Pulsar search method, device and equipment based on artificial intelligence
By extracting and fusing signal data of multiple signal feature types of pulsar candidates, astral classification feature data are generated, which solves the problem of low accuracy in the recognition of pulsar candidates in the prior art and improves the accuracy of pulsar search.
Patent Information
- Application Number
- CN202210190590.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-02-28
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2042-02-28
AI Technical Summary
In the prior art, the recognition accuracy of pulsar candidates is low, making it difficult to effectively search for pulsar signal samples from a large number of pulsar candidates.
By obtaining signal data of multiple signal characteristic types of pulsar candidates, the target signal characteristic data is extracted, and feature fusion is performed to generate celestial classification characteristic data to identify the celestial type of pulsar candidates. The specific method includes inputting signal feature data into the feature fusion network, obtaining self-attention weights for weighted fusion, and generating astral classification feature data.
The accuracy of pulsar search is improved, and the classification accuracy of pulsar candidates is enhanced by characterizing the correlation between signal characteristics of different signal characteristics and signal data.
Smart Images

Figure CN116720151B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of artificial intelligence technology, and specifically to an artificial intelligence-based pulsar search method, apparatus, and device; wherein the device includes a computer device, a computer-readable storage medium, and a computer program product. Background Art
[0002] As the performance of modern pulsar search equipment continues to improve, pulsar search equipment can detect weaker signals, making it possible to detect more pulsars. Pulsar search equipment receives weak signals from hundreds of millions or even billions of light years away, and through signal preprocessing, it will produce tens of millions of pulsar candidates, but most pulsar candidates are often noise signals such as radio frequency interference. How to search for pulsar signal samples from a large number of pulsar candidates has become an urgent problem to be solved.
[0003] In the prior art, the signal data corresponding to the pulsar candidate is usually processed into an image, and a convolutional neural network model is used to classify the pulsar candidate based on the obtained image. However, the recognition accuracy of pulsar candidates identified by this method is low. Summary of the invention
[0004] Based on this, it is necessary to provide an artificial intelligence-based pulsar search method, device, computer equipment, computer-readable storage medium and computer program product to improve the accuracy of pulsar search in response to the above-mentioned technical problems.
[0005] In a first aspect, the present application provides a pulsar search method based on artificial intelligence, the method comprising:
[0006] Acquiring signal data of at least two signal characteristic types of the pulsar candidate;
[0007] Extract target signal feature data of the signal data according to the feature extraction method corresponding to each signal feature type and the signal data; wherein one target signal feature data is extracted based on at least one signal data;
[0008] Perform feature fusion on the target signal feature data to obtain the star classification feature data;
[0009] Identify the stellar type of pulsar candidates based on stellar classification characteristic data.
[0010] In some embodiments of the present application, the number of dimensions of each target signal feature data is the same, and the data length is equal under the same target dimension;
[0011] The target signal feature data is fused to obtain the star classification feature data, including:
[0012] The target signal feature data is stacked based on the target dimension to obtain a signal feature matrix;
[0013] The matrix elements in the signal feature matrix are fused to obtain the star classification feature data.
[0014] In some embodiments of the present application, the matrix elements in the signal feature matrix are fused to obtain the star classification feature data, including:
[0015] Input the signal feature matrix into the feature fusion network;
[0016] Obtain the self-attention weight of each row of signal feature data in the signal feature matrix through the feature fusion network;
[0017] The signal feature data of each row is weightedly fused according to the self-attention weight of each row of signal feature data to generate star classification feature data.
[0018] In some embodiments of the present application, the signal data under at least one signal feature type is a one-dimensional signal sequence, and the signal data under at least two signal feature types are two-dimensional signal images;
[0019] According to the feature extraction method corresponding to each signal feature type and the signal data, target signal feature data of the signal data is extracted, including:
[0020] Extracting target signal feature data of the two-dimensional signal image in the signal data by a feature extraction method corresponding to the two-dimensional signal image;
[0021] Extracting target signal feature data of the one-dimensional signal sequence in the signal data by a feature extraction method corresponding to the one-dimensional signal sequence;
[0022] The two-dimensional signal images are stacked to obtain a stacked signal image, and target signal feature data of the stacked signal image is extracted by a feature extraction method corresponding to the stacked signal image.
[0023] In some embodiments of the present application, the one-dimensional signal sequence includes at least dispersion curve data, and the two-dimensional signal image includes at least a frequency phase distribution image and a time phase distribution image.
[0024] In some embodiments of the present application, extracting target signal feature data of the signal data according to the feature extraction method corresponding to each signal feature type and the signal data includes: extracting target signal feature data of the signal data according to the feature extraction method corresponding to each signal feature type and the signal data through a pulsar search model;
[0025] Performing feature fusion on the target signal feature data to obtain star classification feature data, including: performing feature fusion on the target signal feature data through a pulsar search model to obtain star classification feature data;
[0026] Identifying the star type of the pulsar candidate according to the star classification characteristic data, including: identifying the star type of the pulsar candidate according to the star classification characteristic data using a pulsar search model;
[0027] Prior to obtaining signal data of at least two signal characteristic types of a pulsar candidate, including:
[0028] Obtain signal data samples of at least two signal feature types of the sample star and a star category label;
[0029] Extracting target signal feature samples of the signal data samples according to the feature extraction method corresponding to each signal feature type and the signal data samples through the pulsar search model to be trained;
[0030] The target signal feature samples are fused by the pulsar search model to obtain the star classification feature samples;
[0031] Using the pulsar search model, the predicted star type of the sample star is identified according to the star classification feature sample;
[0032] According to the predicted star type and star category label, the loss value of the pulsar search model is calculated, and the model parameters of the pulsar search model are adjusted based on the loss value until the training end conditions are met.
