Unknown unmanned aerial vehicle signal identification method and device and electronic equipment

Abstract

The embodiment of the invention discloses an unknown unmanned aerial vehicle signal identification method and device and electronic equipment. The method comprises the steps of extracting position information features of a current unmanned aerial vehicle signal based on a pre-constructed target feature extraction model under the condition that the current unmanned aerial vehicle signal to be identified is received; determining a category distance between the current unmanned aerial vehicle signal and a predetermined feature space prototype point of at least one preset signal category based on the position information features; and when the category distance corresponding to each preset signal category is greater than a preset boundary threshold, determining the current unmanned aerial vehicle signal as an unknown unmanned aerial vehicle signal. According to the technical scheme of the embodiment of the invention, aliasing of the unknown unmanned aerial vehicle signal and the known class signal can be avoided, the unknown unmanned aerial vehicle signal can be effectively identified through the class distance, the current black flight and excessive flight conditions can be accurately determined based on the unknown unmanned aerial vehicle signal, and the safety of a transformer substation can be improved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of signal processing technology, and in particular to a method, device and electronic equipment for identifying unknown drone signals. Background Art

[0002] Substations are an important part of the power system, and their safety is crucial to the stable operation of society. In order to effectively ensure the safety of substations, it is necessary to detect drones flying in the area where the substations belong. In the prior art, a drone recognition model is built based on a convolutional neural network, and the type of the received drone signal is determined by the drone recognition model.

[0003] However, in the process of implementing the present invention, it is found that the prior art has at least the following technical problems: the drone recognition model is usually based on the closed set assumption and can only recognize predefined categories in the training set. Due to the characteristics of the activation function layer in the drone recognition model, when the drone recognition model processes unknown drone signals, it will be forced to select from the known categories in the training set, resulting in the aliasing of unknown category signals and known category signals, and it is impossible to determine whether the received drone signal is an unknown category signal, which in turn leads to the inability to detect the illegal and indiscriminate flying of drones that generate unknown category signals, which is not conducive to ensuring the safety of substations. Summary of the invention

[0004] The embodiments of the present invention provide a method, device and electronic device for identifying unknown drone signals to achieve the purpose of improving the safety of substations. The unknown drone signals are effectively identified through category distance, so as to accurately determine the current illegal and indiscriminate flying situations based on the unknown drone signals.

[0005] According to one aspect of the present invention, a method for identifying an unknown drone signal is provided, comprising:

[0006] When receiving a current drone signal to be identified, extracting a location information feature of the current drone signal based on a pre-built target feature extraction model;

[0007] Based on the location information feature, determining a category distance between the current drone signal and a feature space prototype point of at least one predetermined signal category;

[0008] When the category distance corresponding to each of the preset signal categories is greater than a preset boundary threshold, the current drone signal is determined to be an unknown drone signal.

[0009] According to another aspect of the present invention, a device for identifying an unknown drone signal is provided, comprising:

[0010] A feature extraction module is used to extract the location information feature of the current drone signal based on a pre-built target feature extraction model when receiving the current drone signal to be identified;

[0011] A category distance module, configured to determine a category distance between the current drone signal and a category center of at least one preset signal category based on the location information feature;

[0012] The unknown unmanned signal determination module is used to determine the current drone signal as an unknown drone signal when the category distance corresponding to each of the preset signal categories is greater than a preset boundary threshold.

[0013] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0014] at least one processor; and

[0015] a memory communicatively connected to the at least one processor; wherein,

[0016] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the unknown drone signal identification method described in any embodiment of the present invention.

[0017] The technical solution of the embodiment of the present invention, when receiving the current drone signal to be identified, extracts the location information feature of the current drone signal based on a pre-constructed target feature extraction model; and, based on the location information feature, determines the category distance between the current drone signal and the feature space prototype point of at least one predetermined preset signal category; when the category distance corresponding to each preset signal category is greater than the preset boundary threshold, the current drone signal is determined to be an unknown drone signal. It can be seen that the technical solution provided by this embodiment identifies drone signals through category distance, and no activation function layer is used in the signal identification process, so as to avoid the situation where unknown drone signals and known class signals are mixed. The unknown drone signals can be effectively identified through category distance, which is convenient for accurately determining the current illegal flying and indiscriminate flying situations based on unknown drone signals, which is beneficial to improving the safety of substations.

[0018] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0020] Figure 1 is a flow chart of a method for identifying unknown drone signals provided according to an embodiment of the present invention;

[0021] Figure 2 is a flow chart of another unknown drone signal identification method provided according to an embodiment of the present invention;

[0022] Figure 3 is a schematic diagram of the structure of an unknown drone signal identification device provided according to an embodiment of the present invention;

[0023] Figure 4 It is a structural schematic diagram of an electronic device for implementing the unknown drone signal recognition method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0025] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "etc." and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] It should be noted that the collection, collection, updating, analysis, processing, use, transmission, storage and other aspects of user personal information involved in the technical solution of this disclosure are in compliance with the provisions of relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Necessary measures are taken for user personal information to prevent illegal access to user personal information data and maintain the security of user personal information and network security.

