A multi-constraint satellite task planning method and system based on U-Net

Abstract

The application relates to the technical field of satellite task planning, and discloses a multi-constraint satellite task planning method and system based on U-Net. The method comprises the following steps: acquiring orbit parameter information of a target satellite; constructing multi-source constraint adjustment; constructing a unified gridded multi-channel feature set; constructing a three-dimensional sample feature vector based on an orbit period and a snake-shaped arrangement mode; generating a corresponding binary label feature vector; and outputting a satellite task planning prediction result. Compared with the prior art, the method can solve the technical problems that the prior art relies on manual rules to formulate tasks, is difficult to cope with multi-source heterogeneous constraint fusion and large-scale sample parallel processing, and especially under the conditions that task windows overlap densely or ground resources conflict seriously, the prior art cannot realize fast and efficient imaging task planning. Due to the three-dimensional space-time representation of multi-constraint information and the introduction of the U-Net model with a coding and decoding jump structure, the intelligent level of satellite imaging task planning is improved.

Description

Technical Field

[0001] This invention relates to the field of satellite mission planning technology, and in particular to a multi-constraint satellite mission planning method and system based on U-Net. Background Technology

[0002] Currently, satellite imaging mission planning, as a core link connecting user needs with satellite in-orbit operation, directly determines the service capabilities and application value of the satellite and ground systems through its planning efficiency. Scientific and rational planning of satellite imaging missions ensures that satellites maximize their in-orbit performance within limited orbital, payload imaging, and ground receiving resources, avoiding resource waste. Currently, satellite imaging mission planning relies on a manually-led decision-making model. Ground operators typically manually formulate imaging mission sequences based on user-submitted observation requirements, combined with satellite orbital parameters, payload performance indicators, cloud cover forecasts, and available ground receiving resources. This model suffers from problems such as time-consuming manual analysis and adjustment processes, slow response times, uneven resource allocation, omission of key targets, and ineffective imaging. With the surge in the number of satellites and the diversification of user needs, the manual planning model struggles to cope with multi-source constraints and the parallel processing demands of massive missions, hindering the expansion of satellite system service capabilities.

[0003] In recent years, deep learning network models, with their powerful feature extraction and pattern recognition capabilities, have achieved breakthroughs in fields such as computer vision and time series prediction by being trained on massive amounts of historical data. They can perceive multi-dimensional features and adapt to complex application scenarios, providing a new technological path for the intelligent upgrading of satellite imaging mission planning. However, the constraints of satellite imaging missions are multi-source and heterogeneous, including physical constraints related to satellite orbital mechanics, payload operating time constraints, user priority constraints, and ground reception resource constraints. These constraints belong to different dimensions, such as the spatiotemporal dimension represented by cloud cover constraints, the spatial dimension represented by land and sea attributes, and the temporal dimension represented by ground reception resource constraints. There is a disconnect between the traditional deep learning training sample model architecture and the actual needs of satellite mission planning and execution.

[0004] Therefore, how to construct a deep learning framework that adapts to the characteristics of satellite imaging missions and realize the perception of constraint information from multi-source heterogeneous satellite imaging to intelligent mission decision-making has become a technical problem that urgently needs to be solved. Summary of the Invention

[0005] To address the aforementioned technical shortcomings, the purpose of this invention is to propose a U-Net-based multi-constraint satellite mission planning method. This method aims to solve the technical problems in existing technologies, such as reliance on manual rules for mission formulation, difficulty in handling multi-source heterogeneous constraint fusion and large-scale sample parallel processing, and especially the inability to achieve fast and efficient imaging mission planning under conditions of dense overlapping mission windows or severe ground resource conflicts.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention provides a multi-constraint satellite mission planning method based on U-Net. The U-Net-based multi-constraint satellite mission planning method includes: Step S10: Obtain the orbital parameter information of the target satellite, and based on the orbital parameter information, use the nadir point extrapolation mechanism based on the SGP4 orbit prediction model to perform the future nadir point data prediction task and the satellite orbital direction determination task, and output the target satellite sequence data; Step S20: Based on the target satellite sequence data, a multi-source heterogeneous constraint fusion rasterization mechanism with a unified spatiotemporal reference system is used to perform multi-condition constraint data extraction and spatial alignment tasks, and output a multi-channel raster feature set; Step S30: Construct a three-dimensional sample feature vector based on a spatiotemporal mapping reconstruction mechanism based on orbital period segmentation and serpentine sequence encoding using a multi-channel raster feature set; Step S40: Execute the planning imaging binary label generation task based on the 3D sample feature vector, and output the imaging binary label feature vector; Step S50: Based on the three-dimensional sample feature vector and the corresponding imaging binary label feature vector, the training and inference tasks of the U-Net network model are performed using a deep convolutional neural training mechanism driven by U-shaped structure encoding and decoding and composite loss function, and the satellite planning imaging prediction results are output.

