A data processing method and model training method for electromagnetic spectrum sensing
By constructing parameters for multiple satellite constellations, sampling and fusing electromagnetic spectrum data, and generating a fusion benchmark true value with full airspace coverage, the problem of insufficient airspace coverage of electromagnetic spectrum datasets in existing technologies is solved, and the accuracy and robustness of electromagnetic spectrum sensing models are improved.
Patent Information
- Application Number
- CN202611438163.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-16
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies have low spatial coverage in electromagnetic spectrum datasets, resulting in incomplete baseline values for electromagnetic spectrum data, which affects the accuracy and robustness of deep learning models in electromagnetic spectrum perception and prediction.
By acquiring parameters from multiple satellite constellations, multiple satellite constellations are constructed. These constellations are then used to sample electromagnetic spectrum data emitted by medium- and high-orbit space targets, perform spatiotemporal alignment and fusion, generate a fusion benchmark true value covering the entire airspace, and construct a supervised sample set to train the electromagnetic spectrum sensing model.
This improves the integrity and reliability of ground truth data during electromagnetic spectrum sensing model training, and enhances the accuracy and robustness of the model in sensing and predicting electromagnetic spectrum data in the spatial domain.
Smart Images

Figure CN123074615A_ABST
Abstract
Description
Technical Field
[0001] The embodiments in this specification relate to the technical field of data processing, and in particular to a data processing method and model training method for electromagnetic spectrum sensing. Background Technology
[0002] With the widespread application of communication technologies in more and more scenarios, there is a need for accurate sensing and prediction of the electromagnetic spectrum in space, to understand its distribution and efficiently utilize these resources. Deep learning is one approach to electromagnetic spectrum sensing and prediction, using electromagnetic spectrum datasets to train models and then predicting the electromagnetic spectrum map in space based on collected data. However, large quantities of real-world, on-orbit electromagnetic spectrum data are difficult to obtain, making high-fidelity simulation the primary method for constructing electromagnetic spectrum data.
[0003] Existing methods for constructing electromagnetic spectrum datasets through simulation mainly include: simulation based on a single orbital configuration, which can only obtain local coverage data under a specific constellation configuration and cannot generate a baseline true value for full airspace coverage; and spectrum datasets based on measured data, which utilize ground station networks to collect real spectrum data, but the amount is limited.
[0004] In summary, the spatial coverage of electromagnetic spectrum data constructed by existing simulation techniques is low, resulting in insufficient integrity of the baseline truth value of the final electromagnetic spectrum data. Consequently, the accuracy and robustness of the trained model in sensing and predicting electromagnetic spectrum data in the spatial domain are poor. Summary of the Invention
[0005] In view of this, embodiments of this specification provide a data processing method for electromagnetic spectrum sensing. One or more embodiments of this specification also relate to a model training method for electromagnetic spectrum sensing, an electromagnetic spectrum sensing method, a computing device, a computer-readable storage medium, and a computer program product, to address the technical deficiencies existing in the prior art.
[0006] According to a first aspect of the embodiments of this specification, a data processing method for electromagnetic spectrum sensing is provided, comprising: Obtain parameters for multiple satellite constellations and construct satellite constellations corresponding to these parameters. The multiple satellite constellations have different blind spots for observing medium and high orbit space targets. By using multiple satellite constellations, electromagnetic spectrum data emitted by medium- and high-orbit space targets are sampled to obtain sampled electromagnetic spectrum data from multiple satellite constellations; The sampled electromagnetic spectrum data from multiple satellite constellations are spatiotemporally aligned and fused to obtain the fusion reference true value of multiple satellite constellations; Based on the fusion of baseline ground truth values from multiple satellite constellations and sampled electromagnetic spectrum data, a supervised sample set is constructed for training the electromagnetic spectrum sensing model.
[0007] According to a second aspect of the embodiments of this specification, a model training method for electromagnetic spectrum sensing is provided, comprising: A supervised sample set is obtained, wherein the supervised sample set is constructed based on the above-mentioned data processing method for electromagnetic spectrum sensing; The electromagnetic spectrum sensing model is obtained by using sampled electromagnetic spectrum data as training input and fused benchmark true values as supervision labels to perform autoregressive training on the initial neural network model.
