Signal detection method, device, storage medium, electronic device and product

Abstract

This application discloses a signal detection method, device, storage medium, electronic device, and product, relating to the field of artificial intelligence technology. This application can be applied to technical fields such as blockchain, map vehicle networking, etc. The method includes: obtaining signal data corresponding to the signal to be detected; extracting target signal features of the signal to be detected based on the signal data to be detected; classifying the target signal features to obtain first type information corresponding to the signal to be detected; comparing the target signal features with sample signal features to obtain second type information corresponding to the signal to be detected; and determining the signal type of the signal to be detected based on the first type information and the second type information. This application can effectively detect both known and unknown signals, improving the reliability of signal detection.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and specifically to a signal detection method, device, storage medium, electronic device and product. Background Art

[0002] Signal detection involves analyzing various signals to be detected, such as celestial signals. Currently, the commonly used approach in related technologies is to train classification models using supervised learning to detect signals from massive amounts of signal data. Supervised learning methods train classification models using known signals with known patterns. However, these models are generally only effective for classifying known signals.

[0003] However, valuable signals often also include unknown signals that have not been detected by humans. Such unknown signals are often more valuable than known signals that have been observed and their formation reasons have been explained. Under the current scheme, these unknown signals are usually simply blindly classified and cannot be effectively detected, which will cause signal omissions and lead to poor signal detection reliability. Summary of the Invention

[0004] The embodiments of the present application provide a signal detection method, device, storage medium, electronic device and product, which can effectively detect both known signals and unknown signals, effectively improving the reliability of signal detection.

[0005] To solve the above technical problems, the embodiments of the present application provide the following technical solutions:

[0006] According to one embodiment of the present application, a signal detection method includes: obtaining signal data to be detected corresponding to the signal to be detected; extracting target signal features of the signal to be detected based on the signal data to be detected; classifying the target signal features to obtain first type information corresponding to the signal to be detected; comparing the target signal features with sample signal features to obtain second type information corresponding to the signal to be detected; and determining the signal type of the signal to be detected based on the first type information and the second type information.

[0007] According to one embodiment of the present application, a signal detection device includes: a data acquisition module for acquiring signal data to be detected corresponding to the signal to be detected; a feature extraction module for extracting target signal features of the signal to be detected based on the signal data to be detected; a classification module for classifying the target signal features to obtain first type information corresponding to the signal to be detected; a comparison module for comparing the target signal features with sample signal features to obtain second type information corresponding to the signal to be detected; and a determination module for determining the signal type of the signal to be detected based on the first type information and the second type information.

[0008] In one embodiment of the present application, the device also includes a sample signal feature extraction module, and the sample signal feature extraction module includes: a first signal data acquisition unit, used to obtain first signal data corresponding to a first signal sample of a predetermined type, where the predetermined type is a type corresponding to a known signal; and a sample signal feature extraction unit, used to perform feature extraction processing based on the first signal data to obtain the sample signal feature.

[0009] In one embodiment of the present application, the signal type includes one of the predetermined type and the target type, and the target type is the type corresponding to the unknown signal; the sample signal feature extraction unit is used to: input the first signal data into the feature extractor for feature extraction processing to obtain the sample signal features extracted by the feature extractor, wherein the feature extractor is trained based on the second signal data and the first signal data, the second signal data is obtained by adjusting the first signal data, and the second signal data is used to simulate the second signal sample of the target type.

[0010] In one embodiment of the present application, the device also includes a feature extractor training unit, which is used to: input the first signal data into a preset feature extractor to perform intra-class space modeling for the predetermined type to generate a first signal feature for each first signal sample; input the second signal data into a preset feature extractor to perform extra-class space modeling relative to the predetermined type to obtain a second signal feature for each second signal sample; optimize the parameters of the preset feature extractor based on the sample type information obtained by classification processing of the first signal feature and the second signal feature to obtain the trained feature extractor, wherein the optimization direction of the parameter optimization for the intra-class space modeling is consistent with the gravitational direction, and the gravitational direction includes the direction in which each first signal feature is acted upon by the gravitational force of other first signal features.

[0011] In one embodiment of the present application, the feature extraction module is used to: input the signal data to be detected into the feature extractor for feature extraction processing to obtain the target signal feature corresponding to the signal to be detected.

[0012] In one embodiment of the present application, the signal type includes one of a predetermined type and a target type, the predetermined type is a type corresponding to a known signal, and the target type is a type corresponding to an unknown signal; the classification module includes: a feature input unit, used to input the target signal feature into a classifier for classification processing to obtain a classification result corresponding to the signal to be detected, the classifier is trained based on the second signal data and the first signal data, the first signal data corresponds to a first signal sample of a predetermined type, the second signal data is obtained by adjusting the first signal data, and the second signal data is used to simulate a second signal sample of the target type; a result processing unit, used to generate the first type information based on the classification result.

[0013] In one embodiment of the present application, the device also includes a classifier training unit, which is used to: input the first signal feature and the second signal feature into a preset classifier for classification processing to obtain sample type information, wherein the first signal feature is obtained by feature extraction of the first signal data and the second signal data; optimize the parameters of the preset classifier according to the sample type information to obtain the trained classifier.

[0014] In one embodiment of the present application, the classification result is a probability value of the signal to be detected belonging to the predetermined type, and the first type information is a first score that the signal to be detected belongs to the target type; the result processing unit is used to: generate the opposite number of the probability value; and use the opposite number of the probability value as the first score.

[0015] In one embodiment of the present application, the comparison module includes: a similarity calculation unit for calculating the similarity between the target signal feature and the sample signal feature; and an information generation unit for generating second type information corresponding to the signal to be detected based on the similarity.

[0016] In one embodiment of the present application, the sample signal feature corresponds to a first signal sample of a predetermined type, and the second type information is a second score of whether the signal to be detected belongs to the target type; the information generating unit is used to: generate the opposite number of the similarity; and use the opposite number of the similarity as the second score.

[0017] In one embodiment of the present application, the first type information is a first score that the signal to be detected belongs to the target type, and the second type information is a second score that the signal to be detected belongs to the target type; the determination module includes: a type score generation unit, used to generate a type score that the signal to be detected belongs to the target type based on the first score and the second score; a type determination unit, used to determine whether the signal type of the signal to be detected is the target type based on the type score.

[0018] According to one embodiment of the present application, a signal detection method includes: obtaining first signal data of a first signal sample of a predetermined type and second signal data of a second signal sample of a simulated target type, wherein the second signal data is obtained by performing data adjustment processing on the first signal data, and the predetermined type is a type corresponding to a known signal; training a feature extractor and a classifier based on the first signal data and the second signal data; performing feature extraction processing on the signal data to be detected of the signal to be detected by the trained feature extractor to obtain target signal features; classifying the target signal features by the trained classifier to obtain first type information; determining the signal type of the signal to be detected based on the first type information and the second type information, wherein the second type information is obtained by comparing the target signal features with the sample signal features.

[0019] According to one embodiment of the present application, a signal detection device includes: a sample acquisition module for acquiring first signal data of a first signal sample of a predetermined type and second signal data of a second signal sample of a simulated target type, wherein the second signal data is obtained by performing data adjustment processing on the first signal data, and the predetermined type is a type corresponding to a known signal; a training module for training a feature extractor and a classifier based on the first signal data and the second signal data; a first input module for performing feature extraction processing on the signal data to be detected of the signal to be detected by the trained feature extractor to obtain target signal features; a second input module for classifying the target signal features by the trained classifier to obtain first type information; a detection module for determining the signal type of the signal to be detected based on the first type information and the second type information, wherein the second type information is obtained by comparing the target signal features with the sample signal features.

[0020] In one embodiment of the present application, the apparatus further includes a third input module, configured to perform feature extraction processing on the first signal data of the first signal sample by the trained feature extractor to obtain the sample signal feature.

[0021] In one embodiment of the present application, the training module includes: an intra-class space modeling unit, which is used to input the first signal data into a preset feature extractor to perform intra-class space modeling for the predetermined type, and generate a first signal feature for each first signal sample; an out-of-class space modeling unit, which is used to input the second signal data into a preset feature extractor to perform out-of-class space modeling relative to the predetermined type, and obtain a second signal feature for each second signal sample; a sample feature classification unit, which is used to input the first signal feature and the second signal feature into a preset classifier respectively for classification processing to obtain a sample classification result; a type information determination unit, which is used to determine the sample type information of the first signal sample and the second signal sample according to the sample classification result; a parameter optimization unit, which is used to optimize the parameters of the preset feature extractor according to the sample type information, and optimize the parameters of the preset classifier to obtain the trained feature extractor and the classifier.

[0022] In one embodiment of the present application, the parameter optimization unit is used to: perform parameter optimization for the intra-class space modeling according to the sample type information and the target optimization direction, wherein the target optimization direction is consistent with the gravitational direction, and the gravitational direction includes the direction in which each first signal feature is acted upon by the gravitational force of other first signal features.

[0023] In one embodiment of the present application, the sample acquisition module includes: a sample signal data acquisition unit, used to acquire sample signal data of a predetermined type of first signal sample; a folding unit, used to fold the sample signal data to obtain two-dimensional first signal data; an enhancement unit, used to perform data enhancement processing on the first signal data through vertical occlusion to generate the second signal data, and the second signal data is used to simulate the second signal sample of the target type.

[0024] According to another embodiment of the present application, a computer-readable storage medium stores a computer program thereon. When the computer program is executed by a processor of a computer, the computer is caused to execute the method described in the embodiment of the present application.

[0025] According to another embodiment of the present application, an electronic device includes: a memory storing a computer program; and a processor reading the computer program stored in the memory to execute the method described in the embodiment of the present application.

