Multi-source satellite data on-orbit intelligent fusion method, device, equipment and medium
Through the in-orbit intelligent fusion method, deep learning and Kalman filters are used to fusion multi-source satellite data in real time, solving the timeliness problem in traditional methods and improving the accuracy and robustness of the data.
Patent Information
- Application Number
- CN202510133895.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-06
- Publication Date
- 2025-06-13
AI Technical Summary
The traditional multi-source satellite data fusion method has timeliness problems, and the data transmission delay and low ground processing efficiency lead to limited accuracy and robustness of the fusion results.
Multi-source satellite data in orbit intelligent fusion method is adopted, and historical satellite cloud map data are obtained for preprocessing, deep learning prediction models are used to extract features and state prediction, and combined with Kalman filter to fuse the prediction data and observation data to generate the final satellite cloud map data.
It improves the real-time fusion capability of multi-source satellite data, reduces the impact of noise on fusion results, and enhances the accuracy and reliability of fusion results.
Smart Images
Figure CN120147795A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the technical field of satellite data processing, and particularly to a method, device, equipment and medium for intelligent on-orbit fusion of multi-source satellite data.
Background Art
[0002] With the continuous development of space technology, the fusion of multi-source satellite data has become a key means to improve the accuracy and robustness of remote sensing information. The traditional multi-source satellite data fusion adopts the working mode of "satellite data acquisition and downlink + ground fusion processing", however, this mode has obvious timeliness problems.
[0003] On the one hand, due to the limitation of the construction of ground stations outside China, the data transmitted from the satellite to the ground station may be affected by various factors during the process, resulting in data transmission delay. On the other hand, ground fusion processing requires a large amount of time and computing resources, especially when dealing with large-scale multi-source satellite data, the efficiency is often low.
Summary of the Invention
[0004] The embodiments of this application provide a method, device, equipment and medium for intelligent on-orbit fusion of multi-source satellite data, aiming to solve many technical problems existing in the related technologies.
[0005] In a first aspect, the embodiments of this application provide a method for intelligent on-orbit fusion of multi-source satellite data, including:
[0006] Respectively obtain historical satellite cloud map data collected by multiple satellite remote sensors;
[0007] Preprocess each of the historical satellite cloud map data to obtain processed historical satellite cloud map data;
[0008] Use a target deep learning prediction model to perform feature extraction and state prediction on the processed historical satellite cloud map data, and predict the satellite cloud map prediction data corresponding to each satellite remote sensing at the current moment;
[0009] Obtain the satellite cloud map observation data corresponding to the current moment collected by the multiple satellite remote sensors;
[0010] Fuse the satellite cloud map prediction data and the satellite cloud map observation data through a Kalman filter to obtain the final satellite cloud map data corresponding to the current moment.
[0011] In one embodiment, optionally, the method further includes:
[0012] Calculate the error rate corresponding to the target deep learning prediction model according to the final cloud map satellite data and the satellite cloud map prediction data;
[0013] When the error rate is greater than the error rate threshold, retrain the target deep learning prediction model.
[0014] In one embodiment, optionally, the preprocessing of each piece of historical satellite cloud map data to obtain processed historical satellite cloud map data includes:
[0015] Perform normalization processing on each piece of historical satellite cloud map data to map historical satellite cloud map data of different magnitudes into a preset interval, obtaining processed historical satellite cloud map data.
[0016] In one embodiment, optionally, the method further includes:
[0017] Determine the data type and data fusion requirements of the historical satellite cloud map data collected by each satellite remote sensing;
[0018] According to the data type and the data fusion requirements, select a corresponding target deep learning prediction model from multiple deep learning prediction models, where the multiple deep learning prediction models include: convolutional neural network, recurrent neural network, long short-term memory network.
[0019] In one embodiment, optionally, the fusion of the satellite cloud map prediction data and the satellite cloud map observation data through a Kalman filter to obtain the final satellite cloud map data corresponding to the current moment includes:
[0020] According to each piece of satellite cloud map prediction data and the satellite cloud map observation data, calculate the first error covariance corresponding to each satellite cloud map observation data and the second error covariance corresponding to each piece of satellite cloud map prediction data respectively;
[0021] According to each piece of satellite cloud map prediction data, the satellite cloud map observation data, the first error covariance, and the second error covariance, fuse to obtain the final satellite cloud map data corresponding to the current moment.
