A constellation recognition enhancement method and system based on deep learning

Abstract

A constellation recognition enhancement method and system based on deep learning, the method includes data collection, multi-source collaborative data enhancement, dynamic constellation modeling enhancement, constellation cognition enhancement and constellation recognition enhancement. The present invention relates to the field of intelligent stargazing technology, and adopts the method of spectral domain adversarial enhancement and multi-source data fusion to perform multi-source collaborative data enhancement, which improves the data quality while designing a constellation recognition task that combines dynamic constellation modeling and constellation cognition enhancement. It not only identifies stars and constellations, but also analyzes and presents cognitive knowledge and information of constellations; adopts a long-term and short-term convolutional model of a spatiotemporal graph combined with physical model prediction to perform dynamic constellation modeling enhancement; adopts a cross-modal constellation cognition generation model combined with a knowledge graph to perform constellation cognition enhancement.

Description

Technical Field

[0001] The present invention relates to the field of intelligent stargazing technology, and specifically to a constellation recognition enhancement method and system based on deep learning. Background Art

[0002] The constellation recognition enhancement method and system based on deep learning aims to combine multi-source data and deep learning technology, and improve the accuracy and interpretability of constellation recognition by integrating dynamic modeling of stars and cultural semantics. Through steps such as data collection, multi-source collaborative data enhancement, dynamic constellation modeling, and constellation cognitive enhancement, this method can handle the spatiotemporal changes of stars, align visual and text features across modalities, and generate personalized cultural background interpretations. Its role is not only to improve recognition accuracy, but also to provide multi-dimensional and interpretable constellation cognitive assistance to meet the needs of different user groups (such as scholars, educators, and ordinary audiences), and promote the development of constellation recognition technology in the direction of intelligence, personalization, and cross-culturality.

[0003] However, among the existing constellation recognition methods, there are existing constellation recognition methods that mainly perform recognition based on image data from two perspectives: star detection and overall constellation recognition. However, the original data involved in constellation recognition is often seriously interfered by light pollution and has low recognition accuracy under low signal-to-noise ratio conditions, which leads to poor quality of data sources. In addition, since the target output of traditional methods is often a single constellation recognition result, there is also a lack of technical problems in the diverse application of the constellation itself.

[0004] Among the existing dynamic constellation modeling enhancement methods, there are technical problems that the traditional dynamic constellation modeling method relies on pure data-driven and is difficult to capture the laws of long-period motion such as precession and seasonal changes. At the same time, due to the lack of physical law constraints, long-term prediction errors accumulate and the model has poor interpretability.

[0005] Among the existing constellation cognition enhancement methods, there is a technical problem that traditional constellation recognition methods lack cultural semantic associations, and the output results are limited to the positions and names of stars. At the same time, traditional methods are also unable to further generate scientific explanations or cultural narrative content of constellations. Summary of the invention

[0006] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a constellation recognition enhancement method and system based on deep learning. In view of the fact that in the existing constellation recognition methods, the existing constellation recognition methods mainly perform recognition based on image data from two perspectives: star detection and overall constellation recognition. However, the original data involved in constellation recognition is often seriously interfered by light pollution and has low recognition accuracy under low signal-to-noise ratio conditions, which leads to poor quality of data source. In addition, since the target output of traditional methods is often a single constellation recognition result, it also lacks the technical problem of diverse application of the constellation itself. This scheme creatively adopts the method of spectral domain adversarial enhancement and multi-source data fusion to perform multi-source collaborative data enhancement, which improves the data quality and designs a constellation recognition task combining dynamic constellation modeling and constellation cognitive enhancement. It not only recognizes stars and constellations, but also analyzes and presents cognitive knowledge and information of constellations, thereby optimizing the output diversity and result availability of constellation recognition. In view of the fact that in the existing dynamic constellation modeling enhancement methods, the traditional dynamic constellation modeling method relies on pure data drive and is difficult to In order to capture the laws of ultra-long-period motion such as precession of the equinoxes and seasonal changes, and at the same time, due to the lack of physical law constraints, long-term prediction errors accumulate and the model has poor interpretability, this solution creatively adopts a long-term and short-term space-time graph convolution model combined with physical model prediction to enhance dynamic constellation modeling. By designing a physical guidance module as a physical prior constraint, and subsequently introducing space-time graph convolution and residual correction network, the overall accuracy of constellation modeling and the interpretability of prediction results are improved; in view of the technical problems that traditional constellation recognition methods lack cultural semantic associations in existing constellation cognitive enhancement methods, and the output results are limited to the positions and names of stars, and at the same time, traditional methods cannot further generate scientific explanations or cultural narrative content of constellations, this solution creatively adopts a cross-modal constellation cognitive generation model combined with knowledge graphs to enhance constellation cognition. Through knowledge graphs and comparative learning methods, the cultural semantic fusion ability of constellation recognition is improved, and an attempt to collaboratively predict the data graph of multi-source structure of constellation recognition is provided, thereby further realizing the change and practice of intelligent stargazing from traditional constellation recognition to cognitive enhancement.

[0007] The technical solution adopted by the present invention is as follows: The present invention provides a constellation recognition enhancement method based on deep learning, which comprises the following steps:

[0008] Step S1: data collection;

[0009] Step S2: multi-source collaborative data enhancement;

[0010] Step S3: dynamic constellation modeling enhancement;

[0011] Step S4: constellation cognition enhancement;

[0012] Step S5: constellation recognition enhancement.

[0013] Furthermore, in step S1, the data collection is used to collect the original data required for constellation identification enhancement, specifically, to obtain a multi-source star image data set through multi-source data collection;

[0014] The multi-source star data set specifically includes multi-spectral astronomical image data, sensor data, environmental parameter reference data and cultural corpus data.