[0007] Preferably, step S10, which involves acquiring the target satellite's orbital parameter information, performing future nadir point data prediction and satellite orbit direction determination tasks based on the SGP4 orbit prediction model using the nadir point extrapolation mechanism, and outputting the target satellite sequence data, specifically includes: Step S101: Obtain the orbital parameter information of the target satellite; the orbital parameter information includes the number of two rows of roots and the number of six roots in the orbit; Step S102: Based on the orbital parameter information, the SGP4 orbital prediction model is used to calculate the future nadir point data of the target satellite within the preset planned mission period. The future nadir point data includes the nadir point overpass time, nadir point longitude, and nadir point latitude. Step S103: Determine the satellite orbit direction flag based on the temporal change of the sub-satellite latitude. When the time condition is met... Sub-star latitude Less than time +1 sub-star latitude When the satellite's orbital motion direction flag is the ascending orbit direction, set the flag to 1; when Lat(t + 1) < Lat(t) is satisfied, mark the satellite's orbital motion direction flag as the descending orbit direction and set the flag to 0. Step S104: Output the target satellite sequence data including the sub-satellite point overpass time, sub-satellite point longitude, sub-satellite point latitude, and satellite orbital motion direction.

[0008] Preferably, in step S20, the step of performing multi-condition constraint data extraction and spatial alignment tasks based on the target satellite sequence data by using a multi-source heterogeneous constraint fusion rasterization mechanism with a unified spatio-temporal reference system and outputting a multi-channel raster feature set specifically includes: Step S201: Cloud cover constraint stage: Obtain the cloud cover data layer corresponding to the future sub-satellite point data through the cloud cover forecasting system by using a matching rule of rounding up the hourly part. ;

[0009] Among them, is the number of hours corresponding to the matching rule; is the number of minutes corresponding to the matching rule; Based on the cloud cover data layer Construct NetCDF format cloud cover raster data, and by mapping the sub-satellite point longitude and sub-satellite point latitude into the NetCDF format cloud cover raster data, extract the cloud cover forecast value corresponding to the future sub-satellite point data; Output the cloud cover forecast information set, and the cloud cover forecast information set includes the sub-satellite point overpass time, sub-satellite point longitude, sub-satellite point latitude, and cloud cover forecast value; Step S202: Historical imaging coverage constraint stage: Obtain the XML metadata of the satellite's historical imaging records, generate a closed polygon of the historical imaging area through a spatial merging operation based on the XML metadata of the satellite's historical imaging records, and convert the closed polygon of the historical imaging area into historical imaging area raster data consistent with the NetCDF format cloud cover raster data in step S201; When the center point of the historical imaging area raster data is located inside the closed polygon of the historical imaging area, it is judged that the historical imaging has been covered, and output the historical imaging coverage flag ; When the center point of the historical imaging area raster data is located outside the closed polygon of the historical imaging area, it is judged that the historical imaging has not been covered, and output the historical imaging coverage flag ; Output and store the historical imaging coverage situation set, and the historical imaging coverage situation set includes the sub-satellite point overpass time, sub-satellite point longitude, sub-satellite point latitude, and historical imaging coverage flag ; Step S203: User requirement constraint stage: Obtain the closed polygon of the user imaging requirement area, and convert the closed polygon of the user imaging requirement area into user imaging requirement area raster data that is consistent with the NetCDF format cloud cover raster data in step S201. When the center point of the raster data for the user's imaging requirement area is located inside the closed polygon of the user's imaging requirement area, output the requirement priority level flag. ; When the center point of the raster data for the user's imaging demand area is located outside the closed polygon of the user's imaging demand area, output the demand priority level flag. ; Output a set of user requirement information, including the time of the nadir transit, the longitude and latitude of the nadir, and a requirement priority level indicator. ; Step S204: Land-Sea Attribute Constraint Stage: Output and store the land-sea attribute set, which includes the nadir point overpass time, nadir point longitude, nadir point latitude, and land cover indicator. ; Step S205: Ground receiving segment constraint stage: Obtain available receiving window information from the preset ground receiving station resource scheduling system; When the nadir point crosses the top of the satellite and falls within the available receive window information, the output will show whether it is a receive segment flag. ; If the nadir point crosses the top of the satellite but does not fall within the available receive window information, the output should be a receive segment flag. ; Output the ground station data transmission resource set, which includes the nadir overpass time, nadir longitude, nadir latitude, and whether it is a receiving segment flag. ; Finally, by integrating the cloud cover forecast information set, the stored historical imaging coverage set, the user demand information set, the stored land and sea attribute set, and the ground station data transmission resource set, a multi-channel raster feature set is output.

[0010] Preferably, the steps in step S204, the land-sea attribute constraint stage, specifically include: firstly, obtaining the target satellite attributes; If the target satellite is a land observation satellite: continue to acquire the closed polygon of the land vector boundary region, and convert the closed polygon of the land vector boundary region into land vector boundary region raster data consistent with the NetCDF format cloud cover raster data in step S201; when the center point of the land vector boundary region raster data is located inside the closed polygon of the land vector boundary region, output the land cover indicator. When the center point of the land vector boundary region raster data is located outside the closed polygon of the land vector boundary region, output a land cover indicator. ; If the target satellite is a non-land observation satellite: directly generate land vector boundary area raster data with uniform global values, and output land cover indicators. 2; Land cover markers 2. This is used to indicate that the imaging planning of non-land observation satellites is not constrained by land or sea attributes.

