StyleGAN-based satellite image reconstruction radar combined reflectivity method and system

Abstract

The invention discloses a StyleGAN-based satellite image reconstruction radar combined reflectivity method, which comprises the steps of selecting corresponding data in satellite data and radar data, performing preprocessing, and establishing a data set; the method comprises the following steps: constructing an initial model by adopting a network structure combining U-Net and StyleGAN; designing a comprehensive loss function, training the initial model according to the constructed data set, and obtaining a trained radar combined reflectivity reconstruction model; and inputting a real-time satellite image into the radar combined reflectivity reconstruction model to obtain reconstructed radar combined reflectivity. According to the method, the generative adversarial network is utilized to dynamically learn data distribution of radar combined reflectivity, and the detail expressive force of a reconstruction result is increased. Meanwhile, in combination with a style migration technology, capture of the model on small and medium-scale features is enhanced. By integrating the technologies, a set of satellite image reconstruction radar echo model with high accuracy and high detail features is constructed.

Description

Technical Field

[0001] The present invention relates to the technical field of radar echo inversion, and in particular to a method for reconstructing radar combined reflectivity from satellite images based on a style transfer generative adversarial network. Background Art

[0002] Radar reflectivity plays an important role in meteorological monitoring, especially in detecting severe weather phenomena such as severe convection, heavy rain and typhoon. Radar reflectivity can effectively infer precipitation intensity by reflecting the reflection intensity of radar waves from objects in the atmosphere, because reflectivity is closely related to precipitation. The literature [Yu Xiaoding, Zhou Xiaogang, Wang Xiuming. Advances in Thunderstorm and Severe Convective Weather Forecasting Technology [J]. Journal of Meteorology, 2012, 70(03): 311-337.] discloses that radar reflectivity can be converted into precipitation through the ZR relationship (Z is radar reflectivity, R is precipitation rate), making radar a core tool for monitoring precipitation intensity and assessing precipitation areas. In addition, the literature also discloses that radar reflectivity values ​​exceeding 35dBZ usually indicate the occurrence of severe convective weather, which may be accompanied by thunderstorms or other severe weather.

[0003] Although radar combined reflectivity is widely used in meteorological monitoring, it still faces many challenges in its practical application. The literature [Li Bai, Gu Qingtong, Li Ruiyi, et al. Analysis of the monitoring capability of new generation weather radar for severe weather and future development [J]. Meteorology, 2013, 39(03): 265-280.] points out that this technology has problems such as terrain influence, echo pollution, limited monitoring of offshore waters, and difficulty in deployment in remote areas. The literature [Zhang Zhiqing, Lu Feng, Fang Xiang, et al. Application and development of FY-4 satellite [J]. Shanghai Aerospace, 2017, 34(04): 8-19. DOI: 10.19328 / j.cnki.1006-1630.2017.04.002.] proposes that satellite data has significant advantages in coverage, all-weather monitoring, adaptability to complex terrain and multi-spectral observation, which can effectively make up for the limitations of radar. Therefore, the method of converting satellite data into radar reflectivity can expand the application of satellites in disaster monitoring and provide a wider range of radar observation data. However, the literature [Li Chao, Zhuang Xiaoran, Ma Chen, et al. Research on satellite- and radar-based fusion radar reflectivity method and its application in typhoon "Fireworks" observation [J]. Marine Forecast, 2022, 39(05): 84-93.] mentioned that the traditional satellite data reconstruction radar combined reflectivity method usually relies on artificial data feature analysis, and the design based on physical processes is difficult to accurately reflect complex nonlinear processes and data characteristics.

[0004] In recent years, with the rapid development of artificial intelligence technology, especially deep learning technology, its data-driven characteristics have made the processing of large and complex data sets more efficient, and it has gradually become a core tool in the field of remote sensing. For example, the literature [Zhou Kanghui, Zheng Yongguang, Han Lei, et al. Progress in the application of machine learning in severe convection monitoring and forecasting [J]. Meteorology, 2021, 47(03): 274-289.] discusses the application of machine learning in severe convection monitoring.