[0026] According to another embodiment of the present application, a computer program product or computer program includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in various optional implementations described in the embodiments of the present application.

[0027] In an embodiment of the present application, signal data to be detected corresponding to the signal to be detected is obtained; target signal features of the signal to be detected are extracted based on the signal data to be detected; the target signal features are classified and processed to obtain first type information corresponding to the signal to be detected; the target signal features are compared and processed with sample signal features to obtain second type information corresponding to the signal to be detected; and the signal type of the signal to be detected is determined based on the first type information and the second type information.

[0028] In this way, the target signal features of the signal to be detected are extracted, and the first type of information and the second type of information are obtained by classifying and comparing the target signal features. The signal type of the signal to be detected is determined by combining the first type of information and the second type of information, and a classification processing method and a comparison processing method of fusion features are realized to avoid signal omissions caused by simple blind classification. Both known signals and unknown signals can be effectively detected, and the reliability of signal detection can be effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present application, 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 application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0030] Figure 1 A schematic diagram of a system to which embodiments of the present application can be applied is shown.

[0031] Figure 2 A flow chart of a signal detection method according to an embodiment of the present application is shown.

[0032] Figure 3 A schematic diagram based on gravity modeling according to an embodiment of the present application is shown.

[0033] Figure 4 A flow chart of a signal detection method according to another embodiment of the present application is shown.

[0034] Figure 5A flowchart of applying an embodiment of the present application to perform signal detection in a scenario is shown.

[0035] Figure 6 A block diagram of a signal detection device according to an embodiment of the present application is shown.

[0036] Figure 7 A block diagram of a signal detection device according to another embodiment of the present application is shown.

[0037] Figure 8 A block diagram of an electronic device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0038] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0039] It is understandable that in the specific implementation of this application, when it comes to data related to content information and interactive behavior information, when the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0040] Figure 1 Schematic diagram of a system 100 to which embodiments of the present application can be applied is shown. Figure 1 As shown, the system 100 may include a server 101 and / or a terminal 102 .

[0041] Server 101 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, as well as big data and artificial intelligence platforms.

[0042] Terminal 102 can be any device, including but not limited to mobile phones, computers, intelligent voice interaction devices, smart home appliances, in-vehicle terminals, VR / AR devices, smart watches, and computers. In some embodiments, terminal 102 and server 101 can be nodes in a blockchain network. In some embodiments, the signal to be detected can be a relevant signal in a map-based Internet of Vehicles platform.

[0043] In one implementation of this example, the server 101 or the terminal 102 can: obtain the signal data to be detected corresponding to the signal to be detected; extract the target signal characteristics of the signal to be detected based on the signal data to be detected; classify the target signal characteristics to obtain the first type information corresponding to the signal to be detected; compare the target signal characteristics with the sample signal characteristics to obtain the second type information corresponding to the signal to be detected; and determine the signal type of the signal to be detected based on the first type information and the second type information.

[0044] In another embodiment of this example, the server 101 or the terminal 102 may: obtain first signal data of a first signal sample of a predetermined type and second signal data of a second signal sample of a simulated target type, where the second signal data is obtained by performing data adjustment processing on the first signal data; train a feature extractor and a classifier based on the first signal data and the second signal data; perform feature extraction processing on the signal data to be detected of the signal to be detected by the trained feature extractor to obtain target signal features; classify the target signal features by the trained classifier to obtain first type information; determine the signal type of the signal to be detected based on the first type information and the second type information, where the second type information is obtained by comparing the target signal features with the sample signal features.

[0045] Figure 2 The flowchart of the signal detection method according to an embodiment of the present application is schematically shown. The execution subject of the signal detection method can be any device, such as Figure 1 The server 101 or the terminal 102 is shown.

[0046] like Figure 2 As shown, the signal detection method may include steps S210 to S250.

[0047] Step S210, obtaining the signal data to be detected corresponding to the signal to be detected; step S220, extracting the target signal features of the signal to be detected based on the signal data to be detected; step S230, classifying the target signal features to obtain first type information corresponding to the signal to be detected; step S240, comparing the target signal features with the sample signal features to obtain second type information corresponding to the signal to be detected; step S250, determining the signal type of the signal to be detected based on the first type information and the second type information.

[0048] The signal to be detected is a signal to be detected, such as various system signals or celestial body signals. In one embodiment of this example, the signal to be detected is a celestial body signal to be detected. The signal data to be detected is signal data of the signal to be detected. The signal data to be detected can be obtained from a predetermined or specified location according to actual operations.

[0049] The target signal feature may be a feature vector that characterizes the signal to be detected. The target signal feature may be obtained by performing feature extraction processing on the signal to be detected data, that is, the target signal feature corresponding to the signal to be detected is extracted.

[0050] The first type of information may be first information describing the signal type of the signal to be detected, and the first type of information corresponding to the signal to be detected is obtained by classifying the target signal features. The second type of information may be second information describing the signal type of the signal to be detected, and the sample signal features may be a feature vector obtained by extracting features from sample signal data of a specific signal sample. The target signal features are compared with the sample signal features to further analyze the second type of information corresponding to the signal to be detected based on the signal type corresponding to the specific signal sample.

[0051] After obtaining the first type of information and the second type of information, the signal type of the signal to be detected is determined by combining the first type of information and the second type of information. The signal type of the signal to be detected can be detected by integrating the feature classification processing method and the comparison processing method. This avoids the problem of simple blind classification based on the supervised learning classification model in the related art. Only when the signal to be detected is a known signal with a known pattern, can the specific known type of the signal type of the signal to be detected be detected through classification (for example, the periodic signal type emitted by a pulsar or the strong and short-lived signal type emitted by a fast radio burst, etc.), while the problem of missed detection of the signal to be detected with an unknown pattern will be caused by blind classification.

[0052] In the present application, the classification processing method and the comparison processing method of the fusion features are used to detect the signal type of the signal to be detected. For example, when the signal to be detected is a known signal with a known pattern, the first type information and the second type information can be combined to determine the signal type as a predetermined type (also called a normal signal type or a known signal type, etc.). When the signal to be detected is an unknown signal with an unknown pattern, the first type information and the second type information can be combined to determine the signal type as a target type (also called an abnormal signal type or an unknown signal type, etc.). Unknown signals with unknown patterns can also be distinguished without causing signal omission.

[0053] In this way, based on steps S210 to S250, the target signal features of the signal to be detected are extracted, and the first type of information and the second type of information are obtained respectively by classifying and comparing the target signal features. The signal type of the signal to be detected is determined in combination with the first type of information and the second type of information, and the classification processing method and the comparison processing method of the fusion features are realized, which can effectively detect both known signals and unknown signals, thereby effectively improving the reliability of signal detection.

[0054] The following further describes Figure 2 The steps performed under the signal detection embodiment shown correspond to specific embodiments.

[0055] In step S210, the signal data to be detected corresponding to the signal to be detected is obtained.

[0056] In one example, based on a user's selection operation in a signal selection interface, the device may obtain, from a predetermined location, signal data corresponding to the signal to be detected selected in the selection operation. In another example, in response to a user triggering a detection instruction in a detection interface, the device may obtain, from a predetermined location specified in the detection interface, signal data at that location as the signal data corresponding to the signal to be detected. It is understood that the device may obtain signal data according to other optional data acquisition methods.

[0057] In one embodiment, the collected raw signal data of the signal to be detected is folded (i.e., converted from the time domain to the frequency domain) through a fast Fourier transform (FFT) to obtain two-dimensional image data as the signal to be detected data. Based on the two-dimensional image data, the signal to be detected can be used as an image signal to accurately extract target signal features. The two-dimensional image data can include a time-phase distribution diagram and a frequency-phase distribution diagram. It is understood that in other embodiments, the signal to be detected data corresponding to the signal to be detected is the raw signal data.

[0058] In one embodiment, the signal to be detected is a signal of a celestial body to be detected, and the original signal data is an original cosmic signal collected by a radio telescope from space. The original cosmic signal is folded by fast Fourier transform (FFT) to obtain two-dimensional image data. The two-dimensional image data is used as the signal data to be detected corresponding to the celestial body signal to be detected, and the target signal characteristics of the celestial body signal to be detected can be accurately extracted.

[0059] In step S220 , target signal features of the signal to be detected are extracted based on the data of the signal to be detected.

[0060] Inputting the signal data to be detected into the feature extractor can perform feature extraction processing to obtain the signal features extracted by the feature extractor as target signal features. The feature extractor can be an extractor that extracts signal features from the signal data to be detected.

[0061] The feature extractor can be a pre-trained feature extractor based on a neural network such as a convolutional neural network or a deep residual network. For example, the feature extractor can be a feature extractor based on a convolutional neural network. The feature extractor can extract target signal features by performing operations such as convolution and pooling on the input signal data to be detected. In one embodiment of this example, the feature extractor can be a ResNet18 convolutional neural network.

[0062] In one embodiment, in step S240, before comparing the target signal feature with the sample signal feature, the signal detection method also includes: obtaining first signal data corresponding to a first signal sample of a predetermined type, where the predetermined type is a type corresponding to a known signal; performing feature extraction processing based on the first signal data to obtain the sample signal feature.

[0063] The predetermined type is the type corresponding to a known signal. For example, known signals with known patterns (such as periodic signals emitted by pulsars or strong but short-lived signals emitted by fast radio bursts) are uniformly designated as predetermined types (also referred to as normal signal types or known signal types). The first signal sample is a collected known signal sample, such as a periodic signal sample emitted by a pulsar.

