Electromagnetic spectrum radiation source identification method, device, equipment and storage medium

Abstract

The present application discloses a method, device, equipment and storage medium for identifying electromagnetic spectrum radiation sources. The present application pre-trains a first signal classification model, which is obtained by training based on a large number of time-frequency diagrams of audio training signals, wherein the frequency spectra of various types of radiation sources are marked on the time-frequency diagram. The model based on this training can learn the frequency spectrum characteristics of various types of radiation sources in the audio signal detected in a complex electromagnetic environment. Furthermore, after the time-frequency diagram of the actually detected audio signal is input into the model, the target signal generated by the radiation source selected from the audio signal can be obtained, and the result is more accurate. Furthermore, the duration of each target signal is detected in the time-frequency diagram, and a waveform diagram matching the duration of each target signal is selected in the time domain diagram of the audio signal to obtain a narrowband target signal corresponding to each target signal, extract the signal characteristics of each narrowband target signal, and determine the type of target radiation source to which the audio signal belongs.

Description

Technical Field

[0001] The present application relates to the technical field of radiation source identification, and more specifically, to an electromagnetic spectrum radiation source identification method, device, equipment and storage medium. Background Art

[0002] Electromagnetic spectrum radiation source identification technology refers to the feature recognition of the received electromagnetic signal, matching the electromagnetic characteristics of the radiation source with the individual radiation source, so as to identify the type of device that emits the electromagnetic signal.

[0003] The results of radiation source identification play a key role in analyzing the communication network structure and determining the threat level. The existing radiation source identification technology selects the fixed frequency signals at different frequency points in the frequency domain for the detected broadband signal, and then identifies the radiation source type for the selected signal. The fixed frequency signal sorting process mainly adopts an energy-based detection algorithm, that is, it determines the presence or absence of the signal by setting a threshold. However, due to practical problems such as differences in terminal transmission power, nonlinearity and instability of the transmission channel, the traditional energy-based detection algorithm cannot overcome the problem of false alarm or missed alarm caused by the fluctuation of noise power in the broadband channel with time and frequency, resulting in inaccurate fixed frequency signals selected from the broadband signal, which in turn affects the accuracy of the final radiation source type identification result. Summary of the invention

[0004] In view of the above problems, this application is proposed to provide a method, device, equipment and storage medium for identifying electromagnetic spectrum radiation sources, so as to achieve the purpose of improving the accuracy of radiation source type identification results. The specific scheme is as follows:

[0005] A method for identifying electromagnetic spectrum radiation sources, comprising:

[0006] Acquire the detected audio signal and a time-frequency diagram of the audio signal;

[0007] The time-frequency graph is input into a preconfigured first signal classification model to obtain at least one target signal in the time-frequency graph output by the model; the first signal classification model is trained by using a training time-frequency graph obtained by marking the frequency spectra of each type of radiation source in the time-frequency graph of the audio training signal as training data;

[0008] Detecting the duration of each target signal in the time-frequency diagram;

[0009] Selecting a waveform diagram that matches the duration of each target signal in the time domain diagram of the audio signal to obtain a narrowband target signal corresponding to each target signal;

[0010] A signal feature of each of the narrowband target signals is extracted, and the target radiation source type to which the audio signal belongs is determined based on the signal feature.

[0011] Preferably, determining the type of target radiation source to which the audio signal belongs based on the signal feature includes:

[0012] Determine a feature vector representation of the signal feature using a preconfigured second signal classification model, wherein the second signal classification model is configured to extract the feature vector representation of the input signal feature and determine an internal state representation of the corresponding radiation source type based on the feature vector representation;

[0013] A target feature vector representation matching the feature vector representation and its corresponding target radiation source type are searched in a pre-established electromagnetic spectrum library as the target radiation source type to which the audio signal belongs; wherein the electromagnetic spectrum library contains feature vector representations corresponding to each registered known type of radiation source.

[0014] Preferably, the output of the first signal classification model also includes the type of radiation source to which each target signal belongs;

[0015] The determining the type of target radiation source to which the audio signal belongs based on the signal feature includes:

[0016] If it is determined based on the output of the first signal classification model that the radiation source type to which each target signal belongs is not an unknown type, then:

[0017] Determine a feature vector representation of the signal feature using a preconfigured second signal classification model, wherein the second signal classification model is configured to extract the feature vector representation of the input signal feature and determine an internal state representation of the corresponding radiation source type based on the feature vector representation;

[0018] Searching a pre-established electromagnetic spectrum library for a target feature vector representation that matches the feature vector representation and a target radiation source type corresponding thereto as the target radiation source type to which the audio signal belongs; wherein the electromagnetic spectrum library contains a feature vector representation corresponding to each known type of radiation source that has been registered;

[0019] If it is determined based on the output of the first signal classification model that the radiation source type to which a target signal belongs is an unknown type, then:

[0020] The radiation source type confirmed by the user is obtained, and the feature vector representation of the signal feature corresponding to the target signal of the unknown type is determined by using the second signal classification model, and the feature vector representation is added to the electromagnetic spectrum library corresponding to the radiation source type confirmed by the user.

[0021] Preferably, it also includes:

[0022] For a target signal of an unknown type output by the first signal classification model, after extracting the signal features of the target signal and obtaining the radiation source type of the target signal confirmed by the user, the signal features of the target signal are labeled with the radiation source type confirmed by the user as new training data;

[0023] The second signal classification model is updated using the newly added training data.

[0024] Preferably, the time-frequency graph is input into a preconfigured first signal classification model to obtain at least one target signal in the time-frequency graph output by the model, comprising:

[0025] Binarizing the time-frequency graph;

[0026] Perform morphological processing on the binarized time-frequency graph to obtain the contour of the time-frequency graph;

[0027] The binarized time-frequency graph surrounded by the time-frequency graph outline is input into a preconfigured first signal classification model to obtain at least one target signal in the time-frequency graph output by the model.