[0022] In one embodiment, optionally, according to each piece of satellite cloud map prediction data and the satellite cloud map observation data, calculating the first error covariance corresponding to each satellite cloud map observation data and the second error covariance corresponding to each piece of satellite cloud map prediction data respectively includes:
[0023] According to each piece of satellite cloud map prediction data, calculate the average value of satellite cloud map predictions corresponding to all satellite remote sensing;
[0024] According to each piece of satellite cloud map observation data, calculate the average value of satellite cloud map observations corresponding to all satellite remote sensing;
[0025] According to each of the satellite cloud map observation data and the satellite cloud map observation average value, calculate the first error covariance corresponding to each satellite cloud map observation data by using a first calculation formula;
[0026] According to each of the satellite cloud map prediction data and the satellite cloud map prediction average value, calculate the second error covariance corresponding to each satellite cloud map prediction data by using a second calculation formula;
[0027] In one embodiment, optionally, according to each of the satellite cloud map prediction data, the satellite cloud map observation data, the first error covariance, and the second error covariance, fuse to obtain the final satellite cloud map data corresponding to the current moment, including:
[0028] According to the first error covariance and the second error covariance, calculate the first weight corresponding to each satellite cloud map prediction data and the second weight corresponding to each satellite cloud map observation data by using a third calculation formula and a fourth calculation formula respectively;
[0029] Perform weighted summation according to each of the satellite cloud map prediction average value, the satellite cloud map observation average value, the first weight, and the second weight, and fuse to obtain the final satellite cloud map data corresponding to the current moment.
[0030] In a second aspect, an embodiment of the present application provides a multi-source satellite data on-orbit intelligent fusion device, including:
[0031] A first acquisition module, configured to respectively acquire historical satellite cloud map data collected by multiple satellite remote sensors;
[0032] A preprocessing module, configured to preprocess each of the historical satellite cloud map data to obtain processed historical satellite cloud map data;
[0033] A prediction module, configured to perform feature extraction and state prediction on the processed historical satellite cloud map data by using a target deep learning prediction model, and predict the satellite cloud map prediction data corresponding to each satellite remote sensing at the current moment;
[0034] A second acquisition module, configured to acquire the satellite cloud map observation data corresponding to the current moment collected by the multiple satellite remote sensors;
[0035] A fusion module, configured to fuse the satellite cloud map prediction data and the satellite cloud map observation data through a Kalman filter to obtain the final satellite cloud map data corresponding to the current moment.
[0036] In a third aspect, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above multi-source satellite data on-orbit intelligent fusion method are implemented.
[0037] In a fourth aspect, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned multi-source satellite data on-orbit intelligent fusion method are implemented.
[0038] In the solutions implemented by the above-mentioned multi-source satellite data on-orbit intelligent fusion method, device, equipment and medium, historical satellite cloud map data collected by multiple satellite remote sensors are respectively obtained; preprocessing is performed on each of the historical satellite cloud map data to obtain processed historical satellite cloud map data; a target deep learning prediction model is used to perform feature extraction and state prediction on the processed historical satellite cloud map data, and satellite cloud map prediction data corresponding to each satellite remote sensor at the current moment is predicted; satellite cloud map observation data corresponding to the current moment collected by the multiple satellite remote sensors is obtained; the satellite cloud map prediction data and the satellite cloud map observation data are fused through a Kalman filter to obtain the final satellite cloud map data corresponding to the current moment. Through the above technical solutions of the present invention, on the one hand, deep learning can perform feature extraction on multi-source satellite data, provide more accurate prior information for the ensemble Kalman filter, and reduce the influence of noise on the fusion result. On the other hand, the ensemble Kalman filter can optimize and correct the results of deep learning, improving the reliability of the fusion result.
Description of the Drawings
[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.
[0040] Figure 1 Shows a schematic flow chart of a multi-source satellite data on-orbit intelligent fusion method according to an embodiment of the present application.
[0041] Figure 2 Shows a schematic flow chart of step S105 in a multi-source satellite data on-orbit intelligent fusion method according to an embodiment of the present application.
[0042] Figure 3 Shows a schematic flow chart of step S201 in a multi-source satellite data on-orbit intelligent fusion method according to an embodiment of the present application.
