data information of the sub-connection layer corresponding to the type of unmanned aerial vehicle according to the number of sample unmanned aerial vehicles corresponding to the type of unmanned aerial vehicle;

[0153] construct the initial multi-head convolutional neural network model based on the model parameters of the target network layer, the correspondence and the output data information.

[0154] Based on the same inventive concept, the embodiments of the present application also provide a kind of unmanned aerial vehicle individual identification device for realizing the unmanned aerial vehicle individual identification method described above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitation in one or more recommended device embodiments provided below can be referred to the limitation of the unmanned aerial vehicle individual identification method in the above, which will not be repeated here.

[0155] In an embodiment of the present application, as shown in Figure 8 a kind of unmanned aerial vehicle individual identification device 800 is provided, comprising: to-be-measured unmanned aerial vehicle information acquisition module 810, initial prediction information determination module 820 and individual identification module 830, wherein:

[0156] The to-be-measured unmanned aerial vehicle information acquisition module 810 is configured to acquire the type identification information of the to-be-measured unmanned aerial vehicle and the to-be-measured radio frequency signal.

[0157] The initial prediction information determination module 820 is configured to obtain initial prediction information based on the to-be-measured radio frequency signal, the type identification information and the unmanned aerial vehicle individual identification model.

[0158] The individual identification module 830 is configured to determine the individual prediction information of the to-be-measured unmanned aerial vehicle according to the initial prediction information and a preset individual identification condition.

[0159] The unmanned aerial vehicle individual identification model is obtained by training the unmanned aerial vehicle individual identification model training method.

[0160] In an embodiment of the present application, the initial prediction information determination module 820 is specifically configured to:

[0161] determine input information according to the to-be-tested radio frequency signal and the preset radio frequency signal extraction condition;

[0162] input the input information to the unmanned aerial vehicle individual identification model, identify the to-be-tested unmanned aerial vehicle through a sub-connection layer corresponding to the type identification information in the unmanned aerial vehicle individual identification model, and obtain initial prediction information.

[0163] The modules in the training device and the unmanned aerial vehicle individual identification device of the unmanned aerial vehicle individual identification model can be all or part realized by software, hardware, and combinations thereof. The modules can be embedded in or independent of a processor in a computer device in a hardware form, or stored in a memory in a computer device in a software form, so as to be called and executed by the processor to perform the operations corresponding to the modules.

[0164] In an embodiment of the present application, a computer device is provided, which can be a terminal, and an internal structure diagram thereof can be as shown in Figure 9 The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The communication interface of the computer device is configured to perform wired or wireless communication with an external terminal. The wireless communication can be achieved through WIFI, mobile cellular network, NFC (near field communication), or other technologies. The computer program is executed by the processor to implement a training method of a scale problem generation model or a scale problem generation method. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball, or touchpad arranged on the shell of the computer device. The input device can also be an external keyboard, touchpad, or mouse, etc.

[0165] Those skilled in the art can understand that Figure 9 the structure shown in the above

[0166] In an embodiment of the present application, a computer device is provided, which includes a memory and a processor. The memory stores a computer program. The processor executes the computer program to implement the steps in the above method embodiments.

[0167] In an embodiment of the present application, a computer readable storage medium is provided, and a computer program is stored on the computer readable storage medium. The computer program is executed by a processor to implement the steps in the above method embodiments.

[0168] In an embodiment of the present application, a computer program product is provided, and the computer program product comprises a computer program. The computer program is executed by a processor to implement the steps in the above method embodiments.

[0169] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties.