[0015] Further, in step S2, the multi-source collaborative data enhancement is used to comprehensively enhance and preprocess the multi-source data structure in the original data, specifically, based on the multi-source astrological data set, a multi-modal feature fusion method combined with multi-data generative adversarial training is adopted to perform multi-source collaborative data enhancement to obtain constellation recognition optimization data, including the following steps:

[0016] Step S21: spectral domain adversarial enhancement is used to optimize light-polluted and low signal-to-noise ratio images, specifically, the multi-spectral astronomical image data in the multi-source star image data set is subjected to spectral domain adversarial enhancement by adopting a generative adversarial network training method, and a spectral domain enhanced image is obtained by generation;

[0017] Step S22: time-space data alignment, which is used to time-space align the spectral domain optimized image data, the original image data and the sensor data and correct the device jitter and atmospheric refraction, specifically, to synchronize the multispectral astronomical image data, the sensor data, the environmental parameter reference data and the spectral domain enhanced image in the multi-source star image data set, and perform image correction operations to obtain aligned multimodal data;

[0018] Step S23: data generation and expansion, used to generate diverse data simulating different observation conditions, specifically, based on the aligned multimodal data, adopting a generative adversarial network to train a multi-source data generation method, to perform data generation and expansion, and to obtain expanded and enhanced starry sky image data through generation;

[0019] Step S24: multimodal feature fusion, used to construct unified features, specifically, based on the expanded and enhanced starry sky image data, multi-source feature extraction is performed through a pre-trained feature extraction model to obtain image feature data, environmental parameter feature data and sensor coding data, and image, sensor and environmental feature fusion is performed through a cross attention mechanism to obtain multi-source modal unified feature data;

[0020] Step S25: multi-source collaborative data enhancement, specifically, performing data enhancement through the spectral domain adversarial enhancement, the spatiotemporal data alignment, the data generation expansion and the multimodal feature fusion to obtain constellation recognition optimization data;

[0021] The constellation recognition optimization data specifically includes expanded and enhanced starry sky image data, aligned multi-modal data, and multi-source modal unified feature data;

[0022] The expanded and enhanced starry sky image data specifically refers to the starry sky image data that has been generated and expanded to represent light pollution correction and spectral feature restoration under multiple observation conditions;

[0023] The aligned multimodal data specifically refers to the data after the image data, sensor data and environmental data are aligned in time and space;

[0024] The multi-source modality unified feature data specifically refers to a feature set after fusing image, environment and sensor feature data.

[0025] Further, in step S3, the dynamic constellation modeling enhancement is used to construct a dynamic spatiotemporal representation of the constellation and assist constellation recognition. Specifically, according to the constellation recognition optimization data, a spatiotemporal graph convolution long-term and short-term model combined with physical model prediction is used to perform dynamic constellation modeling enhancement to obtain dynamic constellation modeling data, including the following steps:

[0026] Step S31: construct a physical guidance model to generate initialization prediction data of star positions in combination with ephemeris data, and apply the initialization prediction data to the dynamic constellation modeling process, specifically, construct a star position prediction equation based on ephemeris data and sensor data, and introduce the star position prediction equation into the long-term and short-term convolution model of the space-time graph for learning and optimization. The ephemeris data is used to provide theoretical position data of stars, including star right ascension, declination and distance parameters, and the sensor data is used to provide attitude parameters and geographic location parameters of the shooting device. The calculation formula of the star position prediction equation is:

[0027] ;

[0028] In the formula, is the output of the star position prediction equation, which is used to represent the three-dimensional coordinates of the physical predicted position of the i-th star at time t, where t is the time index and i is the star index. is the observation transformation matrix, which is specifically calculated by the attitude parameters and geographic location parameters of the camera in the sensor data, wherein the attitude parameters specifically include the pitch angle, yaw angle and roll angle, and the geographic location parameters specifically include the longitude and latitude and altitude, It is the three-dimensional coordinate of the theoretical position of the star in the ephemeris data;

[0029] Step S32: construct a spatiotemporal graph convolution model to extract the spatiotemporal relationship characteristics between celestial bodies. Specifically, a celestial body relationship graph is constructed by using a partitioning algorithm, and a standard graph convolution model and a standard time series transformer module are constructed to construct the spatiotemporal graph convolution model to obtain the spatiotemporal feature data of celestial bodies.

[0030] Step S33: construct a residual correction module for learning and optimizing the residual between the physical model and the real observation, specifically constructing a residual star position vector, and performing residual correction by constructing a convolutional long short-term memory model to obtain corrected star position data;

[0031] The residual star position vector is used to represent the error between the physical prediction and the actual observation of the star position, and the calculation formula is:

[0032] ;

[0033] In the formula, r i (t) is the residual star position vector, is the three-dimensional coordinate of the real observed star position, is the output of the star position prediction equation;

[0034] The convolutional long short-term memory model includes a convolutional local spatial feature extraction subnet and a long short-term temporal dependency extraction subnet. The convolutional local spatial feature extraction subnet is used to extract spatial features according to the residual star position vector to obtain residual spatial features. The long short-term temporal dependency extraction subnet is used to perform temporal modeling in combination with the residual spatial features to obtain a temporal and spatial hidden state.

[0035] The calculation formula for correcting the star position data is:

[0036] ;

[0037] In the formula, It is to correct the star position data. is the output of the star position prediction equation, is the final output hidden state of the space-time hidden state;

[0038] Step S34: Dynamic constellation modeling model training, specifically, through the construction of the physical guidance model, the construction of the spatiotemporal graph convolution model and the construction of the residual correction module, the dynamic constellation modeling model training is performed to obtain the dynamic constellation modeling model Model PGSTC ;

[0039] Step S35: Dynamic constellation modeling enhancement, specifically, using the dynamic constellation modeling model Model according to the constellation identification optimization data PGSTC , perform dynamic constellation modeling enhancement to obtain dynamic constellation modeling data;

[0040] The dynamic constellation modeling data specifically includes constellation identification star number data, time stamp mark data, star three-dimensional coordinate reference data and star motion trajectory reference data.

[0041] Further, in step S4, the constellation cognition enhancement is used to combine the constellation semantic knowledge with the constellation recognition, specifically, based on the dynamic constellation modeling data and the multi-source astrological data, a cross-modal constellation cognition generation model combined with a knowledge graph is used to perform constellation cognition enhancement to obtain constellation cognition auxiliary reference data, including the following steps:

[0042] Step S41: constructing a cognitive enhancement input layer, specifically, inputting the dynamic constellation modeling data and the cultural corpus data of the multi-source astrological data as raw data of the input layer to obtain an input data set;

[0043] Step S42: constructing a knowledge graph structure, specifically, constructing entities and relationships of the cultural corpus data in the input data set through entity relationship extraction technology, and obtaining constellation cultural semantic information through entity relationship construction, constructing a knowledge graph by taking the dynamic constellation modeling data as constellation physical information, storing the constellation cultural semantic information and the constellation physical information, and obtaining a constellation cognition enhanced knowledge graph structure;