[0033] In some embodiments of the present application, the pulsar search model includes a feature fusion network; the number of dimensions of each target signal feature sample is the same, and the data length is equal under the same target dimension;
[0034] The target signal feature samples are fused through the pulsar search model to obtain the star classification feature samples, including:
[0035] Stack the target signal feature samples based on the target dimension to obtain a sample signal feature matrix;
[0036] Obtain the self-attention weight of each row of signal feature samples in the sample signal feature matrix through the feature fusion network;
[0037] The signal feature samples in each row are weightedly fused according to their self-attention weights to generate star classification feature samples.
[0038] In a second aspect, the present application provides a pulsar search device based on artificial intelligence, the device comprising:
[0039] A signal data acquisition module, used to acquire signal data of at least two signal characteristic types of a pulsar candidate;
[0040] A feature data acquisition module, used to extract target signal feature data of the signal data according to the feature extraction method corresponding to each signal feature type and the signal data; wherein one target signal feature data is extracted based on at least one signal data;
[0041] A feature data fusion module is used to perform feature fusion on the target signal feature data to obtain the star classification feature data;
[0042] The star type classification module is used to identify the star type of the pulsar candidate based on the star classification feature data.
[0043] In a third aspect, the present application further provides a server, the server comprising:
[0044] one or more processors;
[0045] Memory; and
[0046] One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement an artificial intelligence-based pulsar search method.
[0047] In a fourth aspect, the present application also provides a computer-readable storage medium having a computer program stored thereon, and the computer program is loaded by a processor to execute the steps in the artificial intelligence-based pulsar search method.
[0048] In a fifth aspect, an embodiment of the present application provides a computer program product, which includes computer instructions. When the computer program or instructions are executed by a processor, the steps in the pulsar search method based on artificial intelligence are implemented.
[0049] The above-mentioned artificial intelligence-based pulsar search method, device, computer equipment, computer-readable storage medium and computer program product obtain signal data of at least two signal feature types of pulsar candidates; extract target signal feature data of the signal data according to the feature extraction method and signal data corresponding to each signal feature type; wherein one target signal feature data is extracted based on at least one signal data; feature fusion is performed on the target signal feature data to obtain star classification feature data; and the star type of the pulsar candidate is identified according to the star classification feature data. By extracting target signal feature data from the signal data corresponding to multiple signal feature types of the pulsar candidate and feature fusion of the target signal feature data, the obtained star classification feature data can characterize both the signal features of different signal feature types and the correlation between the signal data of different signal feature types. In the subsequent classification of pulsar candidates based on the star classification feature data, the classification accuracy is effectively improved, and the accuracy of the pulsar star search during the pulsar search process is improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. 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.
[0051] Figure 1 is a schematic diagram of a scenario of a pulsar search method based on artificial intelligence in an embodiment of the present application;
[0052] Figure 2 It is a flowchart of a pulsar search method based on artificial intelligence in an embodiment of the present application;
[0053] Figure 3 is a flow chart of the target signal feature data extraction step in an embodiment of the present application;
[0054] Figure 4 is a schematic diagram of a target signal feature data extraction step in an embodiment of the present application;
[0055] Figure 5 It is a flowchart of the steps of obtaining star classification characteristic data in an embodiment of the present application;
[0056] Figure 6 is a flow chart of another step of acquiring star classification characteristic data in an embodiment of the present application;
[0057] Figure 7 is a flowchart of another pulsar search method based on artificial intelligence in an embodiment of the present application;
[0058] Figure 8 is a schematic diagram of a pulsar search method based on artificial intelligence in an embodiment of the present application;
[0059] Fig. 9 It is a flowchart of the pulsar search model training steps in the embodiment of the present application;
[0060] Fig.10 It is a structural schematic diagram of a pulsar search device based on artificial intelligence in an embodiment of the present application;
[0061] Fig.11 It is a schematic diagram of the structure of a computer device in an embodiment of the present application. DETAILED DESCRIPTION
[0062] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.
[0063] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.
[0064] In the description of the present application, the word "for example" is used to mean "used as an example, illustration or explanation". Any embodiment described as "for example" in the present application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any technician in the field to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes will not be elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed in the present application.
[0065] In the embodiments of the present application, artificial intelligence is the theory, method, technology and application system of simulating, extending and expanding human intelligence, perceiving the environment, acquiring knowledge and using knowledge to obtain the best results using digital computers or machines controlled by digital computers. In other words, artificial intelligence is a comprehensive technology of computer science, which attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that the machines have the functions of perception, reasoning and decision-making. Artificial intelligence technology is a comprehensive discipline involving a wide range of fields, including both hardware-level technology and software-level technology. Basic artificial intelligence technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. Artificial intelligence software technology mainly includes several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.
[0066] Machine Learning (ML) is a multi-disciplinary subject that involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.
[0067] With the research and progress of artificial intelligence technology, artificial intelligence technology has been studied and applied in many fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, automatic driving, drones, robots, smart medical care, smart customer service, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role. The solution provided in the embodiment of this application involves artificial intelligence technology, which is specifically described by the following embodiments.
[0068] In the embodiments of the present application, it should be noted that the pulsar search method based on artificial intelligence provided in the embodiments of the present application can be applied to Figure 1In the pulsar search system based on artificial intelligence shown in the figure. Among them, the pulsar search system based on artificial intelligence includes a terminal 100 and a server 200, and the terminal 100 can specifically be a pulsar search device, such as a large radio telescope. The server 200 can be an independent server, or a server network or server cluster composed of servers, which includes but is not limited to a computer, a network host, a single network server, a plurality of network server sets or a cloud server composed of multiple servers. Among them, the cloud server is composed of a large number of computers or network servers based on cloud computing (Cloud Computing). Taking the terminal 100 as a 500-meter aperture spherical radio telescope (Five-hundred-meter Aperture Spherical Radio Telescope, FAST) as an example, FAST conducts sky surveys in a drift scanning manner, receives weak signals from hundreds of millions or even billions of light years away, and sends the received weak signals to the server 200; the server 200 processes the weak signal sent by the terminal 100, generates a pulsar candidate, and searches for pulsars from the pulsar candidate.