[0043] Figure 4 Shows a schematic flow chart of step S202 in a multi-source satellite data on-orbit intelligent fusion method according to another embodiment of the present application.
[0044] Figure 5The block diagram of a multi-source satellite data on-orbit intelligent fusion device according to an embodiment of the present application is shown.
[0045] Figure 6 The block diagram of a computer device according to an embodiment of the present application is shown.
Detailed implementation manners
[0046] In order to better understand the technical solution of the present application, the embodiments of the present application will be described in detail below with reference to the accompanying drawings.
[0047] It should be clear that the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts belong to the scope of protection of the present application.
[0048] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms of "a", "the", and "said" used in the embodiments of the present application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise.
[0049] The following will describe some embodiments of the present application in detail with reference to the accompanying drawings. Without conflict, the following embodiments and the features in the embodiments can be combined with each other.
[0050] It should be noted that the embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence is the theory, method, technology and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.
[0051] Artificial intelligence basic technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. Artificial intelligence software technologies mainly include several major directions such as computer vision technology, robotics, biometric technology, speech processing technology, natural language processing technology, and machine learning / deep learning.
[0052] Please refer to Figure 1 , Figure 1 The schematic flowchart of a multi-source satellite data on-orbit intelligent fusion method according to an embodiment of the present application is shown.
[0053] As Figure 1 shown, the multi-source satellite data on-orbit intelligent fusion method includes:
[0054] Step S101, obtain historical satellite cloud map data collected by multiple satellite remote sensors respectively;
[0055] In this step, various types of original satellite data can be obtained from different satellite sensors, such as optical remote sensing data, microwave remote sensing data, etc. These data sources are diverse and have characteristics such as different resolutions, observation angles, and band ranges.
[0056] Among them, the historical satellite cloud map data can be the satellite cloud map data of multiple satellite remote sensors at time t - 1 (assuming the number is N). It can be assumed that the satellite remote sensing data from each source is a set member, and each set member is represented by Xi, where the value of i ranges from 1 to N.
[0057] Step S102, preprocess each of the historical satellite cloud map data to obtain the preprocessed historical satellite cloud map data;
[0058] In one embodiment, optionally, step S102 includes:
[0059] Perform normalization processing on each of the historical satellite cloud map data to map historical satellite cloud map data of different magnitudes into a preset interval, thereby obtaining the preprocessed historical satellite cloud map data.
[0060] In this step, the preprocessing of the data includes but is not limited to normalization processing, that is, mapping historical satellite cloud map data of different magnitudes into a preset interval to obtain the preprocessed historical satellite cloud map data, which is convenient for subsequent training of the deep learning model and processing of the ensemble Kalman filter. Common normalization methods include min - max normalization, etc. For example, the cloud map data can be uniformly processed into numbers between [0, 1]. 0 represents no cloud, 1 represents having cloud and the cloud layer thickness reaching the maximum value, and a number between 0 and 1 represents having cloud but the cloud layer thickness being in the middle.
[0061] Of course, the preprocessing can also be other processing. For example, it may include uniformly adjusting the data format, such as converting data of different formats into a standard format suitable for subsequent processing (such as matrix form, etc.). Another example is to perform data cleaning to remove obvious incorrect data points, such as outliers caused by sensor failures and other reasons.
[0062] Step S103, use the target deep learning prediction model to perform feature extraction and state prediction on the preprocessed historical satellite cloud map data, and predict the satellite cloud map prediction data corresponding to each satellite remote sensing at the current moment;
[0063] In one embodiment, optionally, the method further includes:
[0064] Determine the data type and data fusion requirements of the historical satellite cloud map data collected by each satellite remote sensing;
[0065] Select the corresponding target deep learning prediction model from multiple deep learning prediction models according to the data type and the data fusion requirements, where the multiple deep learning prediction models include: convolutional neural network, recurrent neural network, long short-term memory network.
[0066] In this embodiment, an appropriate deep learning model architecture can be selected, such as a convolutional neural network (CNN), a recurrent neural network (RNN) and its variants (such as long short-term memory network LSTM, gated recurrent unit GRU, etc.), which is determined according to the specific characteristics of satellite data and the fusion target. Initialize the selected deep learning model, including determining the number of layers of the network, the number of neurons in each layer, the selection of activation functions (such as ReLU, Sigmoid, etc.) and initializing the weight parameters of the model (usually using random initialization or initialization methods based on specific distributions).