[0011] Preferably, step S30, which involves constructing a three-dimensional sample feature vector based on a spatiotemporal mapping reconstruction mechanism using orbital periodic segmentation and serpentine sequence encoding based on a multi-channel raster feature set, specifically includes: Step S301: Obtain the satellite orbit direction flag from the target satellite sequence data, and use the satellite orbit direction flag as the outer index to alternately divide the orbit period segments for ascending or descending orbits, forming the orbit direction dimension; Step S302: In each orbital period segment of the orbital direction dimension, construct a time series dimension according to the time order. The time series dimension includes the ascending orbit start time dimension and the descending orbit start time dimension. Step S303: Fill in the multi-channel raster feature set generated in step S20 at the corresponding positions of the ascending orbit start time dimension and the descending orbit start time dimension in the time series dimension to form a multi-source condition constraint dimension. Step S304: Construct a three-dimensional sample feature vector based on the orbital direction dimension, time series dimension, and multi-source condition constraint dimension using a serpentine sample sequence mapping relationship.

[0012] Preferably, step S40, which involves performing a planning imaging binary label generation task based on the three-dimensional sample feature vector and outputting the imaging binary label feature vector, specifically includes: Step S401: Retrieve the imaging mission start time from the preset satellite historical imaging mission database. and imaging mission end time The imaging task time interval is constructed based on the imaging task start time and imaging task end time. Determine the time point of each sample in the three-dimensional sample feature vector. Is it within the imaging mission time interval? : like If the output plan imaging binary label is 1, then the output plan imaging binary label is 0; otherwise, the output plan imaging binary label is 0. Finally, the imaging binary label feature vector corresponding to the three-dimensional sample feature vector structure is generated by using the serpentine sample sequence mapping relationship.

[0013] Preferably, step S50, which involves training and inferring the U-Net network model based on the three-dimensional sample feature vector and the corresponding imaging binary label feature vector, using a deep convolutional neural training mechanism driven by U-shaped structure encoding and decoding and composite loss function, and outputting the satellite planning imaging prediction result, specifically includes: Step S501: Construct a training sample set based on the three-dimensional sample feature vector and the corresponding imaging binary label feature vector; divide the training sample set into an imaging task sample training set and an imaging task sample validation set in a 7:3 ratio; at the same time, obtain the first satellite planning imaging prediction result template set corresponding to the imaging task sample training set and the second satellite planning imaging prediction result template set corresponding to the imaging task sample validation set. Step S502: Construct the U-Net network model, which includes an input layer, an output layer, an encoder layer, a decoder layer, and a skip connection structure layer. The encoder layer is used for feature compression through convolution and pooling operations; the decoder layer is used for feature restoration through deconvolution operations; the skip connection structure layer is used for low-order and high-order feature fusion; the number of channels in the input layer is set to 6; the output layer generates imaging prediction probabilities in the [0,1] interval using the Sigmoid activation function. If the imaging prediction probability is greater than or equal to 0.5, it is determined to be a planned imaging label; otherwise, it is determined to be a non-planned imaging label. Step S503: Use the imaging task sample training set as the input to the U-Net network model, and use the template set of the first satellite planning imaging prediction results as the output of the U-Net network model; set the loss function of the U-Net network model as a composite loss function composed of cross-entropy (BCE) and Dice coefficients. Simultaneously, the Adam optimizer and StepLR strategy are used to perform weight updates and learning rate adjustments, ultimately outputting the first model weight set. Step S504: Based on the first model weight set, use the imaging task sample validation set as the input to the U-Net network model, and use the second satellite planning imaging prediction result template set as the output of the U-Net network model; set the loss function of the U-Net network model as a composite loss function composed of cross-entropy (BCE) and Dice coefficients. Simultaneously, the Adam optimizer and StepLR strategy are used to perform weight update and learning rate adjustment tasks, and finally output the second model weight set; the model weights with the highest F1 score in the second model weight set are selected and saved to obtain the U-Net network optimization model; Step S505: Apply the U-Net network optimization model to perform the imaging task prediction inference task, and finally output the satellite planning imaging prediction results.

[0014] This invention also provides a U-Net-based multi-constraint satellite mission planning system, comprising: The orbit extrapolation module is used to obtain the orbital parameter information of the target satellite. Based on the orbital parameter information, it uses the nadir point extrapolation mechanism based on the SGP4 orbit prediction model to perform the task of predicting future nadir point data and determining the satellite orbital direction, and outputs the target satellite sequence data. The multi-constraint fusion module is used to perform multi-condition constraint data extraction and spatial alignment tasks based on target satellite sequence data using a multi-source heterogeneous constraint fusion rasterization mechanism with a unified spatiotemporal reference system, and outputs a multi-channel raster feature set. The sample construction module is used to construct three-dimensional sample feature vectors based on a spatiotemporal mapping reconstruction mechanism based on orbital period segmentation and serpentine sequence encoding using a multi-channel raster feature set. The imaging label generation module is used to perform the planning imaging binary label generation task based on the 3D sample feature vector and output the imaging binary label feature vector; The planning model training and inference module is used to train and infer the U-Net network model based on the 3D sample feature vector and the corresponding imaging binary label feature vector. It adopts a deep convolutional neural training mechanism driven by U-shaped structure encoding and decoding and composite loss function to perform the training and inference tasks, and outputs satellite planning imaging prediction results.

