Astronomical data fusion analysis method based on deep learning

Abstract

The invention relates to the field of data processing, in particular to an astronomical data fusion analysis method and system based on deep learning. The method comprises the following steps: acquiring multi-source astronomical data; the multi-source astronomical data aim at the same observation object or the same observation area; inputting the multi-source astronomical data into an adaptive exploration space model to obtain analysis modes corresponding to the multi-source astronomical data; configurable analysis mode parameters and analysis modes are set in the adaptive exploration space model; and performing multi-branch fusion processing on the multi-source astronomical data by adopting the analysis mode to obtain astronomical fusion data corresponding to the multi-source astronomical data. According to the method, related information of different data sources can be dynamically fused, and deep fusion analysis of multi-source astronomical data is realized, so that more accurate and more comprehensive astronomical fusion data is provided for astronomical research.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to an astronomical data fusion analysis method and system based on deep learning. Background Art

[0002] There are many kinds of astronomical data. For example, images and spectral data in the visible light band obtained by optical telescopes are used to observe galaxies, stars, planets and other celestial bodies. Data in the ultraviolet band are usually obtained by special ultraviolet telescopes to help study high-temperature celestial bodies, interstellar media, etc. Data obtained by radio telescopes are used to study celestial bodies and phenomena in the radio band, such as pulsars, radio galaxies and interstellar gases. By splitting the light emitted by celestial bodies, the spectral information of celestial bodies is obtained. These data can be used to analyze the composition, movement, temperature, etc. of celestial bodies. In addition, data from space exploration missions, such as planetary probes and solar probes, provide detailed information about planets, the sun, etc.

[0003] In related technologies, the fusion of multiple astronomical data faces many challenges. However, the difference in the physical properties of multiple astronomical data is one of the main difficulties. Data in different bands, such as visible light, infrared and radio data, reflect the radiation characteristics of celestial bodies in their respective spectral ranges. For example, optical data reveals the visible light radiation and morphology of celestial bodies, infrared data can penetrate interstellar dust and detect hidden celestial bodies, and radio data is used to study low-energy celestial bodies. How to effectively combine the physical information of these different bands is an important problem that needs to be solved in the fusion process.

[0004] In summary, it is urgent to propose a completely new technical solution to solve at least one of the above technical problems. Summary of the invention

[0005] In view of the technical problems existing in the prior art, the present invention provides an astronomical data fusion analysis method and system based on deep learning, which are used to realize the automatic fusion processing of multi-source astronomical data and provide a data basis for the subsequent processing of multi-source astronomical data.

[0006] In a first aspect, the embodiments of the present application provide a method and system for astronomical data fusion analysis based on deep learning, the method comprising:

[0007] Acquiring multi-source astronomical data; the multi-source astronomical data are all for the same observation object or the same observation area;

[0008] Inputting the multi-source astronomical data into an adaptive exploration space model to obtain analysis modes corresponding to the multi-source astronomical data; the adaptive exploration space model is provided with configurable analysis mode parameters and analysis methods;

[0009] The analysis mode is adopted to perform multi-branch fusion processing on the multi-source astronomical data to obtain astronomical fusion data corresponding to the multi-source astronomical data.

[0010] In a second aspect, an embodiment of the present application provides an astronomical data fusion analysis system based on deep learning, the system comprising the following units, wherein:

[0011] A collection unit is configured to acquire multi-source astronomical data; the multi-source astronomical data are all for the same observation object or the same observation area;

[0012] An analysis unit is configured to input the multi-source astronomical data into an adaptive exploration space model to obtain analysis modes corresponding to the multi-source astronomical data; the adaptive exploration space model is provided with configurable analysis mode parameters and analysis methods;

[0013] The fusion unit is configured to adopt the analysis mode to perform multi-branch fusion processing on the multi-source astronomical data to obtain astronomical fusion data corresponding to the multi-source astronomical data.

[0014] In a third aspect, an embodiment of the present application provides an electronic device, the electronic device comprising:

[0015] at least one processor, memory, and input-output unit;

[0016] The memory is used to store a computer program, and the processor is used to call the computer program stored in the memory to execute the astronomical data fusion analysis method based on deep learning of the first aspect.

[0017] In a fourth aspect, a computer-readable storage medium is provided, which includes instructions, which, when executed on a computer, enable the computer to execute the deep learning-based astronomical data fusion analysis method of the first aspect.

[0018] The beneficial effect of the present invention is that it provides a method and system for astronomical data fusion analysis based on deep learning. In this technical solution, multi-source astronomical data are obtained; the multi-source astronomical data are all for the same observation object or the same observation area; the multi-source astronomical data are input into the adaptive exploration space model to obtain the analysis mode corresponding to each of the multi-source astronomical data; the adaptive exploration space model is provided with configurable analysis mode parameters and analysis methods; the analysis mode is used to perform multi-branch fusion processing on the multi-source astronomical data to obtain astronomical fusion data corresponding to the multi-source astronomical data. This method effectively solves the problem of physical property differences of astronomical data in different bands through an adaptive exploration space model, a multi-branch processing network and a real-time fusion mechanism. While maintaining the unique characteristics of each band of data, it can dynamically fuse the relevant information of different data sources to achieve deep fusion analysis of multi-source astronomical data, thereby providing more accurate and comprehensive astronomical fusion data for astronomical research. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flowchart of an astronomical data fusion analysis method and system based on deep learning according to an embodiment of the present application;

[0020] Figure 2 It is a structural schematic diagram of an astronomical data fusion analysis system based on deep learning according to an embodiment of the present application;

[0021] Figure 3 is a schematic diagram of the structure of an electronic device according to an embodiment of the present application;

[0022] Figure 4 It is a structural schematic diagram of a medium device according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] 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.

[0024] 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 of the features. In the description of this application, "plurality" means two or more, unless otherwise clearly and specifically defined.

[0025] In the description of the present application, the term "for example" is used to mean "used as an example, illustration or description". 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 unnecessary details to obscure the description of the present invention. 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.

[0026] There are many kinds of astronomical data. For example, images and spectral data in the visible light band obtained by optical telescopes are used to observe galaxies, stars, planets and other celestial bodies. Data in the ultraviolet band are usually obtained by special ultraviolet telescopes to help study high-temperature celestial bodies, interstellar media, etc. Data obtained by radio telescopes are used to study celestial bodies and phenomena in the radio band, such as pulsars, radio galaxies and interstellar gases. By splitting the light emitted by celestial bodies, the spectral information of celestial bodies is obtained. These data can be used to analyze the composition, movement, temperature, etc. of celestial bodies. In addition, data from space exploration missions, such as planetary probes and solar probes, provide detailed information about planets, the sun, etc.

[0027] In related technologies, the fusion of multiple astronomical data faces many challenges. However, the difference in the physical properties of multiple astronomical data is one of the main difficulties. Data in different bands, such as visible light, infrared and radio data, reflect the radiation characteristics of celestial bodies in their respective spectral ranges. For example, optical data reveals the visible light radiation and morphology of celestial bodies, infrared data can penetrate interstellar dust and detect hidden celestial bodies, and radio data is used to study low-energy celestial bodies. How to effectively combine the physical information of these different bands is an important problem that needs to be solved in the fusion process.