[0064] The first signal data can be two-dimensional image data obtained by folding the original signal data of the first signal sample through fast Fourier transform (FFT) as the first signal data. Based on the two-dimensional image data, the first signal sample can be used as an image signal to accurately extract the sample signal characteristics. The two-dimensional image data may include a time-phase distribution diagram and a frequency-phase distribution diagram.

[0065] The sample signal characteristics correspond to the predetermined type corresponding to the known signal. In the subsequent steps, by comparing the target signal characteristics with the sample signal characteristics, the similarity between the signal to be detected and the known signal can be analyzed, and then the second type information of the signal to be detected can be accurately analyzed in reverse based on the similarity.

[0066] In one embodiment, the signal type includes one of a predetermined type and a target type, where the target type is a type corresponding to the unknown signal; and feature extraction processing is performed based on the first signal data to obtain sample signal features, including:

[0067] The first signal data is input into a feature extractor for feature extraction processing to obtain sample signal features extracted by the feature extractor, wherein the feature extractor is trained based on the second signal data and the first signal data, the second signal data is obtained by adjusting the first signal data, and the second signal data is used to simulate a second signal sample of the target type.

[0068] In this embodiment, the first signal data is data corresponding to a first signal sample of a predetermined type. Since the predetermined type corresponds to a known signal, the first signal sample is a known signal. The second signal data is obtained by adjusting the first signal data. The second signal data is used to simulate a second signal sample of a target type. The target type corresponds to an unknown signal, so the second signal sample is a simulated unknown signal.

[0069] Furthermore, the feature extractor trained by the second signal data and the first signal data will draw a boundary between the known signal and the simulated unknown signal, because the type of the first signal data is a predetermined type (i.e., the type corresponding to the known signal) and the type of the second signal data is a target type (i.e., the type corresponding to the unknown signal). The feature extractor is trained by the second signal data and the first signal data to perform feature extraction. When the extractor parameters in the trained feature extractor are used for feature extraction, the extracted signal features can be used to effectively distinguish between unknown signals and known signals. That is, when the feature extractor uses the trained extractor parameters for feature extraction, the boundary between the known signal and the simulated unknown signal is drawn. The first signal data is then input into the feature extractor for feature extraction processing to obtain sample signal features. The sample signal features are extracted on the basis that the extractor parameters can draw the boundary between the known signal and the simulated unknown signal. Furthermore, in subsequent steps, the second type of information can be further accurately analyzed by comparing the target signal features with the sample signal features.

[0070] In one embodiment, the feature extractor is trained according to the following steps: inputting the first signal data into a preset feature extractor to perform intra-class space modeling for a predetermined type to generate a first signal feature for each first signal sample; inputting the second signal data into a preset feature extractor to perform extra-class space modeling relative to a predetermined type to obtain a second signal feature for each second signal sample; performing parameter optimization on the preset feature extractor based on sample type information obtained by classification processing of the first signal feature and the second signal feature to obtain a trained feature extractor, wherein the optimization direction of the parameter optimization for intra-class space modeling is consistent with the gravitational direction, and the gravitational direction includes the direction in which each first signal feature is acted upon by the gravitational force of other first signal features.

[0071] In this embodiment, when training a feature extractor, the preset feature extractor is trained to perform intra-class space modeling for a predetermined type, that is, the preset feature extractor is trained to perform intra-class space modeling for a known signal, and feature extraction processing is performed on the first signal data through the intra-class space modeling to obtain a first signal feature. Simultaneously, the preset feature extractor is trained to perform extra-class space modeling relative to a predetermined type, that is, the preset feature extractor is trained to perform extra-class space modeling for a simulated unknown signal, and feature extraction processing is performed on the second signal data through the extra-class space modeling to obtain a second signal feature.

[0072] The sample type information obtained by classification processing based on the first signal feature and the second signal feature (for example, the probability that the first signal sample belongs to the predetermined type and the probability that the second signal sample belongs to the predetermined type) can be analyzed, and the loss of the sample type information and the label type information (for example, the probability that the first signal sample belongs to the predetermined type and the probability that the second signal sample belongs to the predetermined type) can be analyzed. The parameters of the preset feature extractor are optimized according to the loss until the training stop condition is met (for example, the number of optimization times reaches a threshold or the classification accuracy is higher than a threshold, etc.) to obtain the trained feature extractor.

[0073] In particular, the optimization direction of parameter optimization for intra-class space modeling is consistent with the gravitational direction, which includes the direction in which each first signal feature is subject to the gravitational force of other first signal features. Specifically, the optimization direction of parameter optimization when optimizing the preset feature extractor is as follows: continuously adjust the extractor parameters based on the sample type information obtained by classification processing of the first signal feature and the second signal feature, determine the direction of the gravitational force on each first signal feature modeled in the intra-class feature space by the preset feature extractor based on the extractor parameters, and control the adjustment of the extractor parameters so that the direction of these gravitational forces meets the predetermined training requirements (for example, the accuracy of the sample type information is higher than the predetermined accuracy or the number of optimizations meets the limited number of times, etc.), and obtain the extractor parameters that meet the requirements. In this way, the feature extractor can implement the intra-class space modeling method based on gravity, so that similar first signal samples are closer in the feature space, thereby further effectively distinguishing the first signal sample (i.e., the known signal) from the second signal sample (i.e., the simulated unknown signal), and further improving the reliability of dividing the boundary between the two.

[0074] In one embodiment, see Figure 3In this application, referring to the gravitational force between point masses, there can be at least one first signal sample, each of which corresponds to a first signal feature. Each first signal feature is considered a "point mass" with unit mass, and gravitational force will be generated between the first signal features. The magnitude of the gravitational force between the first signal features is inversely proportional to the distance between the first signal features. The distance between the first signal features can be measured using methods such as cosine similarity or Euclidean distance of feature vectors. The magnitude of the gravitational force can be calculated based on the gravitational formula.

[0075] See Figure 3 , when modeling the intra-class space, the first signal feature is modeled in the "intra-class feature space", Figure 3 Hollow circles represent feature points to be optimized (i.e., a first signal feature to be optimized), solid diamonds represent feature points corresponding to other predetermined types of first signal samples (i.e., other first signal features). When modeling out-of-class space, the second signal feature is modeled in the "out-of-class feature space," and solid rectangular points represent out-of-class feature points (i.e., second signal features).

[0076] When optimizing parameters for intra-class space modeling, the optimization direction for parameter optimization for modeling the feature points to be optimized (i.e., the first signal features to be optimized) is consistent with the direction of gravity. Specifically, the optimization direction is: continuously adjusting the extractor parameters based on the sample type information obtained by classification processing of the first signal features and the second signal features, and for each feature point to be optimized (i.e., the first signal features to be optimized) obtained by the preset feature extractor based on the extractor parameters in the intra-class feature space modeling, the direction of gravity on the feature point to be optimized (i.e., the first signal features to be optimized) can be determined, and the adjustment of the extractor parameters can be controlled so that the directions of gravity meet the predetermined training requirements (for example, the accuracy of the sample type information is higher than the predetermined accuracy or the number of optimizations meets the limited number, etc.), and the extractor parameters that meet the requirements are obtained. Among them, the direction of gravity includes the direction in which each first signal feature is subject to the gravity of other first signal features. For example, a hollow circle is subject to the gravity of other solid diamond points, and the gravity of other solid diamond points will form a total gravity on the hollow circle. The direction of the total gravity (such as Figure 3 The “gravitational optimization direction” corresponding to the hollow dot shown is the direction in which the hollow dot (i.e., a first signal feature to be optimized) is acted upon by the gravitational force of the solid diamond point (i.e., other first signal features).

[0077] In one embodiment, step S220, extracting target signal features of the signal to be detected based on the signal to be detected data, includes: inputting the signal to be detected data into a feature extractor for feature extraction processing to obtain target signal features corresponding to the signal to be detected.

[0078] In this embodiment, the target signal features are extracted based on the trained feature extractor in the aforementioned embodiment. Since the feature extractor draws a boundary between known signals and simulated unknown signals, the target signal features extracted based on the divided boundary can further improve the analysis accuracy of the first type of information and the second type of information when used for classification processing and comparison processing.

[0079] In step S230, the target signal features are classified to obtain first type information corresponding to the signal to be detected.

[0080] The target signal features are input into a pre-trained classifier, which can classify the target signal features and output a classification result corresponding to the signal to be detected. Based on the classification result, first type information corresponding to the signal to be detected can be generated. In one embodiment of this example, the classifier is a fully connected classifier.

[0081] After the classification result is activated by an activation function (such as a softmax function), the probability that the signal to be detected belongs to a predetermined type (i.e., the type corresponding to the known signal) can be obtained. Based on this probability, the first type information can be further reversely generated (for example, the opposite or inverse of the reversely generated probability, etc.). The first type information can be a score that the signal to be detected belongs to the target type (i.e., the type corresponding to the unknown signal).

[0082] In one embodiment, the signal type includes one of a predetermined type and a target type, the predetermined type is a type corresponding to a known signal, and the target type is a type corresponding to an unknown signal; in step S230, the target signal feature is classified and processed to obtain first type information corresponding to the signal to be detected, including: inputting the target signal feature into a classifier for classification processing to obtain a classification result corresponding to the signal to be detected, the classifier is trained based on the second signal data and the first signal data, the first signal data corresponds to a first signal sample of the predetermined type, the second signal data is obtained by adjusting the first signal data, and the second signal data is used to simulate a second signal sample of the target type; and the first type information is generated based on the classification result.

[0083] In this embodiment, the first signal data is data corresponding to a first signal sample of a predetermined type. Since the predetermined type corresponds to a known signal, the first signal sample is a known signal. The second signal data is obtained by adjusting the first signal data. The second signal data is used to simulate a second signal sample of a target type. The target type corresponds to an unknown signal, so the second signal sample is a simulated unknown signal.