[0027] Figure 1 1 is a flow chart of a method for identifying an unknown drone signal provided according to an embodiment of the present invention. This embodiment is applicable to the case of identifying a currently received drone signal. The method can be performed by an unknown drone signal identification device, which can be implemented in the form of hardware and / or software.

[0028] like Figure 1 As shown, the method of this embodiment may specifically include:

[0029] S110: When a current drone signal to be identified is received, the location information feature of the current drone signal is extracted based on a pre-built target feature extraction model.

[0030] It should be noted that the RF signal generated by the drone can be received by the RF receiver as the current drone signal. In an actual electromagnetic environment, when there is a drone that is not in a hovering state in the scene, the RF signal received by the RF receiver can be expressed as:

[0031] x(t)=x v (t)+x c (t)+J(t)+n(t)

[0032] Where t=1,2,…,T, T is the sampling time of the current drone signal, v x (t) is the downlink video transmission signal sent by the UAV to the UAV controller and received by the RF receiver, which is modulated by orthogonal frequency division multiplexing; v c (t) is the uplink flight control signal sent by the UAV controller to the UAV and received by the RF receiver, which is modulated by frequency hopping; J(t) is the interference signal contained in the current environment, such as a wireless network signal or a Bluetooth signal. n(t) is the hardware inherent noise of the RF receiver, such as Gaussian white noise.

[0033] In this embodiment, after the RF receiver receives the current drone signal, the current drone signal may be sent to an unknown drone signal identification device. In order to determine whether the current drone signal is an unknown drone signal of an unknown category, the features of the current drone signal may be extracted. Specifically, the current drone signal is input into a pre-built target feature extraction model, and the output information of the target feature extraction model is used as the location information feature of the current drone signal. The target feature extraction model may be a neural network model; the location information feature is used to reflect the corresponding relationship between different features in the current drone signal and the signal position. The location signal feature facilitates a better understanding of the sequential relationship between the features in the current drone signal.

[0034] Since the current drone signal is usually coupled in the time domain, it is first necessary to decouple the current drone signal and transform the decoupled signal into 2D to meet the input requirements of the target feature extraction model. Therefore, based on the pre-built target feature extraction model, the specific implementation method of extracting the location information features of the current drone signal includes: performing short-time Fourier transform processing on the current drone signal to obtain the current time-frequency spectrum corresponding to the current drone signal; performing binarization and noise reduction processing on the current time-frequency spectrum, and updating the current time-frequency spectrum based on the time-frequency spectrum obtained after the binarization and noise reduction processing; wherein the current time-frequency spectrum is represented by a vector; based on the target feature extraction model, extracting the location information features of the updated current time-frequency spectrum.

[0035] Specifically, the processing formula for short-time Fourier transform processing of the current drone signal is as follows:

[0036]

[0037] Where X(n,ω) is the current time-frequency spectrum, log represents logarithmic operation, t=1,2,…,T, T is the sampling time of the current UAV signal; W is the window length of the short-time Fourier transform; n = 1, 2, ..., N, N is the number of short-time Fourier transform windows. W is the sliding window length of the short-time Fourier transform, g(t) is the Hamming window function, and |·| is the modulo operation on the complex number.

[0038] Since the current time-frequency spectrum contains the background noise existing in the environment when the signal is collected, in order to enable the target feature extraction model to better focus on the part of the signal generated by the drone, the current time-frequency spectrum can also be subjected to binary denoising. Binarization denoising includes: normalizing the current time-frequency spectrum to obtain a time-frequency spectrum matrix. For each element value in the time-frequency spectrum matrix, the element value is compared with a preset threshold. When the element value is greater than or equal to the preset threshold, the element value in the current time-frequency spectrum matrix is ​​set to 1; when the element value is less than the preset threshold, the element value in the current time-frequency matrix is ​​set to 0. Those skilled in the art can set the size of the preset threshold according to the actual application. Exemplarily, the average value of the element values ​​in the time-frequency spectrum matrix can be used as the preset threshold. The current time-frequency spectrum is updated based on the time-frequency spectrum obtained after the binary denoising process, and the updated current time-frequency spectrum is input into the target feature extraction model, and the information output by the target feature extraction model is used as the location information feature.

[0039] It should be noted that since the uplink flight control signal and downlink video transmission signal of the drone are periodic, the detection of unknown drones can be effectively achieved by extracting the signal intermediate frequency change law of the uplink flight control signal in one cycle.