[0028] Preferably, detecting the duration of each target signal in the time-frequency diagram comprises:

[0029] On the time-frequency diagram, each target signal detected in adjacent frames is matched to obtain the signal state of each target signal in each frame, where the signal state includes continuous, interrupted and new signal;

[0030] Based on the signal state of the target signal in each frame, the start frame and the end frame of each target signal are determined, and based on the time of the start frame and the time of the end frame, the duration of each target signal is obtained.

[0031] Preferably, before extracting the signal feature of each narrowband target signal, the method further comprises:

[0032] Filtering the narrowband target signal to remove blank noise segments in the narrowband target signal to obtain a filtered narrowband target signal;

[0033] The extracting the signal feature of each narrowband target signal comprises:

[0034] The signal features of each filtered narrowband target signal are extracted.

[0035] Preferably, the narrowband target signal is a baseband IQ signal, which is expressed in an IQ array form, and the step of extracting the signal feature of each narrowband target signal includes:

[0036] Convert the narrowband target signal into the frequency domain and extract discrete Fourier features;

[0037] Extracting power spectrum features from the narrowband target signal;

[0038] The IQ array is spliced ​​with the discrete Fourier feature and the power spectrum feature, and the obtained spliced ​​feature is used as the signal feature of the narrowband target signal.

[0039] Preferably, the training process of the second signal classification model includes:

[0040] Obtaining a sample audio signal emitted by a radiation source of known type and its sample time-frequency diagram;

[0041] Inputting the sample time-frequency graph into the first signal classification model to obtain a sample target signal in the sample time-frequency graph;

[0042] Detecting the duration of the sample target signal in the sample time-frequency diagram;

[0043] Selecting a waveform diagram that matches the duration of the sample target signal from the sample time domain diagram to obtain a sample narrowband target signal corresponding to the sample target signal;

[0044] Extracting signal features of sample narrowband target signals, and marking the radiation source type with the extracted signal features to obtain training data;

[0045] The second signal classification model is trained using the training data until a set convergence condition is reached.

[0046] Preferably, the acquiring the detected audio signal comprises:

[0047] Acquire the radio frequency signal detected by the receiver;

[0048] Down-converting the radio frequency signal into an intermediate frequency signal, and converting the intermediate frequency signal into a baseband signal;

[0049] The real part information is extracted from the baseband signal and sampled into an audio signal.

[0050] An electromagnetic spectrum radiation source identification device, comprising:

[0051] A data acquisition unit, used to acquire the detected audio signal and a time-frequency diagram of the audio signal;

[0052] A first signal classification model processing unit is used to input the time-frequency graph into a preconfigured first signal classification model to obtain at least one target signal in the time-frequency graph output by the model; the first signal classification model is trained by using a training time-frequency graph obtained by marking the frequency spectra of each type of radiation source in the time-frequency graph of the audio training signal as training data;

[0053] a duration detection unit, used to detect the duration of each target signal in the time-frequency diagram;

[0054] a narrowband target signal determination unit, configured to select a waveform diagram matching the duration of each target signal in the time domain diagram of the audio signal, and obtain a narrowband target signal corresponding to each target signal;

[0055] A signal feature extraction unit, used to extract the signal feature of each narrowband target signal;

[0056] A radiation source type identification unit is used to determine the target radiation source type to which the audio signal belongs based on the signal feature.

[0057] An electromagnetic spectrum radiation source identification device, comprising: a memory and a processor;

[0058] The memory is used to store programs;

[0059] The processor is used to execute the program to implement the various steps of the electromagnetic spectrum radiation source identification method as described above.

[0060] A storage medium stores a computer program, which, when executed by a processor, implements the various steps of the electromagnetic spectrum radiation source identification method as described above.

[0061] By means of the above technical scheme, the present application pre-trains a first signal classification model, and the first signal classification model is trained based on the time-frequency diagram of a large number of audio training signals, wherein the frequency spectra of various types of radiation sources are marked on the time-frequency diagram. The first signal classification model based on this training can learn the frequency spectrum characteristics of various types of radiation sources in the audio signal detected in a complex electromagnetic environment. Furthermore, after the time-frequency diagram of the actually detected audio signal is input into the trained first signal classification model, the model can output at least one target signal in the time-frequency diagram, and the target signal is the signal generated by the radiation source selected from the audio signal. According to the scheme of the present application, the false alarm or missed alarm problem caused by the selection of the target signal based on the energy detection algorithm is avoided, and the selection result is more accurate.

[0062] On this basis, the present application detects the duration of each target signal in the time-frequency diagram, and then selects the waveform diagram that matches the duration of each target signal in the time domain diagram of the audio signal to obtain the narrowband target signal corresponding to each target signal. By extracting the signal characteristics of each narrowband target signal, the target radiation source type to which the audio signal belongs can be determined. The target radiation source type finally obtained is more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0064] Figure 1 A schematic diagram of a process of identifying an electromagnetic spectrum radiation source provided in an embodiment of the present application;

[0065] Figure 2 A schematic diagram of a time-frequency diagram is illustrated;

[0066] Figure 3 A schematic diagram of a time-frequency diagram annotation method is shown as an example;

[0067] Figure 4 An example of a narrowband target signal schematic diagram is provided;

[0068] Figure 5 An example of a signal schematic diagram after filtering a narrowband target signal is provided;

[0069] Figure 6 An example of a structural schematic diagram of a box differential DOB filter is shown;

[0070] Figure 7 A schematic diagram of the structure of an electromagnetic spectrum radiation source identification device provided in an embodiment of the present application;

[0071] Figure 8 A schematic diagram of the structure of an electromagnetic spectrum radiation source identification device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0072] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0073] The present application provides an electromagnetic spectrum radiation source identification solution, which can be applied to various scenarios where the radiation source type of the detected signal needs to be identified.