[0044] Step S43: constructing a cross-modal alignment network, specifically, constructing a branch feature extraction network based on the constellation cognition enhanced knowledge graph structure, and constructing a cross-modal alignment network through comparative learning to obtain cross-modal alignment feature data; the branch feature extraction network specifically includes a physical data branch, a constellation image branch, and a text cognition branch;

[0045] The contrastive learning is specifically trained by constructing a branch contrastive loss function, and the calculation formula of the branch contrastive loss function is:

[0046] ;

[0047] Where, L cont is the branch contrast loss function, M is the total number of positive samples, m is the positive sample index, and D m is the data feature data corresponding to the mth positive sample, V m is the visual feature data corresponding to the mth positive sample, T m is the text feature data corresponding to the mth positive sample, · is the dot product operator, N is the total number of negative samples, n is the negative sample index, and D n is the data feature data corresponding to the nth negative sample, V n is the visual feature data corresponding to the nth negative sample, T n is the text feature data corresponding to the nth negative sample, is the feature similarity smoothing parameter;

[0048] Step S44: constructing a dynamic constellation cognition enhancement generation network, specifically, based on the constellation cultural semantic information and the cross-modal alignment feature data, combined with a generative artificial intelligence model interface, to assist in the generation and revision of text corpus, and obtain constellation cognition correction data;

[0049] Step S45: constructing a constellation cognition output layer, specifically, using the constellation cognition enhanced knowledge graph structure, the constellation cognition data generated according to the cross-modal alignment feature data, and the constellation cognition correction data as output data of the constellation cognition enhancement model to construct a constellation cognition output layer;

[0050] Step S46: constellation cognitive enhancement model training, specifically, by constructing the cognitive enhancement input layer, constructing the knowledge graph structure, constructing the cross-modal alignment network, constructing the dynamic constellation cognitive enhancement generation network and constructing the constellation cognitive output layer, the constellation cognitive enhancement model training is performed to obtain the constellation cognitive enhancement model Model KGCN ;

[0051] Step S47: constellation recognition enhancement, specifically, using the constellation recognition enhancement model Model according to the dynamic constellation modeling data and the constellation recognition optimization data KGCN , perform constellation cognition enhancement and obtain constellation cognition auxiliary reference data;

[0052] The constellation cognition auxiliary reference data specifically includes a structured constellation cognition knowledge graph and a natural language generated constellation cognition description; the natural language generated constellation cognition description specifically includes constellation cultural knowledge, constellation legends and constellation scientific background.

[0053] Further, in step S5, the constellation recognition enhancement is used to interactively enhance constellation recognition by integrating constellation recognition and cognitive enhancement, specifically, performing constellation recognition and modeling in the starry sky image based on the dynamic constellation modeling data, and performing auxiliary generation of constellation cognitive data based on the constellation cognitive auxiliary reference data to obtain constellation recognition enhanced reference data.

[0054] The present invention provides a constellation recognition enhancement system based on deep learning, comprising a data acquisition module, a multi-source collaborative data enhancement module, a dynamic constellation modeling enhancement module, a constellation cognition enhancement module and a constellation recognition enhancement module;

[0055] The data acquisition module is used for data acquisition, and obtains a multi-source star image data set through data acquisition, and sends the multi-source star image data set to the multi-source collaborative data enhancement module and the constellation cognition enhancement module;

[0056] The multi-source collaborative data enhancement module is used for multi-source collaborative data enhancement, obtains constellation identification optimization data through multi-source collaborative data enhancement, and sends the constellation identification optimization data to the dynamic constellation modeling enhancement module and the constellation cognition enhancement module;

[0057] The dynamic constellation modeling enhancement module is used for dynamic constellation modeling enhancement, obtains dynamic constellation modeling data through dynamic constellation modeling enhancement, and sends the dynamic constellation modeling to the constellation cognition enhancement module and the constellation identification enhancement module;

[0058] The constellation recognition enhancement module is used for constellation recognition enhancement, obtains constellation recognition auxiliary reference data through constellation recognition enhancement, and sends the constellation recognition auxiliary reference data to the constellation recognition enhancement module;

[0059] The constellation recognition enhancement module is used for constellation recognition enhancement, and obtains constellation recognition enhancement reference data through constellation recognition enhancement.

[0060] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0061] (1) Among the existing constellation recognition methods, there are two aspects: the existing constellation recognition methods mainly perform recognition based on image data from two perspectives: star detection and overall constellation recognition. However, the original data involved in constellation recognition is often seriously interfered by light pollution and has low recognition accuracy under low signal-to-noise ratio conditions, which leads to poor quality of data sources. In addition, since the target output of traditional methods is often a single constellation recognition result, there is also a lack of technical problems in the diverse application of constellations themselves. This scheme creatively adopts the method of spectral domain adversarial enhancement and multi-source data fusion to perform multi-source collaborative data enhancement. While improving data quality, it also designs a constellation recognition task that combines dynamic constellation modeling and constellation cognitive enhancement. It not only recognizes stars and constellations, but also analyzes and presents cognitive knowledge and information of constellations, optimizing the output diversity and result availability of constellation recognition.

[0062] (2) In view of the technical problems that the traditional dynamic constellation modeling enhancement methods rely on pure data-driven and are difficult to capture the long-term motion laws such as precession and seasonal changes, and the lack of physical law constraints leads to long-term prediction error accumulation and poor model interpretability, this solution creatively adopts the long-term and short-term space-time graph convolution model combined with physical model prediction to enhance dynamic constellation modeling. By designing a physical guidance module as a physical prior constraint, and subsequently introducing space-time graph convolution and residual correction network, the overall accuracy of constellation modeling and the interpretability of prediction results are improved.

[0063] (3) In view of the technical problems in the existing constellation cognition enhancement methods, the traditional constellation recognition methods lack cultural semantic associations, and the output results are limited to the positions and names of stars. At the same time, the traditional methods are also unable to further generate scientific explanations or cultural narrative content of the constellations. This scheme creatively adopts a cross-modal constellation cognition generation model combined with a knowledge graph to enhance constellation cognition. Through the knowledge graph and comparative learning methods, the cultural semantic fusion ability of constellation recognition is improved, and an attempt is also made to coordinate the prediction of the data graph of the multi-source structure of constellation recognition, thereby further realizing the transformation and practice of intelligent stargazing from traditional constellation recognition to cognitive enhancement. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 A flowchart of a constellation recognition enhancement method based on deep learning provided by the present invention;

[0065] Figure 2 A schematic diagram of a constellation recognition enhancement system based on deep learning provided by the present invention;

[0066] Figure 3 This is a schematic diagram of the process of data preprocessing in step S2;

[0067] Figure 4 This is a schematic diagram of the process of predicting potential fault hazards in step S3;

[0068] Figure 5 This is a flow chart of fault classification and location in step S4.