[0040] In order to be able to extract the signal change law in a cycle, position coding can be added to the current time-frequency spectrum first. Optionally, based on the target feature extraction model, the position information features of the updated current time-frequency spectrum are extracted, including: inputting the updated current time-frequency spectrum into the target feature extraction model, determining the position coding corresponding to the current time-frequency spectrum through the target feature extraction model, adding position coding to the updated current video spectrum to obtain position coding information, extracting the position information features of the position coding information based on the self-attention mechanism module in the target feature extraction model; flattening the position information features, inputting the flattened position information features into the fully connected layer of the target feature extraction model, and updating the output information of the fully connected layer to the position information features.

[0041] Specifically, the process of determining the position code corresponding to the current time-frequency spectrum by the target feature extraction model is as follows: for each position of the current time-frequency spectrum, if the remainder of the signal length corresponding to the position divided by 2 is 1, then the position code at the position is sin(ω i t); the remainder of the signal length at this position divided by 2 is 0, so the position code at this position is cos(ω i t). Where t represents the position index corresponding to the position in the current drone signal, t=1,2,…,T, T is the sampling duration of the current drone signal; ω i represents a preset frequency, which is related to the signal length i; exemplarily, d is the dimension of the current drone signal. Similarly, the position code corresponding to the transpose of the current time-frequency spectrum can be determined.

[0042] In this embodiment, the updated current time-frequency spectrum and the position code corresponding to the current time-frequency spectrum may be superimposed to obtain the angle observation position information in the time domain. Further, the position code corresponding to the transposition of the updated current time-frequency spectrum may be determined, and the transposition of the current time-frequency spectrum and the position code corresponding to the transposition of the current time-frequency spectrum may be superimposed to obtain the angle observation position information in the frequency domain. The angle observation position information in the time domain and the angle observation position information in the frequency domain may be used as the position coding information.

[0043] Furthermore, the position information features of the position encoding information are extracted based on the self-attention mechanism module in the target feature extraction model; the position information features are flattened to flatten the position information features from a two-dimensional matrix to a one-dimensional vector. The flattened position information features are input into the fully connected layer of the target feature extraction model, and the output information of the fully connected layer is updated to the position information features.

[0044] Optionally, the self-attention mechanism module includes a time domain self-attention mechanism sub-module and a frequency domain self-attention mechanism sub-module; the self-attention mechanism module in the target feature extraction model extracts position information features of the position coding information, including: inputting the position coding information into the time domain self-attention mechanism sub-module and the frequency domain self-attention mechanism sub-module respectively, determining the information output by the time domain self-attention mechanism sub-module as the time domain position information feature, and determining the information output by the frequency domain self-attention mechanism sub-module as the frequency domain position information feature; the position information feature is composed of the time domain position information feature and the frequency domain position information feature.

[0045] Specifically, the time domain angle observation position information can be input into the time domain self-attention mechanism submodule to obtain the time domain position information feature; the frequency domain angle observation position information can be input into the frequency domain self-attention mechanism submodule to obtain the frequency domain position information feature. The position information feature is composed of the time domain position information feature and the frequency domain position information feature.

[0046] In this embodiment, the time domain position information features are extracted through the time domain self-attention mechanism sub-module, and the frequency domain position information features are extracted through the frequency domain self-attention mechanism sub-module, so as to understand the position information features from multiple angles more comprehensively and clearly.

[0047] S120: Determine, based on the location information feature, a category distance between the current drone signal and a feature space prototype point of at least one predetermined signal category.

[0048] It should be noted that the preset signal category may be a category corresponding to the sample drone signal included in the training set used when training the target feature extraction model. For example, if the training set contains 10 categories of sample drone signals, then there are also 10 preset signal categories. The feature space prototype point may be any point in the feature space corresponding to the drone signal of the preset signal category. Those skilled in the art may set the feature space prototype point of the preset signal category according to the actual application situation. For example, the feature space prototype point is a space point corresponding to the average vector of the preset signal category, and the average vector is the mean between the feature vectors of the sample drone signal corresponding to the preset signal category.

[0049] In this embodiment, for each preset signal category, the Mahalanobis distance between the position information feature and the feature space prototype point corresponding to the preset signal category is determined, and the Mahalanobis distance is used as the category distance.

[0050] S130: When the category distance corresponding to each preset signal category is greater than a preset boundary threshold, determine the current drone signal as an unknown drone signal.

[0051] In this embodiment, a preset boundary threshold may be determined. Those skilled in the art may determine the preset boundary threshold based on the range of each preset signal category in the feature space according to the actual application. For each preset signal category, if the category distance corresponding to the preset signal category is less than or equal to the preset boundary threshold, it indicates that the current drone signal matches the preset signal category; if the category distance corresponding to the preset signal category is greater than the preset boundary threshold, it indicates that the current drone signal does not match the preset signal category.

[0052] In a specific implementation, it can be determined whether the category distances corresponding to each preset signal category are greater than the preset boundary threshold. If they are all greater, it means that the current drone signal does not match all the preset signal categories. It can be determined that the current drone signal may be an unknown drone signal generated by illegal or indiscriminate flying.