[0015] The present invention also provides a U-Net-based multi-constraint satellite mission planning device, comprising: a memory, a processor, and a U-Net-based multi-constraint satellite mission planning program stored in the memory and executable on the processor. When the U-Net-based multi-constraint satellite mission planning program is executed by the processor, it implements a U-Net-based multi-constraint satellite mission planning method.

[0016] The present invention also provides a computer program product, including a U-Net-based multi-constraint satellite mission planning program, which, when executed by a processor, implements the U-Net-based multi-constraint satellite mission planning method.

[0017] The beneficial effects of this invention are as follows: This invention can construct a unified three-dimensional sample representation structure under the constraint of multi-source heterogeneous imaging, realize the coupling mapping of task planning samples in the dimensions of orbit direction, time series and constraint channel, effectively improve the model's comprehensive perception ability of task continuity, resource conflict and target priority, thereby enhancing the accuracy and scalability of task decision-making in complex imaging scenarios.

[0018] This invention introduces a U-Net network model with a skip connection structure and constructs a composite loss function of cross-entropy and Dice coefficients, which significantly improves the model's accuracy in imaging task recognition in scenarios with small sample sparsity and dense and intertwined target regions. It avoids the problems of uneven resource scheduling, imaging task omission, and task response lag in traditional methods, and improves the imaging performance of satellite systems under limited orbit and ground resources. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the first embodiment of a multi-constraint satellite mission planning method based on U-Net according to the present invention.

[0021] Figure 2 This is a schematic diagram of multi-source heterogeneous constraint information for a first embodiment of a U-Net-based multi-constraint satellite mission planning method of the present invention.

[0022] Figure 3 This is a schematic diagram of the U-Net network model training process in the first embodiment of a U-Net-based multi-constraint satellite mission planning method of the present invention.

[0023] Figure 4 This is a schematic diagram of the device for a U-Net-based multi-constraint satellite mission planning method according to the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Example 1: As Figure 1 The diagram shown is a flowchart of the first embodiment of the U-Net-based multi-constraint satellite mission planning method of the present invention, which presents the first embodiment of the U-Net-based multi-constraint satellite mission planning method of the present invention.

[0026] In the first embodiment, the U-Net-based multi-constraint satellite mission planning method includes: Step S10: Obtain the orbital parameter information of the target satellite, and based on the orbital parameter information, use the nadir point extrapolation mechanism based on the SGP4 orbit prediction model to perform the future nadir point data prediction task and the satellite orbital direction determination task, and output the target satellite sequence data; It should be noted that the "sub-satellite point extrapolation mechanism based on SGP4 orbit prediction model" refers to the continuous time point sampling and prediction of the orbit position in the future period based on the input two rows of satellite root data, and then calculating the latitude and longitude coordinates of the satellite sub-satellite point corresponding to each time point.

[0027] Understandably, this step enables accurate spatial position prediction of the target satellite within the preset imaging planning time window, and, combined with the latitude sequence change trend, generates corresponding orbital direction markers (ascending or descending orbits), providing structural boundaries for the serpentine arrangement mechanism in subsequent sample construction, thereby achieving standardization and controllability of the temporal structure of the planned samples, and improving the consistency and completeness of the model input data.

[0028] It should be understood that, compared to traditional methods that rely solely on static orbital segment numbering or manual determination of orbital periods, this step achieves automatic orbital direction classification without manual intervention through a dynamic judgment mechanism based on the latitude change trend of the sub-satellite point. This improves the automation level of the sample construction process and the ability to model across orbital periods, and avoids model training bias caused by errors in orbital segment division.

[0029] Step S20: Based on the target satellite sequence data, a multi-source heterogeneous constraint fusion rasterization mechanism with a unified spatiotemporal reference system is used to perform multi-condition constraint data extraction and spatial alignment tasks, and output a multi-channel raster feature set; It should be noted that the "multi-source heterogeneous constraint fusion rasterization mechanism" refers to projecting imaging constraint information (including cloud cover forecasts, historical imaging records, user imaging requirement areas, land and sea attributes, and ground receiving segment resource information) with different data formats, spatial resolutions, and temporal granularities onto a standard spatial reference frame that matches the satellite's nadir trajectory, and converting it into a raster structure of the same dimension for subsequent use as multi-channel input features in deep neural network models. This mechanism includes sub-processes such as NetCDF format data parsing, Shapefile to raster conversion, center point and landing point determination, and multi-constraint priority conflict handling.

[0030] Understandably, this step enables the structured integration of multiple constraint data with different spatial and temporal attributes in satellite imaging missions, allowing each satellite trajectory point to obtain its corresponding constraint feature expression in both time and space. This constructs a complete input sample information channel, effectively enhancing the subsequent model's ability to perceive and express complex constraint combinations.

[0031] It should be understood that, compared to the traditional method of using a single constraint or manually selecting some key indicators as task planning conditions, this step realizes a fully rasterized and fully feature-expressed sample input mechanism, which enables the model to simultaneously consider factors such as spatiotemporal dynamic cloud cover changes, ground receiver availability conflicts, historical imaging redundancy and imaging area repetition priority conflicts, thereby improving the task planning model's ability to fit and tolerate complex real-world scenarios.