[0028] It is urgent to propose a completely new technical solution to solve at least one of the above technical problems.

[0029] The embodiment of the present application provides a method and system for astronomical data fusion analysis based on deep learning. In the technical solution, multi-source astronomical data are obtained; the multi-source astronomical data are all for the same observation object or the same observation area; the multi-source astronomical data are input into an adaptive exploration space model to obtain the analysis mode corresponding to each of the multi-source astronomical data; the adaptive exploration space model is provided with configurable analysis mode parameters and analysis methods; the analysis mode is used to perform multi-branch fusion processing on the multi-source astronomical data to obtain astronomical fusion data corresponding to the multi-source astronomical data.

[0030] From a technical point of view, in an embodiment of the present application, the method adopts an adaptive exploration space model, in which configurable analysis mode parameters and analysis methods are set. This model can dynamically adjust the processing method of each data source, and apply special feature extraction and processing strategies to data in different bands (such as visible light, infrared, and radio). For example, for optical data, specific edge detection or morphological analysis methods can be applied to highlight the celestial structure in visible light. For infrared data, a feature extraction network that can penetrate dust is used to focus on the radiation distribution and hidden features of celestial bodies. For radio data, long wavelength characteristics are used to focus on analyzing the physical properties of low-energy objects. This adaptive exploration model can select the optimal analysis method and parameters according to the characteristics of the input data to ensure the correct understanding and processing of information in different bands.

[0031] In an embodiment of the present application, through the design of a multi-branch network, an independent processing branch is established for each band of data (such as visible light, infrared, and radio). Each branch is equipped with a corresponding neural network structure, which is specifically used for the extraction and representation of the modal features. In this way, the data of each band can maximize its specific information. For example, the optical branch focuses on morphology, brightness, and surface structure features. The infrared branch extracts deeply embedded thermal radiation information, which can penetrate dust and detect obscured celestial bodies. The radio branch focuses on studying weak signals from low-energy celestial bodies and celestial bodies at greater distances. The multi-branch structure ensures that astronomical data with different physical properties can be effectively processed and represented separately, while also avoiding the problem of losing key information during direct fusion.

[0032] In an embodiment of the present application, the method uses a real-time fusion module of the data stream to fuse data from different bands in real time by dynamically adjusting the weight and contribution of each branch. In particular, through an adaptive weight mechanism, the fusion method can be flexibly adjusted according to the needs of different scenarios (such as the need to highlight infrared information to detect hidden celestial bodies or emphasize radio data to analyze low-energy celestial bodies). This real-time fusion mechanism can balance the contribution of information from different bands according to the characteristics of the actual observation target (such as galaxies, nebulae, etc.), ensuring that the final fusion result not only retains the key information of each band, but also effectively combines the correlations between different bands.

[0033] In addition, during the fusion process, this method constructs a unified high-dimensional feature space through the feature joint learning mechanism of the neural network, and aligns and aggregates the information of different modalities. Combined with the self-attention mechanism of deep learning (such as the Transformer structure), it can mine potential interactive features and interdependencies for the complex correlations between multi-source data. For example, in the analysis of the core region of galaxies, optical and infrared data may reveal complementary information between each other. Through joint processing, the physical state and evolution process of celestial bodies can be more accurately analyzed. Through adaptive weights and data processing mechanisms, the influence of data from different bands in the fusion process can be dynamically adjusted. For example, when the data quality of a certain band is poor (such as noise interference in radio data), the fusion mechanism will reduce its weight and rely more on information from other bands. This mechanism can effectively improve the robustness of data fusion and ensure the stability of the fusion results.

[0034] In summary, the embodiments of the present application effectively solve the problem of physical property differences of astronomical data in different bands through adaptive exploration space models, multi-branch processing networks and real-time fusion mechanisms. While maintaining the unique characteristics of each band of data, it can dynamically fuse relevant information from different data sources to achieve deep fusion analysis of multi-source astronomical data, thereby providing more accurate and comprehensive astronomical fusion data for astronomical research.

[0035] The astronomical data fusion analysis scheme based on deep learning provided in the embodiment of the present application can also be performed by an electronic device, which can be a server, a server cluster, or a cloud server. The electronic device can also be a terminal device such as a mobile phone, a computer, a tablet computer, a wearable device, or a dedicated device (such as a dedicated terminal device with an astronomical data fusion analysis system based on deep learning). The chips introduced in the above embodiments can also be carried in these electronic devices. Alternatively, these electronic devices can also be installed with a service program for executing an astronomical data fusion analysis scheme based on deep learning.

[0036] Figure 1 A flowchart of an astronomical data fusion analysis method and system based on deep learning is provided in an embodiment of the present application, such as Figure 1 As shown, the method comprises the following steps:

[0037] 101, obtain multi-source astronomical data;

[0038] 102, inputting the multi-source astronomical data into an adaptive exploration space model to obtain analysis modes corresponding to the multi-source astronomical data;

[0039] 103 . Using the analysis mode, perform multi-branch fusion processing on the multi-source astronomical data to obtain astronomical fusion data corresponding to the multi-source astronomical data.

[0040] In the embodiment of the present application, the multi-source astronomical data are all for the same observation object or the same observation area. In the embodiment of the present application, the multi-source astronomical data are all for the same observation object or the same observation area, which specifically refers to the use of different observation equipment and different bands (such as visible light, infrared, radio, etc.) in astronomical observation to observe the same celestial body or the same sky area. This approach ensures that multimodal data obtained from different angles reflects the multiple physical properties of the same celestial body or area.

[0041] The consistency of the observed object means that the data obtained through different observation methods are all for the same celestial body. This consistency can improve the accuracy of data fusion, because all data sources target the same celestial body, although they may come from different bands, but they all reflect the different physical properties of the celestial body. For example, in star observations, visible light data can show the brightness and surface activity of stars, while infrared data can detect the cooling process of the outer layer of stars, and radio data may reveal changes in the magnetic field near stars. For galaxy observations, visible light observations can reveal the morphological structure of galaxies, infrared data can penetrate dust clouds and display the star formation regions inside galaxies, while radio data can provide information on galactic nuclear activity. By processing multi-source data of the same observed object, more complete celestial body information can be obtained, avoiding the misalignment and error problems that may occur when data from different observed objects are fused.

[0042] The consistency of the observation area means that all data come from the same observation area (i.e., the sky area), and the observation data of different bands cover the same astronomical area. This is very important for studying large-scale structures or extended celestial bodies (such as nebulae, star clusters, galaxy clusters, etc.). For nebula observations, visible light data shows the emission spectrum of the nebula, infrared data can penetrate interstellar dust and detect more obscured areas, and radio data can reveal the distribution of low-temperature gases in the nebula. For galaxy cluster observations, galaxy clusters are composed of multiple galaxies, and observations in different bands can respectively reveal the characteristics of different galaxies in the cluster and the distribution of gas and dark matter within the cluster. In this case, data from all bands reflect different physical processes and properties in the same area, ensuring that comprehensive and three-dimensional information about the area can be obtained when data is fused.