[0170] A person of ordinary skill in the art can understand that all or part of the processes in the above method embodiments can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above method embodiments. In the embodiments provided by the present application, any reference to a memory, a database or other medium can include at least one of a non-volatile and a volatile memory. The non-volatile memory can include a read-only memory (ROM), a magnetic tape, a floppy disk, a flash memory, an optical storage, a high-density embedded non-volatile memory, a resistive memory (ReRAM), a magnetoresistive random access memory (MRAM), a ferroelectric memory (FRAM), a phase change memory (PCM), a graphene memory, etc. The volatile memory can include a random access memory (RAM) or an external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as a static random access memory (SRAM) or a dynamic random access memory (DRAM), etc. The database involved in the embodiments provided by the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., but is not limited thereto. The processor involved in the embodiments provided by the present application can be a general processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0171] Any combination of the technical features in the above embodiments can be made, and for the sake of brevity, not all possible combinations are described, however, any combination of the technical features should be considered as within the scope of the present disclosure, as long as the combination is not contradictory.

[0172] The above embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the patent scope of the present application. It should be pointed out that, for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A training method for an individual drone recognition model, characterized in that, The method comprises: obtaining sample set information, the sample set information comprising sample information corresponding to a plurality of unmanned aerial vehicle types, the sample information comprising individual information and radio frequency signals corresponding to each sample unmanned aerial vehicle; training an initial single-head convolutional neural network model based on the sample set information to obtain model parameters of the trained single-head convolutional neural network, and extracting model parameters of a preset network layer from the model parameters of the trained single-head convolutional neural network; constructing an initial multi-head convolutional neural network model based on the preset network layer and a target fully connected layer, the target fully connected layer comprising a plurality of sub-connected layers, the sub-connected layers corresponding one-to-one to the unmanned aerial vehicle types; for each unmanned aerial vehicle type, training the sub-connected layer corresponding to the unmanned aerial vehicle type based on the sample information corresponding to the unmanned aerial vehicle type to obtain an unmanned aerial vehicle individual identification model; wherein the constructing of the initial multi-head convolutional neural network model based on the preset network layer and the target fully connected layer comprises: taking the model parameters of the preset network layer as model parameters of a target network layer in the initial multi-head convolutional neural network model; determining the number of sub-connected layers included in the target fully connected layer according to the total number of the unmanned aerial vehicle types, and establishing a correspondence between the unmanned aerial vehicle types and the sub-connected layers; for each unmanned aerial vehicle type, determining the output data information of the sub-connected layer corresponding to the unmanned aerial vehicle type according to the number of sample unmanned aerial vehicles corresponding to the unmanned aerial vehicle type; constructing the initial multi-head convolutional neural network model based on the model parameters of the target network layer, the correspondence and the output data information; the training of the sub-connected layer corresponding to the unmanned aerial vehicle type based on the sample information corresponding to the unmanned aerial vehicle type for each unmanned aerial vehicle type comprises: for each unmanned aerial vehicle type, inputting initial input information corresponding to the unmanned aerial vehicle type into the initial multi-head convolutional neural network model to obtain second prediction information output by the sub-connected layer corresponding to the unmanned aerial vehicle type; training the sub-connected layer corresponding to the unmanned aerial vehicle type according to the second prediction information and the sample information corresponding to the unmanned aerial vehicle type. 2.The method of claim 1, wherein, the training of the initial single-head convolutional neural network model based on the sample set information to obtain the model parameters of the trained single-head convolutional neural network comprises: determining initial input information according to the radio frequency signals and preset radio frequency signal extraction conditions; inputting the initial input information into the initial single-head convolutional neural network model to obtain first prediction information output by the initial single-head convolutional neural network model; training the initial single-head convolutional neural network model according to the first prediction information and the individual information to obtain the model parameters of the trained single-head convolutional neural network. 3.The method of claim 2, wherein, the determination of the initial input information according to the radio frequency signals and the preset radio frequency signal extraction conditions comprises: performing preset synchronization processing and initial normalization processing on the radio frequency signals to obtain preprocessed radio frequency signals; Add a preset noise signal to the preprocessed radio frequency signal to obtain a noise-added signal, and input the noise-added signal into a preset channel model to obtain a target signal output by the preset channel model; Perform feature extraction on the target signal by using a preset feature extraction method, and determine the initial input information according to a feature extraction result.