[0069] Those skilled in the art will understand that Figure 1 The application environment shown in the figure is only one application scenario of the present application solution and does not constitute a limitation on the application scenario of the present application solution. Other application environments may also include Figure 1 More or less computer equipment as shown in Figure 1 Only one server 200 is shown in the figure. It is understandable that the artificial intelligence-based pulsar search system may also include one or more other servers, which are not specifically limited here. Figure 1 As shown, the artificial intelligence-based pulsar search system may also include a memory for storing data, such as storing pulsar-related signal data.
[0070] It should also be noted that Figure 1 The scenario diagram of the artificial intelligence-based pulsar search system shown is merely an example. The artificial intelligence-based pulsar search system and scenario described in the embodiment of the present invention are intended to more clearly illustrate the technical solution of the embodiment of the present invention, and do not constitute a limitation on the technical solution provided by the embodiment of the present invention. A person of ordinary skill in the art can appreciate that with the evolution of the artificial intelligence-based pulsar search system and the emergence of new business scenarios, the technical solution provided by the embodiment of the present invention is equally applicable to similar technical problems.
[0071] See also Figure 2 The present application embodiment provides a pulsar search method based on artificial intelligence, which is mainly applied to the above Figure 1Taking the server 200 in FIG. 1 as an example, the method includes steps S210 to S240, which are specifically as follows:
[0072] S210, obtaining signal data of at least two signal feature types of a pulsar candidate.
[0073] Among them, a pulsar is a neutron star with strong gravitational force, strong magnetic field and rapid rotation; a pulsar candidate refers to an observation signal suspected to be a pulsar, that is, a pulsar candidate may be an observation signal of a noise signal such as radio frequency interference, or it may be an observation signal of a pulsar signal.
[0074] The signal data of different signal feature types refer to information data of pulsar candidates under different attributes, such as the signal strength of pulsar candidates under different signal frequencies, the signal-to-noise ratio of pulsar candidates, etc. In one embodiment, the signal data of at least two signal feature types include but are not limited to pulse profile curves, time phase distribution images, frequency phase distribution images, dispersion curves, etc.
[0075] Specifically, pulsar signals are very weak and are affected by the interstellar medium during propagation. Therefore, after receiving the original observation signal, the pulsar search device can perform data preprocessing on the received original observation signal to obtain signal data of the original observation signal under different signal feature types as data for subsequent identification of the category of pulsar candidates. More specifically, data preprocessing of the original observation signal received by the pulsar search device can include data processing methods such as shear wave processing, de-dispersion processing, Fourier transform, and signal superposition.
[0076] S220, extracting target signal feature data of the signal data according to the feature extraction method corresponding to each signal feature type and the signal data; wherein one target signal feature data is extracted based on at least one signal data.
[0077] Among them, the target signal characteristic data refers to the information data extracted from the signal data and used to characterize the relevant characteristics of the pulsar candidate.
[0078] Specifically, extracting target signal feature data of signal data may be performed by separately extracting signal data of each signal feature type to obtain target signal feature data corresponding thereto; or signal data belonging to different signal feature types may be superimposed and target signal feature data may be extracted from the superimposed signal data; or target signal feature data may be extracted separately from signal data of each signal feature type and from the superimposed signal data, and the target signal feature data may be extracted from the superimposed signal data, which is not limited here.
[0079] For example, for a pulsar candidate, obtain the signal data corresponding to its signal feature type A and the signal data corresponding to its signal feature type B; first extract the corresponding target signal feature data from the signal data of signal feature type A alone, and extract the corresponding target signal feature data from the signal data of signal feature type B alone; then, superimpose the signal data of signal feature type A and the signal data of signal feature type B, and extract the corresponding target signal data from the superimposed signal data.
[0080] Furthermore, the data dimensions and the signal quantities carried by signal data of different signal feature types are different. Therefore, for signal data of different signal feature types, feature extraction can be performed through different feature extraction methods.
[0081] More specifically, for signal data of different signal feature types, feature extraction can be performed through different feature extraction networks; after obtaining signal data of different signal feature types, the signal data of different signal feature types can be input into the feature extraction network corresponding to the signal feature type for feature extraction. Among them, the feature extraction network is used to map the signal data input into the feature extraction network to a low-dimensional data space, while mining the implicit information in the signal data, so as to use less data to characterize the data information of the signal feature type. It can be understood that the feature extraction network here is a trained neural network model, which can be specifically constructed by a convolutional neural network (CNN), such as VGG, ResNet-50, DenseNet and other neural networks.
[0082] In one embodiment, the amount of information carried by signal data of different signal feature types is different. The data dimension of the target signal feature data output by each feature extraction method (or feature extraction network) can be set to improve the target signal feature data's ability to express feature information.
[0083] In one embodiment, among the signal data of at least two signal feature types, the signal data of at least one signal feature type is a one-dimensional signal sequence, and the signal data of at least two signal feature types are two-dimensional signal images, such as Figure 3 As shown, step S220 may specifically include:
[0084] S310, extracting target signal feature data of the two-dimensional signal image in the signal data by a feature extraction method corresponding to the two-dimensional signal image;
[0085] S320, extracting target signal feature data of the one-dimensional signal sequence in the signal data by using a feature extraction method corresponding to the one-dimensional signal sequence;
[0086] S330, stacking the two-dimensional signal images to obtain a stacked signal image, and extracting target signal feature data of the stacked signal image by a feature extraction method corresponding to the stacked signal image.
[0087] Among them, one-dimensional signal sequence refers to signal data in the shape of 1*N, such as the dispersion curve and pulse profile curve of a pulsar candidate; two-dimensional signal image refers to signal data in the shape of N*N, such as the time phase distribution image and frequency phase distribution image of a pulsar candidate.
[0088] Among them, the two-dimensional signal images are stacked, specifically, the pixel values of each two-dimensional signal image at the same position (pixel point) are added to achieve image superposition to obtain a stacked signal image; or each two-dimensional signal image is superimposed as image data of an image channel to obtain a stacked signal image.