[0069] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0070] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0071] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0072] Example 1, see Figure 1 The present invention provides a constellation recognition enhancement method based on deep learning, which comprises the following steps:

[0073] Step S1: data collection;

[0074] Step S2: multi-source collaborative data enhancement;

[0075] Step S3: dynamic constellation modeling enhancement;

[0076] Step S4: constellation cognition enhancement;

[0077] Step S5: constellation recognition enhancement.

[0078] By performing the above operations, in view of the fact that in the existing constellation recognition methods, the existing constellation recognition methods mainly perform recognition based on image data from two perspectives: star detection and overall constellation recognition. However, the original data involved in constellation recognition are often seriously interfered by light pollution and have low recognition accuracy under low signal-to-noise ratio conditions, which leads to poor quality of data sources. In addition, since the target output of traditional methods is often a single constellation recognition result, it also lacks the technical problem of diverse application of the constellation itself. This scheme creatively adopts the method of spectral domain adversarial enhancement and multi-source data fusion to perform multi-source collaborative data enhancement, which improves the data quality. At the same time, it designs a constellation recognition task that combines dynamic constellation modeling and constellation cognitive enhancement. It not only recognizes stars and constellations, but also analyzes and presents cognitive knowledge and information of constellations, optimizing the output diversity and result availability of constellation recognition.

[0079] Example 2, see Figure 1 and Figure 2 , This embodiment is based on the above embodiment. In step S1, the data collection is used to collect the original data required for constellation recognition enhancement, specifically, to obtain a multi-source star image data set through multi-source data collection;

[0080] The multi-source star image data set specifically includes multi-spectral astronomical image data, sensor data, environmental parameter reference data and cultural corpus data;

[0081] The multispectral astronomical image data is used to provide starry sky visual information and is used for dynamic constellation modeling, and specifically includes image data and attribute data; the image data specifically includes RAW format starry sky image data, multi-band starry sky image data and time series starry sky image data; the attribute data specifically includes exposure time parameters, aperture shooting parameters, image resolution parameters and bit depth parameters;

[0082] The multi-band starry sky image data specifically includes visible light images, infrared images and ultraviolet images;

[0083] The data format of the image data includes FITS format, PNG format and JPEG format;

[0084] The data format of the attribute data is CSV format;

[0085] The sensor data is used to provide acquisition device status and environment information and is used for data enhancement, including positioning data, timestamp data, and IMU data, in CSV format;

[0086] The environmental parameter reference data is used to assist in the dynamic strategy adjustment in the constellation identification process and the image enhancement in the data enhancement process, and includes meteorological data, light pollution reference data and moon phase data; the meteorological data specifically includes temperature, humidity and air pressure data, which are in CSV format; the light pollution reference data specifically includes ambient light intensity data and light pollution level reference data, which are in CSV format; the moon phase data specifically includes moon phase information annotation data and moon position positioning data, which are in CSV format;

[0087] The cultural corpus data is used to provide a knowledge basis for cognitive enhancement of constellations, specifically referring to constellation physical property corpus, constellation myth common sense corpus and constellation astronomy common sense corpus, and is in JSON format.

[0088] Example 3, see Figure 1 , Figure 2 and Figure 3 This embodiment is based on the above embodiment. In step S2, the multi-source collaborative data enhancement is used to comprehensively enhance and preprocess the multi-source data structure in the original data. Specifically, based on the multi-source astrological data set, a multi-modal feature fusion method combined with multi-data generative adversarial training is used to perform multi-source collaborative data enhancement to obtain constellation recognition optimization data, including the following steps:

[0089] Step S21: spectral domain adversarial enhancement is used to optimize light-polluted and low signal-to-noise ratio images, specifically, the multi-spectral astronomical image data in the multi-source star image data set is subjected to spectral domain adversarial enhancement by adopting a generative adversarial network training method, and a spectral domain enhanced image is obtained by generation;

[0090] The generative adversarial network training method specifically adopts the CycleGAN architecture to perform generative adversarial training of dual generative discriminators, and optimizes the recovery of key spectral lines through the spectral attention mechanism;

[0091] The dual generation discriminator includes a light pollution image domain and an ideal observation image domain; the light pollution image domain is specifically used to generate a light polluted starry sky image; the ideal observation image domain is specifically used to generate a starry sky image under ideal observation conditions; through the generative adversarial training, the light pollution images and the ideal observation images are generated in sequence to obtain a light pollution optimized image set;

[0092] By performing spectral attention optimization on the light pollution optimized image set, a spectral domain augmented image is obtained;

[0093] The key spectral lines include H-α line, H-β line, doubly ionized oxygen line and ionized nitrogen line;

[0094] Step S22: time-space data alignment, which is used to time-space align the spectral domain optimized image data, the original image data and the sensor data and correct the device jitter and atmospheric refraction, specifically, to synchronize the multispectral astronomical image data, the sensor data, the environmental parameter reference data and the spectral domain enhanced image in the multi-source star image data set, and perform image correction operations to obtain aligned multimodal data;

[0095] Step S23: data generation and expansion, used to generate diverse data simulating different observation conditions, specifically, based on the aligned multimodal data, adopting a generative adversarial network to train a multi-source data generation method, to perform data generation and expansion, and to obtain expanded and enhanced starry sky image data through generation;

[0096] The method for generating multi-source data through generative adversarial network training specifically refers to training a generative adversarial model using a CycleGAN architecture based on image data, sensor data, and environmental data to generate and expand spatiotemporal alignment feature data;

[0097] The expanded and enhanced starry sky image data specifically includes environmental condition simulation expanded data and star motion simulation expanded data;

[0098] Step S24: multimodal feature fusion, used to construct unified features, specifically, based on the expanded and enhanced starry sky image data, multi-source feature extraction is performed through a pre-trained feature extraction model to obtain image feature data, environmental parameter feature data and sensor coding data, and image, sensor and environmental feature fusion is performed through a cross attention mechanism to obtain multi-source modal unified feature data;

[0099] Step S25: multi-source collaborative data enhancement, specifically, performing data enhancement through the spectral domain adversarial enhancement, the spatiotemporal data alignment, the data generation expansion and the multimodal feature fusion to obtain constellation recognition optimization data;

[0100] The constellation recognition optimization data specifically includes expanded and enhanced starry sky image data, aligned multi-modal data, and multi-source modal unified feature data;

[0101] The expanded and enhanced starry sky image data specifically refers to the starry sky image data that has been generated and expanded to represent light pollution correction and spectral feature restoration under multiple observation conditions;

[0102] The aligned multimodal data specifically refers to the data after the image data, sensor data and environmental data are aligned in time and space;

[0103] The multi-source modality unified feature data specifically refers to a feature set after fusing image, environment and sensor feature data.