[0074] The present application solution can be implemented based on a terminal with data processing capabilities, which can be a mobile phone, computer, server, cloud, etc.

[0075] Next, combine Figure 1 The electromagnetic spectrum radiation source identification method of the present application may include the following steps:

[0076] Step S100: Acquire the detected audio signal and the time-frequency diagram of the audio signal.

[0077] Specifically, in this step, the radio frequency signal detected by the receiver antenna can be used directly as the audio signal, or can be used as the audio signal after being pre-processed.

[0078] An optional method of preprocessing the radio frequency signal detected by the receiver antenna may include:

[0079] The radio frequency signal is down-converted into an intermediate frequency signal, and the intermediate frequency signal is converted into a baseband signal; and real part information is extracted from the baseband signal and sampled into an audio signal.

[0080] Specifically, the RF signals detected by the receiver antenna are generally high-frequency signals. In order to enable the receiver amplifier to work stably and reduce interference, the high-frequency RF signals can be down-converted into intermediate frequency signals and then converted into baseband IQ signals. The baseband IQ signal can be understood as a complex signal of I+Qj, where I: in-phase means in-phase, Q means orthogonal, and the phase difference with I is 90 degrees. In the complex function, I is called the real part and j is the imaginary part.

[0081] After conversion to a baseband signal, real information can be extracted from the baseband signal and then sampled into an audio signal at a set sampling rate.

[0082] In this step, in order to separate the narrowband target signal emitted by the radiation source from the audio signal, the time-frequency diagram of the audio signal is further obtained. The time-frequency diagram is the change of the spectrum of the audio signal on the time axis, such as Figure 2 The example is a time-frequency diagram, in which the horizontal axis represents frequency, the vertical axis represents time, and the amplitude can be represented by different grayscale or color.

[0083] Step S110: input the time-frequency graph into a preconfigured first signal classification model to obtain at least one target signal in the time-frequency graph output by the model.

[0084] The first signal classification model is obtained by training using a training time-frequency graph obtained by marking the frequency spectra of each type of radiation source in the time-frequency graph of the audio training signal as training data.

[0085] In this embodiment, an image processing approach is used to extract narrowband target signals. Specifically, a first signal classification model is trained and the model is used to extract narrowband target signals in a time-frequency diagram.

[0086] When training the first signal classification model, a model pre-training method based on massive time-frequency diagrams can be adopted, wherein the massive data can come from the electromagnetic data accumulated during daily detection. Specifically, the collected electromagnetic data is used as an audio training signal, and a time-frequency diagram of the audio training signal is obtained. The frequency spectrum of each type of radiation source is marked on the time-frequency diagram to obtain training data for training the first signal classification model.

[0087] Among them, rectangular window marking can be used to mark the spectrum of the radiation source on the time-frequency diagram, such as Figure 3 As shown, Figure 3 The spectrum marked by the dotted box in the time-frequency diagram of the example is the spectrum of the target signal corresponding to the radiation source.

[0088] In this embodiment, the radiation source type labels in the training data can have several different labeling forms, for example, the radiation source type labels can be divided into: noise type, several known radiation source types and unknown radiation source types. According to this labeling method, the ability of the model to discover the target signal can be improved, and the risk of identifying irregular noise as the target signal can be reduced.

[0089] Moreover, the first signal classification model based on this training has the ability to discover unknown new types of radiation sources in the electromagnetic spectrum space. When the first signal classification model classifies it as an unknown radiation source type, it can be further confirmed manually. Compared with the existing method of manually screening massive signals to discover new radiation sources, it greatly reduces the data scope of manual screening and saves labor costs.

[0090] For another example, the radiation source type label can be divided into two types: noise type and target signal type.

[0091] The first signal classification model can adopt the Faster-RCNN model structure or other image classification and recognition network structures. When processing the input time-frequency graph, the first signal classification model first extracts the features of the ROI region of interest, then performs a convolution kernel full connection operation on the extracted features, and finally performs confidence prediction through a softmax layer to obtain the final recognition result.

[0092] The first signal classification model trained in the manner of the present application can learn the spectrum characteristics of various types of radiation sources in the audio signal detected in a complex electromagnetic environment, and then can mark the target signal on the time-frequency diagram of the input audio signal, thereby separating the narrowband target signal from the broadband audio signal. The target signal is the signal emitted by the radiation source.

[0093] After the first signal classification model is obtained through training, the time-frequency graph obtained in the previous step can be input into the first signal classification model to obtain at least one target signal in the time-frequency graph output by the model. It is understandable that the detected audio signal may contain signals emitted by one or more radiation sources, and therefore, one or more target signals may also exist in the time-frequency graph output by the model.

[0094] Step S120: Detect the duration of each target signal in the time-frequency diagram.

[0095] Specifically, each target signal may be tracked and detected in the time-frequency diagram, thereby determining the duration of each target signal.

[0096] Step S130: Select a waveform diagram matching the duration of each target signal from the time domain diagram of the audio signal to obtain a narrowband target signal corresponding to each target signal.

[0097] After the duration of each target signal is obtained, a waveform diagram matching the duration of each target signal may be selected in the time domain diagram of the audio signal, thereby obtaining a narrowband target signal corresponding to each target signal in the time domain.

[0098] Step S140: extracting signal features of each of the narrowband target signals, and determining the target radiation source type to which the audio signal belongs based on the signal features.

[0099] Specifically, for each narrowband target signal, signal features may be extracted, and the radiation source type may be identified based on the extracted signal features to determine the target radiation source type to which the audio signal belongs.