[0067] After that, the preprocessed satellite data from each path can be input into the deep learning model for training. During the training process, the model continuously adjusts its weight parameters to learn to extract representative features from the original satellite data. Use labeled data (if available) or adopt unsupervised learning methods (such as autoencoders, etc.) to let the model automatically discover the internal structure and feature patterns in the data. For example, the CNN model may learn features such as textures and edges in satellite images, while the RNN-based models may capture the changing patterns of data in the time series, etc.
[0068] The trained deep learning model outputs the extracted feature vectors for the satellite data from each path. Compared with the original satellite data, these feature vectors have a more compact representation form and contain information that is more valuable for subsequent fusion and analysis. For example, determine the state variable of the cloud map data, use the cloud map data between [0,1] as the cloud layer thickness state variable, and represent it as Xi,t-1. Then use the deep learning prediction model to predict the satellite cloud map at time t, and obtain the state variable Xi,t|t-1 of the cloud map at time t predicted based on time t-1.
[0069] The deep learning model is selected in the present invention because the deep learning model has the following advantages:
[0070] Deep learning has powerful feature extraction capabilities, can improve the data representation method, and provide accurate input for data fusion. Deep learning can automatically learn the features in the data without manual feature engineering, greatly reducing the cost of manual intervention. For example, in an image recognition task, a convolutional neural network (CNN) can automatically learn features in the image, such as edges, textures, colors, etc., thereby improving the accuracy of image recognition.
[0071] Deep learning is highly adaptable and can adapt to different data types and tasks. By adjusting the network structure and parameters, it can flexibly adapt to different application scenarios. For example, in multi-source satellite data fusion, according to the data characteristics of different satellites and task requirements, appropriate deep learning models and parameters can be selected to improve the accuracy and timeliness of data fusion.
[0072] Deep learning is highly robust and can learn high-level features of data, making the model highly robust to data changes and able to handle certain noise and deformations. For example, in multi-source satellite data fusion, since the data of different satellites may have noise and deformations, deep learning can reduce the impact of noise and deformations on the fusion result by learning high-level features of the data.
[0073] Deep learning is easy to expand. By increasing the depth and width of the network, the fitting ability of the model can be increased, and parallel computing technologies such as GPUs can be used to accelerate training and inference. For example, in multi-source satellite data fusion, as the data volume increases and task requirements improve, the depth and width of the deep learning model can be continuously increased to improve the fitting ability and performance of the model.
[0074] Deep learning has strong transferability. Through techniques such as transfer learning, the pre-trained model can be applied to different tasks, greatly improving the efficiency and accuracy of the model. For example, in multi-source satellite data fusion, the deep learning model pre-trained in other fields can be applied to the multi-source satellite data fusion task through techniques such as transfer learning to improve the efficiency and accuracy of the model.
[0075] Among them, the Recurrent Neural Network (RNN) and Long Short-Term Memory Network (LSTM) play important roles in state prediction. RNN can process sequential data with time-dependent relationships. By combining the hidden state of the previous time step and the current input, a new hidden state is generated to capture the dynamic changes in the sequential data. For example, in speech recognition tasks, RNN can process each time step of the sound frame and improve the accuracy of speech recognition by remembering the previous information.
[0076] LSTM is a special RNN structure that can effectively remember long-term dependency information and avoid the problem of gradient vanishing. The key to LSTM lies in the cell state, which is similar to a conveyor belt and can run directly across the entire chain with only a few linear interactions. It is easy for information to flow through and remain unchanged. In natural language processing tasks, LSTM can be used for the training of language models to predict the next word based on the previous words and maintain good performance even when dealing with long texts.
[0077] Step S104, obtain the satellite cloud map observation data corresponding to the current moment collected by the multiple satellites for remote sensing;
[0078] In this step, at time t, the above-mentioned multiple set members at time t are obtained, that is, the satellite cloud map observation data Oi,t of multiple satellites for remote sensing.
[0079] Step S105, fuse the satellite cloud map prediction data and the satellite cloud map observation data through a Kalman filter to obtain the final satellite cloud map data corresponding to the current moment.