[0005] According to the research in the literature [Hilburn, KA; Ebert-Uphoff, I.; Miller, SD Development and Interpretation of a Neural Network-Based Synthetic Radar Reflectivity Estimator Using GOES-R Satellite Observations. J. Appl. Meteor. Climatol. 2020, 60, 1–21.], the model based on deep learning network structure is superior to traditional methods in reconstructing strong radar echoes. Among them, U-Net network is widely used. For example, the U-Net-based radar reflectivity reconstruction algorithm in the literature [Hilburn, KA; Ebert-Uphoff, I.; Miller, SD Development and Interpretation of a Neural Network-Based Synthetic Radar Reflectivity Estimator Using GOES-R Satellite Observations. J. Appl. Meteor. Climatol. 2020, 60, 1–21.] improves the short-term convective scale forecast of high-impact weather disasters and enhances the recognition of the location, shape and intensity of convective systems. The literature [Sun, F.; Li, B.; Min, M.; Qin, D. Deep Learning-Based Radar Composite Reflectivity Factor Estimations from Fengyun-4A Geostationary Satellite Observations. Remote Sens. 2021, 13, 2229.] reconstructs the radar composite reflectivity of China's land and offshore areas based on U-Net using FY-4A satellite observation data. The literature [Yu, X.; Lou, X.; Yan, Y.; Yan, Z.; Cheng, W.; Wang, Z.; Zhao, D.; Xia, J. Radar Echo Reconstruction in Oceanic Area via Deep Learning of Satellite Data. Remote Sens. 2023, 15, 3065.] specifically studies the reconstruction of radar combined reflectivity in the Chinese sea area based on U-Net and Himawari-8 satellite data.

[0006] However, with the deepening of research, the literature [Zhao, J.; Tan, J.; Chen, S.; Huang, Q.; Gao, L.; Li, Y.; Wei, C. Intelligent Reconstruction of Radar Composite Reflectivity Based on Satellite Observations and Deep Learning. Remote Sens. 2024, 16, 275.] pointed out that U-Net has the problem of underestimation when reconstructing strong echoes, and the reconstruction results of the overall radar echo are relatively vague and the details are not rich enough.

[0007] In view of these problems, it is necessary to conduct a more in-depth study on the deep learning network structure to meet the needs of radar combination reflectivity in the inversion process. Summary of the invention

[0008] The present invention aims to solve one of the technical problems existing in the related art at least to a certain extent.

[0009] An object of the present invention is to provide a satellite image reconstruction radar combined reflectivity method based on StyleGAN (style transfer generative adversarial network), using GAN (generative adversarial network) to dynamically learn the data distribution of radar combined reflectivity and increase the detail expression of the reconstruction result; at the same time, combined with Style Transfer (style transfer technology), the model is enhanced to capture small and medium-scale features, and a satellite image reconstruction radar echo model with high accuracy and high detail features is constructed.

[0010] Another object of the present invention is to provide a satellite image reconstruction radar combined reflectivity system based on StyleGAN.

[0011] In order to achieve the above-mentioned object, the present invention provides a method for reconstructing radar combined reflectivity from satellite images based on StyleGAN, comprising:

[0012] Select corresponding data from satellite data and radar data, perform preprocessing, and establish a data set;

[0013] The network structure combining U-Net and StyleGAN is used to build the initial model;

[0014] Design a comprehensive loss function, train the initial model according to the constructed data set, and obtain a trained radar combined reflectivity reconstruction model;

[0015] The real-time satellite image is input into the radar combined reflectivity reconstruction model to obtain the reconstructed radar combined reflectivity.

[0016] A further preferred technical solution of the present invention is that the initial model is constructed as a model framework with U-Net as the basic network structure, the output of U-Net as the generator of the generative adversarial network, and the output of the convolutional network based on style transfer as the discriminator of the generative adversarial network.

[0017] As a preferred approach, SE channel attention mechanism and ResBlock residual block are introduced into U-Net.

[0018] Preferably, the selecting corresponding data from satellite data and radar data, and preprocessing them to establish a data set includes:

[0019] According to the area where the radar combined reflectivity reconstruction method is applied, the spatial resolution and range of the area are determined, and the corresponding reconstruction application grid is established;

[0020] Determine the key factors of model input and screen satellite image data from satellite data;

[0021] Extract basic reflectivity factors from radar data in the reconstruction area, calculate radar combined reflectivity factors, and interpolate them into the reconstruction application grid to obtain radar data;

[0022] The obtained satellite image data and radar data are cleaned, quality controlled and standardized to form a data set for generating a radar combined reflectivity reconstruction model.