[0100] The electromagnetic spectrum radiation source identification method provided in the embodiment of the present application is pre-trained with a first signal classification model, and the first signal classification model is trained based on the time-frequency diagram of a large number of audio training signals, wherein the frequency spectrum of each type of radiation source is marked on the time-frequency diagram. The first signal classification model based on this training can learn the frequency spectrum characteristics of each type of radiation source in the audio signal detected in a complex electromagnetic environment. Furthermore, after the time-frequency diagram of the actually detected audio signal is input into the trained first signal classification model, the model can output at least one target signal in the time-frequency diagram, and the target signal is the signal generated by the radiation source selected from the audio signal. According to the scheme of the present application, the false alarm or missed alarm problem caused by the selection of the target signal based on the energy detection algorithm is avoided, and the selection result is more accurate.

[0101] On this basis, the present application detects the duration of each target signal in the time-frequency diagram, and then selects the waveform diagram that matches the duration of each target signal in the time domain diagram of the audio signal to obtain the narrowband target signal corresponding to each target signal. By extracting the signal characteristics of each narrowband target signal, the target radiation source type to which the audio signal belongs can be determined. The target radiation source type finally obtained is more accurate.

[0102] In some embodiments of the present application, the process of inputting the time-frequency diagram into a preconfigured first signal classification model to obtain at least one target signal in the time-frequency diagram output by the model in the above step S110 is introduced.

[0103] Specifically, the time-frequency graph can be a color or grayscale graph. In order to facilitate the processing of the first signal classification model, the time-frequency graph can be first binarized, and then the binarized time-frequency graph can be further morphologically processed to obtain the contour of the time-frequency graph. On this basis, the binarized time-frequency graph surrounded by the contour of the time-frequency graph can be input into the first signal classification model to obtain at least one target signal in the time-frequency graph output by the model.

[0104] In some embodiments of the present application, the process of detecting the duration of each target signal in the time-frequency diagram in the above step S120 is introduced.

[0105] In this embodiment, each target signal can be tracked and detected in the time-frequency diagram, and then the duration of each target signal can be determined. Specifically:

[0106] S1. On the time-frequency diagram, each target signal detected in adjacent frames is matched to obtain the signal state of each target signal in each frame, where the signal state includes continuous, interrupted and new signal.

[0107] Among them, when the signal state of a target signal detected within a frame is a new signal, it means that the frame in which the target signal is located is the starting frame of the target signal; when the signal state of a target signal detected within a frame is interrupted, it means that the frame in which the target signal is located or the previous frame of the frame is the ending frame of the target signal.

[0108] In an optional manner, the process of matching target signals in adjacent frames may include:

[0109] S11. Obtain a background image. The background image may be an image corresponding to a noise segment selected from a time-frequency graph, or an image corresponding to an average value of multiple noise segments.

[0110] S12, subtract each frame in the time-frequency diagram from the background image in order, if the pixel difference is greater than the set threshold, the pixel is determined to be the target signal, otherwise the current frame is determined to be the background image. If more than N frames are determined to be the background image in succession, the signal is considered to be interrupted. Based on this, the signal status of each target signal in each frame can be obtained.

[0111] S2. Based on the signal state of the target signal in each frame, determine the start frame and the end frame of each target signal, and based on the time of the start frame and the time of the end frame, obtain the duration of each target signal.

[0112] Reference Figure 4 , Figure 4 A schematic diagram of a narrowband target signal is illustrated.

[0113] Depend on Figure 4 It can be seen that after the aforementioned step S130, after selecting the waveform diagram matching the duration of each target signal in the time domain diagram of the audio signal and obtaining the narrowband target signal corresponding to each target signal, the extracted narrowband target signal is an independent signal. However, in actual situations, the narrowband target signal may still contain some blank noise segments, such as Figure 4 The noise segment corresponding to the sampling points 0-200 and 350. The existence of the noise segment will affect the accuracy of subsequent signal feature extraction, and thus affect the accuracy of the final radiation source type recognition result.

[0114] To this end, in this embodiment, before extracting the signal features of each narrowband target signal in step S140, a filtering process can be added to the narrowband target signal to remove the blank noise segment in the narrowband target signal to obtain the filtered narrowband target signal. On this basis, the signal features of each filtered narrowband target signal can be extracted in step S140, thereby ensuring that the extracted signal features do not contain noise features, and then the type of radiation source can be accurately identified based on the signal features.

[0115] To Figure 4 Taking the narrowband target signal of the example as an example, the filtered narrowband target signal is as follows Figure 5 As shown. Figure 5 It can be seen that all noise segments can be filtered out through filtering.

[0116] When filtering a narrowband target signal, a box differential DOB filter can be selected in this embodiment for filtering. The box differential DOB filter can detect signal edges based on amplitude discontinuity, and then detect the leading and trailing edges of the pulse from the noisy signal. The core structure of the box differential DOB filter can refer to Figure 6 As shown:

[0117] Figure 6 In the figure, x(k) represents the input data and h(k) is the constructed DOB filter. Marks ①-⑦ represent seven convolution calculation blocks with similar structures. The calculation process uses the input data and the DOB filter for convolution operation. The DOB filter is constructed as:

[0118]

[0119] Wherein, N is the order of the filter, and this method is used to perform sliding detection on the input data frame by frame.

[0120] In some embodiments of the present application, the process of extracting the signal feature of each narrowband target signal in the above step S140 is introduced.

[0121] The narrowband target signal is a baseband IQ signal, which is expressed in the form of an IQ array. Then the signal features of the narrowband target signal extracted in this embodiment may include any one or more combinations of the following features:

[0122] 1. IQ array corresponding to the narrowband target signal;

[0123] 2. Convert the narrowband target signal into the frequency domain and extract the discrete Fourier features.

[0124] 3. Extract power spectrum features of narrowband target signals.

[0125] The signal feature of the narrowband target signal finally extracted can be any one or a combination of the above three features. For example, the above three features can be spliced, that is, the IQ array is spliced ​​with the discrete Fourier feature and the power spectrum feature, and the obtained spliced ​​feature is used as the signal feature of the narrowband target signal. By splicing the three features as the final signal feature, the information content of the signal feature can be increased, and then when the radiation source type is identified based on the signal feature, the accuracy of the identification result can be improved.