[0032] For example, such as Figure 2 As shown, the imaging constraint data fused in this step includes six types of key information from heterogeneous sources and with inconsistent formats: ① Cloud cover forecast data updated at 3-hour intervals, organized in NetCDF format, providing hourly cloud cover percentage values ​​for the covered area; ② Imaging redundancy hotspot layers constructed based on historical satellite image records, showing the repeated imaging coverage of each region over the past three days; ③ Imaging request geographic target extents exported from the user task management system, expressing the polygon boundaries of the region of interest in vector shapefile format; ④ Global land and ocean mask layers, used to filter out water areas that are not imaging valuable; ⑤ Communication window resource tables from multiple ground stations, structurally describing the visible time windows between each ground station and the target satellite within the next 24 hours; ⑥ Landing point matching rules in the standard georeferenced grid, ensuring precise alignment of orbital points with the grid center. This information fusion process is subsequently used for convolutional feature extraction operations in the U-Net network.

[0033] Step S30: Construct a three-dimensional sample feature vector based on a spatiotemporal mapping reconstruction mechanism based on orbital period segmentation and serpentine sequence encoding using a multi-channel raster feature set; It should be noted that the "spatial-temporal mapping reconstruction mechanism of orbital period segmentation and serpentine sequence encoding" refers to: periodically segmenting the target satellite's nadir trajectory data arranged in chronological order according to its orbital period (e.g., 97 minutes / orbit); within each orbital period, performing a serpentine sorting based on the latitude and longitude direction of the nadir point (odd-numbered orbits are northward ascending orbits, even-numbered orbits are southward descending orbits), ensuring maximum consistency between the temporal order and the spatial trajectory continuity. Subsequently, the multi-channel raster features within each orbital period are horizontally stitched together, and the resulting 3D sample tensor is used as the input for subsequent deep convolutional models.

[0034] Understandably, the 3D feature tensor reconstruction achieved through this mechanism retains both spatial continuity and temporal consistency, enabling the network model to perceive contextual features within the same imaging trajectory and to have a stronger expressive ability for contextual jumps caused by terrain, weather, and mission requirements during trajectory crossing.

[0035] Step S40: Execute the planning imaging binary label generation task based on the 3D sample feature vector, and output the imaging binary label feature vector; It should be noted that the "planning imaging binary labeling" in this step refers to marking each location unit with 0 and 1 to indicate whether it can be used as a valid imaging target, based on the current mission planning strategy and imaging condition constraints.

[0036] Understandably, the binary label vector generated in this step can quickly filter out imaging candidate regions that meet the mission requirements within a large coverage area, effectively avoiding unnecessary resource waste in the subsequent inverse kinematics stage. Especially when facing the long-period orbital conditions of high-orbit satellites, the labeling results have a high degree of spatial continuity and temporal regularity, providing a stable foundation for subsequent mission continuity identification and priority clustering.

[0037] Step S50: Based on the three-dimensional sample feature vector and the corresponding imaging binary label feature vector, the training and inference tasks of the U-Net network model are performed using a deep convolutional neural training mechanism driven by U-shaped structure encoding and decoding and composite loss function, and the satellite planning imaging prediction results are output.

[0038] It should be noted that the "deep convolutional neural training mechanism based on U-shaped structure encoding and decoding and composite loss function" in this step refers to using the 3D sample feature vector constructed in the previous step as the network input layer, and the corresponding imaging binary label feature vector as the supervision signal, inputting it into the U-Net model with symmetric downsampling and upsampling structures to perform end-to-end image region prediction training. This training mechanism uses a combination of composite loss functions, including cross-entropy loss function and boundary-aware loss function, to simultaneously improve the accuracy of region imaging feasibility judgment and the detail of boundary prediction. Skip connections are introduced into the network structure to preserve shallow spatial texture features, while incorporating deep semantic expression capabilities, thereby improving the inference performance of imaging tasks under multi-scale constraints.

[0039] Understandably, this step trains a U-Net network model to learn the mapping relationship for "predicting imaging feasibility from multi-source constrained raster features," constructing an intelligent model that jointly infers the probability of successful imaging from conditions such as orbit, attitude, and target. Unlike traditional rule-driven methods (such as those based on imaging window geometric matching), this method can learn the inherent coupling patterns between complex nonlinear conditions based on historical samples, effectively capturing imaging feasibility patterns under satellite dynamic constraints, thereby significantly improving the intelligence level of imaging planning.

[0040] It should be understood that, compared to traditional heuristic or static priority strategy generation methods, the U-Net deep neural network used in this invention has the ability to automatically learn complex imaging decision patterns, without relying on manually set imaging logic rules. This avoids the shortcomings of traditional strategies in recognizing imaging failure patterns under atypical orbital attitudes or climate constraints. Furthermore, the introduction of a composite loss function allows the model to simultaneously focus on pixel-level accuracy and boundary structure integrity during training, improving the model's ability to distinguish "edge imaging regions" while maintaining overall prediction performance, thus enhancing the engineering applicability of the prediction results.

[0041] For example, such as Figure 3 The figure illustrates the training process of the U-Net network model in this embodiment of the invention. The upper left part of the figure shows the multi-source heterogeneous feature layers used to construct the 3D sample feature vectors, including: cloud cover forecast data, orbital attributes, user imaging requirements, historical successful imaging area coverage, and geographic area spatial coding, etc., which are raster features of multiple dimensions. These features are aligned and rasterized through a unified temporal and spatial reference frame, and then fused to construct a multi-channel input feature tensor, which together with the imaging area binary label matrix shown on the right forms the training sample set.