[0023] Preferably, the data set is divided, and the satellite image data within a period of time and the radar combined reflectivity of the corresponding time and area are defined as a data pair;

[0024] The data pairs are screened again, with the maximum value of the radar combination reflectivity as the standard, and the data pairs with the maximum value of the radar combination reflectivity lower than 15dBZ are eliminated to screen the data with weather processes;

[0025] The retained data pairs are divided into training set, validation set and test set in proportion; the model is trained with the training set to adjust the network weights of the model, the model hyperparameters are fine-tuned with the validation set, and the optimal radar combined reflectivity reconstruction model is obtained with the test set for radar combined reflectivity reconstructed with real-time satellite images.

[0026] Preferably, when training the initial model, the designed comprehensive loss function includes a style transfer loss function, a GAN discrimination loss function, and a weighted combination reflectivity factor loss function;

[0027] In the design of the style transfer loss function, the Gram matrix is ​​used to represent the style features of the image. Let X represent the feature map of the generated image and Y represent the feature map of the specified style image. Then the Gram matrix G Xand G Y Respectively expressed as:

[0028]

[0029] The style transfer loss function is expressed as:

[0030]

[0031] Where i and j represent the rows and columns in the Gram matrix, respectively, C represents the number of channels of the feature map, H and W represent the height and width of the feature map, respectively. The calculation result of the style transfer loss is the difference between the two images in the feature space. The smaller the difference, the more similar the styles are.

[0032] GAN discriminant loss function, expressed as:

[0033] Loss gan=E x～pdata(x) [logD(x)]+E x～Pg(x) [log(1-D(G(x)))]

[0034] Among them, G is the generator, D is the discriminator, Pdata(x) is the true value of the radar combination reflectivity, and Pg(x) is the reconstruction result of the radar combination reflectivity output by U-Net;

[0035] The weighted combined reflectivity factor loss function is expressed as:

[0036]

[0037] in is the reconstruction result of the radar combined reflectivity generated by the radar combined reflectivity reconstruction model, is the true value field of the radar combination reflectivity, and its weight Set to:

[0038]

[0039] Combining the weighted combination reflectivity factor loss function, style transfer loss function and GAN discriminant loss function, the final comprehensive loss function is obtained, which is expressed as

[0040] Loss=Loss weighted+Loss gan+Loss style.

[0041] Preferably, when the real-time satellite image is input into the radar combined reflectivity reconstruction model to obtain the reconstructed radar combined reflectivity, Gaussian smoothing is performed on the edges of the spliced ​​multiple reconstruction areas.

[0042] Another aspect of the present invention provides a satellite image reconstruction radar combined reflectivity system based on StyleGAN, comprising:

[0043] The data acquisition module is used to select corresponding data from satellite data and radar data, perform preprocessing, and establish a data set;

[0044] The model building module is used to build an initial model that combines U-Net and StyleGAN; and is used to design a comprehensive loss function, train the initial model according to the constructed data set, and obtain a trained radar combined reflectivity reconstruction model;

[0045] The radar combination reflectivity reconstruction module is used to input the real-time satellite image into the radar combination reflectivity reconstruction model to obtain the reconstructed radar combination reflectivity.

[0046] Another aspect of the present invention provides a non-transitory computer-readable storage medium having computer instructions stored thereon, which enable a computer to execute the above-mentioned satellite image reconstruction radar combined reflectivity method based on StyleGAN.

[0047] Yet another aspect of the present invention provides an electronic device, comprising: a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus, and the processor calls the logic instructions in the memory to execute the above-mentioned satellite image reconstruction radar combined reflectivity method based on StyleGAN.

[0048] On the other hand, the present invention provides a computer program product, comprising a computer program stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer executes the above-mentioned satellite image reconstruction radar combined reflectivity method based on StyleGAN.