[0126] In some embodiments of the present application, the process of determining the type of target radiation source to which the audio signal belongs based on the signal feature in the above step S140 is introduced.

[0127] This embodiment introduces a solution for identifying the type of radiation source through an electromagnetic spectrum library. Specifically, an electromagnetic spectrum library can be configured, which contains a feature vector representation corresponding to each known type of registered radiation source. The feature vector representation corresponding to each radiation source type can be used as an identity fingerprint of the radiation source type.

[0128] In order to determine the characteristic vector representation corresponding to the radiation source type, the present embodiment further provides a second signal classification model, which is configured to extract the characteristic vector representation of the input signal feature and determine the internal state representation of the corresponding radiation source type based on the characteristic vector representation. The second signal classification model can adopt a residual network ResNet structure. In addition, neural network structures such as Inception, CNN, RNN, DNN, DenseNet, VGG, GoogleNet, LSTM, and AlexNet can also be adopted.

[0129] Based on the second signal classification model, this embodiment provides two optional implementations of the above step S140, which are as follows:

[0130] The first one,

[0131] After the signal features of the narrowband target signal are obtained in step S140, the signal features may be input into a second signal classification model, and a feature vector representation of the signal features may be determined using the second signal classification model.

[0132] Furthermore, a target feature vector representation matching the feature vector representation and its corresponding target radiation source type are searched in a pre-established electromagnetic spectrum library as the target radiation source type to which the audio signal belongs.

[0133] The process of matching the feature vector representations may be to calculate the distance between the two feature vector representations, such as the cosine distance, the Euclidean distance, etc., and use the distance between the vector representations as the matching degree of the two feature vector representations, and then set a matching degree screening threshold to find the target feature vector representation greater than the matching degree threshold. The matching degree threshold may be 0.9 or other set values.

[0134] It can be understood that if a matching target feature vector representation can be found in the electromagnetic spectrum library, it means that the corresponding target radiation source type has been registered in the electromagnetic spectrum library. If a matching target feature vector representation cannot be found in the electromagnetic spectrum library, it means that the radiation source type corresponding to the current audio signal may not be registered in the electromagnetic spectrum library, or the feature vector representation output by the second signal classification model is incorrect. At this time, manual intervention can be performed to manually confirm whether the radiation source to which the current audio signal belongs is indeed not registered in the electromagnetic spectrum library. If so, the radiation source type of the current audio signal confirmed by the user is obtained, and then the feature vector representation determined by the second signal classification model is added to the electromagnetic spectrum library corresponding to the radiation source type confirmed by the user to complete the update of the electromagnetic spectrum library. If it is manually verified that the radiation source to which the current audio signal belongs has been registered in the electromagnetic spectrum library, it means that an error has occurred in the second signal classification model. At this time, the signal feature input to the second signal classification model can be used as a new sample, and the manually verified radiation source type can be used as the sample label of the new sample. The new sample and its sample label are used as training data to update the second signal classification model to improve the performance of the second signal classification model.

[0135] The second type,

[0136] Since the electromagnetic spectrum library may not include the feature vector representations corresponding to all types of radiation sources, it may happen that the feature vector representation output by the second signal classification model cannot find a matching feature vector representation in the electromagnetic spectrum library. To this end, this embodiment provides a solution:

[0137] In this embodiment, for the first signal classification model used to extract the narrowband target signal in the above step, the annotation form of the training data in the training stage can be selected as the noise type, several known radiation source types and unknown radiation source types. Based on this, the first signal classification model can preliminarily determine whether the audio signal belongs to the unknown radiation source type.

[0138] If it is determined based on the output of the first signal classification model that the radiation source types to which each target signal belongs are not unknown types, the target radiation source type to which the audio signal belongs can be determined in combination with the electromagnetic spectrum library in the first manner described above.

[0139] If, based on the output of the first signal classification model, it is determined that the radiation source type to which some target signals belong is an unknown type, then:

[0140] For the target signal that is determined by the first signal classification model to be a radiation source of unknown type:

[0141] Request user intervention to manually confirm whether the radiation source to which the current audio signal belongs is indeed not registered in the electromagnetic spectrum library. If manual verification finds that the radiation source to which the current audio signal belongs has been registered in the electromagnetic spectrum library, then the target radiation source type to which the audio signal belongs can be determined in combination with the electromagnetic spectrum library in the first manner mentioned above.

[0142] If manual verification finds that the radiation source to which the current audio signal belongs is indeed not registered in the electromagnetic spectrum library, the radiation source type confirmed by the user can be obtained, and the second signal classification model can be used to determine the feature vector representation of the signal characteristics corresponding to the unknown type of target signal, and the feature vector representation is added to the electromagnetic spectrum library corresponding to the radiation source type confirmed by the user to complete the update of the electromagnetic spectrum library.

[0143] Furthermore, after the electromagnetic spectrum library is updated, the newly added radiation source types and corresponding feature vector representations in the electromagnetic spectrum library can be used as new training data, and the second signal classification model can be updated using the new training data.

[0144] There are many ways to update the second signal classification model, such as updating the second signal classification model every time the electromagnetic spectrum library is updated. Alternatively, the second signal classification model can be updated using newly added training data when the updated data in the electromagnetic spectrum library accumulates to a certain amount or when the set update time is reached.

[0145] By continuously updating the second signal classification model during use, the recognition effect of the second signal classification model becomes better and better. In some cases, the classification result output by the second signal classification model can be directly used as the target radiation source type to which the audio signal belongs.

[0146] Of course, even if the recognition result of the second signal classification model is not directly used as the final result, with iterative training, the feature vector representation determined by the second signal classification model for the input signal characteristics will be more accurate, and the search for matching feature vector representations and corresponding target radiation source types in the electromagnetic spectrum library based on the feature vector representation will also be more accurate.