[0042] During the network training phase, the U-Net model structure shown below contains multiple convolutional modules, pooling downsampling layers, and transposed convolutional upsampling layers, completing the image prediction task through a symmetrical encoder-decoder structure. The operations such as "conv 3x3 ReLU," "max pool 2x2," and "up-conv 2x2" in each layer represent feature extraction, spatial compression, and restoration operations, respectively. The "skip connection (copy and crop)" mechanism, indicated by the gray arrow, can reintroduce shallow detail information during the upsampling stage, improving the boundary recognition capability for spatially complex regions (such as mountains and densely populated multi-object areas).

[0043] The model training employs a composite loss function that combines pixel-level cross-entropy loss with Dice loss specific to boundary regions. This ensures the accuracy of the overall prediction map while enhancing the ability to preserve the clarity of target boundaries. The final output prediction map not only demonstrates good restoration results in large-scale target areas but also exhibits strong robustness and generalization ability in areas with cloud cover and blurred boundaries.

[0044] Example 2: Furthermore, the present invention provides a U-Net-based multi-constraint satellite mission planning system, employing a U-Net-based multi-constraint satellite mission planning method from the above embodiments, which can solve the technical problem of U-Net-based multi-constraint satellite mission planning. Compared with the prior art, the beneficial effects of the U-Net-based multi-constraint satellite mission planning system provided by the present invention are the same as those of the U-Net-based multi-constraint satellite mission planning method provided in the above embodiments, and other technical features of the U-Net-based multi-constraint satellite mission planning system are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0045] Example 3: This invention provides a multi-constraint satellite mission planning device based on U-Net. Please refer to... Figure 4A U-Net-based multi-constraint satellite mission planning device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to execute the U-Net-based multi-constraint satellite mission planning method described in Embodiment 1 above. The U-Net-based multi-constraint satellite mission planning device in this embodiment may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. This U-Net-based multi-constraint satellite mission planning device is merely an example and should not impose any limitations on the functionality and scope of use of this embodiment. A U-Net-based multi-constraint satellite mission planning device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the U-Net-based multi-constraint satellite mission planning device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An I / O interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows a U-Net-based multi-constraint satellite mission planning device to communicate wirelessly or wiredly with other devices to exchange data. While the figure shows one U-Net-based multi-constraint satellite mission planning device with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0046] Example 4: This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the U-Net-based multi-constraint satellite mission planning method described above. The computer program product provided by this invention can solve a technical problem related to U-Net-based multi-constraint satellite mission planning. Compared with the prior art, the beneficial effects of the computer program product provided by this invention are the same as those of the U-Net-based multi-constraint satellite mission planning method provided in the above embodiments, and will not be repeated here.

[0047] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this invention.

[0048] It should be understood that the various parts disclosed in this invention can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0049] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A multi-constraint satellite mission planning method based on U-Net, characterized in that, The methods include: Step S10: Obtain the orbital parameter information of the target satellite, and based on the orbital parameter information, use the nadir point extrapolation mechanism based on the SGP4 orbit prediction model to perform the future nadir point data prediction task and the satellite orbital direction determination task, and output the target satellite sequence data; Step S20: Based on the target satellite sequence data, a multi-source heterogeneous constraint fusion rasterization mechanism with a unified spatiotemporal reference system is used to perform multi-condition constraint data extraction and spatial alignment tasks, and output a multi-channel raster feature set; Step S30: Construct a three-dimensional sample feature vector based on a spatiotemporal mapping reconstruction mechanism based on orbital period segmentation and serpentine sequence encoding using a multi-channel raster feature set; Step S40: Execute the planning imaging binary label generation task based on the 3D sample feature vector, and output the imaging binary label feature vector; Step S50: Based on the three-dimensional sample feature vector and the corresponding imaging binary label feature vector, the training and inference tasks of the U-Net network model are performed using a deep convolutional neural training mechanism driven by U-shaped structure encoding and decoding and composite loss function, and the satellite planning imaging prediction results are output.

2. The multi-constraint satellite mission planning method based on U-Net as described in claim 1, characterized in that, Step S10 involves obtaining the target satellite's orbital parameter information, and based on this information, employing a nadir point extrapolation mechanism based on the SGP4 orbit prediction model to perform future nadir point data prediction and satellite orbit direction determination tasks, and outputting the target satellite sequence data. Specifically, this includes: Step S101: Obtain the orbital parameter information of the target satellite; the orbital parameter information includes the number of two rows of roots and the number of six roots of the orbit; Step S102: Based on the orbital parameter information, the SGP4 orbital prediction model is used to calculate the future nadir point data of the target satellite within the preset planned mission period. The future nadir point data includes the nadir point overpass time, nadir point longitude, and nadir point latitude. Step S103: Determine the satellite orbit running direction flag based on the temporal variation of the sub-satellite point latitude. When the sub-satellite point latitude at time is less than the sub-satellite point latitude at time + 1 , mark the satellite orbit running direction flag as the ascending orbit direction and set flag to 1; when Lat(t + 1) < Lat(t) is satisfied, mark the satellite orbit running direction flag as the descending orbit direction and set flag to 0; Step S104: Output target satellite sequence data including the nadir overpass time, nadir longitude, nadir latitude, and satellite orbit direction.