[0080] As Figure 2 shown, in one embodiment, optionally, step S105 includes:
[0081] Step S201, respectively calculate the first error covariance corresponding to each satellite cloud map observation data and the second error covariance corresponding to each satellite cloud map prediction data according to each satellite cloud map prediction data and the satellite cloud map observation data;
[0082] As Figure 3 shown, in one embodiment, optionally, step S201 includes:
[0083] Step S301, calculate the average value of the satellite cloud map predictions corresponding to all satellite remote sensing according to each satellite cloud map prediction data;
[0084] Step S302, calculate the average value of the satellite cloud map observations corresponding to all satellite remote sensing according to each satellite cloud map observation data;
[0085] Step S303, calculate the first error covariance corresponding to each satellite cloud map observation data by using a first calculation formula according to each satellite cloud map observation data and the satellite cloud map observation average value;
[0086] Among them, the first calculation formula includes:
[0087] Co = sum(Oi,t - Ot) 2 / N - 1
[0088] Among them, Co represents the first error covariance, Oi,t represents the satellite cloud map observation data, Ot represents the satellite cloud map observation average value, and N represents the number of satellite remote sensing.
[0089] Step S304, calculate the second error covariance corresponding to each satellite cloud map prediction data by using a second calculation formula according to each satellite cloud map prediction data and the satellite cloud map prediction average value;
[0090] Among them, the second calculation formula includes;
[0091] Cx = sum(Xi,t - Xt|t-1) 2 / N - 1
[0092] Wherein, Cx represents the second error covariance, Xi,t represents the satellite cloud image prediction data, Xt|t-1 represents the satellite cloud image prediction average value, and N represents the number of satellite remote sensing.
[0093] Step S202: Based on each of the satellite cloud image prediction data, the satellite cloud image observation data, the first error covariance, and the second error covariance, fuse to obtain the final satellite cloud image data corresponding to the current moment.
[0094] As Figure 4 shown, in one embodiment, optionally, step S202 includes:
[0095] Step S401: Based on the first error covariance and the second error covariance, calculate the first weight corresponding to each satellite cloud image prediction data and the second weight corresponding to each satellite cloud image observation data by using the third calculation formula and the fourth calculation formula respectively;
[0096] Wherein, the third calculation formula includes:
[0097] Px = Co / (Co + Cx)
[0098] Wherein, Px represents the first weight corresponding to the satellite cloud image prediction data, Co represents the first error covariance, and Cx represents the second error covariance.
[0099] The fourth calculation formula includes:
[0100] Po = Cx / (Co + Cx)
[0101] Wherein, Po represents the second weight corresponding to the satellite cloud image observation data, Co represents the first error covariance, and Cx represents the second error covariance.
[0102] Step S402: Perform weighted summation based on each of the satellite cloud image prediction average value, the satellite cloud image observation average value, the first weight, and the second weight to fuse and obtain the final satellite cloud image data Xt corresponding to the current moment.
[0103] Xt = Po * Ot + Px * Xt|t-1
[0104] In this step, the Ensemble Kalman Filter is good at estimating and optimizing the system state, and updating the estimation of the system state in a recursive manner. Using the features extracted by deep learning as the input of the Ensemble Kalman Filter can provide more accurate prior information for state estimation, reduce the impact of noise on the fusion result, and thus improve the accuracy of the fusion result. In satellite cloud image prediction, first, a deep learning model is used to extract the features and motion trends of the cloud image, and then the Ensemble Kalman Filter is used to optimize the prediction result to improve the accuracy and stability of cloud image prediction.
[0105] In one embodiment, optionally, the method further includes:
[0106] According to the final cloud map satellite data and the satellite cloud map prediction data, calculate the error rate corresponding to the target deep learning prediction model;
[0107] When the error rate is greater than the error rate threshold, retrain the target deep learning prediction model.
[0108] In this embodiment, according to the final cloud map satellite data and the satellite cloud map prediction data, calculate the error rate corresponding to the target deep learning prediction model. When the error rate is greater than the error rate threshold, retrain the target deep learning prediction model. In this way, the performance of the deep learning model can be optimized, and its accuracy and stability can be improved. At the same time, the parameter adjustment of the Kalman filter can also be customized according to different application scenarios and data characteristics to meet the needs of on-orbit intelligent fusion of multi-source satellite data, further improving the processing efficiency and accuracy.
[0109] Figure 5 The block diagram of an on-orbit intelligent fusion device for multi-source satellite data according to an embodiment of the present application is shown.