[0147] In some embodiments of the present application, the training process of the second signal classification model is described.

[0148] The training data of the second signal classification model is composed of signal features as training samples, and the radiation source type corresponding to the signal features as sample labels. In order to match the use scenario of the second signal classification model in the aforementioned embodiment, the process of obtaining training data is similar to the process of steps S100-S140 in the embodiment under the aforementioned use scenario, and is specifically as follows:

[0149] S1. Obtain a sample audio signal emitted by a radiation source of known type and its sample time-frequency diagram.

[0150] S2, inputting the sample time-frequency graph into the first signal classification model to obtain a sample target signal in the sample time-frequency graph;

[0151] S3, detecting the duration of the sample target signal in the sample time-frequency diagram;

[0152] S4, selecting a waveform diagram that matches the duration of the sample target signal in the sample time domain diagram to obtain a sample narrowband target signal corresponding to the sample target signal;

[0153] S5, extracting signal features of the sample narrowband target signal, and marking the radiation source type with the extracted signal features to obtain training data;

[0154] S6. Using the training data to train a second signal classification model until a set convergence condition is reached.

[0155] The above-mentioned training process of the second signal classification model is applicable to the initial training process of the second signal classification model and the subsequent iterative updating process.

[0156] The electromagnetic spectrum radiation source identification device provided in an embodiment of the present application is described below. The electromagnetic spectrum radiation source identification device described below and the electromagnetic spectrum radiation source identification method described above can be referenced to each other.

[0157] See also Figure 7 , Figure 7 This is a schematic diagram of the structure of an electromagnetic spectrum radiation source identification device disclosed in an embodiment of the present application.

[0158] like Figure 7 As shown, the device may include:

[0159] A data acquisition unit 11, used to acquire the detected audio signal and the time-frequency diagram of the audio signal;

[0160] The first signal classification model processing unit 12 is used to input the time-frequency graph into a preconfigured first signal classification model to obtain at least one target signal in the time-frequency graph output by the model; the first signal classification model is trained by using a training time-frequency graph obtained by marking the frequency spectra of each type of radiation source in the time-frequency graph of the audio training signal as training data;

[0161] A duration detection unit 13, used to detect the duration of each target signal in the time-frequency diagram;

[0162] A narrowband target signal determination unit 14 is used to select a waveform diagram matching the duration of each target signal in the time domain diagram of the audio signal to obtain a narrowband target signal corresponding to each target signal;

[0163] A signal feature extraction unit 15, configured to extract a signal feature of each narrowband target signal;

[0164] The radiation source type identification unit 16 is used to determine the target radiation source type to which the audio signal belongs based on the signal feature.

[0165] Optionally, the process of the radiation source type identification unit determining the target radiation source type to which the audio signal belongs based on the signal feature may include:

[0166] Determine a feature vector representation of the signal feature using a preconfigured second signal classification model, wherein the second signal classification model is configured to extract the feature vector representation of the input signal feature and determine an internal state representation of the corresponding radiation source type based on the feature vector representation;

[0167] A target feature vector representation matching the feature vector representation and its corresponding target radiation source type are searched in a pre-established electromagnetic spectrum library as the target radiation source type to which the audio signal belongs; wherein the electromagnetic spectrum library contains feature vector representations corresponding to each registered known type of radiation source.

[0168] Optionally, the output of the first signal classification model may further include the type of radiation source to which each target signal belongs; based on this, the process of the radiation source type identification unit determining the type of target radiation source to which the audio signal belongs based on the signal feature may include:

[0169] If it is determined based on the output of the first signal classification model that the radiation source type to which each target signal belongs is not an unknown type, then:

[0170] Determine a feature vector representation of the signal feature using a preconfigured second signal classification model, wherein the second signal classification model is configured to extract the feature vector representation of the input signal feature and determine an internal state representation of the corresponding radiation source type based on the feature vector representation;

[0171] Searching a pre-established electromagnetic spectrum library for a target feature vector representation that matches the feature vector representation and a target radiation source type corresponding thereto as the target radiation source type to which the audio signal belongs; wherein the electromagnetic spectrum library contains a feature vector representation corresponding to each known type of radiation source that has been registered;

[0172] If it is determined based on the output of the first signal classification model that the radiation source type to which a target signal belongs is an unknown type, then:

[0173] The radiation source type confirmed by the user is obtained, and the feature vector representation of the signal feature corresponding to the target signal of the unknown type is determined by using the second signal classification model, and the feature vector representation is added to the electromagnetic spectrum library corresponding to the radiation source type confirmed by the user.

[0174] Optionally, the device of the present application may also include: a second signal classification model updating unit, which is used to extract the signal characteristics of the target signal of an unknown type output by the first signal classification model, and after obtaining the radiation source type of the target signal confirmed by the user, mark the signal characteristics of the target signal with the radiation source type confirmed by the user as new training data; and use the new training data to update the second signal classification model.

[0175] Optionally, the process in which the first signal classification model processing unit inputs the time-frequency graph into a preconfigured first signal classification model to obtain at least one target signal in the time-frequency graph output by the model may include:

[0176] Binarizing the time-frequency graph;

[0177] Perform morphological processing on the binarized time-frequency graph to obtain the contour of the time-frequency graph;

[0178] The binarized time-frequency graph surrounded by the time-frequency graph outline is input into a preconfigured first signal classification model to obtain at least one target signal in the time-frequency graph output by the model.

[0179] Optionally, the process of the duration detection unit detecting the duration of each target signal in the time-frequency diagram may include:

[0180] On the time-frequency diagram, each target signal detected in adjacent frames is matched to obtain the signal state of each target signal in each frame, where the signal state includes continuous, interrupted and new signal;

[0181] Based on the signal state of the target signal in each frame, the start frame and the end frame of each target signal are determined, and based on the time of the start frame and the time of the end frame, the duration of each target signal is obtained.