[0104] Example 4, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the dynamic constellation modeling enhancement is used to construct a dynamic spatiotemporal representation of the constellation and assist constellation recognition. Specifically, according to the constellation recognition optimization data, a spatiotemporal graph convolution long-term and short-term model combined with physical model prediction is used to perform dynamic constellation modeling enhancement to obtain dynamic constellation modeling data, including the following steps:

[0105] Step S31: construct a physical guidance model to generate initialization prediction data of star positions in combination with ephemeris data, and apply the initialization prediction data to the dynamic constellation modeling process, specifically, construct a star position prediction equation based on ephemeris data and sensor data, and introduce the star position prediction equation into the long-term and short-term convolution model of the space-time graph for learning and optimization. The ephemeris data is used to provide theoretical position data of stars, including star right ascension, declination and distance parameters, and the sensor data is used to provide attitude parameters and geographic location parameters of the shooting device. The calculation formula of the star position prediction equation is:

[0106] ;

[0107] In the formula, is the output of the star position prediction equation, which is used to represent the three-dimensional coordinates of the physical predicted position of the i-th star at time t, where t is the time index and i is the star index. is the observation transformation matrix, which is specifically calculated by the attitude parameters and geographic location parameters of the camera in the sensor data, wherein the attitude parameters specifically include the pitch angle, yaw angle and roll angle, and the geographic location parameters specifically include the longitude and latitude and altitude, It is the three-dimensional coordinate of the theoretical position of the star in the ephemeris data;

[0108] Step S32: construct a spatiotemporal graph convolution model to extract the spatiotemporal relationship characteristics between celestial bodies. Specifically, a celestial body relationship graph is constructed by using a partitioning algorithm, and a standard graph convolution model and a standard time series transformer module are constructed to construct the spatiotemporal graph convolution model to obtain the spatiotemporal feature data of celestial bodies.

[0109] The triangulation algorithm specifically refers to a Delaunay triangulation algorithm, and the star space relationship graph structure data is constructed by adopting the Delaunay triangulation algorithm; the star space relationship graph structure data specifically includes a node set and an edge set, the node set is used to represent the star, and the edge set is used to represent the spatial adjacency relationship between the star;

[0110] Step S33: construct a residual correction module for learning and optimizing the residual between the physical model and the real observation, specifically constructing a residual star position vector, and performing residual correction by constructing a convolutional long short-term memory model to obtain corrected star position data;

[0111] The residual star position vector is used to represent the error between the physical prediction and the actual observation of the star position, and the calculation formula is:

[0112] ;

[0113] In the formula, r i (t) is the residual star position vector, is the three-dimensional coordinate of the real observed star position, is the output of the star position prediction equation;

[0114] The convolutional long short-term memory model includes a convolutional local spatial feature extraction subnet and a long short-term temporal dependency extraction subnet. The convolutional local spatial feature extraction subnet is used to extract spatial features according to the residual star position vector to obtain residual spatial features. The long short-term temporal dependency extraction subnet is used to perform temporal modeling in combination with the residual spatial features to obtain a temporal and spatial hidden state.

[0115] The calculation formula of the residual space feature is:

[0116] ;

[0117] In the formula, is the residual space feature, CNN(·) is the standard convolution feature extraction operation function, r i (t) is the residual star position vector;

[0118] The calculation formula of the time-space hidden state is:

[0119] ;

[0120] In the formula, is the final output hidden state of the space-time hidden state, LSTM(·) is the standard long short-term memory neural network function, is the residual space feature, is the time-space hidden state at the previous time t-1;

[0121] The calculation formula for correcting the star position data is:

[0122] ;

[0123] In the formula, It is to correct the star position data. is the output of the star position prediction equation, is the final output hidden state of the space-time hidden state;

[0124] Step S34: Dynamic constellation modeling model training, specifically, through the construction of the physical guidance model, the construction of the spatiotemporal graph convolution model and the construction of the residual correction module, the dynamic constellation modeling model training is performed to obtain the dynamic constellation modeling model Model PGSTC ;

[0125] Step S35: Dynamic constellation modeling enhancement, specifically, using the dynamic constellation modeling model Model according to the constellation identification optimization data PGSTC , perform dynamic constellation modeling enhancement to obtain dynamic constellation modeling data;

[0126] The dynamic constellation modeling data specifically includes constellation identification star number data, time stamp mark data, star three-dimensional coordinate reference data and star motion trajectory reference data.

[0127] By performing the above operations, in view of the technical problems in the existing dynamic constellation modeling enhancement methods, the traditional dynamic constellation modeling methods rely on pure data-driven, which are difficult to capture the ultra-long period motion laws such as precession and seasonal changes. At the same time, due to the lack of physical law constraints, long-term prediction errors accumulate and the model interpretability is poor. This scheme creatively adopts the space-time graph convolution long-term and short-term model combined with the physical model prediction to enhance the dynamic constellation modeling. By designing a physical guidance module as a physical prior constraint, and subsequently introducing the space-time graph convolution and residual correction network, the overall accuracy of constellation modeling and the interpretability of the prediction results are improved.