3. The multi-constraint satellite mission planning method based on U-Net as described in claim 2, characterized in that, Step S20, which involves performing multi-condition constraint data extraction and spatial alignment tasks based on the target satellite sequence data using a multi-source heterogeneous constraint fusion rasterization mechanism with a unified spatiotemporal reference system, and outputting a multi-channel raster feature set, specifically includes: Step S201: Cloud Cover Constraint Stage: Obtain the cloud cover data layer corresponding to the future nadir point data using the cloud cover forecasting system with an hourly part rounded up matching rule. ; in, For the number of hours corresponding to the matching rule; The number of minutes corresponding to the matching rule; Based on cloud data layer Construct cloud cover raster data in NetCDF format, and extract cloud cover forecast values ​​corresponding to future nadir point data by mapping the longitude and latitude of the nadir point to the NetCDF format cloud cover raster data. Output a set of cloud cover forecast information, which includes the time of the nadir transit, the longitude of the nadir, the latitude of the nadir, and the forecast cloud cover value; Step S202: Historical Imaging Coverage Constraint Stage: Obtain XML metadata of satellite historical imaging records, generate closed polygons of historical imaging areas through spatial merging operation based on the XML metadata of satellite historical imaging records, and convert the closed polygons of historical imaging areas into historical imaging area raster data consistent with the NetCDF format cloud cover raster data in step S201. When the center point of the historical imaging area raster data is located inside the closed polygon of the historical imaging area, it is determined that the historical imaging has been covered, and a historical imaging coverage flag is output. ; When the center point of the historical imaging area raster data is located outside the closed polygon of the historical imaging area, it is determined that the historical imaging is not covered, and a historical imaging coverage flag is output. ; Output a set of historical imaging coverage data, including the time of nadir overpass, longitude, latitude, and historical imaging coverage markers. ; Step S203: User requirement constraint stage: Obtain the closed polygon of the user imaging requirement area, and convert the closed polygon of the user imaging requirement area into user imaging requirement area raster data that is consistent with the NetCDF format cloud cover raster data in step S201. When the center point of the raster data for the user's imaging requirement area is located inside the closed polygon of the user's imaging requirement area, output the requirement priority level flag. ; When the center point of the raster data for the user's imaging demand area is located outside the closed polygon of the user's imaging demand area, output the demand priority level flag. ; Output a set of user requirement information, including the time of the nadir transit, the longitude and latitude of the nadir, and a requirement priority level indicator. ; Step S204: Land-Sea Attribute Constraint Stage: Output and store the land-sea attribute set, which includes the nadir point overpass time, nadir point longitude, nadir point latitude, and land cover indicator. ; Step S205: Ground receiving segment constraint stage: Obtain available receiving window information from the preset ground receiving station resource scheduling system; When the nadir point crosses the top of the satellite and falls within the available receive window information, the output will show whether it is a receive segment flag. ; If the nadir point crosses the top of the satellite but does not fall within the available receive window information, the output should be a receive segment flag. ; Output the ground station data transmission resource set, which includes the nadir overpass time, nadir longitude, nadir latitude, and whether it is a receiving segment flag. ; Finally, by integrating the cloud cover forecast information set, the stored historical imaging coverage set, the user demand information set, the stored land and sea attribute set, and the ground station data transmission resource set, a multi-channel raster feature set is output.

4. The multi-constraint satellite mission planning method based on U-Net as described in claim 2, characterized in that, In step S204, the steps of the land-sea attribute constraint stage specifically include: first, obtaining the target satellite attributes; If the target satellite is a land observation satellite: continue to acquire the closed polygon of the land vector boundary region, and convert the closed polygon of the land vector boundary region into land vector boundary region raster data consistent with the NetCDF format cloud cover raster data in step S201; when the center point of the land vector boundary region raster data is located inside the closed polygon of the land vector boundary region, output the land cover indicator. When the center point of the land vector boundary region raster data is located outside the closed polygon of the land vector boundary region, output a land cover indicator. ; If the target satellite is a non-land observation satellite: directly generate land vector boundary area raster data with uniform global values, and output land cover indicators. 2; Land cover markers 2. This is used to indicate that the imaging planning of non-land observation satellites is not constrained by land or sea attributes.

5. The multi-constraint satellite mission planning method based on U-Net as described in claim 1, characterized in that, Step S30, which involves constructing a three-dimensional sample feature vector based on a spatiotemporal mapping reconstruction mechanism using orbital period segmentation and serpentine sequence encoding based on a multi-channel raster feature set, specifically includes: Step S301: Obtain the satellite orbit direction flag from the target satellite sequence data, and use the satellite orbit direction flag as the outer index to alternately divide the orbit period segments for ascending or descending orbits, forming the orbit direction dimension; Step S302: In each orbital period segment of the orbital direction dimension, construct a time series dimension according to the time order. The time series dimension includes the ascending orbit start time dimension and the descending orbit start time dimension. Step S303: Fill in the multi-channel raster feature set generated in step S20 at the corresponding positions of the ascending orbit start time dimension and the descending orbit start time dimension in the time series dimension to form a multi-source condition constraint dimension. Step S304: Construct a three-dimensional sample feature vector based on the orbital direction dimension, time series dimension, and multi-source condition constraint dimension using a serpentine sample sequence mapping relationship.