[0084] The classifier trained by the second signal data and the first signal data will draw a boundary between known signals and simulated unknown signals, and then input the target signal features into the trained classifier for classification processing to obtain the first type of information. The first type of information is generated based on the dividing boundary, further improving the accuracy of the first type of information.

[0085] In one embodiment, the classifier is trained according to the following steps: the first signal feature and the second signal feature are respectively input into a preset classifier for classification processing to obtain sample type information, wherein the first signal feature is obtained by feature extraction of the first signal data and the second signal data; and the preset classifier is optimized according to the sample type information to obtain a trained classifier.

[0086] The preset classifier classifies the first signal feature and the second signal feature to obtain sample type information (for example, the probability that the first signal sample belongs to the predetermined type and the probability that the second signal sample belongs to the predetermined type). The loss of the sample type information and the label type information (for example, the probability that the first signal sample belongs to the predetermined type and the probability that the second signal sample belongs to the predetermined type) can be analyzed, and the parameters of the preset classifier are optimized according to the loss until the training stop condition is met (for example, the number of optimization times reaches a threshold or the classification accuracy is higher than a threshold, etc.) to obtain the trained classifier.

[0087] In one embodiment, the classification result is a probability value of the signal to be detected belonging to a predetermined type, and the first type information is a first score of the signal to be detected belonging to the target type; generating the first type information based on the classification result includes: generating the opposite number of the probability value; and using the opposite number of the probability value as the first score.

[0088] In this embodiment, the classification result is the probability value that the signal to be detected belongs to a predetermined type (i.e., the type corresponding to the known signal). By generating the opposite number of the probability value as the first score, the applicant has found that generating the first score in this reverse generation method can extremely accurately determine the signal type of the signal to be detected.

[0089] In step S240 , the target signal feature is compared with the sample signal feature to obtain second type information corresponding to the signal to be detected.

[0090] In one embodiment, step S240 compares the target signal feature with the sample signal feature to obtain the second type of information corresponding to the signal to be detected. This can be: subtracting the target signal feature from the sample signal feature to obtain the gap between the two, and determining the second type of information corresponding to the signal to be detected based on the gap. For example, a preset score corresponding to the range of the gap is used as the second type of information corresponding to the signal to be detected.

[0091] In one embodiment, step S240 compares the target signal feature with the sample signal feature to obtain the second type of information corresponding to the signal to be detected, including: calculating the similarity between the target signal feature and the sample signal feature; and generating the second type of information corresponding to the signal to be detected based on the similarity.

[0092] In this embodiment, the method for calculating the similarity between the target signal feature and the sample signal feature can be calculating cosine similarity or Euclidean distance. In one embodiment of this example, the method for calculating similarity is calculating cosine similarity. By calculating the distance between the target signal feature and the sample signal feature using the similarity measure, the difference between the two can be accurately reflected. Based on the similarity, the second type of information corresponding to the signal to be detected can be accurately generated.

[0093] In one embodiment, the sample signal feature corresponds to a first signal sample of a predetermined type, and the second type information is a second score of whether the signal to be detected belongs to the target type; determining the second type information corresponding to the signal to be detected based on the similarity includes: generating the opposite number of the similarity; and using the opposite number of the similarity as the second score.

[0094] In this embodiment, the opposite number of the similarity is generated as the second score. The applicant has found that the signal type of the signal to be detected can be determined very accurately by generating the second score in this reverse generation manner.

[0095] In step S250 , the signal type of the signal to be detected is determined based on the first type information and the second type information.

[0096] In one embodiment, the first type information is a first score for whether the signal to be detected belongs to the target type, and the second type information is a second score for whether the signal to be detected belongs to the target type; step S250, based on the first type information and the second type information, determines the signal type of the signal to be detected, including: if the first score is higher than a first score threshold and the second score is higher than a second score threshold, then the signal type of the signal to be detected is the target type; otherwise, the signal type of the signal to be detected is a predetermined type.

[0097] In one embodiment, the first type information is a first score for whether the signal to be detected belongs to the target type, and the second type information is a second score for whether the signal to be detected belongs to the target type; step S250, based on the first type information and the second type information, determines the signal type of the signal to be detected, including: generating a type score for whether the signal to be detected belongs to the target type based on the first score and the second score; and determining whether the signal type of the signal to be detected is the target type based on the type score.

[0098] The method of generating a type score for the signal to be detected belonging to the target type based on the first score and the second score can be to directly sum or multiply the first score and the second score, or to weight the summation according to a predetermined weighting coefficient, etc., and use the result as the type score for the signal to be detected belonging to the target type.

[0099] Whether the signal type of the signal to be detected is the target type is determined according to the type score. If the type score is higher than a predetermined score threshold, the signal type of the signal to be detected is the target type.

[0100] Figure 4 The flowchart of the signal detection method according to an embodiment of the present application is schematically shown. The execution subject of the signal detection method can be any device, such as Figure 1 The server 101 or the terminal 102 is shown.

[0101] like Figure 4 As shown, the signal detection method may include steps S310 to S350.

[0102] Step S310, obtaining first signal data of a first signal sample of a predetermined type and second signal data of a second signal sample of a simulated target type, where the second signal data is obtained by performing data adjustment processing on the first signal data; step S320, training a feature extractor and a classifier based on the first signal data and the second signal data; step S330, performing feature extraction processing on the signal data to be detected of the signal to be detected by the trained feature extractor to obtain target signal features; step S340, classifying the target signal features by the trained classifier to obtain first type information; step S350, determining the signal type of the signal to be detected based on the first type information and the second type information, where the second type information is obtained by comparing the target signal features with the sample signal features.

[0103] The first signal data is data corresponding to a first signal sample of a predetermined type. The predetermined type may be a type corresponding to a known signal, i.e., the first signal sample may be a known signal. The second signal data is obtained by adjusting the first signal data. The second signal data is used to simulate a second signal sample of a target type. The target type may be a type corresponding to an unknown signal, i.e., the second signal sample is a simulated unknown signal.

[0104] The feature extractor and classifier trained using the second signal data and the first signal data can demarcate a boundary between known signals and simulated unknown signals. The feature extractor can then extract target signal features based on the demarcation of the boundary. Inputting the target signal features into the trained classifier for classification can yield accurate first-type information. It will be appreciated that in some embodiments, the feature extractor and classifier can form a detection model.

[0105] The second type of information may be second information describing the signal type of the signal to be detected, and the sample signal feature may be a feature vector obtained by extracting features from the sample signal data of a specific signal sample. By comparing and processing the target signal feature with the sample signal feature, the second type of information corresponding to the signal to be detected may be further analyzed based on the signal type corresponding to the specific signal sample.

[0106] After obtaining the first type of information and the second type of information, the signal type of the signal to be detected is determined by combining the first type of information and the second type of information. The signal type of the signal to be detected can be detected by integrating the feature classification processing method and the comparison processing method. This avoids the problem of simple blind classification based on the supervised learning classification model in the related art. Only when the signal to be detected is a known signal with a known pattern, can the specific known type of the signal type of the signal to be detected be detected through classification (for example, the periodic signal type emitted by a pulsar or the strong and short-lived signal type emitted by a fast radio burst, etc.), while the problem of missed detection of the signal to be detected with an unknown pattern will be caused by blind classification.

[0107] In the present application, the classification processing method and the comparison processing method of the fusion features are used to detect the signal type of the signal to be detected. For example, when the signal to be detected is a known signal with a known pattern, the first type information and the second type information can be combined to determine the signal type as a predetermined type (also called a normal signal type or a known signal type, etc.). When the signal to be detected is an unknown signal with an unknown pattern, the first type information and the second type information can be combined to determine the signal type as a target type (also called an abnormal signal type or an unknown signal type, etc.). Unknown signals with unknown patterns can also be distinguished without causing signal omission.

[0108] In this way, based on steps S310 to S350, the target signal features of the signal to be detected are extracted, and the first type of information and the second type of information are obtained respectively by classifying and comparing the target signal features. The signal type of the signal to be detected is determined in combination with the first type of information and the second type of information, and a classification processing method and a comparison processing method of fusion features are realized to avoid signal omissions caused by simple blind classification, and both known signals and unknown signals can be effectively detected, thereby effectively improving the reliability of signal detection.

[0109] In one embodiment, before determining the signal type of the signal to be detected based on the first type information and the second type information, the method further includes: performing feature extraction processing on the first signal data of the first signal sample through a trained feature extractor to obtain sample signal features.

[0110] The feature extractor trained with the second signal data and the first signal data can draw a boundary between known signals and simulated unknown signals. The first signal data is input into the feature extractor for feature extraction processing to obtain sample signal features. The sample signal features are extracted based on the divided boundaries. Furthermore, by comparing the target signal features with the sample signal features, the second type of information can be further accurately analyzed.

[0111] In one embodiment, training a feature extractor and a classifier based on the first signal data and the second signal data includes:

[0112] The first signal data is input into a preset feature extractor for modeling the intra-class space of a predetermined type to generate a first signal feature for each first signal sample; the second signal data is input into a preset feature extractor for modeling the extra-class space of a predetermined type to obtain a second signal feature for each second signal sample; the first signal feature and the second signal feature are respectively input into a preset classifier for classification processing to obtain a sample classification result; based on the sample classification result, the sample type information of the first signal sample and the second signal sample is determined; the parameters of the preset feature extractor are optimized based on the sample type information, and the parameters of the preset classifier are optimized to obtain a trained feature extractor and classifier.