[0043] When multi-source data acquired for the same observation object or region are fused, the data in each band provides different levels of information about the celestial body or region, allowing for a more comprehensive understanding of astronomical phenomena. Data from different bands can complement each other and help make up for information blind spots that may exist in a single band. For example, infrared data can reveal celestial bodies obscured by dust, while visible light cannot do this. By fusing multimodal data from the same object or region, the different physical properties of celestial bodies can be analyzed in multiple dimensions to form a richer physical description. For example, when studying galaxies, optical data can provide information about their morphology, while radio data can provide information about their activity status. Since observation data from different bands often have different noise and errors, the overall observation accuracy can be improved by fusing these data. For example, some bands may be more stable in the high energy or low energy range, and the observation error can be reduced by fusing data from different bands.

[0044] In the study of celestial evolution, when studying the evolution of celestial bodies, data in different bands reflect the activities of celestial bodies at different stages. For example, the formation process of early stars may be revealed through infrared and radio data, while the mature stage of stars may be mainly recorded through visible light observations.

[0045] In the large-scale structure of the universe, when observing galaxy clusters or the large-scale structure of the universe, multimodal observations can reveal the distribution of matter at different scales, such as galaxies, gas, and dark matter, forming a comprehensive understanding of the structure of the universe.

[0046] The multi-source astronomical data described in the embodiments of the present application are all for the same observation object or the same observation area. By using multi-modal data obtained from different bands, the fusion of these data can overcome the limitations of single-band observation and provide more comprehensive astronomical information. This fusion method can effectively improve the accuracy, depth and breadth of astronomical data analysis, allowing astronomers to better understand the physical properties of celestial bodies or celestial regions and their evolution process.

[0047] 101, obtain multi-source astronomical data. In an embodiment of the present application, step 101 is intended to obtain various types of astronomical data from different observation devices, bands or platforms, and these data come from the same observation object or area. Multi-source astronomical data include data in different bands such as visible light, infrared, radio, ultraviolet, X-rays, etc., and each type of data reflects different physical properties of celestial bodies, such as brightness, morphology, energy state, etc. These data are collected through different ground-based or space telescopes, and are stored in a database after processing and calibration. Acquiring multi-source data helps to fully understand the characteristics of celestial bodies, improve observation accuracy, and support the study of complex astronomical phenomena.

[0048] Each type of astronomical data reflects the physical properties of celestial bodies in different bands. For example, infrared data can reveal stars hidden in dust clouds, while visible light data can show the surface brightness of stars. Data in different bands also differ in spatial and temporal resolution. For example, ground-based telescopes have low spatial resolution due to the influence of the atmosphere, while space telescopes have higher resolution. Observations in different bands are affected by different noises. For example, radio data are susceptible to ground-based radio interference, while infrared data are susceptible to thermal noise. Therefore, the acquisition process of multi-source data also needs to consider noise processing and data quality optimization.

[0049] By acquiring data in different bands, the physical state of celestial bodies can be revealed from multiple dimensions. For example, visible light and infrared data can simultaneously understand the surface and internal state of a star. The complementarity of multi-source data helps to improve the accuracy of observation results. For example, observations of the same star in the infrared and radio bands can verify each other, thereby improving the reliability of the data. Many astronomical phenomena (such as supernova explosions, galaxy formation, etc.) involve multiple physical processes. Acquiring multi-source data helps to study these complex astronomical phenomena more comprehensively.

[0050] In the embodiment of the present application, the adaptive exploration space model is provided with configurable analysis mode parameters and analysis methods. The purpose of this setting is to provide flexible and efficient analysis capabilities under different exploration needs and data backgrounds. This configuration mechanism can adjust and optimize the behavior of the model according to different application scenarios and observation targets, ensuring high-precision and efficient analysis under complex data sets.

[0051] Analysis mode parameters refer to the core parameters used by the adaptive exploration space model under different operating conditions. These parameters determine how the model processes input data and performs exploration and analysis. These parameters can be configured according to different needs, mainly including:

[0052] Spatial resolution determines the spatial accuracy used by the model during exploration. Different astronomical phenomena have different requirements for resolution. Larger resolution is suitable for large-scale astronomical structures (such as galaxy clusters), while smaller resolution is suitable for high-precision observations (such as observations of individual stars).

[0053] Time resolution: When performing time series analysis on dynamically changing astronomical phenomena (such as supernova explosions or stellar activities), different time interval parameters need to be configured in order to capture the changing process of celestial bodies.

[0054] Band selection: According to the specific astronomical research object, different electromagnetic bands (such as infrared, radio, X-ray, etc.) are selected for analysis. This parameter can be configured to use only a single band or multi-band data fusion.

[0055] Noise filtering parameters are used to configure the degree of noise processing in data preprocessing. Different observation devices or bands may face different noise sources, which requires flexible setting of noise filtering intensity to improve data quality.

[0056] Analysis methods refer to the specific methods used by the adaptive exploration space model to process and analyze multi-source astronomical data. Different analysis methods can be optimized for specific research goals or data types. Exemplary methods include:

[0057] Statistical analysis: Quantitative analysis of the overall characteristics of the observed data, such as calculating the mean, variance, standard deviation and other statistical quantities. This method is suitable for the preliminary analysis of large-scale survey data or the study of the properties of celestial groups.

[0058] Frequency domain analysis: By performing Fourier transform or wavelet transform on astronomical signals, the periodic change characteristics in the signals can be extracted. This is suitable for studying periodic celestial bodies such as pulsars.

[0059] Time domain analysis: Analysis of the characteristics of celestial bodies such as brightness and position that change with time. It is mainly applied to astronomical phenomena with obvious time changes, such as variable stars and supernova explosions.

[0060] Multimodal data fusion analysis: Comprehensive analysis of astronomical data from different bands is suitable for comprehensive research on complex celestial systems. For example, combining infrared, visible light, and radio data to analyze the multi-layer structure of galaxies.

[0061] Machine learning analysis: Introducing machine learning algorithms (such as classification, regression or clustering algorithms) to perform pattern recognition or predictive analysis on astronomical data. This method is suitable for discovering potential astronomical laws or performing automated data classification under large-scale data sets.

[0062] The adaptive exploration space model can dynamically adjust the above parameters and analysis methods according to data input and analysis requirements. The model continuously optimizes its configuration to adapt to different types of data sources and analysis tasks, ensuring that the computational efficiency is improved while ensuring accuracy. For example, when the input data has high noise, the model can automatically increase the noise filtering intensity; when the observed object is a rapidly changing celestial body, the model will automatically switch to the time domain analysis mode to capture instantaneous changes.

[0063] When studying the evolution of stars or galaxies, we can compare data from different periods through multimodal data fusion and time domain analysis to obtain detailed trajectories of celestial body evolution. We can fuse multi-band data (such as visible light, infrared, radio, etc.) of the same observation object to obtain the multi-dimensional physical properties of the celestial body or region. We can use machine learning models to automatically classify celestial bodies such as galaxies, quasars, pulsars, etc. for a large amount of sky survey data.

[0064] In the embodiment of the present application, the adaptive exploration space model can be flexibly adjusted according to different observation requirements and data types by setting configurable analysis mode parameters and analysis methods. Through the adaptive mechanism, the model can dynamically optimize the analysis process and improve the efficiency and accuracy of data processing, thereby better meeting the analysis needs of multi-source data in the field of astronomy.