[0128] Example 5, see Figure 1 , Figure 2 and Figure 5This embodiment is based on the above embodiment. In step S4, the constellation cognition enhancement is used to combine the constellation semantic knowledge with the constellation recognition. Specifically, based on the dynamic constellation modeling data and the multi-source astrological data, a cross-modal constellation cognition generation model combined with a knowledge graph is used to perform constellation cognition enhancement to obtain constellation cognition auxiliary reference data, including the following steps:

[0129] Step S41: constructing a cognitive enhancement input layer, specifically, inputting the dynamic constellation modeling data and the cultural corpus data of the multi-source astrological data as raw data of the input layer to obtain an input data set;

[0130] Step S42: constructing a knowledge graph structure, specifically, constructing entities and relationships of the cultural corpus data in the input data set through entity relationship extraction technology, and obtaining constellation cultural semantic information through entity relationship construction, constructing a knowledge graph by taking the dynamic constellation modeling data as constellation physical information, storing the constellation cultural semantic information and the constellation physical information, and obtaining a constellation cognition enhanced knowledge graph structure;

[0131] Step S43: constructing a cross-modal alignment network, specifically, constructing a branch feature extraction network based on the constellation cognition enhanced knowledge graph structure, and constructing a cross-modal alignment network through comparative learning to obtain cross-modal alignment feature data; the branch feature extraction network specifically includes a physical data branch, a constellation image branch, and a text cognition branch;

[0132] The physical data branch specifically uses the three-dimensional coordinate reference data of the stars in the dynamic constellation modeling data as data feature data;

[0133] The constellation image branch specifically uses the ViT model to extract features from the constellation image to obtain visual feature data;

[0134] The text recognition branch specifically uses the BERT model constellation cultural semantic information to extract features and obtain text feature data;

[0135] The contrastive learning is specifically trained by constructing a branch contrastive loss function, and the calculation formula of the branch contrastive loss function is:

[0136] ;

[0137] Where, L cont is the branch contrast loss function, M is the total number of positive samples, m is the positive sample index, and D m is the data feature data corresponding to the mth positive sample, V m is the visual feature data corresponding to the mth positive sample, T mis the text feature data corresponding to the mth positive sample, · is the dot product operator, N is the total number of negative samples, n is the negative sample index, and D n is the data feature data corresponding to the nth negative sample, V n is the visual feature data corresponding to the nth negative sample, T n is the text feature data corresponding to the nth negative sample, is the feature similarity smoothing parameter;

[0138] Step S44: constructing a dynamic constellation cognition enhancement generation network, specifically, based on the constellation cultural semantic information and the cross-modal alignment feature data, combined with a generative artificial intelligence model interface, to assist in the generation and revision of text corpus, and obtain constellation cognition correction data;

[0139] Step S45: constructing a constellation cognition output layer, specifically, using the constellation cognition enhanced knowledge graph structure, the constellation cognition data generated according to the cross-modal alignment feature data, and the constellation cognition correction data as output data of the constellation cognition enhancement model to construct a constellation cognition output layer;

[0140] Step S46: constellation cognitive enhancement model training, specifically, by constructing the cognitive enhancement input layer, constructing the knowledge graph structure, constructing the cross-modal alignment network, constructing the dynamic constellation cognitive enhancement generation network and constructing the constellation cognitive output layer, the constellation cognitive enhancement model training is performed to obtain the constellation cognitive enhancement model Model KGCN ;

[0141] Step S47: constellation recognition enhancement, specifically, using the constellation recognition enhancement model Model according to the dynamic constellation modeling data and the constellation recognition optimization data KGCN , perform constellation cognition enhancement and obtain constellation cognition auxiliary reference data;

[0142] The constellation cognition auxiliary reference data specifically includes a structured constellation cognition knowledge graph and a natural language generated constellation cognition description; the natural language generated constellation cognition description specifically includes constellation cultural knowledge, constellation legends and constellation scientific background.

[0143] By performing the above operations, in view of the technical problems in the existing constellation cognition enhancement methods, the traditional constellation recognition methods lack cultural semantic associations, the output results are limited to the positions and names of stars, and the traditional methods are also unable to further generate scientific explanations or cultural narrative content of the constellations. This scheme creatively adopts a cross-modal constellation cognition generation model combined with a knowledge graph to enhance constellation cognition. Through the knowledge graph and comparative learning methods, the cultural semantic fusion ability of constellation recognition is improved, and an attempt is also provided for the collaborative prediction of the data graph of the multi-source structure of constellation recognition, thereby further realizing the change and practice of intelligent stargazing from traditional constellation recognition to cognitive enhancement.

[0144] Example 6, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S5, the constellation recognition enhancement is used to interactively enhance constellation recognition by integrating constellation recognition and cognitive enhancement. Specifically, constellation recognition and modeling are performed in the starry sky image based on the dynamic constellation modeling data, and constellation cognitive data is auxiliary generated based on the constellation cognitive auxiliary reference data to obtain constellation recognition enhancement reference data.

[0145] Embodiment 7, see Figure 1 and Figure 2 , this embodiment is based on the above embodiment. The present invention provides a deep learning-based constellation recognition enhancement system, including a data acquisition module, a multi-source collaborative data enhancement module, a dynamic constellation modeling enhancement module, a constellation cognition enhancement module and a constellation recognition enhancement module;

[0146] The data acquisition module is used for data acquisition, and obtains a multi-source star image data set through data acquisition, and sends the multi-source star image data set to the multi-source collaborative data enhancement module and the constellation cognition enhancement module;

[0147] The multi-source collaborative data enhancement module is used for multi-source collaborative data enhancement, obtains constellation identification optimization data through multi-source collaborative data enhancement, and sends the constellation identification optimization data to the dynamic constellation modeling enhancement module and the constellation cognition enhancement module;

[0148] The dynamic constellation modeling enhancement module is used for dynamic constellation modeling enhancement, obtains dynamic constellation modeling data through dynamic constellation modeling enhancement, and sends the dynamic constellation modeling to the constellation cognition enhancement module and the constellation identification enhancement module;

[0149] The constellation recognition enhancement module is used for constellation recognition enhancement, obtains constellation recognition auxiliary reference data through constellation recognition enhancement, and sends the constellation recognition auxiliary reference data to the constellation recognition enhancement module;

[0150] The constellation recognition enhancement module is used for constellation recognition enhancement, and obtains constellation recognition enhancement reference data through constellation recognition enhancement.