[0053] Furthermore, after determining the category distance between the current drone signal and the category center of at least one preset signal category based on the location information feature, it also includes: when the category distance corresponding to any preset signal category is less than or equal to a preset boundary threshold, determining the minimum distance in the category distance, and determining the preset signal category corresponding to the minimum distance as the current signal category of the current drone signal.

[0054] Specifically, if there is only one category distance that is less than or equal to the preset boundary threshold, the preset signal category corresponding to the category distance can be used as the current signal category of the current drone signal. If there are multiple category distances that are all less than or equal to the preset boundary threshold, they can be sorted according to the category distances, and the minimum distance among the category distances that are less than or equal to the preset boundary threshold can be determined, and the preset signal category corresponding to the minimum distance can be determined as the current signal category.

[0055] This embodiment provides a method for determining the current signal category in response to a situation where the category distance is less than or equal to a preset boundary threshold, so as to facilitate rapid and accurate determination of the current signal category.

[0056] The technical solution of the embodiment of the present invention, when receiving the current drone signal to be identified, extracts the location information feature of the current drone signal based on a pre-constructed target feature extraction model; and, based on the location information feature, determines the category distance between the current drone signal and the feature space prototype point of at least one predetermined preset signal category; when the category distance corresponding to each preset signal category is greater than the preset boundary threshold, the current drone signal is determined to be an unknown drone signal. It can be seen that the technical solution provided by this embodiment identifies drone signals through category distance, and no activation function layer is used in the signal identification process, so as to avoid the situation where unknown drone signals and known class signals are mixed. The unknown drone signals can be effectively identified through category distance, which is convenient for accurately determining the current illegal flying and indiscriminate flying situations based on unknown drone signals, which is beneficial to improving the safety of substations.

[0057] Figure 2 It is a flowchart of another unknown drone signal identification method provided according to an embodiment of the present invention. Based on the above embodiment, this embodiment optionally further includes: obtaining a training set before extracting the location information feature of the current drone signal based on a pre-built target feature extraction model; training the feature extraction model to be trained based on sample drone signals and preset signal categories, and taking the convergence of the target loss function corresponding to the feature extraction model to be trained as the training goal to obtain a target feature extraction model. Among them, the explanations of the terms that are the same or corresponding to the above embodiments are not repeated here. Figure 2 As shown, the method includes:

[0058] S210, obtaining a training set; wherein the training set includes multiple preset signal categories and sample drone signals corresponding to the preset signal categories; based on the sample drone signals and the preset signal categories, training the feature extraction model to be trained, taking the convergence of the target loss function corresponding to the feature extraction model to be trained as the training goal, and obtaining the target feature extraction model.

[0059] In this embodiment, the feature extraction model to be trained is trained based on the sample drone signal and the preset signal category. The feature extraction model to be trained may be a neural network model. The target loss function may include a cross entropy loss function. The model parameters of the feature extraction model to be trained may be adjusted by stochastic gradient descent until the target loss function corresponding to the feature extraction model to be trained converges, that is, the training target is achieved. The feature extraction model to be trained that achieves the training target is used as the target feature extraction model.

[0060] It should be noted that in the process of training the target feature extraction model, in order to achieve better learning effect of the feature extraction model to be trained, the discrimination of the feature vector can be improved and the spatial distribution characteristics can be optimized. Specifically, based on the sample drone signal and the preset signal category, the feature extraction model to be trained is trained, including: for each preset signal category, based on the feature extraction model to be trained, determining the first training distance between the preset complementary point corresponding to the preset signal category and the sample drone signal, and determining the first loss function corresponding to the preset signal category based on the first training distance; determining the target loss function based on the first loss function corresponding to each preset signal category, and training the feature extraction model to be trained based on the target loss function.

[0061] In this embodiment, a preset complementary point can be set for each preset signal category, so that the difference between the out-of-class samples and the preset complementary point is smaller, while the difference between the in-class samples and the preset complementary point is larger. Among them, the preset complementary point is a spatial point outside the feature space corresponding to the sample drone signal of the preset signal category. It should be noted that according to the properties of the complementary point, the probability that the drone signal is of any category is proportional to the distance between the drone signal and the complementary point of the category. The greater the distance, the greater the probability; conversely, the smaller the distance, the smaller the probability. Therefore, the distance between the sample drone signal of each known category and each preset complementary point can be maximized through the normalized exponential function and the cross entropy loss.

[0062] For example, the Euclidean distance and the dot product distance between the location information feature corresponding to the sample drone signal and the preset complementary point can be determined, and the difference between the Euclidean distance of 1 / h and the dot product distance is determined as the first training distance between the sample drone signal and the preset complementary point. Wherein, h is the dimension of the feature space corresponding to the sample drone signal.