[0065] As an optional embodiment, in 102, the multi-source astronomical data is input into the adaptive exploration space model to obtain the analysis mode corresponding to each of the multi-source astronomical data, which can be implemented as follows:

[0066] A multimodal feature extraction layer is used to extract multimodal astronomical features from the multi-source astronomical data; the multimodal astronomical features include at least one of the following: observed space features, observed calorific value change features, and radio time series features;

[0067] Using a projection layer, the multi-source astronomical data are projected into an adaptive exploration space respectively; the adaptive exploration space is constructed by a pre-configured astronomical space model;

[0068] Using a routing module, the multi-modal astronomical features are transmitted to corresponding analysis units;

[0069] Using an analysis layer, learning from the multi-modal astronomical features corresponding to each analysis unit to obtain the modal attribute features corresponding to each of the multi-modal astronomical features;

[0070] A configuration layer is adopted to match the analysis modes corresponding to the multi-modal astronomical features based on the modal attribute characteristics; and configurable analysis mode parameters and analysis methods are set in the analysis unit.

[0071] In this optional embodiment 102, the analysis process for multi-source astronomical data is implemented by an adaptive exploration space model, and the specific steps include multi-modal feature extraction, projection, routing, analysis and configuration. The following example illustrates this process:

[0072] Suppose we analyze a galaxy, which has been observed by different astronomical equipment and bands, and has obtained multi-source astronomical data including visible light data, infrared data, and radio data. These data are input into the adaptive exploration space model, and the following steps are performed:

[0073] The multimodal feature extraction layer extracts multimodal features from multi-source astronomical data:

[0074] Observational spatial characteristics: Visible light data provides spatial information such as the morphology and brightness distribution of galaxies, such as the size and structure of galaxies (such as spiral arms and core regions).

[0075] Observe the characteristics of thermal value changes: Infrared data reveal the temperature changes of dust clouds and star forming regions in the galaxy, reflecting the thermal state of the galaxy.

[0076] Radio timing characteristics: Radio data provide characteristics of physical activity in galaxies by observing activities in the center of galaxies, such as changes in radio jets or pulsars.

[0077] The projection layer then projects these multi-source data into an adaptive exploration space, which is constructed from a pre-configured astronomical space model. Through this model, the data is displayed in a higher-dimensional way, fusing data from different bands to gain a deeper understanding of the overall properties of galaxies. For example, visible light data is projected into the morphological space of galaxies; infrared data is projected into the thermodynamic space of temperature changes; and radio data is projected into the temporal dynamic space of radio activity.

[0078] Then, the routing module transmits these different modal features (such as galaxy morphology, calorific value distribution, and radio activity features) to different analysis units. For example, morphological features are transmitted to a unit dedicated to analyzing galaxy structure, calorific value variation features are transmitted to a unit analyzing interstellar medium and star formation, and radio features are transmitted to a unit analyzing high-energy astrophysical phenomena.

[0079] Then, each analysis unit performs corresponding analysis based on the received features. The Galaxy Morphology Unit learns the structural features in the visible light data and analyzes the brightness distribution of the spiral arms and core of the galaxy. The Star Formation and Thermal State Analysis Unit learns the temperature characteristics of the star formation region from the infrared data and analyzes the activity of these regions. The Radio Analysis Unit learns the time series variation characteristics in the radio data and analyzes whether the radio activity in the center of the galaxy is related to the black hole.

[0080] The configuration layer matches the most appropriate analysis mode according to the various modal attribute characteristics learned in the analysis unit (such as the morphology of the galaxy, the activity of star formation, and the frequency of radio activity). For example, if the morphology of the galaxy shows a tight spiral structure and the infrared data show significant star formation activity, the analysis mode will focus more on the study of star formation and evolution. If strong radio jet features are detected in the radio data and change rapidly, the configured analysis mode may focus on the analysis of black hole activity and jets.

[0081] Each analysis unit will further use different analysis methods (such as timing analysis, thermodynamic analysis, etc.) according to these matching patterns to generate the final analysis results. For example, if the galaxy is an active galactic nucleus (AGN), multimodal feature extraction can reveal the presence of radio jet activity in its center, and infrared data show that the dust cloud near the center of the galaxy is heating up. The configuration layer will match these features to the AGN activity analysis mode, and the model will further use high-energy radio timing analysis, focusing on the intensity changes of radio jets, and thermodynamic analysis to study the relationship between temperature changes in the central region of the galaxy and black hole activity.

[0082] Through the above process, the adaptive exploration space model can flexibly and accurately process multi-source astronomical data and generate in-depth analysis results for specific astronomical phenomena.

[0083] Further optionally, in the above steps, a multimodal feature extraction layer is used to extract multimodal astronomical features from the multi-source astronomical data, which can be implemented as follows:

[0084] Identifying data modality types contained in the multi-source astronomical data;

[0085] Generate a corresponding feature extraction branch based on the identified data modality type to obtain a feature extraction network; wherein the feature extraction branch includes at least an independent extraction branch and a mixed extraction branch;

[0086] Inputting the multi-source astronomical data into the feature extraction network;

[0087] In the feature extraction network, each independent extraction branch is used to extract a set of independent astronomical data features of the corresponding modality, and the mixed extraction branch is used to extract a set of mixed astronomical data features of at least two modalities;

[0088] The extracted independent astronomical data features and mixed astronomical data features are input into a cross-modal interaction layer, and interactive astronomical data features are extracted from the independent astronomical data features and the mixed astronomical data features; the cross-modal interaction layer is dynamically adjusted based on the structural branches in the feature extraction network.

[0089] In a further optional embodiment, the multimodal feature extraction layer can be implemented by constructing a network architecture based on feature extraction branches to extract features for different types of astronomical data. The following is an exemplary illustration showing how to extract independent and mixed astronomical data features through a feature extraction network and dynamically adjust and interact in the cross-modal interaction layer.

[0090] Suppose a star is studied using visible light, infrared, and radio data, which reflect different physical properties of the star. First, identify that the input multi-source astronomical data belongs to multiple modalities, such as the following: Visible light data is used to show the brightness, color, and surface characteristics of the star. Infrared data is used to reflect the temperature distribution inside or around the star, especially interstellar dust and star formation regions. Radio data is used to reveal low-energy electromagnetic radiation in or around the star, especially magnetic fields and electromagnetic activities.

[0091] Based on the identified data modality type, the system generates corresponding feature extraction branches and constructs a feature extraction network. In this network, independent extraction branches generate independent feature extraction branches for each modality, process the data of each modality separately, and extract its unique physical features. Mixed extraction branches generate mixed extraction branches for data of at least two modalities, extract comprehensive features across modalities, and capture the complementarity between modalities.

[0092] Then, each modal data is input into the corresponding independent extraction branch. The visible light independent extraction branch extracts the brightness, color and morphological characteristics of the star surface, such as star spots, luminosity changes, etc. The infrared independent extraction branch extracts the thermodynamic characteristics around the star, especially the temperature distribution of the dust cloud or the calorific value changes in the star formation region. The radio independent extraction branch extracts radio activity characteristics, such as the temporal changes of the stellar magnetic field and the pulsed radio signal. At the same time, the combination of multi-modal data is processed in the mixed extraction branch. The visible light infrared mixed extraction branch combines the optical characteristics of the star surface with the thermal state of the dust cloud around it to extract the correlation characteristics between the stellar luminosity and calorific value changes. The infrared radio mixed extraction branch combines the thermal characteristics and radio signals to extract the interactive characteristics between the thermal activity and magnetic field activity of the star. The independent astronomical data features and mixed astronomical data features extracted by the cross-modal interaction layer are input into the cross-modal interaction layer. This interaction layer is used to combine the information between different modes and further extract interactive features from these features.