[0110] As Figure 5 shown, in a second aspect, an embodiment of the present application provides an on-orbit intelligent fusion device 50 for multi-source satellite data, including:
[0111] A first acquisition module 51, configured to respectively acquire historical satellite cloud map data collected by multiple satellite remote sensing;
[0112] A preprocessing module 52, configured to preprocess each piece of the historical satellite cloud map data to obtain preprocessed historical satellite cloud map data;
[0113] A prediction module 53, configured to use a target deep learning prediction model to extract features and perform state prediction on the preprocessed historical satellite cloud map data, and predict satellite cloud map prediction data corresponding to each satellite remote sensing at the current moment;
[0114] A second acquisition module 54, configured to acquire satellite cloud map observation data corresponding to the current moment collected by the multiple satellites for remote sensing;
[0115] A fusion module 55, configured to fuse the satellite cloud map prediction data and the satellite cloud map observation data through a Kalman filter to obtain the final satellite cloud map data corresponding to the current moment.
[0116] In one embodiment, optionally, the apparatus further includes:
[0117] A calculation module, configured to calculate an error rate corresponding to the target deep learning prediction model according to the final satellite cloud map data and the satellite cloud map prediction data;
[0118] A training module, configured to retrain the target deep learning prediction model when the error rate is greater than an error rate threshold.
[0119] In one embodiment, optionally, the preprocessing module is configured to:
[0120] Perform normalization processing on each piece of historical satellite cloud map data to map historical satellite cloud map data of different magnitudes into a preset interval, so as to obtain processed historical satellite cloud map data.
[0121] In one embodiment, optionally, the apparatus further includes:
[0122] A determination module, configured to determine the data type and data fusion requirements of the historical satellite cloud map data collected by each satellite for remote sensing;
[0123] A selection module, configured to select a corresponding target deep learning prediction model from multiple deep learning prediction models according to the data type and the data fusion requirements, where the multiple deep learning prediction models include: a convolutional neural network, a recurrent neural network, and a long short-term memory network.
[0124] In one embodiment, optionally, the fusion module includes:
[0125] A data calculation unit, configured to calculate a first error covariance corresponding to each satellite cloud map observation data and a second error covariance corresponding to each satellite cloud map prediction data according to each satellite cloud map prediction data and the satellite cloud map observation data;
[0126] A data fusion unit, configured to fuse to obtain the final satellite cloud map data corresponding to the current moment according to each satellite cloud map prediction data, the satellite cloud map observation data, the first error covariance, and the second error covariance.
[0127] In one embodiment, optionally, the data calculation unit is configured to:
[0128] Based on each of the satellite cloud map prediction data, calculate the average value of the satellite cloud map predictions corresponding to all satellite remote sensing;
[0129] Based on each of the satellite cloud map observation data, calculate the average value of the satellite cloud map observations corresponding to all satellite remote sensing;
[0130] Based on each of the satellite cloud map observation data and the satellite cloud map observation average value, use the first calculation formula to calculate the first error covariance corresponding to each satellite cloud map observation data;
[0131] Based on each of the satellite cloud map prediction data and the satellite cloud map prediction average value, use the second calculation formula to calculate the second error covariance corresponding to each satellite cloud map prediction data;
[0132] In one embodiment, optionally, the data fusion unit is configured to:
[0133] Based on the first error covariance and the second error covariance, calculate the first weight corresponding to each satellite cloud map prediction data and the second weight corresponding to each satellite cloud map observation data by using the third calculation formula and the fourth calculation formula respectively;
[0134] Perform weighted summation according to each of the satellite cloud map prediction average value, the satellite cloud map observation average value, the first weight, and the second weight to fuse and obtain the final satellite cloud map data corresponding to the current moment.
[0135] In the solutions implemented by the above multi-source satellite data in-orbit intelligent fusion method, device, equipment, and medium, historical satellite cloud map data collected by multiple satellite remote sensing are respectively obtained; each of the historical satellite cloud map data is preprocessed to obtain the preprocessed historical satellite cloud map data; a target deep learning prediction model is used to perform feature extraction and state prediction on the preprocessed historical satellite cloud map data, and the satellite cloud map prediction data corresponding to each satellite remote sensing at the current moment is predicted; the satellite cloud map observation data corresponding to the current moment collected by the multiple satellite remote sensing are obtained; the satellite cloud map prediction data and the satellite cloud map observation data are fused through a Kalman filter to obtain the final satellite cloud map data corresponding to the current moment. Through the above technical solutions of the present invention, on the one hand, deep learning can perform feature extraction on multi-source satellite data, provide more accurate prior information for the ensemble Kalman filter, and reduce the influence of noise on the fusion result. On the other hand, the ensemble Kalman filter can optimize and correct the results of deep learning, improving the reliability of the fusion result.