[0113] When training the feature extractor, the preset feature extractor is trained to perform intra-class space modeling for a predetermined type, that is, the preset feature extractor is trained to perform intra-class space modeling for a known signal, and feature extraction processing is performed on the first signal data through the intra-class space modeling to obtain a first signal feature. Simultaneously, the preset feature extractor is trained to perform extra-class space modeling relative to the predetermined type, that is, the preset feature extractor is trained to perform extra-class space modeling for a simulated unknown signal, and feature extraction processing is performed on the second signal data through the extra-class space modeling to obtain a second signal feature.

[0114] The preset classifier classifies the first signal feature and the second signal feature to obtain sample type information (for example, the probability that the first signal sample belongs to the predetermined type and the probability that the second signal sample belongs to the predetermined type). The loss of the sample type information and the label type information (for example, the probability that the first signal sample is calibrated to belong to the predetermined type and the probability that the second signal sample is calibrated to belong to the predetermined type) can be analyzed, and the parameters of the preset feature extractor and the preset classifier are optimized according to the loss until the training stop condition is met (for example, the number of optimization times reaches a threshold or the classification accuracy is higher than a threshold, etc.) to obtain the trained feature extractor and classifier.

[0115] In one embodiment, parameters of a preset feature extractor are optimized based on sample type information, including: based on the sample type information, parameter optimization is performed for intra-class space modeling according to a target optimization direction, wherein the target optimization direction is consistent with the gravitational direction, and the gravitational direction includes the direction in which each first signal feature is acted upon by the gravitational force of other first signal features.

[0116] The optimization direction of parameters for intra-class space modeling is consistent with the gravitational direction. The gravitational direction includes the direction in which each first signal feature is acted upon by the gravitational force of other first signal features. This enables the feature extractor to implement a gravity-based intra-class space modeling method so that similar first signal samples are closer in the feature space, thereby further effectively distinguishing the first signal sample (i.e., the known signal) from the second signal sample (i.e., the simulated unknown signal), and further improving the reliability of dividing the boundary between the two.

[0117] In one embodiment, see Figure 3 In this application, referring to the gravitational force between point masses, there can be at least one first signal sample, each of which corresponds to a first signal feature. Each first signal feature is considered a "point mass" with unit mass, and gravitational force will be generated between the first signal features. The magnitude of the gravitational force between the first signal features is inversely proportional to the distance between the first signal features. The distance between the first signal features can be measured using methods such as cosine similarity or Euclidean distance of feature vectors. The magnitude of the gravitational force can be calculated based on the gravitational formula.

[0118] See Figure 3 , when modeling the intra-class space, the first signal feature is modeled in the "intra-class feature space", Figure 3 Hollow circles represent feature points to be optimized (i.e., a first signal feature to be optimized), solid diamonds represent feature points corresponding to other predetermined types of first signal samples (i.e., other first signal features). When modeling out-of-class space, the second signal feature is modeled in the "out-of-class feature space," and solid rectangular points represent out-of-class feature points (i.e., second signal features).

[0119] When optimizing parameters for intra-class space modeling, the optimization direction for optimizing parameters for the modeling of the feature points to be optimized (i.e., the first signal features to be optimized) is consistent with the direction of gravity. The direction of gravity includes the direction in which each first signal feature is affected by the gravity of other first signal features. For example, a hollow dot is affected by the gravity of other solid diamond points, and the gravity of other solid diamond points will form a total gravity on the hollow dot. The direction of the total gravity (e.g., Figure 3 The “gravitational optimization direction” corresponding to the hollow dot shown is the direction in which the hollow dot (i.e., a first signal feature to be optimized) is acted upon by the gravitational force of the solid diamond point (i.e., other first signal features).

[0120] In one embodiment, the first signal data is first signal data obtained by obtaining a first signal sample of a predetermined type and second signal data of a second signal sample of a simulated target type, including:

[0121] Acquire sample signal data of a first signal sample of a predetermined type; fold the sample signal data to obtain two-dimensional first signal data; perform data enhancement processing on the two-dimensional first signal data by vertically occluding in the time dimension and the frequency dimension to generate second signal data, and the second signal data is used to simulate a second signal sample of the target type.

[0122] The sample signal data of the first signal sample may be the collected original signal data, and the sample signal data may be folded by fast Fourier transform (FFT) to obtain two-dimensional first signal data (i.e., two-dimensional image data). The two-dimensional first signal data (i.e., two-dimensional image data) may be, for example, Figure 5 The image data corresponding to the noise (normal signal) shown in FIG. The two-dimensional first signal data is subjected to data enhancement processing by vertical occlusion to generate second signal data, for example, as shown in FIG. Figure 5 As shown, part of the image data corresponding to the noise (normal signal) is vertically shielded to obtain the image data corresponding to the "abnormal signal", i.e., the second signal data. In this way, the applicant has found that the second signal data can effectively simulate the second signal sample of the target type.

[0123] In particular, when the signal to be detected is a signal of a celestial body to be detected, the original signal data is the original cosmic signal collected from space by a radio telescope, and the original cosmic signal is folded by fast Fourier transform (FFT) to obtain two-dimensional first signal data (i.e., two-dimensional image data), and the first signal data (i.e., two-dimensional image data) may include a time-phase distribution diagram and a frequency-phase distribution diagram, and the two-dimensional first signal data (i.e., two-dimensional image data) is vertically blocked, and part of the data of a predetermined width and a predetermined height in the first signal data (i.e., two-dimensional image data) is blocked. The way of blocking the part of the data may be to reset the part of the data to data of a predetermined pixel value, and then, the second signal data of the second signal sample of the simulated target type may be constructed in the time dimension and the frequency dimension. For example, Figure 5 As shown, vertically masking a portion of the 2D image data corresponding to the "noise (normal) signal" yields the 2D image data corresponding to the "abnormal signal," or second signal data. In this way, when the detected celestial body signal is of the target type, it effectively meets the criteria used by astronomical experts for identifying unknown celestial body signals of scientific value.

[0124] The above embodiment is further described below with reference to a process of performing signal detection in an application scenario. Figure 5 The following figure shows the flow chart of signal detection in this scenario. In this scenario, known signals with known patterns in celestial body signals are regarded as "noise signals", and unknown signals with unknown patterns are regarded as "abnormal signals".

[0125] Among them, the pulsar signal is separated from the "noise signal" as an abnormal signal for testing. The abnormal signal for testing can be called the "pulsar (abnormal) signal". In this scenario, the "pulsar (abnormal) signal" is the "signal to be detected".

[0126] The signal other than the pulsar signal in the "noise signal" can be called the "noise (normal) signal". The "noise (normal) signal" is used as the first signal sample of the predetermined type. The first signal data corresponding to the "noise (normal) signal" is adjusted to obtain the second signal data. The second signal data is used to simulate the "abnormal signal". The "abnormal signal" is used as the second signal sample of the target type, that is, the predetermined type is the type corresponding to the known signal, and the target type is the type corresponding to the unknown signal.

[0127] See Figure 5 The signal detection process in this scenario may include steps S410 to S480.

[0128] Step S410 , obtaining first signal data of a “noise (normal) signal” and second signal data of a simulated “abnormal signal”, where the second signal data is obtained by performing data adjustment processing on the first signal data.

[0129] Among them, step S410, obtaining the first signal data of the "noise (normal) signal" and the second signal data of the simulated "abnormal signal", specifically includes: step S411, obtaining the sample signal data of the "noise (normal) signal" (that is, the original cosmic signal); folding the sample signal data to obtain two-dimensional first signal data; step S412, performing data enhancement processing on the first signal data through vertical occlusion to generate second signal data.

[0130] The first signal data can be obtained by folding the sample signal data of the "noise (normal) signal" collected by the radio telescope (i.e., the original cosmic signal) through fast Fourier transform (FFT) to obtain the two-dimensional first signal data. The two-dimensional first signal data is as follows: Figure 5 The two-dimensional image data corresponding to the “noise (normal) signal” shown in FIG. 1 may include a time-phase distribution diagram and a frequency-phase distribution diagram.

[0131] Further, such as Figure 5As shown, part of the two-dimensional image data corresponding to the "noise (normal) signal" is vertically blocked to obtain the two-dimensional image data corresponding to the "abnormal signal", namely the second signal data, and then the second signal data simulating the "abnormal signal" is constructed in the time dimension and the frequency dimension. In this way, when the signal of the celestial body to be detected is detected as the target type, the signal of the celestial body to be detected can effectively meet the judgment of astronomical experts on unknown celestial body signals with scientific value.

[0132] Step S420: training a feature extractor and a classifier based on the first signal data and the second signal data.

[0133] Among them, step S420, training a feature extractor and a classifier based on the first signal data and the second signal data, specifically includes steps S421 to S424.

[0134] Step S421: input the first signal data into a preset feature extractor to perform intra-class space modeling for a predetermined type, and generate a first signal feature for each first signal sample.

[0135] See Figure 5 , preset feature extractor (such as Figure 5 The feature extractor shown in FIG. 1 performs intra-class space modeling in the “intra-class feature space” to generate the first signal feature of each first signal sample (ie, Figure 5 Normal features shown).

[0136] Step S422: input the second signal data into a preset feature extractor to perform modeling relative to a predetermined type of out-of-class space, and obtain a second signal feature of each second signal sample.

[0137] See Figure 5 , the preset feature extractor performs out-of-class space modeling outside the “in-class feature space” to generate the second signal feature of each second signal sample (i.e. Figure 5 abnormal features shown).

[0138] Step S423: Input the first signal feature and the second signal feature into a preset classifier (such as Figure 5 The auxiliary classifier shown in the figure) performs classification processing to obtain a sample classification result; according to the sample classification result, determines the sample type information of the first signal sample and the second signal sample;

[0139] Step S424 , optimizing parameters of the preset feature extractor and the preset classifier according to the sample type information, so as to obtain a trained feature extractor and classifier.