[0008] According to a third aspect of the embodiments of this specification, an electromagnetic spectrum sensing method is provided, comprising: The electromagnetic spectrum data emitted by medium- and high-orbit space targets in the airspace to be measured is sampled using a satellite constellation to obtain the electromagnetic spectrum data to be measured. The electromagnetic spectrum data to be measured is input into the electromagnetic spectrum sensing model, and the electromagnetic spectrum map corresponding to the spatial domain to be measured is output. The electromagnetic spectrum sensing model is trained based on the above-mentioned model training method for electromagnetic spectrum sensing.
[0009] According to a fourth aspect of the embodiments of this specification, a computing device is provided, comprising: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor, they implement the above-described data processing method for electromagnetic spectrum sensing, model training method for electromagnetic spectrum sensing, and steps of the above-described electromagnetic spectrum sensing method.
[0010] According to a fifth aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores a computer program / instructions, which, when executed by a processor, implement the steps of the above-described data processing method for electromagnetic spectrum sensing, the model training method for electromagnetic spectrum sensing, and the electromagnetic spectrum sensing method described above.
[0011] According to a sixth aspect of the embodiments of this specification, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described data processing method for electromagnetic spectrum sensing, the model training method for electromagnetic spectrum sensing, and the electromagnetic spectrum sensing method described above.
[0012] In one embodiment of this specification, multiple satellite constellation parameters are acquired, and satellite constellations corresponding to these parameters are constructed. These multiple satellite constellations have different observation blind zones for medium- and high-orbit space targets. Electromagnetic spectrum data emitted by medium- and high-orbit space targets are sampled using these multiple satellite constellations to obtain sampled electromagnetic spectrum data for multiple satellite constellations. The sampled electromagnetic spectrum data from multiple satellite constellations are then spatiotemporally aligned and fused to obtain a fusion reference ground truth for the multiple satellite constellations. Based on the fusion reference ground truth and the sampled electromagnetic spectrum data, a supervisory sample set is constructed for training the electromagnetic spectrum sensing model. It is possible to construct satellite constellations with different spatial geometric distributions using different constellation parameters, enabling each satellite constellation to complement each other in the observation blind zone. By using multi-constellation collaborative sampling, the observation blind zone that exists when observing a single constellation is compensated for. This allows the generated fusion reference ground truth to have complete spatial coverage characteristics across the entire airspace, eliminating spatial gaps in the reference ground truth. This improves the integrity and reliability of the ground truth data used in training the electromagnetic spectrum sensing model, enabling the deep learning model to learn and infer the spectrum distribution based on complete spatial context information. This enhances the accuracy and robustness of the trained model in sensing and predicting electromagnetic spectrum data in the airspace. Attached Figure Description
[0013] Figure 1 This is a flowchart of a data processing method for electromagnetic spectrum sensing provided in one embodiment of this specification; Figure 2 This is a schematic diagram of the radiation of a medium-to-high orbit space target antenna provided in one embodiment of this specification; Figure 3 This is a schematic diagram of the radiation of a low-orbit satellite antenna provided in one embodiment of this specification; Figure 4 This specification provides an embodiment of an antenna pattern for a medium-to-high orbit space target. Figure 5 This specification provides an embodiment of a low-Earth orbit satellite antenna pattern. Figure 6 This is a schematic diagram of the received signal power distribution of a low-Earth orbit satellite constellation provided in one embodiment of this specification; Figure 7 This is a schematic diagram of the equivalent isotropic radiated power of a low-Earth orbit satellite constellation provided in one embodiment of this specification; Figure 8 This is a schematic diagram of the orbital altitude field of a low-Earth orbit satellite constellation provided in one embodiment of this specification; Figure 9 This is a schematic diagram of the position matrix of a medium-to-high orbit space target provided in one embodiment of this specification; Figure 10This is a flowchart of a model training method for electromagnetic spectrum sensing provided in one embodiment of this specification; Figure 11 This is a flowchart of an electromagnetic spectrum sensing method provided in one embodiment of this specification; Figure 12 This is a structural block diagram of a computing device provided in one embodiment of this specification. Detailed Implementation
[0014] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.