[0089] After the above-mentioned pulsar search device receives the original observation signal, it often performs data preprocessing on the received original observation signal. After the data preprocessing, the signal data obtained includes but is not limited to a one-dimensional signal sequence and a two-dimensional signal image; specifically, each one-dimensional signal sequence can be individually input into a feature extraction network corresponding to the one-dimensional signal sequence to obtain target signal feature data corresponding to the one-dimensional signal sequence; and each two-dimensional signal image can be individually input into a feature extraction network corresponding to the two-dimensional signal image to obtain target signal feature data corresponding to the two-dimensional signal image; then, for two-dimensional signal images with the same data format, the two-dimensional signal images can be stacked to obtain a stacked signal image, and the stacked signal image can be input into a feature extraction network corresponding to the stacked signal image to extract the target signal data corresponding to the stacked signal image.
[0090] For example, the signal data of at least two signal feature types includes a one-dimensional signal sequence and two two-dimensional signal images, such as Figure 4 As shown, the one-dimensional signal sequence is a one-dimensional sequence with a length of 100, and the two two-dimensional signal images are both 64*64 two-dimensional images; for the one-dimensional signal sequence, the one-dimensional signal sequence is input into the feature extraction network 1, and the corresponding target signal feature data is obtained through the feature extraction network 1; for the two-dimensional signal image A and the two-dimensional signal image B, the two-dimensional signal image A and the two-dimensional signal image B can be respectively input into the feature extraction network 2 and the feature extraction network 4 to extract the corresponding target signal feature data; at the same time, the two-dimensional signal image A and the two-dimensional signal image B are stacked to obtain a stacked signal image with a channel number of 2 and a data dimension of 64*64, and then the stacked signal image with a channel number of 2 and a data dimension of 64*64 is input into the feature extraction network 3 to extract the corresponding target signal feature data.
[0091] S230, performing feature fusion on the target signal feature data to obtain star classification feature data.
[0092] Feature fusion refers to integrating the information contained in each target information feature data to obtain star classification feature data that can more accurately express the relevant features of the pulsar candidate. Feature fusion of target signal feature data can be performed by vector concatenation of each target signal feature data to obtain star classification feature data; or multiple target signal feature data can be mapped through a fully connected layer to obtain star classification feature data.
[0093] Furthermore, the feature dimensions of the target signal feature data output by each feature extraction method (or feature extraction network) can be set to have the same number of dimensions and the same data length under the same target dimension; therefore, Figure 5 As shown, in one embodiment, step S230 may further include:
[0094] S510, stacking target signal feature data based on the target dimension to obtain a signal feature matrix.
[0095] S520, fusing the matrix elements in the signal feature matrix to obtain star classification feature data.
[0096] The matrix element refers to the data constituting the signal feature matrix, and specifically may be a row vector constituting the signal feature matrix, or a column vector constituting the signal feature matrix.
[0097] After obtaining multiple target signal feature data with the same number of dimensions and equal data length under the same target dimension, the target signal feature data can be first stacked based on the target dimension with equal data length to obtain a signal feature matrix; then, the matrix elements to be fused are determined from the signal feature matrix, and then the matrix elements are weightedly fused to obtain the star classification feature data.
[0098] Specifically, after obtaining the signal feature matrix, the signal feature matrix can be input into the feature fusion network, and the matrix elements are weighted and fused through the feature fusion network to obtain the star classification feature data. The feature fusion network here is a trained neural network model, which can be specifically constructed by a convolutional neural network, such as a fully connected layer, a convolutional layer, and other neural networks. Through the feature fusion network, the correlation relationship between the target signal feature data corresponding to different signal feature types can be learned from the signal feature matrix, and more feature information that can characterize the star type of the pulsar candidate can be obtained, so as to improve the accuracy of the subsequent classification of the pulsar candidate.
[0099] Furthermore, since the target signal feature data corresponding to different signal feature types have different abilities to express the information related to the star type of the pulsar candidate. Figure 6 As shown, in one embodiment, the matrix elements in the signal feature matrix are fused to obtain the star classification feature data, including:
[0100] S610, inputting the signal feature matrix into a feature fusion network;
[0101] S620, obtaining the self-attention weight of each row of signal feature data in the signal feature matrix through a feature fusion network;
[0102] S630, performing weighted fusion on each row of signal feature data according to the self-attention weight of each row of signal feature data to generate star classification feature data.
[0103] The row signal feature data refers to the row vector of the signal feature matrix. It can be understood that each row signal feature data is associated with a target feature data corresponding to a signal feature type. For example, Figure 4 As shown, for a one-dimensional signal sequence, 1*16 dimensional target signal feature data can be extracted, for two two-dimensional signal images, 2*16 dimensional target signal feature data can be extracted separately, and for a stacked signal image obtained by stacking two two-dimensional signal images, 3*16 dimensional target signal feature data can be extracted; after obtaining each target signal feature data, data stacking can be performed based on dimensions with equal data length to obtain an 8*16 signal feature matrix, in which the 1*16 dimensional matrix elements in each row are associated with target feature data corresponding to a signal feature type.
[0104] Among them, the self-attention weight refers to the probability distribution of attention allocation to each row of signal feature data. Since signal data of different signal feature types all come from pulsar candidates, there is a certain correlation between signal data of any signal feature type, and the fusion of signal data of different signal feature types has different effects on the classification of pulsar candidates. When the combination of signal data of two or more signal feature types has a greater influence on the classification of pulsar candidates, the self-attention weight of the corresponding row of signal feature data is greater. When the signal data of two or more signal feature types has a smaller influence on the classification of pulsar candidates, the self-attention weight of the corresponding row of signal feature data is smaller.
[0105] Specifically, the self-attention weight of each row of signal feature data is obtained through the feature fusion network. When weighted fusion is performed on each row of signal feature data, the correlation between different parts of the entire input is noticed through the self-attention weight mechanism, so as to increase the row signal feature data related to the classification of pulsar candidates in the signal feature matrix, and suppress the row signal feature data in the signal feature matrix that are irrelevant to or have little correlation with the classification of pulsar candidates. This makes the acquired star classification feature data characterize both the signal features of different signal feature types and the correlation between signal data of different signal feature types. Pulsar candidates are classified based on the star classification feature data, thereby effectively improving the accuracy of pulsar candidate classification.