[0140] Among them, the parameters of the preset feature extractor are optimized according to the sample type information, specifically including: according to the sample type information, according to the target optimization direction, parameter optimization is performed for intra-class space modeling, wherein the target optimization direction is consistent with the gravity direction, and the gravity direction includes the direction in which each first signal feature is affected by the gravity of other first signal features.

[0141] The preset classifier classifies the first signal feature and the second signal feature to obtain sample type information (such as the probability that the first signal sample belongs to a predetermined type and the probability that the second signal sample belongs to a predetermined type), analyzes the sample type information and the label type information (i.e. Figure 5 The loss of the "auxiliary classification label" shown, such as the probability of the first signal sample being calibrated to belong to the predetermined type and the probability of the second signal sample being calibrated to belong to the predetermined type), is used in this scenario. Figure 5 The cross entropy loss shown is used to optimize the parameters of the preset feature extractor and the preset classifier through "back propagation" according to the loss until the training stop condition is met (for example, the number of optimizations reaches a threshold or the classification accuracy is higher than a threshold, etc.) to obtain the trained feature extractor and classifier.

[0142] See Figure 3 Referring to the gravitational attraction between point masses, there can be at least one "noise (normal) signal," each of which corresponds to a first signal feature. Considering each first signal feature as a point mass with unit mass, gravitational attraction occurs between the first signal features. The magnitude of the gravitational attraction between the first signal features is inversely proportional to the distance between them. The distance between the first signal features can be measured using methods such as the cosine similarity of feature vectors or the Euclidean distance. The magnitude of the gravitational attraction can be calculated using the gravitational formula.

[0143] See Figure 3 , when modeling the intra-class space, the first signal feature is modeled in the "intra-class feature space", Figure 3 Hollow circles represent feature points to be optimized (i.e., a first signal feature to be optimized), solid diamonds represent feature points corresponding to other predetermined types of "noise (normal) signals" (i.e., other first signal features). When modeling out-of-class space, the second signal feature is modeled in the "out-of-class feature space," and solid rectangles represent out-of-class feature points (i.e., second signal features).

[0144] When optimizing parameters for intra-class space modeling, the optimization direction for optimizing parameters for the modeling of the feature points to be optimized (i.e., the first signal features to be optimized) is consistent with the direction of gravity. The direction of gravity includes the direction in which each first signal feature is affected by the gravity of other first signal features. For example, a hollow dot is affected by the gravity of other solid diamond points, and the gravity of other solid diamond points will form a total gravity on the hollow dot. The direction of the total gravity (e.g., Figure 3 The “gravitational optimization direction” corresponding to the hollow dot shown is the direction in which the hollow dot (i.e., a first signal feature to be optimized) is acted upon by the gravitational force of the solid diamond point (i.e., other first signal features).

[0145] Step S430: Perform feature extraction based on the first signal data to obtain sample signal features. The process of performing feature extraction based on the first signal data to obtain sample signal features specifically includes: inputting the first signal data into a trained feature extractor to perform feature extraction and obtain sample signal features extracted by the feature extractor. Figure 5 The trained feature extractor can then perform feature extraction in the "intra-class feature space" to obtain sample signal features. In this scenario, the feature extractor uses the ResNet18 convolutional neural network, specifically its 17 convolutional layers.

[0146] Step S440, obtain the signal data to be detected corresponding to the "pulsar (abnormal) signal". Figure 5 The two-dimensional image data corresponding to the "pulsar (abnormal) signal" shown in FIG. The original signal data of the "pulsar (abnormal) signal" collected by the radio telescope is folded by fast Fourier transform (FFT) to obtain the following Figure 5 The two-dimensional image data corresponding to the “pulsar (abnormal) signal” shown is used as the signal data to be detected. The two-dimensional image data may include a time-phase distribution diagram and a frequency-phase distribution diagram.

[0147] Step S450: extract the target signal features of the "pulsar (abnormal) signal" based on the signal data to be detected. The target signal features are as follows: Figure 5 The "Test Features" shown.

[0148] Among them, extracting the target signal features of the signal to be detected based on the signal data to be detected specifically includes: inputting the signal data to be detected into the trained feature extractor for feature extraction processing to obtain the target signal features (i.e., test features) corresponding to the signal to be detected, that is, through the trained feature extractor, performing feature extraction processing on the signal data to be detected of the signal to be detected to obtain the target signal features (i.e., test features).

[0149] Step S460 classifies the target signal features to obtain first type information corresponding to "pulsar (anomaly) signal." Specifically, the trained classifier is used to classify the target signal features (i.e., test features) to obtain the first type information. Step S460 specifically includes: inputting the target signal features into the classifier for classification processing to obtain a classification result corresponding to "pulsar (anomaly) signal," and generating the first type information based on the classification result.

[0150] The classification result is a probability value of the signal to be detected belonging to a predetermined type, and the first type information is a first score of whether the "pulsar (abnormal) signal" belongs to the target type; generating the first type information according to the classification result includes: generating the opposite number of the probability value; using the opposite number of the probability value as the first score (such as Figure 5 “Normal Category Scoring” shown).

[0151] In this scenario, the classifier is a fully connected classifier. After the classification result output by the fully connected classifier is activated by an activation function (such as a softmax function), the probability value of the "pulsar (abnormal) signal" belonging to the predetermined type (that is, the type corresponding to the known signal) can be obtained. Based on the probability value, the opposite number of the probability value is reversely generated as the first score of the "pulsar (abnormal) signal" belonging to the target type (that is, the type corresponding to the unknown signal).

[0152] Step S470: The target signal characteristics (such as Figure 5 The test features shown in FIG. 1 are compared with the sample signal features to obtain the second type of information corresponding to the "pulsar (anomaly) signal." Comparing the target signal features with the sample signal features to obtain the second type of information corresponding to the "pulsar (anomaly) signal" specifically includes: calculating the similarity between the target signal features and the sample signal features; and generating the second type of information corresponding to the "pulsar (anomaly) signal" based on the similarity.

[0153] The second type of information is the second score that the "pulsar (abnormal) signal" belongs to the target type (i.e., the type corresponding to the unknown signal); determining the second type of information corresponding to the "pulsar (abnormal) signal" based on the similarity, including: generating the opposite number of the similarity; and using the opposite number of the similarity as the second score.

[0154] When generating the inverse of the similarity, specifically, the “maximum similarity” among the similarities between the target signal feature and each sample signal feature is determined, and the inverse of the “maximum similarity” is generated as the second score.

[0155] Step S480: Determine the signal type of the "pulsar (abnormal) signal" based on the first type information and the second type information. Determine the signal type of the signal to be detected based on the first type information and the second type information, including: generating a type score (i.e., Figure 5 The “candidate abnormality score” is shown in FIG5 ); and whether the signal type of the “pulsar (abnormal) signal” is the target type is determined according to the type score.

[0156] In this scenario, by applying the embodiments of the present application, the signal type of the "pulsar (abnormal) signal" used for testing can be accurately detected as the target type, while the "noise (normal) signal" can be accurately detected as the predetermined type.

[0157] In this way, by applying the embodiments of the present application in this scenario, at least the following beneficial effects are achieved: when the "pulsar (abnormal) signal" used for testing is replaced with other signals to be detected (i.e., abnormal signals), the target signal features are extracted for the signal to be detected (i.e., abnormal signals), and the first type of information and the second type of information are obtained by classifying and comparing the target signal features, respectively. The signal type of the signal to be detected is determined by combining the first type of information and the second type of information, and a classification processing method and a comparison processing method of fusion features are implemented to avoid signal omissions caused by simple blind classification, and both known signals (i.e., noise (normal) signals) and unknown signals (i.e., abnormal signals) can be effectively detected, thereby effectively improving the reliability of signal detection.

[0158] Specifically, in this scenario, by applying the embodiments of the present application, a self-supervised anomaly detection method for detecting celestial signals is formed as a whole. Compared with the solutions based on supervised learning methods in related technologies, in this scenario, a large number of collected "noise signals" are used as training sets to train feature extractors and classifiers, and "abnormal signals" are identified using anomaly detection methods. Compared with traditional supervised learning signal search methods, both known and unknown signals can be effectively detected. In the actual celestial signal search process, more unknown signals that do not exist in the training set can be recalled.

[0159] To facilitate better implementation of the signal detection method provided in the embodiment of the present application, the embodiment of the present application also provides a signal detection device based on the above signal detection method. The meanings of the terms herein are the same as those in the above signal detection method, and the specific implementation details can be referred to the description in the method embodiment. Figure 6 A block diagram of a signal detection device according to an embodiment of the present application is shown. Figure 7 A block diagram of a signal detection device according to another embodiment of the present application is shown.

[0160] like Figure 6 As shown, the signal detection device 500 may include a data acquisition module 510 , a feature extraction module 520 , a classification module 530 , a comparison module 540 , and a determination module 550 .

[0161] The data acquisition module 510 can be used to obtain the signal data to be detected corresponding to the signal to be detected; the feature extraction module 520 can be used to extract the target signal features of the signal to be detected based on the signal data to be detected; the classification module 530 can be used to classify the target signal features to obtain the first type of information corresponding to the signal to be detected; the comparison module 540 can be used to compare the target signal features with the sample signal features to obtain the second type of information corresponding to the signal to be detected; the determination module 550 can be used to determine the signal type of the signal to be detected based on the first type of information and the second type of information.

[0162] In one embodiment of the present application, the device 500 also includes a sample signal feature extraction module, which includes: a first signal data acquisition unit, used to obtain first signal data corresponding to a first signal sample of a predetermined type, where the predetermined type is a type corresponding to a known signal; and a sample signal feature extraction unit, used to perform feature extraction processing based on the first signal data to obtain the sample signal feature.