[0015] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.
[0016] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."
[0017] Furthermore, it should be noted that the data involved in one or more embodiments of this specification (including but not limited to data used for analysis, stored data, displayed data, etc.) are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.
[0018] First, the terms and concepts used in one or more embodiments of this specification will be explained.
[0019] STK (Systems Tool Kit): A commercial satellite orbit simulation software.
[0020] Equivalent Isotropically Radiated Power (EIRP): The radiated power of a transmitter antenna relative to an omnidirectional antenna in the direction of maximum radiation, used to characterize the intensity of electromagnetic signals emitted by a space target.
[0021] Low Earth Orbit (LEO): Satellites operate in an orbital region at a relatively low altitude above the Earth's surface, approximately 200-2000 km.
[0022] Geostationary Orbit (GEO): A circular orbit in which the satellite's orbital period is the same as the Earth's rotation period and the orbital inclination is zero. The orbital altitude is approximately 35,786 km.
[0023] Medium Earth Orbit (MEO): The orbital region where satellites operate at altitudes between low Earth orbit and geostationary orbit, approximately 2000-35786 km.
[0024] Celestrak: A public ephemeris library that provides publicly available TLE (two-line orbital elements) data.
[0025] Ground Truth: Standard reference data used for comparison in supervised learning.
[0026] Computing devices: Computing devices are computer systems designed to perform one or more specific tasks. Compared to personal mainframes, they are weaker in performance but have significant advantages in terms of size and power consumption. They are commonly used in various electronic and mechanical control systems.
[0027] As communication technologies are increasingly used in more and more scenarios, there is a need to accurately sense and predict the electromagnetic spectrum in space, and to understand the distribution of the electromagnetic spectrum in space in order to make efficient use of electromagnetic spectrum resources in space.
[0028] However, existing electromagnetic spectrum sensing research mainly relies on ground-based measured data and lacks high-fidelity simulation datasets specifically designed for low-Earth orbit (LEO) satellite constellations sensing space targets. This results in a lack of fundamental data support for the training and validation of related algorithms. The baseline ground truth of the electromagnetic spectrum of space targets (i.e., the complete spectrum distribution covering the entire space and all time periods) cannot be directly obtained in actual observations, and existing methods lack a systematic strategy for generating the baseline ground truth. LEO constellations with a single orbital inclination have significant blind spots in their coverage of space targets, resulting in uneven and incomplete spatial distribution of spectrum observation data, making it difficult to support supervised learning training of full-domain spectrum reconstruction algorithms. Simulation datasets lack systematic design for scenarios with multiple target densities and multiple orbital types, resulting in insufficient generalization ability and an inability to effectively evaluate the performance of algorithms under different space conditions. The original simulation data consists of sparse observations with irregular distributions, which do not match the regular gridded input format required by deep learning models, and lack systematic preprocessing and standardization procedures.
[0029] Low Earth Orbit (LEO) satellite constellations, with their advantages of wide coverage and low latency, have become an important platform for conducting wide-area electromagnetic spectrum sensing of medium and high Earth Orbit targets. However, due to the dynamic changes in the observation geometry of LEO constellations over time, the short time of a single overpass, and the need for multi-orbital plane coordination for global coverage, there are significant gaps in the observation coverage of a single constellation configuration at a specific moment. To train deep learning-based spectrum reconstruction and prediction algorithms, it is necessary to construct a paired dataset containing sparse observation inputs and complete ground truth spectrum values. However, real on-orbit data is difficult to obtain in large quantities and involves sensitive issues related to spectrum management, making high-fidelity simulation the current main approach.
[0030] The existing work on constructing relevant simulation datasets mainly includes: Simulation based on a single orbital configuration: This method uses a LEO constellation with a fixed orbital inclination to simulate coverage of a specific target and generate signal received power data. However, this method can only obtain local coverage data under a specific constellation configuration and cannot generate true baseline values for full airspace coverage. (Source: Related satellite spectrum sensing simulation literature)
[0031] Spectrum dataset based on measured data: Real spectrum data is collected using ground station networks. However, the amount of data is limited, and it cannot cover multi-target density scenarios. Furthermore, it is difficult to systematically expand due to actual deployment limitations.