6. The multi-constraint satellite mission planning method based on U-Net as described in claim 5, characterized in that, Step S40, which involves performing a planning imaging binary label generation task based on the 3D sample feature vector and outputting the imaging binary label feature vector, specifically includes: Step S401: Retrieve the imaging mission start time from the preset satellite historical imaging mission database. and imaging mission end time The imaging task time interval is constructed based on the imaging task start time and imaging task end time. Determine the time point of each sample in the three-dimensional sample feature vector. Is it within the imaging mission time interval? : like If the output plan imaging binary label is 1, then the output plan imaging binary label is 0; otherwise, the output plan imaging binary label is 0. Finally, the imaging binary label feature vector corresponding to the three-dimensional sample feature vector structure is generated by using the serpentine sample sequence mapping relationship.

7. The multi-constraint satellite mission planning method based on U-Net as described in claim 1, characterized in that, Step S50, based on the three-dimensional sample feature vector and the corresponding imaging binary label feature vector, employs a deep convolutional neural training mechanism driven by U-structure encoding and decoding and a composite loss function to perform the training and inference tasks of the U-Net network model, and outputs the satellite planning imaging prediction results. Specifically, this includes: Step S501: Construct a training sample set based on the three-dimensional sample feature vector and the corresponding imaging binary label feature vector; divide the training sample set into an imaging task sample training set and an imaging task sample validation set in a 7:3 ratio; at the same time, obtain the first satellite planning imaging prediction result template set corresponding to the imaging task sample training set and the second satellite planning imaging prediction result template set corresponding to the imaging task sample validation set. Step S502: Construct the U-Net network model, which includes an input layer, an output layer, an encoder layer, a decoder layer, and a skip connection structure layer. The encoder layer is used for feature compression through convolution and pooling operations; the decoder layer is used for feature restoration through deconvolution operations; the skip connection structure layer is used for low-order and high-order feature fusion; the number of channels in the input layer is set to 6; the output layer generates imaging prediction probabilities in the [0,1] interval using the Sigmoid activation function. If the imaging prediction probability is greater than or equal to 0.5, it is determined to be a planned imaging label; otherwise, it is determined to be a non-planned imaging label. Step S503: Use the imaging task sample training set as the input to the U-Net network model, and use the template set of the first satellite planning imaging prediction results as the output of the U-Net network model; set the loss function of the U-Net network model as a composite loss function composed of cross-entropy (BCE) and Dice coefficients. Simultaneously, the Adam optimizer and StepLR strategy are used to perform weight updates and learning rate adjustments, ultimately outputting the first model weight set. Step S504: Based on the first model weight set, use the imaging task sample validation set as the input to the U-Net network model, and use the second satellite planning imaging prediction result template set as the output of the U-Net network model; set the loss function of the U-Net network model as a composite loss function composed of cross-entropy (BCE) and Dice coefficients. Simultaneously, the Adam optimizer and StepLR strategy are used to perform weight update and learning rate adjustment tasks, and finally output the second model weight set; the model weights with the highest F1 score in the second model weight set are selected and saved to obtain the U-Net network optimization model; Step S505: Apply the U-Net network optimization model to perform the imaging task prediction inference task, and finally output the satellite planning imaging prediction results.

8. A U-Net-based multi-constraint satellite mission planning system, applied to the U-Net-based multi-constraint satellite mission planning method according to any one of claims 1 to 7, characterized in that, The U-Net-based multi-constraint satellite mission planning system includes: The orbit extrapolation module is used to obtain the orbital parameter information of the target satellite. Based on the orbital parameter information, it uses the nadir point extrapolation mechanism based on the SGP4 orbit prediction model to perform the task of predicting future nadir point data and determining the satellite orbital direction, and outputs the target satellite sequence data. The multi-constraint fusion module is used to perform multi-condition constraint data extraction and spatial alignment tasks based on target satellite sequence data using a multi-source heterogeneous constraint fusion rasterization mechanism with a unified spatiotemporal reference system, and outputs a multi-channel raster feature set. The sample construction module is used to construct three-dimensional sample feature vectors based on a spatiotemporal mapping reconstruction mechanism based on orbital period segmentation and serpentine sequence encoding using a multi-channel raster feature set. The imaging label generation module is used to perform the planning imaging binary label generation task based on the 3D sample feature vector and output the imaging binary label feature vector; The planning model training and inference module is used to train and infer the U-Net network model based on the 3D sample feature vector and the corresponding imaging binary label feature vector. It adopts a deep convolutional neural training mechanism driven by U-shaped structure encoding and decoding and composite loss function to perform the training and inference tasks, and outputs satellite planning imaging prediction results.

9. A multi-constraint satellite mission planning device based on U-Net, characterized in that, The U-Net-based multi-constraint satellite mission planning device includes: a memory, a processor, and a U-Net-based multi-constraint satellite mission planning program stored in the memory and executable on the processor. When the U-Net-based multi-constraint satellite mission planning program is executed by the processor, it implements a U-Net-based multi-constraint satellite mission planning method according to any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes a U-Net-based multi-constraint satellite mission planning program, which, when executed by a processor, implements a U-Net-based multi-constraint satellite mission planning method according to any one of claims 1 to 7.

Images