[0136] In a third aspect, a computer device is provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned multi-source satellite data on-orbit intelligent fusion method are implemented.
[0137] In a fourth aspect, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned multi-source satellite data on-orbit intelligent fusion method are implemented.
[0138] It should be noted that those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working processes of the above-described multi-source satellite data on-orbit intelligent fusion device and each module can refer to the corresponding processes in the foregoing embodiments of the multi-source satellite data on-orbit intelligent fusion method, and will not be elaborated herein.
[0139] It should be noted that those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working processes of the above-described model training device and each module can refer to the corresponding processes in the foregoing embodiments of the multi-source satellite data on-orbit intelligent fusion method, and will not be elaborated herein.
[0140] The above-mentioned multi-source satellite data on-orbit intelligent fusion device can be implemented in the form of a computer program, and this computer program can run on a computer device as shown in Figure 6 shown.
[0141] Figure 6 The block diagram of a computer device according to an embodiment of the present application is shown.
[0142] Referring to Figure 6 , this computer device includes a processor, a memory, and a network interface connected through a system bus. Among them, the memory can include a storage medium and an internal memory.
[0143] The storage medium can store an operating system and a computer program. This computer program includes program instructions, and when the program instructions are executed, the processor can be enabled to execute any one of the multi-source satellite data on-orbit intelligent fusion methods provided in the embodiments of the present application.
[0144] The processor is used to provide computing and control capabilities to support the operation of the entire computer device.
[0145] The internal memory provides an environment for the operation of the computer program in the storage medium. When this computer program is executed by the processor, the processor can be enabled to execute any one of the multi-source satellite data on-orbit intelligent fusion methods. The storage medium can be non-volatile or volatile.
[0146] The network interface is used for network communication, such as sending the assigned tasks, etc. Those skilled in the art can understand that Figure 6 The structure shown in Figure 6 is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have a different component layout.
[0147] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.
[0148] In addition, an embodiment of this application provides a computer-readable storage medium storing computer-executable instructions, and the computer-executable instructions are used to execute the steps of the method in the embodiment of the first aspect.
[0149] It should be noted that the functions or steps that the above computer-readable storage medium or electronic device can achieve can be correspondingly referred to the relevant descriptions in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.
[0150] It should be understood that the term "and / or" used herein is only an association relationship describing associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " herein generally represents an "or" relationship between the associated objects before and after.
[0151] It should be understood that although the terms first, second, etc. may be used to describe the setting units in the embodiments of this application, these setting units should not be limited to these terms. These terms are only used to distinguish the setting units from each other. For example, without departing from the scope of the embodiments of this application, the first setting unit may also be referred to as the second setting unit, and similarly, the second setting unit may also be referred to as the first setting unit.
[0152] Depending on the context, as used herein, the word "if" can be interpreted as "when" or "while" or "in response to determining" or "in response to detecting". Similarly, depending on the context, the phrase "if determined" or "if detected (stated condition or event)" can be interpreted as "when determined" or "in response to determining" or "when detected (stated condition or event)" or "in response to detecting (stated condition or event)".
[0153] In several embodiments provided by the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point, the displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of the devices or units can be in electrical, mechanical, or other forms.
[0154] In addition, in each embodiment of the present application, each functional unit can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a hardware plus software functional unit.
[0155] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.
[0156] The above-described embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the protection scope of the present invention.
Claims
1. A multi-source satellite data on-orbit intelligent fusion method, characterized in that: The method comprises: Acquire historical satellite cloud image data collected by multiple satellite remote sensing sources respectively; Preprocessing each of the historical satellite cloud image data to obtain processed historical satellite cloud image data; Use the target deep learning prediction model to perform feature extraction and state prediction on the processed historical satellite cloud image data, and predict the satellite cloud image prediction data corresponding to each satellite remote sensing at the current moment; Obtaining satellite cloud image observation data corresponding to the current moment collected by remote sensing of the plurality of satellites; The satellite cloud image prediction data and the satellite cloud image observation data are fused through a Kalman filter to obtain final satellite cloud image data corresponding to the current moment.