[0093] Interactive astronomical data features can capture the correlation between different modal data of stars, such as whether changes in surface brightness are related to changes in the thermal value of its surroundings, or whether magnetic field activity affects the optical characteristics of the star's surface.

[0094] The cross-modal interaction layer is dynamically adjusted based on the structural branches of the feature extraction network. For example, the system can dynamically adjust the extracted feature types and the structure of the branches based on the quality of the input data or the research objectives: if the radio data shows that the magnetic field activity of the star is very strong, the system may strengthen the weight of the radio and infrared mixed branches to further analyze the impact of magnetic field activity on the environment around the star. If the infrared data shows that the temperature around the star fluctuates greatly, the system will prioritize the analysis of the interaction between infrared and visible light to explore the relationship between the brightness of the star and its thermal state.

[0095] For example, when studying a pulsar with a strong magnetic field, the timing characteristics of the pulse signal are extracted through an independent radio branch, while the relationship between magnetic field activity and the heating of the material around the star is analyzed through a mixed infrared and radio branch. The cross-modal interaction layer dynamically adjusts the analysis, giving priority to the correlation between the temporal changes of the pulse signal and the fluctuations of the infrared thermal value.

[0096] Through this flexible feature extraction and interactive analysis mechanism, the system can more comprehensively analyze multi-source astronomical data, reveal the complex relationship between different modal data of celestial objects, and provide astronomers with valuable scientific insights.

[0097] Further optionally, in the above steps, a projection layer is used to project the multi-source astronomical data into the adaptive exploration space respectively, which can be implemented as follows:

[0098] Independent astronomical data features, mixed astronomical data features, and interactive astronomical data features are projected into an adaptive exploration space according to a multimodal dimension to obtain a plurality of astronomical data feature points; and a plurality of corresponding astronomical data exploration paths are constructed by fitting the plurality of astronomical data feature points.

[0099] Among them, the intersections between multiple astronomical data exploration paths correspond to interactive astronomical data features or mixed astronomical data features, and the independent feature points in multiple astronomical data exploration paths correspond to independent astronomical data features. In the adaptive exploration space, the independent feature points of multi-source astronomical data are projected on their respective exploration paths, such as the brightness feature points on the visible light data path, the temperature feature points on the infrared data path, etc. When there is a correlation or interaction between data of different modes, these exploration paths will produce intersections in space. For example, the intersection of infrared and radio data represents the mutual influence of high-energy activity and temperature changes in galaxies. The intersection corresponds to mixed or interactive astronomical data features, reflecting the complex physical connection between different modes.

[0100] Taking the example of studying galaxy evolution as an example, the above steps first require the input of visible light, infrared and radio data into the adaptive exploration space. The brightness distribution characteristics of visible light data, the temperature distribution characteristics of infrared data, and the timing signal characteristics of radio data are projected as independent astronomical data feature points, forming three corresponding exploration paths. On these paths, each independent feature point reflects the individual physical properties of the galaxy in this mode, such as the brightness change and temperature structure of the galaxy.

[0101] Next, the interaction features of the mixed infrared and radio data are projected into the same exploration space to form path intersections, which are marked as the relationship between the high-energy activity inside the galaxy and the surrounding temperature changes. The intersections represent the interaction characteristics between these modes, revealing, for example, the connection between black hole activity and star formation regions. Through these intersections, astronomers can deeply analyze the complex physical relationship between different modes.

[0102] Further optionally, in the above steps, a routing module is used to transmit the multimodal astronomical features to the corresponding analysis unit, which can be implemented as follows: inputting astronomical data feature points in multiple astronomical data exploration paths into the routing module; and allocating each astronomical data feature point to the corresponding analysis unit through the adaptive weight and routing allocation matrix in the routing module.

[0103] Specifically, in this process, the system first inputs the feature points in the multimodal astronomical data exploration path into the routing module, and then allocates each feature point to the corresponding analysis unit through the adaptive weight and routing allocation matrix in the module. The adaptive weight is used to dynamically adjust the allocation according to the importance of the data features to ensure that key features are prioritized, while the routing allocation matrix is ​​responsible for reasonably distributing the feature points of different modes to specific analysis units for in-depth modal analysis or cross-modal correlation research.

[0104] Exemplarily, the routing module is expressed as the following formula: In the above formula, F i is the i-th astronomical data feature point, W i is the modal feature weight corresponding to the i-th astronomical data feature point, T ij is the degree of fit between the i-th astronomical data feature point and the j-th analysis unit, α is a specific parameter used to adjust the sensitivity of the activation function. The lower the specific parameter value of α, the more balanced the adaptive distribution of the astronomical data feature points. represents the probability of the i-th astronomical data feature point in the routing assignment matrix being assigned to the j-th analysis unit, A j is the function of the j-th analysis unit processing the i-th astronomical data feature point received, β j is the importance weight of the j-th analysis unit in the output value, and O is the comprehensive output value corresponding to the j-th analysis unit.

[0105] The purpose of the above formula is to describe how to dynamically allocate multiple astronomical data feature points to different analysis units based on their modal feature weights and fitness through the routing module. The formula calculates the probability of each astronomical data feature point being allocated to each analysis unit through adaptive weights, specific parameters of activation function sensitivity, and routing allocation matrix. Then, each analysis unit will perform specific processing on the received feature points and output a comprehensive result based on their importance weights. This method ensures that different feature points can be efficiently and reasonably allocated to the most appropriate analysis units, improving the accuracy and efficiency of the overall analysis.

[0106] Further optionally, in the above steps, an analysis layer is used to learn the modal attribute features corresponding to each of the multimodal astronomical features from the multimodal astronomical features corresponding to each analysis unit, which can be implemented as follows: through each analysis unit, the corresponding modal attribute features are learned from the different modal astronomical features received by each analysis unit.

[0107] Here, the modal attribute features are used to reflect the corresponding astronomical data source characteristics. Exemplarily, the astronomical data source characteristics include at least one of the following: spatial distribution characteristics, time series characteristics, spectral characteristics, frequency spectrum characteristics, and cross-modal interaction characteristics.

[0108] In this scenario, modal attribute features are used to reflect the characteristics of each astronomical data source. The following are several examples to illustrate the characteristics of different astronomical data sources and their modal attribute features:

[0109] Spatial distribution characteristics: By analyzing the optical images of galaxy clusters, their distribution patterns in the universe can be obtained. Modal attribute characteristics can include the position coordinates and density distribution of galaxies. For example, by using optical modal data to analyze the distribution of galaxies in three-dimensional space, their spatial distribution characteristics can be obtained.

[0110] Timing characteristics: By analyzing the radio pulse signals emitted by pulsars, the characteristics of their periodic changes are obtained. Modal attribute characteristics can include pulse period, time delay, etc. For example, by using a radio telescope to observe the radio signals of pulsars, its timing characteristics, such as periodic changes and intensity fluctuations, can be obtained.