[0063] After the first training distance is determined, the first training distance may be normalized by a normalized exponential function, and the first training distance may be updated based on the value obtained after the normalization. The normalization formula is as follows:

[0064]

[0065] Where l1(k) represents the first training distance corresponding to the kth sample drone signal, p1(k) represents the normalized value of l1(k); i∈{1,2,…,K}, K represents the number of sample drone signals, which is a positive integer. l1(i) represents the first training distance corresponding to the i-th sample drone signal.

[0066] Furthermore, for each sample drone signal corresponding to each preset signal category, the first training distance between the sample drone signal and the preset complementary point corresponding to the preset signal category is determined, and based on the first training distance, the first sample loss function corresponding to the sample drone signal is determined; the sum of the first sample loss functions of each sample drone signal corresponding to the preset signal category is determined as the first loss function corresponding to the preset signal category. The first loss function corresponding to each preset signal category can be accumulated, and the accumulated function can be used as the target loss function. The convergence of the target loss function is used as the goal to train the feature extraction model to be trained. Optionally, the calculation formula of the first sample loss function is:

[0067] L1=-log p

[0068] Wherein, p represents the updated first training distance, and L1 represents the first sample loss function corresponding to the updated first training distance.

[0069] This embodiment constructs a sparse feature space by presetting complementarity, so that the feature vectors learned by the feature extraction model to be trained have higher discrimination and better spatial distribution characteristics, which is convenient for improving the training effect.

[0070] In order to make the spatial distribution of feature vectors of drone signals of the same category as close as possible, thereby minimizing the differences between samples within the category, a prototype point learning method can be added on the basis of complementary point learning. Specifically, feature space prototype points can be predefined; feature space prototype points include the center point of the spatial range to which the sample drone signals corresponding to each preset signal category belong. Optionally, the feature space prototype point can be a spatial point corresponding to the average vector of the preset signal category, and the average vector is the mean between the feature vectors of the sample drone signals corresponding to the preset signal category; or, the feature space prototype point can be a spatial point corresponding to the intermediate vector corresponding to the preset signal category.

[0071] In a specific implementation, before determining the target loss function based on the first loss function corresponding to each preset signal category, it also includes: for each preset signal category, based on the feature extraction model to be trained, determining the feature space prototype point of the sample drone signal corresponding to the preset signal category, determining the second training distance between the feature space prototype point and the sample drone signal, and determining the second loss function corresponding to the preset signal category based on the second training distance; wherein the feature space prototype point is the space point corresponding to the average vector of the preset signal category, and the average vector is the mean between the feature vectors of the sample drone signal corresponding to the preset signal category; determining the target loss function based on the first loss function corresponding to each preset signal category, including: determining the target loss function based on the first loss function and the second loss function corresponding to each preset signal category.

[0072] In this embodiment, the optimization goal of the feature space prototype point is to minimize the distance between each sample drone signal and the corresponding feature space prototype point. However, due to the use of complementary point learning, for each preset signal category, there is a situation where the sample drone signals corresponding to the preset signal category are all located near the circle boundary corresponding to the preset signal category, and the circle boundary is a circle edge with the preset complementary point corresponding to the preset signal category as the center and the preset distance as the radius. Since the feature space prototype point is the spatial point corresponding to the average vector of the preset signal category, it is possible that the preset complementary point of the preset signal category coincides with the feature space prototype point, resulting in the invalidation of the prototype point learning strategy and making the optimization of the feature extraction model to be trained more difficult. Therefore, by restricting the feature space prototype points, a more accurate aggregation direction can be provided for the feature space.

[0073] In a specific implementation, for each sample drone signal of each preset signal category, the sample position information feature of the sample drone signal is extracted based on the feature extraction model to be trained, the second training distance between the sample position information feature and the prototype point of the feature space is determined, and the second sample loss function is determined based on the second training distance; the sum of the second sample loss functions of each sample drone signal corresponding to each preset signal category is determined as the second loss function corresponding to the preset signal category. In this embodiment, the calculation formula of the second sample loss function is:

[0074] L2=l2-β(l3-R)

[0075] Wherein, L2 represents the second training distance, l2 is the distance between the feature space prototype point and the preset complementary point corresponding to the preset signal category, β is a preset hyperparameter, and the value range is (0,1); R is the preset distance; L2 is the second sample loss function. Those skilled in the art can determine the preset distance according to the actual application situation, and this embodiment does not limit this.

[0076] In a specific implementation, the sum of the first loss function of each preset signal category is determined to obtain a first sum value; the sum of the second loss function of each preset signal category is determined to obtain a second sum value. The sum of the first sum value and the second sum value is determined as the target loss function. Through this embodiment, a sparse feature space is constructed by combining complementary point learning and prototype point learning, so that the feature vector learned by the feature extraction model to be trained has higher discrimination and better spatial distribution characteristics, which is conducive to improving the training effect of the feature extraction model to be trained.