2. The method according to claim 1, characterized in that The method further comprises: Calculate the error rate corresponding to the target deep learning prediction model according to the final cloud image satellite data and the satellite cloud image prediction data; When the error rate is greater than an error rate threshold, the target deep learning prediction model is retrained.
3. The method according to claim 1, characterized in that The preprocessing of each of the historical satellite cloud image data to obtain the processed historical satellite cloud image data includes: Each of the historical satellite cloud image data is normalized to map the historical satellite cloud image data of different magnitudes into a preset interval to obtain processed historical satellite cloud image data.
4. The method according to claim 1, characterized in that: The method further comprises: Determine the data type and data fusion requirements of historical satellite cloud image data collected by each satellite remote sensing; According to the data type and the data fusion requirement, a corresponding target deep learning prediction model is selected from a plurality of deep learning prediction models, wherein the plurality of deep learning prediction models include: a convolutional neural network, a recurrent neural network, and a long short-term memory network.
5. The method according to claim 1, characterized in that The fusing of the satellite cloud image prediction data and the satellite cloud image observation data by a Kalman filter to obtain the final satellite cloud image data corresponding to the current moment includes: According to each of the satellite cloud image prediction data and the satellite cloud image observation data, respectively calculating a first error covariance corresponding to each of the satellite cloud image observation data and a second error covariance corresponding to each of the satellite cloud image prediction data; According to each of the satellite cloud image prediction data, the satellite cloud image observation data, the first error covariance and the second error covariance, the final satellite cloud image data corresponding to the current moment is obtained by fusion.
6. The method according to claim 5, characterized in that According to each of the satellite cloud image prediction data and the satellite cloud image observation data, respectively calculating a first error covariance corresponding to each of the satellite cloud image observation data and a second error covariance corresponding to each of the satellite cloud image prediction data, including: According to each of the satellite cloud image prediction data, the satellite cloud image prediction average value corresponding to all satellite remote sensing is calculated; According to each of the satellite cloud image observation data, the average value of the satellite cloud image observation corresponding to all satellite remote sensing is calculated; Calculate the first error covariance corresponding to each satellite cloud image observation data using a first calculation formula according to each of the satellite cloud image observation data and the satellite cloud image observation average value; According to each of the satellite cloud image prediction data and the satellite cloud image prediction average value, a second calculation formula is used to calculate the second error covariance corresponding to each satellite cloud image prediction data.
7. The method according to claim 5, characterized in that According to each of the satellite cloud image prediction data, the satellite cloud image observation data, the first error covariance and the second error covariance, the final satellite cloud image data corresponding to the current moment is obtained by fusing, including: According to the first error covariance and the second error covariance, respectively calculate the first weight corresponding to each satellite cloud image prediction data and the second weight corresponding to each satellite cloud image observation data using the third calculation formula and the fourth calculation formula; A weighted sum is performed based on the predicted average value of each satellite cloud image, the observed average value of the satellite cloud image, the first weight, and the second weight, and the final satellite cloud image data corresponding to the current moment is obtained by fusion.
8. An on-orbit intelligent fusion device for multi-source satellite data, characterized in that: include: The first acquisition module is used to respectively acquire historical satellite cloud image data collected by multiple satellite remote sensing; A preprocessing module, used for preprocessing each of the historical satellite cloud image data to obtain processed historical satellite cloud image data; A prediction module is used to extract features and predict the status of the processed historical satellite cloud image data using a target deep learning prediction model, and predict the satellite cloud image prediction data corresponding to each satellite remote sensing at the current moment; A second acquisition module is used to acquire the satellite cloud image observation data corresponding to the current moment collected by the multiple satellite remote sensing; The fusion module is used to fuse the satellite cloud image prediction data and the satellite cloud image observation data through a Kalman filter to obtain the final satellite cloud image data corresponding to the current moment.
9. A computer device, characterized in that: include: at least one processor; and, a memory communicatively coupled to the at least one processor; The memory stores instructions executable by the at least one processor, wherein the instructions are configured to execute the method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: Computer executable instructions are stored, and the computer executable instructions are used to execute the method according to any one of claims 1 to 7.