[0032] Data sets constructed using existing technologies have the following drawbacks: (1) Incomplete baseline true value: Single or random orbit configuration cannot achieve full airspace coverage without blind spots, and the generated spectrum true value has obvious gaps; (2) Insufficient scene generalization: The lack of multi-objective density gradient design makes it impossible to systematically evaluate the performance of the algorithm under different numbers of objects; (3) Incomplete coverage of orbit types: It does not systematically cover a variety of typical orbits such as geostationary orbit (GEO) and medium Earth orbit (MEO), and the representativeness of the data is limited; (4) Lack of standardized preprocessing process: The original simulation data is irregularly distributed and lacks a unified gridding, normalization and multimodal alignment scheme, making it difficult to use directly for supervised learning.
[0033] In summary, the existing technology produces electromagnetic spectrum data with low spatial coverage, resulting in insufficient integrity of the reference truth value of the final electromagnetic spectrum data.
[0034] To address the above problems, this specification provides a data processing method for electromagnetic spectrum sensing. One or more embodiments of this specification also relate to a model training method for electromagnetic spectrum sensing, an electromagnetic spectrum sensing method, a computing device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments.
[0035] See Figure 1 , Figure 1 A flowchart of a data processing method for electromagnetic spectrum sensing according to an embodiment of this specification is shown, including the following specific steps: Step 102: Obtain parameters for multiple satellite constellations and construct satellite constellations corresponding to these parameters. The multiple satellite constellations have different blind spots for observing medium and high orbit space targets.
[0036] A satellite constellation is a satellite system consisting of multiple satellites deployed according to specific orbital parameters; for example, it could be a low Earth orbit (LEO) satellite constellation.
[0037] Satellite constellation parameters are a set of parameters for the orbits of satellites in a satellite constellation, such as orbital altitude, orbital inclination, and right ascension of the ascending node.
[0038] Medium and high Earth orbit space targets are devices that can emit electromagnetic signals and operate at geostationary orbit or medium Earth orbit altitudes. For example, they can be geostationary orbit communication satellites, meteorological satellites, etc.
[0039] A blind spot is a spatial region in which a satellite constellation cannot effectively observe a space target under specific spatiotemporal conditions. For example, it could be a coverage gap in the satellite constellation.
[0040] Optionally, one way to obtain multiple satellite constellation parameters is to set up an orbit configuration parameter scanning task in a satellite orbit simulation tool, generate multiple sets of constellation configuration configuration files in batches according to preset combinations of orbit inclination, orbital altitude, and right ascension of the ascending node, and read and export the satellite constellation parameters from each set of configuration files. Another approach is to select satellite numbers of different orbit types (low Earth orbit, medium Earth orbit, and geostationary orbit) from a publicly available ephemeris database, and obtain the orbit inclination, orbital altitude, and right ascension of the ascending node for each satellite at the current epoch by calling the ephemeris database's orbital element query interface. This specification does not limit the scope of this approach.
[0041] Optionally, one way to construct a satellite constellation corresponding to multiple satellite constellation parameters is to import the acquired sets of satellite constellation parameters into the scene configuration module of the orbit simulation tool. The simulation tool then automatically calculates the phase distribution of each satellite node within the same orbital plane based on the orbital inclination, orbital altitude, and right ascension of the ascending node in each set of parameters, and completes the orbital deployment of all satellite nodes. Another approach is to use a numerical integral model of satellite orbital dynamics, with each set of satellite constellation parameters as the initial orbital element input, to integrally calculate the position and velocity vectors of each satellite node at each moment during the simulation period. Based on the calculation results, the instantaneous orbital positions of each satellite node are deployed in the simulation space. This specification does not limit the implementation of this approach.