[0077] Furthermore, for any preset signal category, continuously increasing the distance between the sample drone signal of the preset signal category and the preset complementary point will also simultaneously expand the spatial distribution range of each sample drone signal of the preset signal category, increasing the sparsity of the sample distribution. Therefore, it is necessary to limit the distance between the sample drone signals corresponding to each preset signal category. Specifically, the farthest distance threshold can be determined in advance. For each sample drone signal of each preset signal category, the difference between the first training distance and the preset farthest distance threshold is determined. In order to limit the distance between the sample drone signals corresponding to each preset signal category, a third sample loss function can be set; the sum of each third sample loss function is determined as the third loss function. The determination formula of the third sample loss function is:

[0078] L3=max(Q,0)

[0079] The third sample loss function is L3, and Q represents the difference between the first training distance and the preset maximum distance threshold.

[0080] In this embodiment, in order to improve the training effect of the feature extraction model to be trained, a target loss function may be determined based on the first loss function, the second loss function and the third loss function. Exemplarily, the target loss function may be the sum of the first loss function, the second loss function and the third loss function of a preset multiple. The third loss function of the preset multiple is of the same order of magnitude as the first loss function and the second loss function.

[0081] S220: When a current drone signal to be identified is received, the location information feature of the current drone signal is extracted based on a pre-built target feature extraction model.

[0082] S230: Determine, based on the location information feature, a category distance between the current drone signal and a feature space prototype point of at least one predetermined signal category.

[0083] S240: When the category distance corresponding to each preset signal category is greater than a preset boundary threshold, determine the current drone signal as an unknown drone signal.

[0084] In this embodiment, the feature extraction model to be trained is jointly trained based on complementary point learning and prototype point learning, and the feature space is iteratively optimized to ensure the sparsity of the distribution of UAV signal features.

[0085] Figure 3 : is a schematic diagram of the structure of an unknown drone signal recognition device provided according to an embodiment of the present invention, and the device is used to execute the unknown drone signal recognition method provided by any of the above embodiments. The device and the unknown drone signal recognition method of the above embodiments belong to the same inventive concept, and the details not described in detail in the embodiment of the unknown drone signal recognition device can refer to the embodiment of the above unknown drone signal recognition method. Figure 3 As shown, the device comprises:

[0086] The feature extraction module 10 is used to extract the location information feature of the current drone signal based on a pre-built target feature extraction model when receiving the current drone signal to be identified;

[0087] A category distance module 11, used to determine the category distance between the current drone signal and the category center of at least one preset signal category based on the location information feature;

[0088] The unknown unmanned signal determination module 12 is used to determine the current drone signal as an unknown drone signal when the category distance corresponding to each preset signal category is greater than a preset boundary threshold.

[0089] Based on any optional technical solution in the embodiment of the present invention, optionally, the feature extraction module 10 includes:

[0090] The short-time Fourier transform submodule is used to perform short-time Fourier transform processing on the current UAV signal to obtain the current time-frequency spectrum corresponding to the current UAV signal;

[0091] A denoising submodule, used to perform binarization denoising processing on the current time-frequency spectrum, and update the current time-frequency spectrum based on the time-frequency spectrum obtained after the binarization denoising processing; wherein the current time-frequency spectrum is represented by a vector;

[0092] The feature extraction submodule is used to extract the updated position information features of the current time-frequency spectrum based on the target feature extraction model.

[0093] Based on any optional technical solution in the embodiments of the present invention, optionally, the feature extraction submodule includes:

[0094] A feature extraction unit is used to input the updated current time-frequency spectrum into the target feature extraction model, determine the position code corresponding to the current time-frequency spectrum through the target feature extraction model, add the position code to the updated current video spectrum to obtain the position code information, and extract the position information feature of the position code information based on the self-attention mechanism module in the target feature extraction model;

[0095] The flattening operation unit is used to flatten the position information feature, input the flattened position information feature into the fully connected layer of the target feature extraction model, and update the output information of the fully connected layer to the position information feature.

[0096] On the basis of any optional technical solution in the embodiment of the present invention, optionally, the self-attention mechanism module includes a time domain self-attention mechanism submodule and a frequency domain self-attention mechanism submodule;

[0097] Feature extraction unit, including:

[0098] The information input subunit is used to input the position encoding information into the time domain self-attention mechanism submodule and the frequency domain self-attention mechanism submodule respectively, determine the information output by the time domain self-attention mechanism submodule as the time domain position information feature, and determine the information output by the frequency domain self-attention mechanism submodule as the frequency domain position information feature; the position information feature is composed of the time domain position information feature and the frequency domain position information feature.

[0099] Based on any optional technical solution in the embodiments of the present invention, optionally, the device further includes:

[0100] The current signal category determination module is used to determine the category distance between the current drone signal and the feature space prototype point of at least one preset signal category based on the location information feature, and when the category distance corresponding to any preset signal category is less than or equal to the preset boundary threshold, determine the minimum distance in the category distance, and determine the preset signal category corresponding to the minimum distance as the current signal category of the current drone signal.