[0163] In one embodiment of the present application, the signal type includes one of the predetermined type and the target type, and the target type is the type corresponding to the unknown signal; the sample signal feature extraction unit is used to: input the first signal data into the feature extractor for feature extraction processing to obtain the sample signal features extracted by the feature extractor, wherein the feature extractor is trained based on the second signal data and the first signal data, the second signal data is obtained by adjusting the first signal data, and the second signal data is used to simulate the second signal sample of the target type.

[0164] In one embodiment of the present application, the device 500 also includes a feature extractor training unit, which is used to: input the first signal data into a preset feature extractor to perform intra-class space modeling for the predetermined type to generate a first signal feature for each first signal sample; input the second signal data into a preset feature extractor to perform extra-class space modeling relative to the predetermined type to obtain a second signal feature for each second signal sample; optimize the parameters of the preset feature extractor according to the sample type information obtained by classifying the first signal feature and the second signal feature to obtain the trained feature extractor, wherein the optimization direction of the parameter optimization for the intra-class space modeling is consistent with the gravitational direction, and the gravitational direction includes the direction in which each first signal feature is acted upon by the gravitational force of other first signal features.

[0165] In one embodiment of the present application, the feature extraction module is used to: input the signal data to be detected into the feature extractor for feature extraction processing to obtain the target signal feature corresponding to the signal to be detected.

[0166] In one embodiment of the present application, the signal type includes one of a predetermined type and a target type, the predetermined type is a type corresponding to a known signal, and the target type is a type corresponding to an unknown signal; the classification module includes: a feature input unit, used to input the target signal feature into a classifier for classification processing to obtain a classification result corresponding to the signal to be detected, the classifier is trained based on the second signal data and the first signal data, the first signal data corresponds to a first signal sample of a predetermined type, the second signal data is obtained by adjusting the first signal data, and the second signal data is used to simulate a second signal sample of the target type; a result processing unit, used to generate the first type information based on the classification result.

[0167] In one embodiment of the present application, the device 500 also includes a classifier training unit, which is used to: input the first signal feature and the second signal feature into a preset classifier for classification processing to obtain sample type information, wherein the first signal feature is obtained by feature extraction of the first signal data and the second signal data; optimize the parameters of the preset classifier according to the sample type information to obtain the trained classifier.

[0168] In one embodiment of the present application, the classification result is a probability value of the signal to be detected belonging to the predetermined type, and the first type information is a first score that the signal to be detected belongs to the target type; the result processing unit is used to: generate the opposite number of the probability value; and use the opposite number of the probability value as the first score.

[0169] In one embodiment of the present application, the comparison module includes: a similarity calculation unit for calculating the similarity between the target signal feature and the sample signal feature; and an information generation unit for generating second type information corresponding to the signal to be detected based on the similarity.

[0170] In one embodiment of the present application, the sample signal feature corresponds to a first signal sample of a predetermined type, and the second type information is a second score of whether the signal to be detected belongs to the target type; the information generating unit is used to: generate the opposite number of the similarity; and use the opposite number of the similarity as the second score.

[0171] In one embodiment of the present application, the first type information is a first score that the signal to be detected belongs to the target type, and the second type information is a second score that the signal to be detected belongs to the target type; the determination module includes: a type score generation unit, used to generate a type score that the signal to be detected belongs to the target type based on the first score and the second score; a type determination unit, used to determine whether the signal type of the signal to be detected is the target type based on the type score.

[0172] According to one embodiment of the present application, a signal detection method includes: obtaining first signal data of a first signal sample of a predetermined type and second signal data of a second signal sample of a simulated target type, wherein the second signal data is obtained by performing data adjustment processing on the first signal data, and the predetermined type is a type corresponding to a known signal; training a feature extractor and a classifier based on the first signal data and the second signal data; performing feature extraction processing on the signal data to be detected of the signal to be detected by the trained feature extractor to obtain target signal features; classifying the target signal features by the trained classifier to obtain first type information; determining the signal type of the signal to be detected based on the first type information and the second type information, wherein the second type information is obtained by comparing the target signal features with the sample signal features.

[0173] like Figure 7 As shown, the signal detection device 600 may include a sample acquisition module 610 , a training module 620 , a first input module 630 , a second input module 640 , and a detection module 650 .

[0174] The sample acquisition module 610 can be used to obtain first signal data of a first signal sample of a predetermined type and second signal data of a second signal sample of a simulated target type, where the second signal data is obtained by performing data adjustment processing on the first signal data, and the predetermined type is the type corresponding to the known signal; the training module 620 can be used to train a feature extractor and a classifier based on the first signal data and the second signal data; the first input module 630 can be used to perform feature extraction processing on the signal data to be detected of the signal to be detected through the trained feature extractor to obtain target signal features; the second input module 640 can be used to classify the target signal features through the trained classifier to obtain first type information; the detection module 650 can be used to determine the signal type of the signal to be detected based on the first type information and the second type information, where the second type information is obtained by comparing the target signal features with the sample signal features.

[0175] In one embodiment of the present application, the apparatus 600 further includes a third input module configured to perform feature extraction processing on the first signal data of the first signal sample by using the trained feature extractor to obtain the sample signal feature.

[0176] In one embodiment of the present application, the training module includes: an intra-class space modeling unit, which is used to input the first signal data into a preset feature extractor to perform intra-class space modeling for the predetermined type, and generate a first signal feature for each first signal sample; an out-of-class space modeling unit, which is used to input the second signal data into a preset feature extractor to perform out-of-class space modeling relative to the predetermined type, and obtain a second signal feature for each second signal sample; a sample feature classification unit, which is used to input the first signal feature and the second signal feature into a preset classifier respectively for classification processing to obtain a sample classification result; a type information determination unit, which is used to determine the sample type information of the first signal sample and the second signal sample according to the sample classification result; a parameter optimization unit, which is used to optimize the parameters of the preset feature extractor according to the sample type information, and optimize the parameters of the preset classifier to obtain the trained feature extractor and the classifier.

[0177] In one embodiment of the present application, the parameter optimization unit is used to: perform parameter optimization for the intra-class space modeling according to the sample type information and the target optimization direction, wherein the target optimization direction is consistent with the gravitational direction, and the gravitational direction includes the direction in which each first signal feature is acted upon by the gravitational force of other first signal features.

[0178] In one embodiment of the present application, the sample acquisition module includes: a sample signal data acquisition unit, used to acquire sample signal data of a predetermined type of first signal sample; a folding unit, used to fold the sample signal data to obtain two-dimensional first signal data; an enhancement unit, used to perform data enhancement processing on the first signal data through vertical occlusion to generate the second signal data, and the second signal data is used to simulate the second signal sample of the target type.

[0179] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiment of the application, the features and functions of two or more modules or units described above can be concretized in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be concretized.

[0180] In addition, an embodiment of the present application further provides an electronic device, which may be a terminal or a server, such as Figure 8 , which shows a schematic diagram of the structure of the electronic device involved in the embodiment of the present application, specifically:

[0181] The electronic device may include one or more processing core processors 701, one or more computer-readable storage media memories 702, a power supply 703, an input unit 704 and other components. Those skilled in the art will understand that Figure 8 The electronic device structure shown in the figure does not constitute a limitation of the electronic device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange components differently.

[0182] Processor 701 is the control center of the electronic device, connecting the various components of the entire computer device using various interfaces and lines. By running or executing software programs and / or modules stored in memory 702 and accessing data stored in memory 702, it performs various functions of the computer device and processes data, thereby detecting the electronic device. Optionally, processor 701 may include one or more processing cores; preferably, processor 701 may integrate an application processor and a modem processor, wherein the application processor primarily processes the operating system, user interfaces, and application programs, while the modem processor primarily handles wireless communications. It is understood that the modem processor may not be integrated into processor 701.

[0183] The memory 702 can be used to store software programs and modules. The processor 701 executes various functional applications and data processing by running the software programs and modules stored in the memory 702. The memory 702 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 702 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage device. Accordingly, the memory 702 may also include a memory controller to provide the processor 701 with access to the memory 702.

[0184] The electronic device also includes a power supply 703 for supplying power to various components. Preferably, the power supply 703 can be logically connected to the processor 701 via a power management system, thereby enabling the power management system to manage charging, discharging, and power consumption. The power supply 703 can also include one or more DC or AC power supplies, a recharging system, a power failure detection circuit, a power converter or inverter, a power status indicator, and other arbitrary components.

[0185] The electronic device may further include an input unit 704, which may be configured to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0186] Although not shown, the electronic device may further include a display unit, etc., which will not be described in detail herein. Specifically, in this embodiment, the processor 701 in the electronic device loads the executable files corresponding to one or more computer program processes into the memory 702 according to the following instructions, and the processor 701 runs the computer program stored in the memory 702, thereby implementing the various functions of the aforementioned embodiments of the present application.

[0187] In one embodiment, the processor 701 may execute: obtaining signal data to be detected corresponding to the signal to be detected; extracting target signal features of the signal to be detected based on the signal data to be detected; classifying the target signal features to obtain first type information corresponding to the signal to be detected; comparing the target signal features with sample signal features to obtain second type information corresponding to the signal to be detected; and determining the signal type of the signal to be detected based on the first type information and the second type information.

[0188] For example, in one embodiment, the processor 701 can execute: obtaining first signal data of a first signal sample of a predetermined type and second signal data of a second signal sample of a simulated target type, where the second signal data is obtained by performing data adjustment processing on the first signal data, and the predetermined type is the type corresponding to a known signal; training a feature extractor and a classifier based on the first signal data and the second signal data; performing feature extraction processing on the signal data to be detected of the signal to be detected by the trained feature extractor to obtain target signal features; classifying the target signal features by the trained classifier to obtain first type information; determining the signal type of the signal to be detected based on the first type information and the second type information, where the second type information is obtained by comparing the target signal features with the sample signal features.