[0042] For example, in the orbit simulation scenario, six low-Earth orbit (LEO) satellite constellations are set up. The six LEO satellite constellations have different orbital inclinations. Each LEO satellite constellation contains the same number of satellite nodes, and the satellite nodes are evenly distributed in the same orbital plane. The orbital planes of the six LEO satellite constellations are staggered in space, so that the observation coverage gap formed by any one satellite constellation for geosynchronous orbit targets can be filled by the observation coverage areas of the other two satellite constellations.
[0043] Indicatively, the equipment and link settings for each space target scenario are consistent in terms of system parameter configuration. The low-Earth orbit satellite is equipped with a 31×31 element rectangular phased array antenna, capable of beamforming and electronic reconfiguration; the space target is equipped with a high-gain parabolic directional antenna with a radius of 1 meter to compensate for long-distance transmission path loss. The system operates in the Ku band with a center frequency of 14.5 GHz, an element spacing of half a wavelength, a signal bandwidth of 96 MHz, and uses Quadrature Phase Shift Keying (QPSK) modulation. The communication link modeling strictly follows the propagation model specified by relevant institutions to ensure that the simulation results conform to the characteristics of the actual space communication environment.
[0044] Simulation scenarios from different perspectives, such as Figure 2and Figure 3 As shown, where, Figure 2 This specification illustrates a schematic diagram of the radiation from a medium-to-high orbit space target antenna according to one embodiment. Figure 3 This diagram illustrates the radiation of a low-Earth orbit (LEO) satellite antenna according to one embodiment of this specification. The blue sphere represents the Earth, and the LEO satellite constellation is presented as an irregular cone. The top of the constellation is orange and gradually transitions to blue towards the bottom, providing full coverage in LEO. Space targets are represented as objects with an orange-red bottom, a yellow center, and a blue top. EUTELSAT 16A and EUTELSAT 117WB represent geostationary orbits. The more reddish the area, the greater the gain in that direction. Figure 4 This specification shows the radiation pattern of a medium-to-high orbit space target antenna according to one embodiment, illustrating the radiation pattern of a high-gain parabolic antenna used on a space target (Directivity represents the directivity coefficient in the figure), with power ranging from 0 degrees to 360°, the highest power at 90°, and the radiated power decreasing as it deviates from the main lobe direction in other angular directions; Figure 5 This specification illustrates a radiation pattern of a low-Earth orbit (LEO) satellite antenna according to one embodiment, showing the radiation pattern of a 31×31 element phased array antenna used on a LEO satellite. The power is symmetrically distributed with the 0° direction as the center in the range of -90° to 90°. The radiated power reaches its highest value in the 0° direction (i.e., the direction of the antenna normal). As the beam direction deviates from the normal direction, the radiated power gradually decreases, and the power drops to its lowest value near ±90°.
[0045] In step 102, multiple satellite constellation parameters are obtained, and satellite constellations corresponding to the multiple satellite constellation parameters are constructed. The multiple satellite constellations have different observation blind spots for medium and high orbit space targets, which lays the foundation for subsequent sampling of electromagnetic spectrum data emitted by medium and high orbit space targets using multiple satellite constellations.
[0046] Step 104: Using multiple satellite constellations, sample the electromagnetic spectrum data emitted by medium and high orbit space targets to obtain sampled electromagnetic spectrum data from multiple satellite constellations.
[0047] Optionally, sampling electromagnetic spectrum data emitted by medium- and high-orbit space targets using multiple satellite constellations to obtain sampled electromagnetic spectrum data from multiple satellite constellations can be achieved by: calculating the instantaneous spatial distance and relative line-of-sight angle between each satellite in each constellation and each space target at each sampling time; selecting satellite-target pairs that are currently observable based on preset visibility criteria; and calculating the signal power value at the satellite receiver based on the electromagnetic wave free-space propagation loss formula and the equivalent isotropic radiation power parameter of the space target for each pair. Another implementation method is to pre-calculate the visibility time windows for all satellite-target pairs throughout the simulation period based on the orbit prediction results of each satellite constellation and the orbit prediction results of the space target. Within each visibility window, the spatial distance and received signal power between the satellite and the target are calculated by interpolation according to a preset sampling time interval. This specification does not limit this approach in the embodiments.