[0106] Still see Figure 4 After obtaining the characteristic data of each target signal, the data can be stacked based on the dimension with equal data length to obtain an 8*16 signal feature matrix; for the signal feature matrix, the self-attention weight of each row of signal feature data in the signal feature matrix is first obtained through the feature fusion network, and then the signal feature numbers of each row are weightedly fused according to the self-attention weight of the signal feature numbers of each row to generate 1*16 dimensional star classification feature data.
[0107] S240, identifying the star type of the pulsar candidate according to the star classification feature data.
[0108] After obtaining the star classification feature data, the pulsar candidate can be classified according to the star classification feature data to obtain the star type of the pulsar candidate. The star type can include pulsar category and non-pulsar category.
[0109] Specifically, the star type of the pulsar candidate is identified according to the star classification feature data. Specifically, the star classification feature data corresponding to the pulsar candidate is input into a trained classifier, and the pulsar candidate is classified by using the classifier; wherein the classifier can be a classifier based on machine learning, such as an SVM (Suppor Vector Machine) classifier, or a classifier based on deep learning, such as a classifier built based on a convolutional neural network model.
[0110] In the above-mentioned artificial intelligence-based pulsar search method, signal data of at least two signal feature types of pulsar candidates are obtained; target signal feature data of the signal data are extracted according to the feature extraction method and signal data corresponding to each signal feature type; wherein one target signal feature data is extracted based on at least one signal data; feature fusion is performed on the target signal feature data to obtain star classification feature data; and the star type of the pulsar candidate is identified according to the star classification feature data. By extracting target signal feature data from signal data corresponding to multiple signal feature types and feature fusion of the target signal feature data, the obtained star classification feature data can characterize both the signal features of different signal feature types and the correlation between signal data of different signal feature types. In the subsequent classification of pulsar candidates based on the star classification feature data, the classification accuracy is effectively improved, and the accuracy of the pulsar star search during the pulsar search process is improved.
[0111] In one embodiment, Figure 7 As shown, a pulsar search method based on artificial intelligence includes:
[0112] S710, acquiring signal data of at least two signal characteristic types of a pulsar candidate;
[0113] S720, extracting target signal feature data of the signal data according to the feature extraction method corresponding to each signal feature type and the signal data through the pulsar search model;
[0114] S730, performing feature fusion on the target signal feature data through a pulsar search model to obtain star classification feature data;
[0115] S740, identifying the star type of the pulsar candidate according to the star classification feature data using a pulsar search model.
[0116] The pulsar search model is used to classify pulsar candidates to determine the star type of the pulsar candidate. The pulsar search model here is a trained neural network model, which can be constructed by a convolutional neural network.
[0117] Specifically, after obtaining signal data of at least two signal feature types of a pulsar candidate, the signal data of each signal feature type is used as the input of a pulsar search model. The pulsar search model is used to extract target signal feature data of the signal data, perform feature fusion on the target signal feature data, and identify the stellar type of the pulsar candidate based on the stellar classification feature data obtained by feature fusion. Finally, the pulsar search model outputs the stellar type of the pulsar candidate.
[0118] It can be understood that the specific limitations on the steps of extracting target signal feature data of signal data, performing feature fusion on the target signal feature data, and identifying the star type of pulsar candidate based on the star classification feature data obtained by feature fusion can be found in the limitations in the above embodiments, and will not be elaborated here one by one.
[0119] Further, see Figure 8 In one embodiment, the pulsar search model includes a plurality of feature extraction networks, feature fusion networks and classifiers corresponding to different signal feature types respectively.
[0120] Specifically, after the signal data of each signal feature type is input into the pulsar search model as input data, the target signal feature data is first extracted from the signal data through the feature extraction network corresponding to each signal feature type, wherein one target signal feature data is extracted based on at least one signal data; then, each target signal feature data is used as the input of the feature fusion network, the feature fusion network performs feature fusion on the target signal feature data, and outputs the star classification feature data; finally, the star classification feature data is used as the input of the classifier, and the classifier classifies the pulsar candidates based on the star classification feature data to obtain the star type of the pulsar candidate.
[0121] Furthermore, in order to classify pulsar candidates and identify their star types, the pulsar search model needs to be trained in advance. Fig. 9 As shown, in one embodiment, before acquiring signal data of at least two signal characteristic types of a pulsar candidate, the method includes:
[0122] S910, obtaining signal data samples of at least two signal feature types of a sample star and a star category label.
[0123] The signal data samples of the sample star refer to data samples used to train the pulsar search model, and may include signal data samples whose star type is pulsar and signal data samples whose star type is non-pulsar.
[0124] The star category label is used to identify the star type of the sample star. Specifically, the star category label can be "0" or "1"; when the star type of the sample star is a pulsar, its corresponding star category label is marked as "1", and when the star type of the sample star is not a pulsar, its corresponding star category label is marked as "0".
[0125] S920, extracting target signal feature samples of the signal data samples according to the feature extraction method corresponding to each signal feature type and the signal data samples through the pulsar search model to be trained.
[0126] Specifically, the pulse search model may include multiple feature extraction networks corresponding to different signal feature types; after obtaining signal data samples of the sample star with different signal feature types, the signal data samples can be used as input data of the feature extraction network and input into the feature extraction network to extract the target signal feature samples.
[0127] Among them, the feature extraction network can be a pre-trained network model with certain initialization parameters. The training process can be omitted in the entire training process of the pulsar search model, saving model training time and improving model training efficiency.
[0128] S930, performing feature fusion on the target signal feature samples through a pulsar search model to obtain star classification feature samples.
[0129] The pulse search model may include a feature fusion network, and the feature fusion network in the pulse search model may be used to perform feature fusion on the target signal feature samples to obtain the star classification feature samples.