[0111] Spectral characteristics: By analyzing the spectral data of a star, its elemental composition, temperature, velocity and other characteristics can be obtained. Modal attribute characteristics can include the position, intensity and width of the spectral line. For example, when analyzing the spectral lines of a star, it is found that it has a strong hydrogen spectral line feature, indicating that it is mainly composed of hydrogen, which is its spectral characteristic.

[0112] Spectral characteristics: By analyzing the spectrum distribution in radio astronomy data, different radio sources in the universe can be identified. Modal attribute characteristics can include frequency range, signal strength, etc. For example, when studying cosmic microwave background radiation, its spectrum characteristics, including frequency distribution and intensity fluctuations, can be analyzed through radio spectrum data.

[0113] Cross-modal interaction characteristics: By combining data from different modes such as optical, infrared, and X-rays, the multi-band characteristics of celestial bodies are identified to reflect their cross-modal behavior. Modal attribute characteristics can include common features observed in different modes. For example, by analyzing the optical, X-ray, and radio signals of a black hole, it is found that the data from each mode all show the phenomenon of rapid accretion of matter, reflecting its cross-modal interaction characteristics.

[0114] These modal property characteristics help to gain a deep understanding of the physical and chemical properties of astronomical data sources and enhance the study and interpretation of cosmic phenomena.

[0115] In the above steps, the analysis layer is set up to perform in-depth processing on the multi-modal astronomical features received by each analysis unit. Specifically, the analysis layer extracts and learns the attribute characteristics of different modes from each analysis unit, such as brightness in the optical mode, temperature in the infrared mode, and electromagnetic wave characteristics in the radio mode. Each analysis unit learns the unique attributes of the specific modal data it receives through pattern recognition and feature extraction algorithms. Finally, the analysis layer summarizes the attribute characteristics of each mode to provide a basis for subsequent comprehensive analysis or cross-modal fusion.

[0116] Further optionally, in the above steps, the configuration layer is used to match the analysis modes corresponding to the multi-modal astronomical features based on the modal attribute features, which can be implemented as follows:

[0117] Aggregate the tag information in the modal attribute features corresponding to different data sources into multiple groups of comprehensive attribute features; each group of comprehensive attribute features is used to represent the characteristic analysis characteristics corresponding to at least one group of the multi-modal astronomical features; and construct a corresponding analysis model based on each group of comprehensive attribute features. In the embodiment of the present application, the analysis model at least includes: an analysis model prototype, analysis model parameters, and an analysis model parameter adjustment strategy.

[0118] The modal attribute features in different astronomical data sources usually carry tag information, such as observation type, data source (optical, infrared, radio, etc.) and physical phenomenon (pulsar, galaxy, black hole, etc.). The configuration layer aggregates the attribute features into comprehensive attribute features through this tag information.

[0119] Suppose there is a multimodal observation data of a pulsar, including radio data (timing characteristics), X-ray data (energy characteristics) and optical data (spectral characteristics). The modal attribute features of these data sources carry different labeling information (such as "pulse period", "high-energy radiation", "spectral line position"), and are aggregated through the configuration layer to obtain a set of comprehensive attribute features that reflect the multimodal behavior of the pulsar. Each set of comprehensive attribute features represents a specific analytical feature of the multimodal astronomical features. These features are used to determine which features are related to the target analysis task (such as astronomical object classification, feature detection). In the case of pulsars, comprehensive attribute features may include "pulse period consistency", "high-energy X-ray burst intensity" and "spectral line redshift characteristics". These features are used to represent the analytical characteristics of the celestial body, such as its timing behavior and high-energy radiation properties, to help scientists further analyze its physical properties. The configuration layer builds an appropriate analysis model for each set of data based on each set of comprehensive attribute features. The main components of the analysis model include analysis model prototype, analysis model parameters and parameter adjustment strategy. For example, for pulsars, the analysis model prototype can use a timing analysis model (such as Fourier transform) to analyze the periodicity of the pulse signal. The analysis model parameters may include the pulse period range, noise level, etc., which are used to optimize the analysis accuracy. The analysis model parameter adjustment strategy can be that if the observed data noise is high, the model can adaptively adjust the parameters, such as increasing the strength of the filter to reduce interference and improve the analysis accuracy of the signal.

[0120] It is understood that the analytical model prototype provides an analytical framework or algorithm for processing specific types of astronomical data. For example, time series analysis, spectrum analysis, spectral line fitting, etc. The analytical model parameters are used to adjust the model to adapt to different data situations. For example, different astronomical objects have different pulse periods, and the model can set different parameters to deal with these differences. The analytical model parameter adjustment strategy can dynamically adjust the model parameters based on real-time data feedback. For example, when the noise level of optical data increases, the parameters of the filter are adjusted to ensure accurate extraction of target features.

[0121] Suppose an active galactic nucleus (AGN) is being analyzed. Based on this, during the aggregation of modal attribute features, there are different marker information in the optical, infrared and radio data of the AGN, such as the radiation intensity in the optical mode, the temperature distribution in the infrared mode and the jet activity characteristics in the radio mode. These features are aggregated by configuring layers to form a set of comprehensive attribute features. These comprehensive attribute features may represent the energy release, jet activity and surrounding matter accretion characteristics of the AGN, which are very critical for studying the physical mechanism of AGN. Based on these comprehensive attribute features, a model suitable for AGN analysis is systematically constructed. The prototype of the analysis model may be a jet analysis model based on radio data, with parameters including jet intensity, jet velocity, etc. The parameter adjustment strategy may be to dynamically adapt to the background noise in the radio data to more clearly distinguish the jet characteristics.

[0122] In this way, the configuration layer builds corresponding analysis modes according to the modal attribute characteristics of different astronomical data sources, thus improving the processing and analysis capabilities of multimodal astronomical data.

[0123] As an optional embodiment, the analysis mode is adopted in 103 to perform multi-branch fusion processing on the multi-source astronomical data to obtain astronomical fusion data corresponding to the multi-source astronomical data, which can be implemented as follows:

[0124] Constructing a multi-branch processing network based on the analysis mode; the multi-branch processing network at least includes a processing branch corresponding to each of the analysis modes, and the branch structure and processing flow of the processing branch are determined by the analysis mode;

[0125] Inputting the multi-source astronomical data into the multi-branch processing network to obtain corresponding multiple groups of multi-source astronomical data analysis result data streams;

[0126] A data stream real-time fusion module is used to perform real-time fusion processing on multiple groups of multi-source astronomical data analysis result data streams according to their respective dynamic weights to obtain the astronomical fusion data; the dynamic weight is determined by the feedback score of the corresponding processing branch.

[0127] In step 103, first, based on different analysis modes, the system constructs a multi-branch processing network. Each processing branch represents an analysis mode, and its branch structure and processing flow are determined by the characteristics of the analysis mode. This allows the system to use different analysis methods for different astronomical data sources (such as optical, radio, infrared, etc.). For example, time series analysis is used for pulsar signal analysis, and spectrum analysis is used for spectral data processing. Each branch independently processes its own data source to ensure that the processing is highly targeted and accurate. Secondly, after the multi-source astronomical data is input into the multi-branch processing network, it is independently analyzed by each processing branch to generate multiple groups of analysis result data streams. These data streams correspond to the analysis results of each data source. For example, when analyzing an active galactic nucleus (AGN), the optical data passes through the spectral analysis branch, the radio data passes through the radio jet analysis branch, and the infrared data passes through the energy distribution analysis branch, and finally the analysis results under different modes are obtained. In this way, the spectral characteristics of the AGN, the intensity of the radio jet, the distribution of infrared radiation and other aspects can be obtained respectively.