[0048] For example, for multiple space targets in geosynchronous orbit, six low-Earth orbit satellite constellations with different orbital inclinations are used to continuously collect signal power data at the receivers of each satellite according to a preset sampling time interval. Each satellite constellation independently completes signal acquisition of all space targets within the same sampling time, resulting in six sets of sampled electromagnetic spectrum data that complement each other from the perspective of space observation. Each set of sampled electromagnetic spectrum data carries the signal power distribution information of that constellation under its own observation coverage conditions.
[0049] In step 104, multiple satellite constellations are used to sample the electromagnetic spectrum data emitted by medium- and high-orbit space targets, obtaining sampled electromagnetic spectrum data from multiple satellite constellations. This allows for the complementary use of data collected from different observation directions by multiple constellations to obtain spectrum information for the same space region, even when there are gaps in the observation coverage of a single constellation. This provides a data foundation for subsequent spatiotemporal alignment and fusion of sampled electromagnetic spectrum data from multiple satellite constellations.
[0050] Step 106: Perform spatiotemporal alignment and fusion of the sampled electromagnetic spectrum data of multiple satellite constellations to obtain the fusion reference true value of multiple satellite constellations.
[0051] Spatiotemporal alignment fusion involves merging electromagnetic spectrum data from different satellite constellations at the same sampling time under a unified spatial grid coordinate system to obtain fused spectrum data covering the entire airspace at that time.
[0052] The fusion reference true value is the complete electromagnetic spectrum distribution data covering the entire airspace and all time periods obtained by spatiotemporally aligning and fusing the sampled electromagnetic spectrum data of multiple satellite constellations at the same sampling time.
[0053] Optionally, the spatiotemporal alignment and fusion of sampled electromagnetic spectrum data from multiple satellite constellations to obtain the fusion reference truth value of multiple satellite constellations can be achieved by direct splicing, weighted fusion (weighted by satellite elevation angle), or maximum value fusion. The embodiments in this specification do not limit this.
[0054] For example, for six sets of sampled electromagnetic spectrum data collected by six low-Earth orbit satellite constellations at a certain sampling time, the six sets of data are spatially aligned according to a spatial distribution grid constructed with a unified longitude and latitude dimension. For each spatial grid point, all sampled values in the six constellations that have observation conditions for that grid point are extracted. The sampled values are numerically fused according to their corresponding observation geometric weights to obtain the full-grid fused reference true value covering the entire geosynchronous orbit zone at that sampling time. The above fusion operation is repeated for all sampling times to obtain the fused reference true value sequence at each sampling time.
[0055] In step 106, the sampled electromagnetic spectrum data of multiple satellite constellations are spatiotemporally aligned and fused to obtain the fused reference true value of multiple satellite constellations. This can integrate the original data of complementary observations of multiple constellations into reference data with complete airspace coverage characteristics, eliminating the gaps in the true value data under a single constellation configuration.
[0056] Step 108: Based on the fusion of baseline ground truth values and sampled electromagnetic spectrum data from multiple satellite constellations, construct a supervised sample set for training the electromagnetic spectrum sensing model.
[0057] Optionally, one approach to constructing a supervised sample set for training an electromagnetic spectrum sensing model based on fused baseline ground truth and sampled electromagnetic spectrum data from multiple satellite constellations could be to use sampled electromagnetic spectrum data from a single satellite constellation as model input samples and the corresponding fused baseline ground truth as supervised labels, thus forming input-output paired data. Another approach could be to preprocess the sampled data and fused baseline ground truth using a regularized grid to generate a supervised sample set containing input feature tensors and label tensors. This specification does not limit the scope of this approach.
[0058] For example, the sampled electromagnetic spectrum data of one set of low-Earth orbit satellite constellations is organized according to a spatial grid as input data for training samples, and the fusion benchmark ground value corresponding to the spatiotemporal range of the sampled data is used as the annotation data for the training samples. By pairing the spatial mapping relationship between the sampled data and the fusion benchmark ground value, a supervision sample set containing multiple samples is constructed.