[0130] Furthermore, the number of dimensions of each target signal feature sample is the same, and the data length is equal under the same target dimension; step S930 may specifically include: stacking the target signal feature samples based on the target dimension to obtain a sample signal feature matrix; obtaining the self-attention weights of each row of signal feature samples in the sample signal feature matrix through a feature fusion network; performing weighted fusion of each row of signal feature samples according to the self-attention weights of each row of signal feature samples to generate star classification feature samples.
[0131] Among them, after obtaining the target signal feature samples, the target signal feature samples can be stacked based on the target dimension with equal data length to obtain the sample signal feature matrix; then, the row signal feature samples of each row in the sample signal feature matrix are taken as an input item and input into the feature fusion network, and the self-attention weights of each row signal feature sample are learned through the feature fusion network; finally, the row signal feature samples are weightedly fused based on the self-attention weights of the row signal feature samples to obtain the star classification feature samples.
[0132] S940, identifying the predicted star type of the sample star according to the star classification feature sample through a pulsar search model.
[0133] S950, calculating the loss value of the pulsar search model according to the star prediction type and the star category label, and adjusting the model parameters of the pulsar search model based on the loss value until the training end condition is met.
[0134] Specifically, after obtaining the predicted star type, the difference between the predicted star type and the star category label can be calculated through the loss function to obtain the loss value of the pulsar search model; wherein the loss function can specifically be a cross entropy loss function.
[0135] After the loss value of the pulsar search model is obtained, the model parameters in the pulsar search model can be adjusted based on the loss value until the training is completed. It can be understood that the training end condition here can be adjusted or set according to actual needs. For example, when the loss value meets the convergence condition, it can be considered that the training end condition is met; or when the number of training times reaches a preset number, it can be considered that the training end condition is met, which is not limited here.
[0136] Combine the following Figure 8 The artificial intelligence-based pulsar search method provided in this application is further explained.
[0137] Step 1: Acquisition of signal data:
[0138] FAST uses drift scanning to survey the sky and receives weak signals from hundreds of millions or even billions of light years away as raw observation data. Through data preprocessing, at least two signal characteristic types of signal data of pulsar candidates are obtained from the raw observation data. Among them, at least two signal characteristic types of signal data include frequency phase distribution images, time phase distribution images, and dispersion curve diagrams.
[0139] The time phase distribution image is obtained by accumulating the observation data of the pulsar candidate in different frequency domains, which reflects the intensity of the signal during the observation time. The pulsar signal is periodic, and the signal will continue to repeat during the entire observation time. On the time phase distribution image, the greater the signal intensity, the darker the color. If it is a pulsar signal, a vertical line corresponding to the peak position of the pulse profile curve will be formed during the entire observation time, that is, there is an intermittent or continuous but non-through vertical line in the time phase distribution image.
[0140] The frequency phase distribution image is obtained by accumulating the observation data of the pulsar candidate during the observation time, which reflects the intensity of the signal at different frequencies. Since the pulsar radio radiation is broadband, the typical pulse signal should appear in most of the observed frequency bands. If it is a pulsar signal, the frequency phase distribution image has vertical lines corresponding to the peaks in most frequencies.
[0141] The dispersion curve reflects the change in the pulse curve signal-to-noise ratio when different dispersion values are used for de-dispersion. When the pulse signal passes through the interstellar medium, dispersion will occur. When the correct value is used for de-dispersion, the pulse signal-to-noise ratio will be the largest. If it is a pulsar signal, the dispersion curve will conform to the Lorentz distribution or Poisson distribution.
[0142] Specifically, the frequency phase distribution image and the time phase distribution image are respectively determined as two-dimensional signal images; and dispersion curve data are obtained from the dispersion curve diagram to convert the dispersion curve diagram into a one-dimensional sequence to obtain a one-dimensional signal sequence.
[0143] Step 2, feature data extraction:
[0144] The dispersion curve data is input into the first feature extraction network (such as Figure 8 In the feature extraction network 1) shown, target signal feature data corresponding to the dispersion curve data is obtained through the first feature network;
[0145] The frequency phase distribution image is input into the second feature extraction network (such as Figure 8 In the feature extraction network 2) shown in the figure, the target signal feature data corresponding to the frequency phase distribution image is obtained through the second feature network;
[0146] The time phase distribution image is input into the fourth feature extraction network (such as Figure 8 In the feature extraction network 4) shown in the figure, the target signal feature data corresponding to the time phase distribution image is obtained through the fourth feature network;
[0147] The frequency phase distribution image and the time phase distribution image are stacked to obtain a stacked signal image, and the obtained stacked signal image is input into the third feature extraction network (such as Figure 8 In the feature extraction network 3) shown, the target signal feature data corresponding to the stacked signal image is obtained through the third feature network.
[0148] Specifically, the first feature extraction network includes a three-layer neural network, including at least a hidden layer and a ReLU activation layer. The input data of the first feature extraction network has a data length of a one-dimensional sequence of 100, and the output data is a 16-dimensional vector;
[0149] The second feature extraction network or the fourth feature extraction network uses ResNet18 as the backbone network; the input data of the second feature extraction network or the fourth feature extraction network are both single-channel 64*64 grayscale images, and the output data are both a 2*16 matrix;
[0150] The third feature extraction network uses ResNet18 as the backbone network. The input data of the third feature extraction network is a 2-channel 64*64 grayscale image, and the output data is a 3*16 matrix.
[0151] Therefore, for the one-dimensional signal sequence of dispersion curve data, 1*16 dimensional target signal feature data can be extracted; for the two-dimensional signal images of frequency phase distribution image and time phase distribution image, 2*16 dimensional target signal feature data can be extracted respectively; for the stacked signal image obtained by stacking the frequency phase distribution image and the time phase distribution image, 3*16 dimensional target signal feature data can be extracted.
[0152] Step 3: Signal feature fusion:
[0153] The target signal feature data obtained in step 2 can be stacked into an 8*16 signal feature matrix; then, the row signal feature data of each row in the signal feature matrix is used as input data of the feature fusion network, and the star classification feature data is obtained through the feature fusion network.