[0049] Beneficial effects: The present invention proposes a method and system for reconstructing radar combined reflectivity from satellite images based on StyleGAN (style transfer generative adversarial network). The generative adversarial network is used to dynamically learn the data distribution of radar combined reflectivity, thereby increasing the detail expression of the reconstruction results. At the same time, the style transfer technology is combined to enhance the model's capture of small and medium-scale features. Compared with the U-Net model, the present invention also introduces the SE channel attention mechanism and the ResBlock residual block, which effectively improves the training stability and reconstruction accuracy. In order to cope with the problem of data imbalance, the model constructed by the present invention adopts a weighted radar echo loss function; by integrating the above technologies, a set of satellite image reconstruction radar echo models with high accuracy and high detail features is constructed. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flowchart of the method for reconstructing radar combined reflectivity from satellite images based on StyleGAN according to the present invention;

[0051] Figure 2 Schematic diagram of StyleGAN (Style Transfer Generative Adversarial Network) in an embodiment of the present invention;

[0052] Figure 3 : is a comparison chart of radar echo reconstruction results based on the U-Net model in the embodiment and the radar combined reflectivity reconstruction model constructed by the present invention; wherein Figure 3 (a) is the TS score, Figure 3 (b) Score Bias;

[0053] Figure 4 A comparison chart of radar echo reconstruction effects for a land severe convection case using a U-Net-based model in the embodiment and a radar combined reflectivity reconstruction model constructed by the present invention;

[0054] Figure 5 It is a comparison chart of radar echo reconstruction effects for a high wind case at sea by the U-Net-based model in the embodiment and the radar combined reflectivity reconstruction model constructed by the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments, and they should not be understood as limitations on the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In the description of the present invention, it should be understood that the terms used are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0056] Combine the following Figure 1-Figure 3 The present invention describes a method and system for reconstructing radar combined reflectivity from satellite images based on StyleGAN.

[0057] Embodiment 1: This embodiment provides a method for reconstructing radar combined reflectivity from satellite images based on StyleGAN. It mainly includes four parts, such as Figure 1 As shown, the specific steps are:

[0058] S1. Select corresponding data from satellite data and radar data, perform preprocessing, and establish a data set; including:

[0059] First, determine the spatial resolution and scope of the reconstruction application, and establish the corresponding application grid. Then, make a satellite image reconstruction radar combined reflectivity data set with space-time matching of satellite data and radar data. In this embodiment, the data set uses the data of the sunflower 8 satellite, and after screening, a total of 8 channels of data are selected, including 07, 08, 09, 10, 13, 14, 15, and 16 channels. These channels mainly provide mid- and lower-level water vapor and cloud cover information as key factors for model input. For radar data, the basic reflectivity factor of the S-band polarization radar is extracted from the reconstruction grid range, the radar combined reflectivity factor is calculated after quality control, and it is interpolated to each grid point of the reconstruction application grid.

[0060] After generating satellite images and radar echo data, the overall data set is cleaned, quality controlled and standardized, and finally a data set for reconstructing radar combination reflectivity is formed. The data set is divided at the same time point, with each 30 minutes as a data pair. The data pair includes satellite images and radar combination reflectivity of the corresponding time and area. The data pairs are screened again, and the maximum value of the radar combination reflectivity is used as the standard. The data pairs with the maximum value of the radar combination reflectivity lower than 15dBZ are eliminated, and data with weather processes are screened. Finally, the training set, validation set and test set are established for the construction of the model for satellite image reconstruction of radar combination reflectivity.

[0061] S2. Use the network structure combining U-Net and StyleGAN to build the initial model.

[0062] In this embodiment, the initial model is based on the U-Net deep learning model and is established using the Style Transfer-based Generative Adversarial Network (StyleGAN) method as a learning framework to solve the problem of small-scale information loss that is prone to occur in the U-Net reconstruction model.

[0063] In addition, in order to improve U-Net's feature extraction capabilities for multi-channel satellite data and better distinguish the importance of features between different channels for reconstructing radar echoes, the SE (Squeeze-and-Excitation) channel attention mechanism is introduced. At the same time, the effective and commonly used ResBlock residual block in convolution is added to increase the stability of training.

[0064] Finally, if Figure 2 As shown in the figure, the initial model is constructed as a model framework with SEResUnet as the basic network structure, the output of SEResUnet as the generator of the generative adversarial network, and the output of the convolutional network based on style transfer as the discriminator of the generative adversarial network.

[0065] S3. Design a comprehensive loss function, train the initial model according to the constructed data set, and obtain a trained radar combined reflectivity reconstruction model. The designed comprehensive loss function includes style transfer loss function, GAN discriminant loss function and weighted combined reflectivity factor loss function.

[0066] In the design of the style transfer loss function, the style loss is a loss function used to measure the style similarity between the generated image and the specified style image. This embodiment uses the Gram matrix to represent the style features of the image. The calculation method of the Gram matrix is ​​to regard the number of channels of the feature map as a dimension, reshape the feature map into a two-dimensional matrix, and then calculate the product of the transposed matrix of the matrix and itself, and the result is the Gram matrix.