[0101] Based on any optional technical solution in the embodiments of the present invention, optionally, the device further includes:

[0102] A training set acquisition module is used to acquire a training set before extracting the location information feature of the current drone signal based on a pre-built target feature extraction model; wherein the training set includes a plurality of preset signal categories and sample drone signals corresponding to the preset signal categories;

[0103] The model training module is used to train the feature extraction model to be trained based on sample drone signals and preset signal categories, and takes the convergence of the target loss function corresponding to the feature extraction model to be trained as the training goal to obtain the target feature extraction model.

[0104] Based on any optional technical solution in the embodiments of the present invention, optionally, the model training module includes:

[0105] A first loss function determination submodule is used to determine, for each preset signal category, a preset complementary point corresponding to the preset signal category and a first training distance between the sample drone signals based on the feature extraction model to be trained, and determine a first loss function corresponding to the preset signal category based on the first training distance; wherein the preset complementary point is a spatial point outside the feature space corresponding to the sample drone signal of the preset signal category;

[0106] The target loss function determination submodule is used to determine the target loss function based on the first loss function corresponding to each preset signal category, and train the feature extraction model to be trained based on the target loss function.

[0107] Based on any optional technical solution in the embodiments of the present invention, optionally, the model training module further includes:

[0108] A second loss function determination submodule is used to determine, for each preset signal category, a feature space prototype point of a sample drone signal corresponding to the preset signal category based on a feature extraction model to be trained, determine a second training distance between the feature space prototype point and the sample drone signal, and determine a second loss function corresponding to the preset signal category based on the second training distance before determining a target loss function based on the first loss function corresponding to each preset signal category; wherein the feature space prototype point is a space point corresponding to an average vector of the preset signal category, and the average vector is a mean value between feature vectors of the sample drone signal corresponding to the preset signal category;

[0109] The target loss function determination submodule includes:

[0110] The target loss function determination unit is used to determine the target loss function based on the first loss function and the second loss function corresponding to each preset signal category.

[0111] The technical solution of the embodiment of the present invention, when receiving the current drone signal to be identified, extracts the location information feature of the current drone signal based on a pre-constructed target feature extraction model; and, based on the location information feature, determines the category distance between the current drone signal and the feature space prototype point of at least one predetermined preset signal category; when the category distance corresponding to each preset signal category is greater than the preset boundary threshold, the current drone signal is determined to be an unknown drone signal. It can be seen that the technical solution provided by this embodiment identifies drone signals through category distance, and no activation function layer is used in the signal identification process, so as to avoid the situation where unknown drone signals and known class signals are mixed. The unknown drone signals can be effectively identified through category distance, which is convenient for accurately determining the current illegal flying and indiscriminate flying situations based on unknown drone signals, which is beneficial to improving the safety of substations.

[0112] It is worth noting that in the above-mentioned embodiment of the unknown drone signal identification device, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0113] Figure 4 : is a schematic diagram of the structure of an electronic device that implements the unknown drone signal identification method of an embodiment of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0114] like Figure 4As shown, the electronic device 20 includes at least one processor 21, and a memory connected to the at least one processor 21, such as a read-only memory (ROM) 22, a random access memory (RAM) 23, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 21 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 22 or the computer program loaded from the storage unit 28 to the random access memory (RAM) 23. In RAM23, various programs and data required for the operation of the electronic device 20 can also be stored. The processor 21, ROM22 and RAM23 are connected to each other through a bus 24. An input / output (I / O) interface 25 is also connected to the bus 24.

[0115] A number of components in the electronic device 20 are connected to the I / O interface 25, including: an input unit 26, such as a keyboard, a mouse, etc.; an output unit 27, such as various types of displays, speakers, etc.; a storage unit 28, such as a disk, an optical disk, etc.; and a communication unit 29, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 29 allows the electronic device 20 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0116] The processor 21 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 21 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 21 executes the various methods and processes described above, such as the unknown drone signal recognition method.

[0117] In some embodiments, the unknown drone signal identification method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 28. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 20 via the ROM 22 and / or the communication unit 29. When the computer program is loaded into the RAM 23 and executed by the processor 21, one or more steps of the unknown drone signal identification method described above may be performed. Alternatively, in other embodiments, the processor 21 may be configured to execute the unknown drone signal identification method in any other appropriate manner (e.g., by means of firmware).

[0118] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0119] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0120] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0121] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0122] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0123] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0124] This embodiment also provides a computer program product, including a computer program, which, when executed by a processor, implements the unknown drone signal identification method provided in any embodiment of the present application.