[0154] Among them, the feature fusion network is constructed using a 2-layer 2-head Transformer network. The data dimension of its input data is 16, the token dimension is 8, and the output data is a 16-dimensional vector.
[0155] Specifically, in the stacked 8*16 signal feature matrix, the 16-dimensional row signal feature data of each row is used as the input of the Transformer network. The self-attention mechanism of the Transformer network is used to learn the correlation between the signal feature data of each row, and then the self-attention weight of each row of signal feature data is obtained. The signal feature samples of each row are weightedly fused according to the self-attention weight of each row of signal feature samples, so that the feature data of each target signal are fully integrated and the global characteristics of the pulsar candidate are effectively obtained.
[0156] Step 4, star type identification:
[0157] The star classification feature data is input into the classifier, and the prediction score of the pulsar candidate belonging to the pulsar type is obtained through the classifier; if the prediction score is greater than the preset score threshold, the pulsar candidate is determined to be a pulsar.
[0158] Among them, the classifier includes a fully connected layer network and an activation layer network. The star classification feature data output by the feature fusion network is output through the fully connected layer network to output the prediction values belonging to different star types, and the prediction values are input into the sigmoid activation layer network. The sigmoid activation layer network outputs the prediction score of the pulsar candidate belonging to the pulsar type.
[0159] In order to better implement the pulsar search method based on artificial intelligence provided in the embodiment of the present application, on the basis of the pulsar search method based on artificial intelligence provided in the embodiment of the present application, a pulsar search device based on artificial intelligence is also provided in the embodiment of the present application, such as Fig.10 As shown, the pulsar search device 1000 based on artificial intelligence includes:
[0160] A signal data acquisition module 1010 is used to acquire signal data of at least two signal characteristic types of a pulsar candidate;
[0161] The feature data acquisition module 1020 is used to extract target signal feature data of the signal data according to the feature extraction method corresponding to each signal feature type and the signal data; wherein one target signal feature data is extracted based on at least one signal data;
[0162] The feature data fusion module 1030 is used to perform feature fusion on the target signal feature data to obtain the star classification feature data;
[0163] The star type classification module 1040 is used to identify the star type of the pulsar candidate according to the star classification feature data.
[0164] In some embodiments of the present application, the number of dimensions of each target signal feature data is the same, and the data length is equal under the same target dimension; the feature data fusion module is specifically used to stack the target signal feature data based on the target dimension to obtain a signal feature matrix; and to fuse the matrix elements in the signal feature matrix to obtain star classification feature data.
[0165] In some embodiments of the present application, the feature data fusion module is specifically used to input the signal feature matrix into the feature fusion network; obtain the self-attention weight of each row of signal feature data in the signal feature matrix through the feature fusion network; perform weighted fusion on each row of signal feature data according to the self-attention weight of each row of signal feature data to generate star classification feature data.
[0166] In some embodiments of the present application, the signal data under at least one signal feature type is a one-dimensional signal sequence, and the signal data under at least two signal feature types are two-dimensional signal images; the feature data acquisition module is specifically used to extract the target signal feature data of the two-dimensional signal image in the signal data by a feature extraction method corresponding to the two-dimensional signal image; extract the target signal feature data of the one-dimensional signal sequence in the signal data by a feature extraction method corresponding to the one-dimensional signal sequence; stack the two-dimensional signal images to obtain a stacked signal image, and extract the target signal feature data of the stacked signal image by a feature extraction method corresponding to the stacked signal image.
[0167] In some embodiments of the present application, the one-dimensional signal sequence includes at least dispersion curve data, and the two-dimensional signal image includes at least a frequency phase distribution image and a time phase distribution image.
[0168] In some embodiments of the present application, a feature data acquisition module is specifically used to extract target signal feature data of signal data according to the feature extraction method corresponding to each signal feature type and signal data through a pulsar search model; a feature data fusion module is specifically used to perform feature fusion on the target signal feature data through a pulsar search model to obtain star classification feature data; a star type classification module is specifically used to identify the star type of a pulsar candidate according to the star classification feature data through a pulsar search model; the pulsar search device based on artificial intelligence also includes a model training module, which is specifically used to: extract target signal feature samples of signal data samples according to the feature extraction method corresponding to each signal feature type and signal data samples through a pulsar search model to be trained; perform feature fusion on the target signal feature samples through a pulsar search model to obtain star classification feature samples; identify the predicted star type of the sample star through the pulsar search model according to the star classification feature samples; calculate the loss value of the pulsar search model according to the predicted star type and the star category label, and adjust the model parameters of the pulsar search model based on the loss value until the training end condition is met.
[0169] In some embodiments of the present application, the pulsar search model includes a feature fusion network; the number of dimensions of each target signal feature sample is the same, and the data length is equal under the same target dimension; the model training module is specifically used to stack the target signal feature samples based on the target dimension to obtain a sample signal feature matrix; the self-attention weights of each row of signal feature samples in the sample signal feature matrix are obtained through the feature fusion network; the signal feature samples in each row are weightedly fused according to the self-attention weights of the signal feature samples in each row to generate star classification feature samples.
[0170] The specific definition of the pulsar search device based on artificial intelligence can be found in the definition of the pulsar search method based on artificial intelligence above, which will not be repeated here. Each module in the above-mentioned pulsar search device based on artificial intelligence can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
[0171] In some embodiments of the present application, the pulsar search device 1000 based on artificial intelligence can be implemented in the form of a computer program. The computer program can be used in Fig.11 The computer device shown in the figure is run on the computer device. The memory of the computer device can store various program modules constituting the artificial intelligence-based pulsar search device 1000, for example, Fig.10The signal data acquisition module 1010, feature data acquisition module 1020, feature data fusion module 1030 and star type classification module 1040 are shown. The computer program composed of various program modules enables the processor to execute the steps of the pulsar search method based on artificial intelligence in various embodiments of the present application described in this specification.