[0067] Let X represent the feature map of the generated image and Y represent the feature map of the specified style image, then the Gram matrix G X and G Y Respectively expressed as:

[0068] G X =XX T

[0069] G Y =YY T

[0070] The style transfer loss function is expressed as:

[0071]

[0072] Among them, i and j represent the rows and columns in the Gram matrix respectively, C represents the number of channels of the feature map, H and W represent the height and width of the feature map respectively; the calculation result of the style transfer loss is the difference between the two images in the feature space. The smaller the difference, the more similar the styles are.

[0073] GAN is one of the deep learning methods, which mainly consists of two parts, namely the generator G (Generator) with an encoding-decoding structure and the discriminator D (Discriminator). The generator G is a radar echo reconstruction model. By learning the distribution of the real observed image, the reconstructed echo is made more realistic, hoping to deceive the discriminator D. The discriminator D needs to distinguish the true and false of the reconstructed echo and the real echo input from the generator, and identify the real echo as correctly as possible. In this way, a kind of adversarial training is formed between the generator G and the discriminator D. Therefore, GAN is optimized and trained by the adversarial loss of the generated network.

[0074] GAN discriminant loss function, expressed as:

[0075] Lossgan=E x～Pdata(x)[logD(x)]+E x～Pg(x) [log(1-D(G(x)))]

[0076] Among them, G is the generator, D is the discriminator, Pdata(x) is the true value of the radar combination reflectivity, and Pg(x) is the radar combination reflectivity reconstruction result output by U-Net.

[0077] The weighted combined reflectivity factor loss function is a loss function designed based on the distribution of radar combined reflectivity factor data. It aims to optimize the reconstruction effect of the model for radar echoes with higher thresholds but fewer numbers in a weighted manner.

[0078] The weighted combined reflectivity factor loss function is expressed as:

[0079]

[0080] in is the reconstruction result of the radar combined reflectivity generated by the radar combined reflectivity reconstruction model, is the true value field of the radar combination reflectivity, and its weight Set to:

[0081]

[0082] Combining the weighted combination reflectivity factor loss function, style transfer loss function and GAN discriminant loss function, the final comprehensive loss function is obtained, which is expressed as

[0083] Loss=Loss weighted+Loss gan+Loss style.

[0084] S4. After the model is built, the optimal reconstruction model is obtained through model training and verification based on the constraints of the loss function. After the real-time satellite image is input into the model, the radar combined reflectivity can be reconstructed in real time every 10 minutes. Since the resolution of the model input is fixed, when the model is directly applied to the reconstruction of radar echoes in a larger range, regional splicing problems may be encountered. Therefore, during the reconstruction process, the edges of the spliced ​​multiple reconstruction areas need to be Gaussian smoothed to ensure a smooth transition and avoid edge discontinuity problems.

[0085] In order to reflect the improvement effect of the algorithm of the present invention, the basic U-Net was selected as a control, and the radar echo reconstruction experiments of the offshore typhoon case and the land severe convection case were carried out.

[0086] Two schemes are used for comparison. Scheme 1: Use basic U-Net to reconstruct radar echoes based on the sunflower 8 satellite. Scheme 2: Use the satellite image reconstruction radar combined reflectivity model based on style transfer generative adversarial network (StyleGAN) constructed in this embodiment to reconstruct radar echoes using satellite data from the sunflower 8 satellite.

[0087] The radar echo reconstruction results in September 2022 were selected for evaluation, and the reconstruction effect with radar echo observations on land was measured according to the TS scores and Bias scores of 15dZ, 25dBZ, 35dBZ and 45dBZ. The calculation formulas for TS scores and Bias scores are:

[0088] TS=NA / NA+NB+NC

[0089] BIAS=NA+NB / NA+NC

[0090] Among them, NA is the number of correct radar echo reconstruction, NB is the number of false alarms in radar echo reconstruction, and NC is the number of missed alarms in radar echo reconstruction. The closer TS is to 1, the better the prediction effect.

[0091] Evaluation of radar echo reconstruction effect Figure 3 As shown, the method of this embodiment has an excellent reconstruction effect for the area where the radar echo is greater than 35dBZ, the TS score is improved to 154%, and the Bias score above 45dBZ is improved from 0.67 to 0.97. The forecast results have a good balance between false alarms and missed alarms.