4. A method for identifying an individual drone, characterized in that, The method comprises: obtaining type identification information of a to-be-tested unmanned aerial vehicle and a to-be-tested radio frequency signal; obtaining initial prediction information based on the to-be-tested radio frequency signal, the type identification information, and an unmanned aerial vehicle individual identification model; determining individual prediction information of the to-be-tested unmanned aerial vehicle according to the initial prediction information and a preset individual identification condition; The unmanned aerial vehicle individual identification model is trained by the unmanned aerial vehicle individual identification model training method in any one of claims 1 to 3. 5.The method of claim 4, wherein, The method comprises: determining input information according to the to-be-tested radio frequency signal and a preset radio frequency signal extraction condition; inputting the input information into the unmanned aerial vehicle individual identification model, identifying the to-be-tested unmanned aerial vehicle by a sub-connection layer corresponding to the type identification information in the unmanned aerial vehicle individual identification model, and obtaining initial prediction information. 6.A device for training a UAV individual identification model, characterized in that, The device comprises: an initial acquisition module configured to acquire sample set information, the sample set information comprising sample information corresponding to a plurality of unmanned aerial vehicle types, the sample information comprising individual information and a radio frequency signal corresponding to each sample unmanned aerial vehicle; a model parameter determination module configured to train an initial single-head convolutional neural network model based on the sample set information, obtain model parameters of the trained single-head convolutional neural network, and extract model parameters of a preset network layer from the model parameters of the trained single-head convolutional neural network; a model construction module configured to construct an initial multi-head convolutional neural network model based on the preset network layer and a target fully connected layer, the target fully connected layer comprising a plurality of sub-connection layers, and the sub-connection layers corresponding one-to-one to the unmanned aerial vehicle types; a training module configured to, for each unmanned aerial vehicle type, train a sub-connection layer corresponding to the unmanned aerial vehicle type by using sample information corresponding to the unmanned aerial vehicle type, and obtain an unmanned aerial vehicle individual identification model; The model construction module is specifically configured to: use the model parameters of the preset network layer as model parameters of a target network layer in the initial multi-head convolutional neural network model; determine a number of sub-connection layers included in the target fully connected layer according to a total number of the unmanned aerial vehicle types, and establish a correspondence between the unmanned aerial vehicle types and the sub-connection layers; for each unmanned aerial vehicle type, determine output data information of the sub-connection layer corresponding to the unmanned aerial vehicle type according to a number of sample unmanned aerial vehicles corresponding to the unmanned aerial vehicle type; construct the initial multi-head convolutional neural network model based on the model parameters of the target network layer, the correspondence, and the output data information. The training module is specifically configured to: For each type of unmanned aerial vehicle, the initial input information corresponding to the unmanned aerial vehicle type is input into the initial multi-head convolutional neural network model to obtain second prediction information output by a sub-connection layer corresponding to the unmanned aerial vehicle type; The sub-connection layer corresponding to the unmanned aerial vehicle type is trained according to the second prediction information and sample information corresponding to the unmanned aerial vehicle type.

7. An unmanned aerial vehicle individual identification device, characterized by, The apparatus comprises: A to-be-tested unmanned aerial vehicle information acquisition module configured to acquire type identification information of a to-be-tested unmanned aerial vehicle and to-be-tested radio frequency signals; An initial prediction information determination module configured to obtain initial prediction information based on the to-be-tested radio frequency signals, the type identification information, and an unmanned aerial vehicle individual identification model; An individual identification module configured to determine individual prediction information of the to-be-tested unmanned aerial vehicle according to the initial prediction information and preset individual identification conditions. The unmanned aerial vehicle individual identification model is trained by the training method of the unmanned aerial vehicle individual identification model according to any one of claims 1 to 3.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 3 or any one of claims 4 to 5.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method according to any one of claims 1 to 3 or any one of claims 4 to 5.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method according to any one of claims 1 to 3 or any one of claims 4 to 5.

Images