[0172] For example, Fig.11 The computer device shown can be Fig.10 The signal data acquisition module 1010 in the pulsar search device 1000 based on artificial intelligence shown in the figure performs step S210. The computer device can perform step S220 through the feature data acquisition module 1020. The computer device can perform step S230 through the feature data fusion module 1030. The computer device can perform step S240 through the star type classification module 1040. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external computer device through a network connection. When the computer program is executed by the processor, a pulsar search method based on artificial intelligence is implemented.
[0173] Those skilled in the art will understand that Fig.11 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0174] In some embodiments of the present application, a computer device is provided, comprising one or more processors; a memory; and one or more applications, wherein the one or more applications are stored in the memory and configured to execute the steps of the above-mentioned artificial intelligence-based pulsar search method by the processor. The steps of the artificial intelligence-based pulsar search method here can be the steps of the artificial intelligence-based pulsar search method in each of the above-mentioned embodiments.
[0175] In some embodiments of the present application, a computer-readable storage medium is provided, which stores a computer program, and the computer program is loaded by a processor, so that the processor executes the steps of the pulsar search method based on artificial intelligence. The steps of the pulsar search method based on artificial intelligence here can be the steps of the pulsar search method based on artificial intelligence in each of the above embodiments.
[0176] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Any reference to memory, storage, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).
[0177] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0178] The above is a detailed introduction to an artificial intelligence-based pulsar search method, device, computer equipment and storage medium provided in the embodiments of the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on the present invention.
Claims
1. A pulsar search method based on artificial intelligence, characterized in that: include: Acquire signal data samples of at least two signal feature types of the sample star and a star category label; extract target signal feature samples of the signal data samples according to the feature extraction method corresponding to each signal feature type and the signal data samples through the pulsar search model to be trained; The target signal feature samples are subjected to feature fusion by the pulsar search model to obtain star classification feature samples; the star prediction type of the sample star is identified by the pulsar search model according to the star classification feature samples; the loss value of the pulsar search model is calculated according to the star prediction type and the star category label, and the model parameters of the pulsar search model are adjusted based on the loss value until the training end condition is met; Acquiring signal data of at least two signal characteristic types of the pulsar candidate; Extracting target signal feature data of the signal data according to the feature extraction method corresponding to each signal feature type and the signal data through the pulsar search model; wherein one target signal feature data is extracted based on at least one signal data; Performing feature fusion on the target signal feature data through the pulsar search model to obtain star classification feature data; The pulsar search model is used to identify the star type of the pulsar candidate according to the star classification feature data.
2. The method according to claim 1, characterized in that The target signal feature data have the same number of dimensions and the same data length under the same target dimension; The step of performing feature fusion on the target signal feature data to obtain star classification feature data includes: Stacking the target signal feature data based on the target dimension to obtain a signal feature matrix; The matrix elements in the signal feature matrix are merged to obtain the star classification feature data.
3. The method according to claim 2, characterized in that The step of fusing the matrix elements in the signal feature matrix to obtain the star classification feature data includes: Inputting the signal feature matrix into a feature fusion network; Obtaining the self-attention weight of each row of signal feature data in the signal feature matrix through the feature fusion network; The row signal feature data are weightedly fused according to the self-attention weight of each row signal feature data to generate the star classification feature data.
4. The method according to claim 1, characterized in that: The signal data under at least one of the signal feature types is a one-dimensional signal sequence, and the signal data under at least two of the signal feature types are two-dimensional signal images; The extracting target signal feature data of the signal data according to the feature extraction method corresponding to each signal feature type and the signal data includes: Extracting target signal feature data of the two-dimensional signal image in the signal data by a feature extraction method corresponding to the two-dimensional signal image; Extracting target signal feature data of the one-dimensional signal sequence in the signal data by a feature extraction method corresponding to the one-dimensional signal sequence; The two-dimensional signal images are stacked to obtain a stacked signal image, and target signal feature data of the stacked signal image is extracted by a feature extraction method corresponding to the stacked signal image.
5. The method according to claim 4, characterized in that The one-dimensional signal sequence includes at least dispersion curve data, and the two-dimensional signal image includes at least a frequency phase distribution image and a time phase distribution image.
6. The method according to claim 1, characterized in that The pulsar search model includes a feature fusion network; the number of dimensions of the target signal feature samples is the same, and the data length is equal under the same target dimension; The step of performing feature fusion on the target signal feature samples through the pulsar search model to obtain star classification feature samples includes: Stacking the target signal feature samples based on the target dimension to obtain a sample signal feature matrix; Obtaining the self-attention weight of each row of signal feature samples in the sample signal feature matrix through the feature fusion network; The row signal feature samples are weightedly fused according to the self-attention weights of the row signal feature samples to generate the star classification feature samples.
7. A pulsar search device based on artificial intelligence, characterized in that: The device comprises: A model training module is used to obtain signal data samples of at least two signal feature types of a sample star and a star category label; extract target signal feature samples of the signal data samples according to the feature extraction method corresponding to each signal feature type and the signal data sample through a pulsar search model to be trained; perform feature fusion on the target signal feature samples through the pulsar search model to obtain star classification feature samples; identify the star prediction type of the sample star according to the star classification feature samples through the pulsar search model; calculate the loss value of the pulsar search model according to the star prediction type and the star category label, and adjust the model parameters of the pulsar search model based on the loss value until the training end condition is met; A signal data acquisition module, used to acquire signal data of at least two signal characteristic types of a pulsar candidate; A feature data acquisition module, used to extract target signal feature data of the signal data according to the feature extraction method corresponding to each signal feature type and the signal data through the pulsar search model; wherein one target signal feature data is extracted based on at least one signal data; A feature data fusion module, used to perform feature fusion on the target signal feature data through the pulsar search model to obtain star classification feature data; A star type classification module is used to identify the star type of the pulsar candidate according to the star classification feature data through the pulsar search model.
8. A computer device, characterized in that: The computer device comprises: one or more processors; Memory; and One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the artificial intelligence-based pulsar search method according to any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program is loaded by a processor to execute the steps in the artificial intelligence-based pulsar search method as described in any one of claims 1 to 6.
10. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the steps of the pulsar search method based on artificial intelligence as claimed in any one of claims 1 to 6 are implemented.