[0189] Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be accomplished by a computer program, or by controlling related hardware through a computer program. The computer program may be stored in a computer-readable storage medium and loaded and executed by a processor.

[0190] To this end, an embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored. The computer program can be loaded by a processor to execute the steps of any method provided in the embodiment of the present application.

[0191] The computer-readable storage medium may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0192] Since the computer program stored in the computer-readable storage medium can execute the steps of any method provided in the embodiments of the present application, the beneficial effects that can be achieved by the method provided in the embodiments of the present application can be achieved. Please refer to the previous embodiments for details and will not be repeated here.

[0193] According to one aspect of the present application, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various optional implementations of the above-described embodiments of the present application.

[0194] Those skilled in the art will readily conceive of other embodiments of the present application after considering the specification and practicing the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present application that follow the general principles of this application and include common knowledge or customary techniques in the art that are not disclosed herein.

[0195] It should be understood that the present application is not limited to the embodiments that have been described above and shown in the accompanying drawings, but various modifications and changes may be made without departing from the scope thereof.

Claims

1. A signal detection method, characterized in that: include: Obtaining signal data to be detected corresponding to the signal to be detected; extracting target signal features of the signal to be detected based on the signal to be detected data; Classify the target signal features to obtain first type information corresponding to the signal to be detected; Comparing the target signal feature with the sample signal feature to obtain second type information corresponding to the signal to be detected; determining a signal type of the signal to be detected based on the first type information and the second type information; The signal type includes one of a predetermined type and a target type, wherein the target type is a type corresponding to the unknown signal; Before comparing the target signal feature with the sample signal feature, the method further includes: Acquire first signal data corresponding to a first signal sample of a predetermined type, where the predetermined type is a type corresponding to a known signal; The first signal data is input into a feature extractor for feature extraction processing to obtain the sample signal features extracted by the feature extractor, wherein the feature extractor is trained based on the second signal data and the first signal data, the second signal data is obtained by data adjustment of the first signal data, and the second signal data is used to simulate a second signal sample of the target type.

2. The method according to claim 1, characterized in that The feature extractor is trained according to the following steps: Inputting the first signal data into a preset feature extractor to perform intra-class space modeling for the predetermined type, and generating a first signal feature for each first signal sample; Inputting the second signal data into a preset feature extractor to perform modeling relative to the predetermined type of out-of-class space to obtain a second signal feature of each second signal sample; According to the sample type information obtained by classifying the first signal feature and the second signal feature, the parameters of the preset feature extractor are optimized to obtain the trained feature extractor, wherein: The optimization direction of the parameters optimized for the intra-class space modeling is consistent with the gravitational direction, and the gravitational direction includes the direction in which each first signal feature is acted upon by the gravitational force of other first signal features.

3. The method according to claim 2, characterized in that The extracting the target signal feature of the signal to be detected based on the signal to be detected data includes: The signal data to be detected is input into the feature extractor for feature extraction processing to obtain the target signal feature corresponding to the signal to be detected.

4. The method according to any one of claim 2, characterized in that The signal type includes one of a predetermined type and a target type, the predetermined type is a type corresponding to a known signal, and the target type is a type corresponding to an unknown signal; The classifying and processing the target signal feature to obtain the first type of information corresponding to the signal to be detected includes: Inputting the target signal feature into a classifier for classification processing to obtain a classification result corresponding to the signal to be detected, wherein the classifier is trained based on the second signal data and the first signal data, the first signal data corresponds to a first signal sample of a predetermined type, and the second signal data is obtained by adjusting the first signal data, and the second signal data is used to simulate a second signal sample of the target type; The first type of information is generated according to the classification result.

5. The method according to claim 4, characterized in that The classifier is trained according to the following steps: Inputting the first signal feature and the second signal feature into a preset classifier for classification processing to obtain sample type information, wherein the first signal feature is obtained by feature extraction of the first signal data and the second signal data; Parameters of the preset classifier are optimized according to the sample type information to obtain the trained classifier.

6. The method according to claim 4, characterized in that The classification result is a probability value of the signal to be detected belonging to the predetermined type, and the first type information is a first score of the signal to be detected belonging to the target type; Generating the first type of information according to the classification result includes: generating the opposite of the probability value; The inverse of the probability value is used as the first score.

7. The method according to any one of claim 2, characterized in that The comparing and processing the target signal feature with the sample signal feature to obtain the second type of information corresponding to the signal to be detected includes: Calculating the similarity between the target signal feature and the sample signal feature; Second type information corresponding to the signal to be detected is generated according to the similarity.

8. The method according to claim 7, characterized in that The sample signal feature corresponds to a first signal sample of a predetermined type, and the second type information is a second score of the signal to be detected belonging to the target type; The determining, according to the similarity, the second type of information corresponding to the signal to be detected includes: generating the opposite of the similarity; The inverse of the similarity is used as the second score.

9. The method according to any one of claim 2, characterized in that The first type information is a first score of whether the signal to be detected belongs to the target type, and the second type information is a second score of whether the signal to be detected belongs to the target type; The determining the signal type of the signal to be detected based on the first type information and the second type information includes: generating a type score for the signal to be detected belonging to a target type according to the first score and the second score; It is determined whether the signal type of the signal to be detected is the target type according to the type score.

10. A signal detection method, characterized in that: include: Acquire first signal data of a first signal sample of a predetermined type and second signal data of a second signal sample of a simulated target type, wherein the second signal data is obtained by performing data adjustment processing on the first signal data, and the predetermined type is a type corresponding to a known signal; training a feature extractor and a classifier based on the first signal data and the second signal data; Performing feature extraction processing on the signal data to be detected by the trained feature extractor to obtain target signal features; Classifying the target signal features by the trained classifier to obtain first type information; determining a signal type of the signal to be detected based on the first type information and the second type information, wherein the second type information is obtained by comparing a characteristic of the target signal with a characteristic of a sample signal; The signal type includes one of the predetermined type and the target type, and the target type is a type corresponding to the unknown signal; Before comparing the target signal feature with the sample signal feature, the method further includes: Acquire first signal data corresponding to a first signal sample of a predetermined type, where the predetermined type is a type corresponding to a known signal; The first signal data is input into a feature extractor for feature extraction processing to obtain the sample signal features extracted by the feature extractor, wherein the feature extractor is trained based on the second signal data and the first signal data, the second signal data is obtained by data adjustment of the first signal data, and the second signal data is used to simulate a second signal sample of the target type.

11. The method according to claim 10, characterized in that The training of a feature extractor and a classifier based on the first signal data and the second signal data includes: Inputting the first signal data into a preset feature extractor to perform intra-class space modeling for the predetermined type, and generating a first signal feature for each first signal sample; Inputting the second signal data into a preset feature extractor to perform modeling relative to the predetermined type of out-of-class space to obtain a second signal feature of each second signal sample; Inputting the first signal feature and the second signal feature into a preset classifier for classification processing to obtain a sample classification result; Determining sample type information of the first signal sample and the second signal sample according to the sample classification result; Parameters of the preset feature extractor are optimized according to the sample type information, and parameters of the preset classifier are optimized to obtain the trained feature extractor and classifier.

12. The method according to claim 11, characterized in that Optimizing parameters of the preset feature extractor according to the sample type information includes: According to the sample type information, parameter optimization is performed for the intra-class space modeling according to a target optimization direction, wherein the target optimization direction is consistent with a gravitational direction, and the gravitational direction includes a direction in which each first signal feature is acted upon by gravitational forces of other first signal features.

13. The method according to claim 11, characterized in that The acquiring first signal data of a first signal sample of a predetermined type and second signal data of a second signal sample of a simulated target type comprises: acquiring sample signal data of a first signal sample of a predetermined type; Folding the sample signal data to obtain two-dimensional first signal data; The first signal data is subjected to data enhancement processing through vertical occlusion to generate the second signal data, where the second signal data is used to simulate a second signal sample of the target type.

14. A signal detection device, characterized in that: include: A data acquisition module, used to acquire the signal data to be detected corresponding to the signal to be detected; a feature extraction module, configured to extract target signal features of the signal to be detected based on the signal data to be detected; the signal type includes one of a predetermined type and a target type, the target type being a type corresponding to an unknown signal; Before comparing the target signal feature with the sample signal feature, the method further includes: obtaining first signal data corresponding to a first signal sample of a predetermined type, where the predetermined type is a type corresponding to a known signal; inputting the first signal data into a feature extractor for feature extraction processing to obtain the sample signal feature extracted by the feature extractor, wherein the feature extractor is trained based on second signal data and the first signal data, the second signal data is obtained by data adjustment of the first signal data, and the second signal data is used to simulate a second signal sample of the target type; A classification module, configured to classify the target signal features to obtain first type information corresponding to the signal to be detected; a comparison module, configured to compare the target signal feature with the sample signal feature to obtain second type information corresponding to the signal to be detected; A determination module is used to determine the signal type of the signal to be detected based on the first type information and the second type information.

15. A computer-readable storage medium, characterized in that A computer program is stored thereon, and when the computer program is executed by a processor of a computer, the computer is caused to execute the method according to any one of claims 1 to 13.

16. An electronic device, characterized in that: include: a memory storing a computer program; A processor reads a computer program stored in a memory to execute the method according to any one of claims 1 to 13.

17. A computer program product, characterized in that The computer program product comprises a computer program, which implements the method of any one of claims 1 to 13 when executed by a processor.

Images