[0182] Optionally, the device of the present application may also include: a filtering unit, used to filter the narrowband target signal before extracting the signal characteristics of each narrowband target signal to remove the blank noise segment in the narrowband target signal to obtain a filtered narrowband target signal.

[0183] Optionally, the narrowband target signal is a baseband IQ signal, which is expressed in an IQ array form, and the process of extracting the signal feature of each narrowband target signal by the signal feature extraction unit may include:

[0184] Convert the narrowband target signal into the frequency domain and extract discrete Fourier features;

[0185] Extracting power spectrum features from the narrowband target signal;

[0186] The IQ array is spliced ​​with the discrete Fourier feature and the power spectrum feature, and the obtained spliced ​​feature is used as the signal feature of the narrowband target signal.

[0187] Optionally, the apparatus of the present application may further include a second signal classification model training unit, which is used to train the second signal classification model. The training process may include:

[0188] Obtaining a sample audio signal emitted by a radiation source of known type and its sample time-frequency diagram;

[0189] Inputting the sample time-frequency graph into the first signal classification model to obtain a sample target signal in the sample time-frequency graph;

[0190] Detecting the duration of the sample target signal in the sample time-frequency diagram;

[0191] Selecting a waveform diagram that matches the duration of the sample target signal from the sample time domain diagram to obtain a sample narrowband target signal corresponding to the sample target signal;

[0192] Extracting signal features of sample narrowband target signals, and marking the radiation source type with the extracted signal features to obtain training data;

[0193] The second signal classification model is trained using the training data until a set convergence condition is reached.

[0194] Optionally, the process of the data acquisition unit acquiring the detected audio signal may include:

[0195] Acquire the radio frequency signal detected by the receiver;

[0196] Down-converting the radio frequency signal into an intermediate frequency signal, and converting the intermediate frequency signal into a baseband signal;

[0197] The real part information is extracted from the baseband signal and sampled into an audio signal.

[0198] The electromagnetic spectrum radiation source identification device provided in the embodiment of the present application can be applied to electromagnetic spectrum radiation source identification equipment, such as terminals: mobile phones, computers, etc. Optionally, Figure 8 The hardware structure diagram of the electromagnetic spectrum radiation source identification device is shown in FIG. Figure 8 ,The hardware structure of the electromagnetic spectrum radiation source identification device may include: at least one processor 1, at least one communication interface 2, at least one memory 3 and at least one communication bus 4;

[0199] In the embodiment of the present application, the number of the processor 1, the communication interface 2, the memory 3, and the communication bus 4 is at least one, and the processor 1, the communication interface 2, and the memory 3 communicate with each other through the communication bus 4;

[0200] The processor 1 may be a central processing unit CPU, or an application-specific integrated circuit ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present invention, etc.;

[0201] The memory 3 may include a high-speed RAM memory, and may also include a non-volatile memory, such as at least one disk memory;

[0202] The memory stores a program, and the processor can call the program stored in the memory, wherein the program is used to:

[0203] Acquire the detected audio signal and a time-frequency diagram of the audio signal;

[0204] The time-frequency graph is input into a preconfigured first signal classification model to obtain at least one target signal in the time-frequency graph output by the model; the first signal classification model is trained by using a training time-frequency graph obtained by marking the frequency spectra of each type of radiation source in the time-frequency graph of the audio training signal as training data;

[0205] Detecting the duration of each target signal in the time-frequency diagram;

[0206] Selecting a waveform diagram that matches the duration of each target signal in the time domain diagram of the audio signal to obtain a narrowband target signal corresponding to each target signal;

[0207] A signal feature of each of the narrowband target signals is extracted, and the target radiation source type to which the audio signal belongs is determined based on the signal feature.

[0208] Optionally, the detailed functions and extended functions of the program may refer to the above description.

[0209] The embodiment of the present application further provides a storage medium, which may store a program suitable for execution by a processor, wherein the program is used to:

[0210] Acquire the detected audio signal and a time-frequency diagram of the audio signal;

[0211] The time-frequency graph is input into a preconfigured first signal classification model to obtain at least one target signal in the time-frequency graph output by the model; the first signal classification model is trained by using a training time-frequency graph obtained by marking the frequency spectra of each type of radiation source in the time-frequency graph of the audio training signal as training data;

[0212] Detecting the duration of each target signal in the time-frequency diagram;

[0213] Selecting a waveform diagram that matches the duration of each target signal in the time domain diagram of the audio signal to obtain a narrowband target signal corresponding to each target signal;

[0214] A signal feature of each of the narrowband target signals is extracted, and the target radiation source type to which the audio signal belongs is determined based on the signal feature.

[0215] Optionally, the detailed functions and extended functions of the program may refer to the above description.

[0216] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0217] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can refer to each other.

[0218] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying electromagnetic spectrum radiation sources, characterized in that: include: Acquire the detected audio signal and a time-frequency diagram of the audio signal; The time-frequency graph is input into a preconfigured first signal classification model to obtain at least one target signal in the time-frequency graph output by the model; the first signal classification model is trained by using a training time-frequency graph obtained by marking the frequency spectra of each type of radiation source in the time-frequency graph of the audio training signal as training data; Detecting the duration of each target signal in the time-frequency diagram; Selecting a waveform diagram that matches the duration of each target signal in the time domain diagram of the audio signal to obtain a narrowband target signal corresponding to each target signal; A signal feature of each of the narrowband target signals is extracted, and the target radiation source type to which the audio signal belongs is determined based on the signal feature.