[0151] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0152] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0153] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A constellation recognition enhancement method based on deep learning, characterized by: The method comprises the following steps: Step S1: data collection, obtaining a multi-source star data set; Step S2: Multi-source collaborative data enhancement to obtain constellation recognition optimization data; Step S3: Dynamic constellation modeling enhancement, using the space-time graph convolution long-term and short-term model combined with physical model prediction to perform dynamic constellation modeling enhancement to obtain dynamic constellation modeling data, including the following steps: Step S31: Constructing a physical guidance model, constructing a star position prediction equation based on ephemeris data and sensor data, and introducing the star position prediction equation into the space-time graph convolution long-term and short-term model for learning and optimization; Step S32: Constructing a space-time graph convolution model; Step S33: Constructing a residual correction module, by constructing a residual star position vector and constructing a convolution long-term and short-term memory model to perform residual correction; Step S34: Dynamic constellation modeling model training; Step S35: Dynamic constellation modeling enhancement; Step S4: constellation cognition enhancement, using a cross-modal constellation cognition generation model combined with a knowledge graph to perform constellation cognition enhancement and obtain constellation cognition auxiliary reference data, including the following steps: Step S41: constructing a cognition enhancement input layer; Step S42: constructing a knowledge graph structure, by taking the dynamic constellation modeling data as constellation physical information, constructing a knowledge graph, storing constellation cultural semantic information and the constellation physical information, and obtaining a constellation cognition enhancement knowledge graph structure; Step S43: constructing a cross-modal alignment network, by constructing a branch feature extraction network including a physical data branch, a constellation image branch and a text cognition branch, and by contrastive learning, constructing the cross-modal alignment network; Step S44: constructing a dynamic constellation cognition enhancement generation network; Step S45: constructing a constellation cognition output layer; Step S46: constellation cognition enhancement model training; Step S47: constellation cognition enhancement; Step S5: constellation recognition enhancement, obtaining constellation recognition enhancement reference data.

2. The constellation recognition enhancement method based on deep learning according to claim 1, characterized in that: In step S1, the multi-source star image data set specifically includes multi-spectral astronomical image data, sensor data, environmental parameter reference data and cultural corpus data.

3. The constellation recognition enhancement method based on deep learning according to claim 2, characterized in that: In step S2, the multi-source collaborative data enhancement is used to comprehensively enhance and preprocess the multi-source data structure in the original data. Specifically, based on the multi-source astrological data set, a multi-modal feature fusion method combined with multi-data generative adversarial training is used to perform multi-source collaborative data enhancement to obtain constellation recognition optimization data, including the following steps: Step S21: spectral domain adversarial enhancement, specifically, performing spectral domain adversarial enhancement on the multispectral astronomical image data in the multi-source star image data set by using a generative adversarial network training method, and obtaining a spectral domain enhanced image by generation; Step S22: aligning the spatiotemporal data, specifically performing timestamp synchronization and image correction operations on the multispectral astronomical image data, sensor data, environmental parameter reference data and the spectral domain enhanced image in the multi-source star image data set to obtain aligned multimodal data; Step S23: data generation and expansion, used to generate diverse data simulating different observation conditions, specifically, based on the aligned multimodal data, adopting a generative adversarial network to train a multi-source data generation method, to perform data generation and expansion, and to obtain expanded and enhanced starry sky image data through generation; Step S24: multimodal feature fusion, used to construct unified features, specifically, based on the expanded and enhanced starry sky image data, multi-source feature extraction is performed through a pre-trained feature extraction model to obtain image feature data, environmental parameter feature data and sensor coding data, and image, sensor and environmental feature fusion is performed through a cross attention mechanism to obtain multi-source modal unified feature data; Step S25: multi-source collaborative data enhancement, specifically, performing data enhancement through the spectral domain adversarial enhancement, the spatiotemporal data alignment, the data generation expansion and the multimodal feature fusion to obtain constellation recognition optimization data; The constellation recognition optimization data specifically includes expanded and enhanced starry sky image data, aligned multi-modal data, and multi-source modal unified feature data; The expanded and enhanced starry sky image data specifically refers to the starry sky image data that has been generated and expanded to represent light pollution correction and spectral feature restoration under multiple observation conditions; The aligned multimodal data specifically refers to the data after the image data, sensor data and environmental data are aligned in time and space; The multi-source modality unified feature data specifically refers to a feature set after fusing image, environment and sensor feature data.

4. The constellation recognition enhancement method based on deep learning according to claim 3, characterized in that: In step S3, the dynamic constellation modeling enhancement is used to construct a dynamic spatiotemporal representation of the constellation and assist constellation recognition. Specifically, based on the constellation recognition optimization data, a spatiotemporal graph convolution long-term and short-term model combined with physical model prediction is used to perform dynamic constellation modeling enhancement to obtain dynamic constellation modeling data, including the following steps: Step S31: construct a physical guidance model to generate initialization prediction data of star positions in combination with ephemeris data, and apply the initialization prediction data to the dynamic constellation modeling process, specifically, construct a star position prediction equation based on ephemeris data and sensor data, and introduce the star position prediction equation into the long-term and short-term convolution model of the space-time graph for learning and optimization. The ephemeris data is used to provide theoretical position data of stars, including star right ascension, declination and distance parameters, and the sensor data is used to provide attitude parameters and geographic location parameters of the shooting device. The calculation formula of the star position prediction equation is: In the formula, is the output of the star position prediction equation, which is used to represent the three-dimensional coordinates of the physical predicted position of the i-th star at time t, where t is the time index, i is the star index, and T obs (t) is the observation transformation matrix, which is specifically calculated by the attitude parameters and geographic location parameters of the camera in the sensor data, wherein the attitude parameters specifically include the pitch angle, yaw angle and roll angle, and the geographic location parameters specifically include the latitude, longitude and altitude, It is the three-dimensional coordinate of the theoretical position of the star in the ephemeris data; Step S32: construct a spatiotemporal graph convolution model to extract the spatiotemporal relationship characteristics between celestial bodies. Specifically, a celestial body relationship graph is constructed by using a partitioning algorithm, and a standard graph convolution model and a standard time series transformer module are constructed to construct the spatiotemporal graph convolution model to obtain the spatiotemporal feature data of celestial bodies. Step S33: construct a residual correction module for learning and optimizing the residual between the physical model and the real observation, specifically constructing a residual star position vector, and performing residual correction by constructing a convolutional long short-term memory model to obtain corrected star position data; The residual star position vector is used to represent the error between the physical prediction and the actual observation of the star position, and the calculation formula is: In the formula, r i (t) is the residual star position vector, is the three-dimensional coordinate of the real observed star position, is the output of the star position prediction equation; The convolutional long short-term memory model includes a convolutional local spatial feature extraction subnet and a long short-term temporal dependency extraction subnet. The convolutional local spatial feature extraction subnet is used to extract spatial features according to the residual star position vector to obtain residual spatial features. The long short-term temporal dependency extraction subnet is used to perform temporal modeling in combination with the residual spatial features to obtain a temporal and spatial hidden state. The calculation formula for correcting the star position data is: In the formula, It is to correct the star position data. is the output of the star position prediction equation, is the final output hidden state of the space-time hidden state; Step S34: Dynamic constellation modeling model training, specifically, through the construction of the physical guidance model, the construction of the spatiotemporal graph convolution model and the construction of the residual correction module, the dynamic constellation modeling model training is performed to obtain the dynamic constellation modeling model Model PGSTC ; Step S35: Dynamic constellation modeling enhancement, specifically, using the dynamic constellation modeling model Model according to the constellation identification optimization data PGSTC , perform dynamic constellation modeling enhancement and obtain dynamic constellation modeling data.