[0092] In addition, choose Figure 4 An example of severe convection over land on September 13, 2022 is shown, as well as Figure 5 The effect is demonstrated by using a case of a typhoon at sea on July 10, 2022. In the case, the method of this embodiment has outstanding performance on the structure of radar echo and above the 35dBZ threshold.

[0093] The comparison of prediction results of various schemes shows that the method of the present invention has excellent performance.

[0094] Embodiment 2: This embodiment provides a satellite image reconstruction radar combined reflectivity system based on StyleGAN, including:

[0095] The data acquisition module is used to select corresponding data from satellite data and radar data, perform preprocessing, and establish a data set;

[0096] The model building module is used to build an initial model that combines U-Net and StyleGAN; and is used to design a comprehensive loss function, train the initial model according to the constructed data set, and obtain a trained radar combined reflectivity reconstruction model;

[0097] The radar combination reflectivity reconstruction module is used to input the real-time satellite image into the radar combination reflectivity reconstruction model to obtain the reconstructed radar combination reflectivity.

[0098] Embodiment 3: This embodiment provides a non-transitory computer-readable storage medium having computer instructions stored thereon, the computer instructions causing a computer to execute a satellite image reconstruction radar combined reflectivity method based on StyleGAN, the method comprising the following steps:

[0099] The data acquisition module is used to select corresponding data from satellite data and radar data, perform preprocessing, and establish a data set;

[0100] The model building module is used to build an initial model that combines U-Net and StyleGAN; and is used to design a comprehensive loss function, train the initial model according to the constructed data set, and obtain a trained radar combined reflectivity reconstruction model;

[0101] The radar combination reflectivity reconstruction module is used to input the real-time satellite image into the radar combination reflectivity reconstruction model to obtain the reconstructed radar combination reflectivity.

[0102] Embodiment 4: This embodiment provides an electronic device, which may include: a processor, a communications interface, a memory, and a communication bus, wherein the processor, the communications interface, and the memory communicate with each other through the communication bus. The processor may call the logic instructions in the memory to execute a satellite image reconstruction radar combined reflectivity method based on StyleGAN, and the method includes the following steps:

[0103] The data acquisition module is used to select corresponding data from satellite data and radar data, perform preprocessing, and establish a data set;

[0104] The model building module is used to build an initial model that combines U-Net and StyleGAN; and is used to design a comprehensive loss function, train the initial model according to the constructed data set, and obtain a trained radar combined reflectivity reconstruction model;

[0105] The radar combination reflectivity reconstruction module is used to input the real-time satellite image into the radar combination reflectivity reconstruction model to obtain the reconstructed radar combination reflectivity.

[0106] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0107] Embodiment 5: This embodiment provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute a satellite image reconstruction radar combined reflectivity method based on StyleGAN, and the method includes the following steps:

[0108] The data acquisition module is used to select corresponding data from satellite data and radar data, perform preprocessing, and establish a data set;

[0109] The model building module is used to build an initial model that combines U-Net and StyleGAN; and is used to design a comprehensive loss function, train the initial model according to the constructed data set, and obtain a trained radar combined reflectivity reconstruction model;

[0110] The radar combination reflectivity reconstruction module is used to input the real-time satellite image into the radar combination reflectivity reconstruction model to obtain the reconstructed radar combination reflectivity.

[0111] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0112] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A satellite image reconstruction radar combined reflectivity method based on StyleGAN, characterized in that: include: Select corresponding data from satellite data and radar data, perform preprocessing, and establish a data set; The network structure combining U-Net and StyleGAN is used to build the initial model; Design a comprehensive loss function, train the initial model according to the constructed data set, and obtain a trained radar combined reflectivity reconstruction model; The real-time satellite image is input into the radar combined reflectivity reconstruction model to obtain the reconstructed radar combined reflectivity.

2. The satellite image reconstruction radar combined reflectivity method based on StyleGAN according to claim 1, characterized in that: The initial model is constructed as a model framework with U-Net as the basic network structure, the output of U-Net as the generator of the generative adversarial network, and the output of the convolutional network based on style transfer as the discriminator of the generative adversarial network.

3. The satellite image reconstruction radar combined reflectivity method based on StyleGAN according to claim 1, characterized in that: The SE channel attention mechanism and ResBlock residual block are introduced in U-Net.