2. The method according to claim 1, characterized in that The determining the type of target radiation source to which the audio signal belongs based on the signal feature includes: Determine a feature vector representation of the signal feature using a preconfigured second signal classification model, wherein the second signal classification model is configured to extract the feature vector representation of the input signal feature and determine an internal state representation of the corresponding radiation source type based on the feature vector representation; A target feature vector representation matching the feature vector representation and its corresponding target radiation source type are searched in a pre-established electromagnetic spectrum library as the target radiation source type to which the audio signal belongs; wherein the electromagnetic spectrum library contains feature vector representations corresponding to each registered known type of radiation source.

3. The method according to claim 1, characterized in that The output of the first signal classification model also includes the type of radiation source to which each target signal belongs; The determining the type of target radiation source to which the audio signal belongs based on the signal feature includes: If it is determined based on the output of the first signal classification model that the radiation source type to which each target signal belongs is not an unknown type, then: Determine a feature vector representation of the signal feature using a preconfigured second signal classification model, wherein the second signal classification model is configured to extract the feature vector representation of the input signal feature and determine an internal state representation of the corresponding radiation source type based on the feature vector representation; Searching a pre-established electromagnetic spectrum library for a target feature vector representation that matches the feature vector representation and a target radiation source type corresponding thereto as the target radiation source type to which the audio signal belongs; wherein the electromagnetic spectrum library contains a feature vector representation corresponding to each known type of radiation source that has been registered; If it is determined based on the output of the first signal classification model that the radiation source type to which a target signal belongs is an unknown type, then: The radiation source type confirmed by the user is obtained, and the feature vector representation of the signal feature corresponding to the target signal of the unknown type is determined by using the second signal classification model, and the feature vector representation is added to the electromagnetic spectrum library corresponding to the radiation source type confirmed by the user.

4. The method according to claim 3, characterized in that Also includes: For a target signal of an unknown type output by the first signal classification model, after extracting the signal features of the target signal and obtaining the radiation source type of the target signal confirmed by the user, the signal features of the target signal are labeled with the radiation source type confirmed by the user as new training data; The second signal classification model is updated using the newly added training data.

5. The method according to claim 1, characterized in that Inputting the time-frequency graph into a preconfigured first signal classification model to obtain at least one target signal in the time-frequency graph output by the model, comprising: Binarizing the time-frequency graph; Perform morphological processing on the binarized time-frequency graph to obtain the contour of the time-frequency graph; The binarized time-frequency graph surrounded by the time-frequency graph outline is input into a preconfigured first signal classification model to obtain at least one target signal in the time-frequency graph output by the model.

6. The method according to claim 1, characterized in that Detecting the duration of each target signal in the time-frequency diagram includes: On the time-frequency diagram, each target signal detected in adjacent frames is matched to obtain the signal state of each target signal in each frame, where the signal state includes continuous, interrupted and new signal; Based on the signal state of the target signal in each frame, the start frame and the end frame of each target signal are determined, and based on the time of the start frame and the time of the end frame, the duration of each target signal is obtained.

7. The method according to claim 1, characterized in that Before extracting the signal feature of each narrowband target signal, the method further comprises: Filtering the narrowband target signal to remove blank noise segments in the narrowband target signal to obtain a filtered narrowband target signal; The extracting the signal feature of each narrowband target signal comprises: The signal features of each filtered narrowband target signal are extracted.

8. The method according to claim 1, characterized in that The narrowband target signal is a baseband IQ signal, which is expressed in an IQ array form. The extracting of the signal feature of each narrowband target signal includes: Convert the narrowband target signal into the frequency domain and extract discrete Fourier features; Extracting power spectrum features from the narrowband target signal; The IQ array is spliced ​​with the discrete Fourier feature and the power spectrum feature, and the obtained spliced ​​feature is used as the signal feature of the narrowband target signal.

9. The method according to claim 2, characterized in that: The training process of the second signal classification model includes: Obtaining a sample audio signal emitted by a radiation source of known type and its sample time-frequency diagram; Inputting the sample time-frequency graph into the first signal classification model to obtain a sample target signal in the sample time-frequency graph; Detecting the duration of the sample target signal in the sample time-frequency diagram; Selecting a waveform diagram that matches the duration of the sample target signal in the sample time domain diagram to obtain a sample narrowband target signal corresponding to the sample target signal; Extracting signal features of sample narrowband target signals, and marking the radiation source type with the extracted signal features to obtain training data; The second signal classification model is trained using the training data until a set convergence condition is reached.

10. The method according to any one of claims 1 to 9, characterized in that: The step of acquiring the detected audio signal comprises: Acquire the radio frequency signal detected by the receiver; Down-converting the radio frequency signal into an intermediate frequency signal, and converting the intermediate frequency signal into a baseband signal; The real part information is extracted from the baseband signal and sampled into an audio signal.

11. An electromagnetic spectrum radiation source identification device, characterized in that: include: A data acquisition unit, used to acquire the detected audio signal and a time-frequency diagram of the audio signal; A first signal classification model processing unit is used to input the time-frequency graph into a preconfigured first signal classification model to obtain at least one target signal in the time-frequency graph output by the model; the first signal classification model is trained by using a training time-frequency graph obtained by marking the frequency spectra of each type of radiation source in the time-frequency graph of the audio training signal as training data; a duration detection unit, used to detect the duration of each target signal in the time-frequency diagram; a narrowband target signal determination unit, configured to select a waveform diagram matching the duration of each target signal in the time domain diagram of the audio signal, and obtain a narrowband target signal corresponding to each target signal; A signal feature extraction unit, used to extract the signal feature of each narrowband target signal; A radiation source type identification unit is used to determine the target radiation source type to which the audio signal belongs based on the signal feature.

12. An electromagnetic spectrum radiation source identification device, characterized in that: include: Memory and processor; The memory is used to store programs; The processor is used to execute the program to implement each step of the electromagnetic spectrum radiation source identification method as described in any one of claims 1 to 10.

13. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, each step of the electromagnetic spectrum radiation source identification method as described in any one of claims 1 to 10 is implemented.

Images