[0128] Then, the data stream real-time fusion module fuses these multiple groups of analysis result data streams. The core mechanism of fusion is based on the dynamic weights of each data stream, which is determined by the feedback score of each processing branch. The feedback score can reflect the analysis accuracy and data credibility of the branch. Assuming that in the process of analyzing AGN, the signal of radio data is very clear and the feedback score is high, the data stream weight of the radio analysis branch is higher; while the infrared data has poor observation conditions, more noise, and a lower feedback score, so its weight is relatively small. Through this dynamic adjustment, the system can highlight more representative data sources in real time to avoid noise affecting the overall analysis results.

[0129] Finally, by fusing the various data streams, the system ultimately obtains comprehensive astronomical fusion data. This data combines the essence of multi-source data and provides a more comprehensive and accurate result. For example, in the analysis of AGN, the final astronomical fusion data may contain comprehensive information on the characteristics of AGN's radio jet activity, infrared radiation characteristics, and optical spectrum characteristics. The fused data can not only describe the physical mechanism of AGN more clearly, but also provide a more complete analysis of astronomical phenomena than a single data source.

[0130] When studying an active galactic nucleus, it is often necessary to combine data from different bands for comprehensive analysis. For example, optical data provides the radiation intensity and spectral characteristics of AGN; radio data provides the intensity and structure of AGN jet activity; infrared data reveals the heating of dust around AGN and the energy distribution. Optical data enters the spectral analysis model branch, radio data enters the radio jet analysis model branch, and infrared data enters the energy distribution analysis model branch. Each processing branch is analyzed independently to generate the corresponding data stream. The feedback score of the radio jet branch is high, indicating that the jet signal is strong and the analysis result is relatively reliable; the infrared data has a low score due to more noise. According to the feedback score, a higher weight is assigned to the radio jet data stream, and its analysis results are used as the main reference. Combining the characteristics of different bands, a complete analysis picture of AGN can be obtained, describing its radio jet activity, optical radiation characteristics and infrared radiation distribution. Through this multi-branch fusion analysis mode, more valuable information can be extracted from complex astronomical data, reducing the limitations of single-modal data, and enhancing the understanding of multi-modal astronomical phenomena.

[0131] In an embodiment of the present application, multi-source astronomical data are obtained; the multi-source astronomical data are all for the same observation object or the same observation area; the multi-source astronomical data are input into an adaptive exploration space model to obtain analysis modes corresponding to the multi-source astronomical data; the adaptive exploration space model is provided with configurable analysis mode parameters and analysis methods; the multi-source astronomical data are subjected to multi-branch fusion processing using the analysis mode to obtain astronomical fusion data corresponding to the multi-source astronomical data. In this solution, the problem of differences in physical properties of astronomical data in different bands is effectively solved through an adaptive exploration space model, a multi-branch processing network and a real-time fusion mechanism. While maintaining the unique characteristics of each band of data, it is possible to dynamically fuse relevant information from different data sources to achieve deep fusion analysis of multi-source astronomical data, thereby providing more accurate and comprehensive astronomical fusion data for astronomical research.

[0132] In another embodiment of the present application, an astronomical data fusion analysis system based on deep learning is also provided. Figure 2 The astronomical data fusion analysis system based on deep learning includes the following units:

[0133] A collection unit is configured to acquire multi-source astronomical data; the multi-source astronomical data are all for the same observation object or the same observation area;

[0134] An analysis unit is configured to input the multi-source astronomical data into an adaptive exploration space model to obtain analysis modes corresponding to the multi-source astronomical data; the adaptive exploration space model is provided with configurable analysis mode parameters and analysis methods;

[0135] The fusion unit is configured to adopt the analysis mode to perform multi-branch fusion processing on the multi-source astronomical data to obtain astronomical fusion data corresponding to the multi-source astronomical data.

[0136] Further optionally, the analysis unit inputs the multi-source astronomical data into the adaptive exploration space model to obtain the analysis mode corresponding to each of the multi-source astronomical data, and is configured as follows:

[0137] A multimodal feature extraction layer is used to extract multimodal astronomical features from the multi-source astronomical data; the multimodal astronomical features include at least one of the following: observed space features, observed calorific value change features, and radio time series features;

[0138] Using a projection layer, the multi-source astronomical data are projected into an adaptive exploration space respectively; the adaptive exploration space is constructed by a pre-configured astronomical space model;

[0139] Using a routing module, the multi-modal astronomical features are transmitted to corresponding analysis units;

[0140] Using an analysis layer, learning from the multi-modal astronomical features corresponding to each analysis unit to obtain the modal attribute features corresponding to each of the multi-modal astronomical features;

[0141] A configuration layer is adopted to match the analysis modes corresponding to the multi-modal astronomical features based on the modal attribute characteristics; and configurable analysis mode parameters and analysis methods are set in the analysis unit.

[0142] Further optionally, the analysis unit uses a multimodal feature extraction layer to extract multimodal astronomical features from the multi-source astronomical data, and is configured to:

[0143] Identifying data modality types contained in the multi-source astronomical data;

[0144] Generate a corresponding feature extraction branch based on the identified data modality type to obtain a feature extraction network; wherein the feature extraction branch includes at least an independent extraction branch and a mixed extraction branch;

[0145] Inputting the multi-source astronomical data into the feature extraction network;

[0146] In the feature extraction network, each independent extraction branch is used to extract a set of independent astronomical data features of the corresponding modality, and the mixed extraction branch is used to extract a set of mixed astronomical data features of at least two modalities;

[0147] The extracted independent astronomical data features and mixed astronomical data features are input into a cross-modal interaction layer, and interactive astronomical data features are extracted from the independent astronomical data features and the mixed astronomical data features; the cross-modal interaction layer is dynamically adjusted based on the structural branches in the feature extraction network.

[0148] Further optionally, the analysis unit uses a projection layer to project the multi-source astronomical data into the adaptive exploration space respectively, and is configured as follows:

[0149] Projecting independent astronomical data features, mixed astronomical data features, and interactive astronomical data features into an adaptive exploration space according to a multimodal dimension to obtain a plurality of astronomical data feature points;

[0150] Multiple astronomical data exploration paths are constructed by fitting multiple astronomical data feature points;

[0151] The intersection points between the multiple astronomical data exploration paths correspond to interactive astronomical data features or mixed astronomical data features, and the independent feature points in the multiple astronomical data exploration paths correspond to independent astronomical data features.