4. The satellite image reconstruction radar combined reflectivity method based on StyleGAN according to claim 1, characterized in that: The method of selecting corresponding data from satellite data and radar data, performing preprocessing, and establishing a data set includes: According to the area where the radar combined reflectivity reconstruction method is applied, the spatial resolution and range of the area are determined, and the corresponding reconstruction application grid is established; Determine the key factors of model input and screen satellite image data from satellite data; Extract basic reflectivity factors from radar data in the reconstruction area, calculate radar combined reflectivity factors, and interpolate them into the reconstruction application grid to obtain radar data; The obtained satellite image data and radar data are cleaned, quality controlled and standardized to form a data set for generating a radar combined reflectivity reconstruction model.

5. The satellite image reconstruction radar combined reflectivity method based on StyleGAN according to claim 4 is characterized in that: It also includes dividing the data set, defining the satellite image data within a period of time and the radar combined reflectivity of the corresponding time and area as a data pair; The data pairs are screened again, with the maximum value of the radar combination reflectivity as the standard, and the data pairs with the maximum value of the radar combination reflectivity lower than 15dBZ are eliminated to screen the data with weather processes; Divide the reserved data pairs into training set, validation set and test set in proportion; The model is trained with the training set to adjust the network weights of the model, the model hyperparameters are fine-tuned with the validation set, and the optimal radar combined reflectivity reconstruction model is obtained with the test set for radar combined reflectivity reconstructed with real-time satellite images.

6. The satellite image reconstruction radar combined reflectivity method based on StyleGAN according to claim 2, characterized in that: When training the initial model, the designed comprehensive loss function includes style transfer loss function, GAN discriminant loss function and weighted combination reflectivity factor loss function; In the design of the style transfer loss function, the Gram matrix is ​​used to represent the style features of the image. Let X represent the feature map of the generated image and Y represent the feature map of the specified style image. Then the Gram matrix G X and G " Respectively expressed as: The style transfer loss function is expressed as: Where i and j represent the rows and columns in the Gram matrix, respectively, C represents the number of channels of the feature map, H and W represent the height and width of the feature map, respectively. The calculation result of the style transfer loss is the difference between the two images in the feature space. The smaller the difference, the more similar the styles are. GAN discriminant loss function, expressed as: Loss gan E (～P+,-,(() [log D(x)]+E (～P0(() [log(1-DG(x)))] Among them, G is the generator, D is the discriminator, Pdata(x) is the true value of the radar combination reflectivity, and Pg(x) is the reconstruction result of the radar combination reflectivity output by U-Net; The weighted combined reflectivity factor loss function is expressed as: in is the reconstruction result of the radar combined reflectivity generated by the radar combined reflectivity reconstruction model, is the true value field of the radar combined reflectivity, and its weight Set to: Combining the weighted combination reflectivity factor loss function, style transfer loss function and GAN discriminant loss function, the final comprehensive loss function is obtained, which is expressed as Loss=Loss weighted+Loss gan+Loss style.

7. The satellite image reconstruction radar combined reflectivity method based on StyleGAN according to claim 1, characterized in that: When the real-time satellite image is input into the radar combined reflectivity reconstruction model to obtain the reconstructed radar combined reflectivity, Gaussian smoothing is performed on the edges of the spliced ​​multiple reconstruction areas.

8. A satellite image reconstruction radar combined reflectivity system based on StyleGAN, characterized in that: include: The data acquisition module is used to select corresponding data from satellite data and radar data, perform preprocessing, and establish a data set; The model building module is used to build an initial model that combines U-Net and StyleGAN; and is used to design a comprehensive loss function, train the initial model according to the constructed data set, and obtain a trained radar combined reflectivity reconstruction model; The radar combination reflectivity reconstruction module is used to input the real-time satellite image into the radar combination reflectivity reconstruction model to obtain the reconstructed radar combination reflectivity.

9. A non-transitory computer-readable storage medium having computer instructions stored thereon, the computer instructions enabling a computer to execute the satellite image reconstruction radar combined reflectivity method based on StyleGAN as described in any one of claims 1-7.

10. An electronic device, comprising: A processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus, and the processor calls the logic instructions in the memory to execute the satellite image reconstruction radar combined reflectivity method based on StyleGAN according to any one of claims 1 to 7.

Images