[0125] In the process of implementation, the computer program product can be written in one or more programming languages ​​or a combination thereof to perform the computer program code of the present invention, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0126] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0127] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying unknown drone signals, characterized in that: include: When receiving a current drone signal to be identified, extracting a location information feature of the current drone signal based on a pre-built target feature extraction model; Based on the location information feature, determining a category distance between the current drone signal and a feature space prototype point of at least one predetermined signal category; When the category distance corresponding to each of the preset signal categories is greater than a preset boundary threshold, the current drone signal is determined to be an unknown drone signal.

2. The method according to claim 1, characterized in that The method of extracting the location information feature of the current drone signal based on the pre-built target feature extraction model includes: Performing short-time Fourier transform processing on the current drone signal to obtain a current time-frequency spectrum corresponding to the current drone signal; Performing binarization denoising processing on the current time-frequency spectrum, and updating the current time-frequency spectrum based on the time-frequency spectrum obtained after the binarization denoising processing; wherein the current time-frequency spectrum is represented by a vector; Based on the target feature extraction model, the updated position information feature of the current time-frequency spectrum is extracted.

3. The method according to claim 2, characterized in that The step of extracting the updated position information feature of the current time-frequency spectrum based on the target feature extraction model includes: Inputting the updated current time-frequency spectrum into the target feature extraction model, determining the position code corresponding to the current time-frequency spectrum through the target feature extraction model, adding the position code to the updated current video spectrum to obtain position code information, and extracting the position information feature of the position code information based on the self-attention mechanism module in the target feature extraction model; The position information feature is flattened, the flattened position information feature is input into a fully connected layer of the target feature extraction model, and the output information of the fully connected layer is updated to the position information feature.

4. The method according to claim 3, characterized in that The self-attention mechanism module includes a time-domain self-attention mechanism submodule and a frequency-domain self-attention mechanism submodule; The extracting the position information feature of the position encoding information based on the self-attention mechanism module in the target feature extraction model includes: Inputting the position encoding information into the time domain self-attention mechanism submodule and the frequency domain self-attention mechanism submodule respectively, determining the information output by the time domain self-attention mechanism submodule as the time domain position information feature, and determining the information output by the frequency domain self-attention mechanism submodule as the frequency domain position information feature; The location information feature is composed of the time domain location information feature and the frequency domain location information feature.

5. The method according to claim 1, characterized in that After determining the category distance between the current drone signal and a feature space prototype point of at least one predetermined signal category based on the location information feature, the method further includes: When the category distance corresponding to any one of the preset signal categories is less than or equal to the preset boundary threshold, the minimum distance among the category distances is determined, and the preset signal category corresponding to the minimum distance is determined as the current signal category of the current drone signal.

6. The method according to claim 1, characterized in that Before extracting the location information feature of the current drone signal based on the pre-built target feature extraction model, the method further includes: Acquire a training set; wherein the training set includes a plurality of preset signal categories and sample drone signals corresponding to the preset signal categories; Based on the sample UAV signal and the preset signal category, the feature extraction model to be trained is trained, and the convergence of the target loss function corresponding to the feature extraction model to be trained is taken as the training goal to obtain the target feature extraction model.

7. The method according to claim 6, characterized in that The training of the feature extraction model to be trained based on the sample drone signal and the preset signal category includes: For each of the preset signal categories, based on the feature extraction model to be trained, determine a first training distance between a preset complementary point corresponding to the preset signal category and the sample drone signal, and based on the first training distance, determine a first loss function corresponding to the preset signal category; wherein the preset complementary point is a spatial point outside the feature space corresponding to the sample drone signal of the preset signal category; The target loss function is determined based on the first loss function corresponding to each of the preset signal categories, and the feature extraction model to be trained is trained based on the target loss function.

8. The method according to claim 7, characterized in that Before determining the target loss function based on the first loss function corresponding to each of the preset signal categories, the method further includes: For each of the preset signal categories, based on the feature extraction model to be trained, determine the feature space prototype point of the sample drone signal corresponding to the preset signal category, determine the second training distance between the feature space prototype point and the sample drone signal, and determine the second loss function corresponding to the preset signal category based on the second training distance; wherein the feature space prototype point is the space point corresponding to the average vector of the preset signal category, and the average vector is the mean between the feature vectors of the sample drone signals corresponding to the preset signal category; The determining the target loss function based on the first loss function corresponding to each of the preset signal categories includes: The target loss function is determined based on the first loss function and the second loss function corresponding to each of the preset signal categories.

9. An unknown drone signal recognition device, characterized in that: include: A feature extraction module is used to extract the location information feature of the current drone signal based on a pre-built target feature extraction model when receiving the current drone signal to be identified; A category distance module, configured to determine a category distance between the current drone signal and a category center of at least one preset signal category based on the location information feature; The unknown unmanned signal determination module is used to determine the current drone signal as an unknown drone signal when the category distance corresponding to each of the preset signal categories is greater than a preset boundary threshold.

10. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the unknown drone signal identification method described in any one of claims 1-8.

Images