[0152] Further optionally, the analysis unit, using a routing module, transmits the multimodal astronomical feature to a corresponding analysis unit, and is configured as follows:

[0153] Inputting astronomical data feature points in a plurality of astronomical data exploration paths into a routing module;

[0154] Allocate each astronomical data feature point to the corresponding analysis unit through the adaptive weight and routing allocation matrix in the routing module;

[0155] Among them, the routing module is expressed as the following formula: Among them, F i is the i-th astronomical data feature point, W i is the modal feature weight corresponding to the i-th astronomical data feature point, T ij is the degree of fit between the i-th astronomical data feature point and the j-th analysis unit, α is a specific parameter used to adjust the sensitivity of the activation function. The lower the specific parameter value of α, the more balanced the adaptive distribution of the astronomical data feature points. represents the probability of the i-th astronomical data feature point in the routing assignment matrix being assigned to the j-th analysis unit, A j is the function of the j-th analysis unit processing the i-th astronomical data feature point received, β j is the importance weight of the j-th analysis unit in the output value, and O is the comprehensive output value corresponding to the j-th analysis unit.

[0156] Further optionally, the analysis unit, using the analysis layer, learns from the multi-modal astronomical features corresponding to each analysis unit to obtain the modal attribute features corresponding to each of the multi-modal astronomical features, and is configured as follows:

[0157] Through each analysis unit, corresponding modal attribute features are learned from the different modal astronomical features received; the modal attribute features are used to reflect the corresponding astronomical data source characteristics; the astronomical data source characteristics include at least one of the following: spatial distribution characteristics, time series characteristics, spectral characteristics, frequency spectrum characteristics, and cross-modal interaction characteristics;

[0158] The analysis unit, using the configuration layer, matches the analysis modes corresponding to the multi-modal astronomical features based on the modal attribute features, and is configured as follows:

[0159] Aggregating the tag information in the modal attribute features corresponding to different data sources into multiple groups of comprehensive attribute features; each group of comprehensive attribute features is used to represent the feature analysis characteristics corresponding to at least one group of the multi-modal astronomical features;

[0160] A corresponding analysis model is constructed based on each set of comprehensive attribute features; the analysis model at least includes: an analysis model prototype, analysis model parameters, and an analysis model parameter adjustment strategy.

[0161] Further optionally, the fusion unit adopts the analysis mode to perform multi-branch fusion processing on the multi-source astronomical data to obtain astronomical fusion data corresponding to the multi-source astronomical data, and is configured as follows:

[0162] Constructing a multi-branch processing network based on the analysis mode; the multi-branch processing network at least includes a processing branch corresponding to each of the analysis modes, and the branch structure and processing flow of the processing branch are determined by the analysis mode;

[0163] Inputting the multi-source astronomical data into the multi-branch processing network to obtain corresponding multiple groups of multi-source astronomical data analysis result data streams;

[0164] A data stream real-time fusion module is used to perform real-time fusion processing on multiple groups of multi-source astronomical data analysis result data streams according to their respective dynamic weights to obtain the astronomical fusion data; the dynamic weight is determined by the feedback score of the corresponding processing branch.

[0165] The system can implement various steps in the above method embodiment, which will not be expanded here.

[0166] In the embodiments of the present application, automatic fusion processing of multi-source astronomical data is achieved, providing a data basis for subsequent processing of multi-source astronomical data.

[0167] See also Figure 3 , Figure 3Schematic diagram of an electronic device provided by an embodiment of the present invention. Figure 3 As shown, an embodiment of the present invention provides an electronic device 500, including a memory 510, a processor 5102, and a computer program 511 stored in the memory 510 and executable on the processor 5102. When the processor 5102 executes the computer program 511, the following steps are implemented: acquiring multi-source astronomical data; the multi-source astronomical data are all for the same observation object or the same observation area; inputting the multi-source astronomical data into an adaptive exploration space model to obtain analysis modes corresponding to the multi-source astronomical data; the adaptive exploration space model is provided with configurable analysis mode parameters and analysis methods; using the analysis mode, multi-branch fusion processing is performed on the multi-source astronomical data to obtain astronomical fusion data corresponding to the multi-source astronomical data.

[0168] See also Figure 4 , Figure 4 A schematic diagram of an embodiment of a computer-readable storage medium provided in an embodiment of the present invention. Figure 4 As shown, this embodiment provides a computer-readable storage medium 600, on which a computer program 611 is stored. When the computer program 611 is executed by a processor, the following steps are implemented: obtaining multi-source astronomical data; the multi-source astronomical data are all for the same observation object or the same observation area; inputting the multi-source astronomical data into an adaptive exploration space model to obtain analysis modes corresponding to the multi-source astronomical data; the adaptive exploration space model is provided with configurable analysis mode parameters and analysis methods; using the analysis mode, multi-branch fusion processing is performed on the multi-source astronomical data to obtain astronomical fusion data corresponding to the multi-source astronomical data.

[0169] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and for parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0170] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0171] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0172] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0173] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0174] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0175] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for astronomical data fusion analysis based on deep learning, characterized in that: The method comprises: Acquiring multi-source astronomical data; the multi-source astronomical data are all for the same observation object or the same observation area; Inputting the multi-source astronomical data into an adaptive exploration space model to obtain analysis modes corresponding to the multi-source astronomical data; the adaptive exploration space model is provided with configurable analysis mode parameters and analysis methods; The analysis mode is adopted to perform multi-branch fusion processing on the multi-source astronomical data to obtain astronomical fusion data corresponding to the multi-source astronomical data.

2. The astronomical data fusion analysis method based on deep learning according to claim 1 is characterized in that: The step of inputting the multi-source astronomical data into the adaptive exploration space model to obtain the analysis modes corresponding to the multi-source astronomical data comprises: A multimodal feature extraction layer is used to extract multimodal astronomical features from the multi-source astronomical data; the multimodal astronomical features include at least one of the following: observed space features, observed calorific value change features, and radio time series features; Using a projection layer, the multi-source astronomical data are projected into an adaptive exploration space respectively; the adaptive exploration space is constructed by a pre-configured astronomical space model; Using a routing module, the multi-modal astronomical features are transmitted to corresponding analysis units; Using an analysis layer, learning from the multi-modal astronomical features corresponding to each analysis unit to obtain the modal attribute features corresponding to each of the multi-modal astronomical features; A configuration layer is adopted to match the analysis modes corresponding to the multi-modal astronomical features based on the modal attribute characteristics; and configurable analysis mode parameters and analysis methods are set in the analysis unit.

3. An astronomical data fusion analysis system based on deep learning, characterized in that: The system comprises the following units, wherein: A collection unit is configured to acquire multi-source astronomical data; the multi-source astronomical data are all for the same observation object or the same observation area; An analysis unit is configured to input the multi-source astronomical data into an adaptive exploration space model to obtain analysis modes corresponding to the multi-source astronomical data; the adaptive exploration space model is provided with configurable analysis mode parameters and analysis methods; The fusion unit is configured to adopt the analysis mode to perform multi-branch fusion processing on the multi-source astronomical data to obtain astronomical fusion data corresponding to the multi-source astronomical data.

4. An electronic device, characterized in that: include: Memory for storing computer software programs; A processor is used to read and execute the computer software program, thereby implementing the astronomical data fusion analysis method based on deep learning as described in any one of claims 1 or 2.

5. A non-transitory computer-readable storage medium, characterized in that: The storage medium stores a computer software program, and when the computer software program is executed by the processor, the deep learning-based astronomical data fusion analysis method as described in any one of claims 1 or 2 is implemented.

Images