5. The constellation recognition enhancement method based on deep learning according to claim 4, characterized in that: In step S3, the dynamic constellation modeling data specifically includes constellation identification star number data, timestamp mark data, star three-dimensional coordinate reference data and star motion trajectory reference data.

6. The constellation recognition enhancement method based on deep learning according to claim 5, characterized in that: In step S4, the constellation cognition enhancement is used to combine the constellation semantic knowledge with the constellation recognition, specifically, based on the dynamic constellation modeling data and the multi-source astrological data, a cross-modal constellation cognition generation model combined with a knowledge graph is used to perform constellation cognition enhancement to obtain constellation cognition auxiliary reference data, including the following steps: Step S41: constructing a cognitive enhancement input layer, specifically, inputting the dynamic constellation modeling data and the cultural corpus data of the multi-source astrological data as raw data of the input layer to obtain an input data set; Step S42: constructing a knowledge graph structure, specifically, constructing entities and relationships of the cultural corpus data in the input data set through entity relationship extraction technology, and obtaining constellation cultural semantic information through entity relationship construction, constructing a knowledge graph by taking the dynamic constellation modeling data as constellation physical information, storing the constellation cultural semantic information and the constellation physical information, and obtaining a constellation cognition enhanced knowledge graph structure; Step S43: constructing a cross-modal alignment network, specifically, constructing a branch feature extraction network based on the constellation cognition enhanced knowledge graph structure, and constructing a cross-modal alignment network through comparative learning to obtain cross-modal alignment feature data; the branch feature extraction network specifically includes a physical data branch, a constellation image branch, and a text cognition branch; The contrastive learning is specifically trained by constructing a branch contrastive loss function, and the calculation formula of the branch contrastive loss function is: Where, L cont is the branch contrast loss function, M is the total number of positive samples, m is the positive sample index, and D m is the data feature data corresponding to the mth positive sample, V m is the visual feature data corresponding to the mth positive sample, T m is the text feature data corresponding to the mth positive sample, · is the dot product operator, N is the total number of negative samples, n is the negative sample index, D n is the data feature data corresponding to the nth negative sample, V n is the visual feature data corresponding to the nth negative sample, T n is the text feature data corresponding to the nth negative sample, τ is the feature similarity smoothing parameter; Step S44: constructing a dynamic constellation cognition enhancement generation network, specifically, based on the constellation cultural semantic information and the cross-modal alignment feature data, combined with a generative artificial intelligence model interface, to assist in the generation and revision of text corpus, and obtain constellation cognition correction data; Step S45: constructing a constellation cognition output layer, specifically, using the constellation cognition enhanced knowledge graph structure, the constellation cognition data generated according to the cross-modal alignment feature data, and the constellation cognition correction data as output data of the constellation cognition enhancement model to construct a constellation cognition output layer; Step S46: constellation cognitive enhancement model training, specifically, by constructing the cognitive enhancement input layer, constructing the knowledge graph structure, constructing the cross-modal alignment network, constructing the dynamic constellation cognitive enhancement generation network and constructing the constellation cognitive output layer, the constellation cognitive enhancement model training is performed to obtain the constellation cognitive enhancement model Model KGCN ; Step S47: constellation recognition enhancement, specifically, using the constellation recognition enhancement model Model according to the dynamic constellation modeling data and the constellation recognition optimization data KGCN , perform constellation cognition enhancement and obtain constellation cognition auxiliary reference data.

7. The constellation recognition enhancement method based on deep learning according to claim 6, characterized in that: In step S4, the constellation cognition auxiliary reference data specifically includes a structured constellation cognition knowledge graph and a constellation cognition description generated by natural language; the constellation cognition description generated by natural language specifically includes constellation cultural knowledge, constellation legends and constellation scientific background; In step S5, the constellation recognition enhancement is used to interactively enhance constellation recognition by integrating constellation recognition and cognitive enhancement, specifically, performing constellation recognition and modeling in the starry sky image based on the dynamic constellation modeling data, and performing auxiliary generation of constellation cognitive data based on the constellation cognitive auxiliary reference data to obtain constellation recognition enhancement reference data.

8. A constellation recognition enhancement system based on deep learning, used to implement a constellation recognition enhancement method based on deep learning as claimed in any one of claims 1 to 7, characterized in that: It includes data acquisition module, multi-source collaborative data enhancement module, dynamic constellation modeling enhancement module, constellation cognition enhancement module and constellation recognition enhancement module.

9. The deep learning-based constellation recognition enhancement system according to claim 8, characterized in that: The data acquisition module is used for data acquisition, and obtains a multi-source star image data set through data acquisition, and sends the multi-source star image data set to the multi-source collaborative data enhancement module and the constellation cognition enhancement module; The multi-source collaborative data enhancement module is used for multi-source collaborative data enhancement, obtains constellation identification optimization data through multi-source collaborative data enhancement, and sends the constellation identification optimization data to the dynamic constellation modeling enhancement module and the constellation cognition enhancement module; The dynamic constellation modeling enhancement module is used for dynamic constellation modeling enhancement, obtains dynamic constellation modeling data through dynamic constellation modeling enhancement, and sends the dynamic constellation modeling to the constellation cognition enhancement module and the constellation identification enhancement module; The constellation recognition enhancement module is used for constellation recognition enhancement, obtains constellation recognition auxiliary reference data through constellation recognition enhancement, and sends the constellation recognition auxiliary reference data to the constellation recognition enhancement module; The constellation recognition enhancement module is used for constellation recognition enhancement, and obtains constellation recognition enhancement reference data through constellation recognition enhancement.

Images