[0059] In step 108, a supervised sample set for training the electromagnetic spectrum sensing model is constructed based on the fusion of the baseline ground truth and sampled electromagnetic spectrum data from multiple satellite constellations. This set provides structured training data with complete ground truth annotations for the deep learning model, enabling the model to learn the mapping rules between sparse observations and complete spectrum distributions within a supervised learning framework.
[0060] In the embodiments of this specification, multiple satellite constellation parameters are acquired, and satellite constellations corresponding to these parameters are constructed. These multiple satellite constellations have different observation blind zones for medium- and high-orbit space targets. Electromagnetic spectrum data emitted by medium- and high-orbit space targets are sampled using these multiple satellite constellations to obtain sampled electromagnetic spectrum data for multiple satellite constellations. The sampled electromagnetic spectrum data from multiple satellite constellations are then spatiotemporally aligned and fused to obtain a fusion reference ground truth for multiple satellite constellations. Based on the fusion reference ground truth and the sampled electromagnetic spectrum data from multiple satellite constellations, a supervisory sample set is constructed for training the electromagnetic spectrum sensing model. It is possible to construct satellite constellations with different spatial geometric distributions using different constellation parameters, enabling each satellite constellation to complement each other in the observation blind zone. By using multi-constellation collaborative sampling, the observation blind zone that exists when observing a single constellation is compensated for. This allows the generated fusion reference ground truth to have complete spatial coverage characteristics across the entire airspace, eliminating spatial gaps in the reference ground truth. This improves the integrity and reliability of the ground truth data used in training the electromagnetic spectrum sensing model, enabling the deep learning model to learn and infer the spectrum distribution based on complete spatial context information. This enhances the accuracy and robustness of the trained model in sensing and predicting electromagnetic spectrum data in the airspace.
[0061] In one optional embodiment of this specification, prior to step 102, the method further includes: Construct multiple space target scenarios, in which medium and high orbit space targets with different densities are constructed in each space target scenario; Step 104 includes: By using multiple satellite constellations, electromagnetic spectrum data emitted by medium- and high-orbit space targets in various space target scenarios are sampled to obtain sampled electromagnetic spectrum data corresponding to multiple satellite constellations in each space target scenario.
[0062] The space target scenario is a simulation scenario that includes a specific number of space targets with a specific orbital distribution.
[0063] Optionally, one way to construct multiple space target scenarios is to select a corresponding number of space targets from a public ephemeris database according to a preset number of targets and configure their orbital numbers. Another way is to design requirements based on a gradient of the number of targets, and progressively add space targets to a basic scenario to construct a progressive space target scenario. This specification does not limit the embodiments in this way.
[0064] For example, six progressive spatial target scenarios can be designed based on the number of targets. Each scenario adds targets to the previous scenario, as follows: Scenario 1: 3 space targets (targets 1-3); Scenario 2: 6 space targets (targets 1-6); Scenario 3: 9 space targets (targets 1-9); Scenario 4: 12 space targets (targets 1-12); Scenario 5: 15 space targets (targets 1-15); Scenario 6: 18 space targets (targets 1-18). The number of targets in each scenario increases sequentially, forming a systematic coverage in terms of target density. The selected space targets cover typical orbit types such as geostationary orbit, medium Earth orbit, and inclined geostationary orbit, with orbital inclination ranging from 0° to 57° and average motion ranging from 1.00 to 2.01 revolutions per day. They also include different functional types such as communication and navigation, ensuring the representativeness and generalization ability of the dataset from both orbital and functional characteristics.
[0065] In the embodiments of this specification, multiple space target scenarios are constructed, in which medium- and high-orbit space targets with different densities are present. Multiple satellite constellations are used to sample the electromagnetic spectrum data emitted by the medium- and high-orbit space targets in each space target scenario, obtaining sampled electromagnetic spectrum data corresponding to each space target scenario for multiple satellite constellations. This enables the formation of simulation scenarios with multi-gradient coverage in the target density dimension. By applying the same set of satellite constellations to collect data in scenarios with different target densities, the diversity of the supervised sample set in the dimension of the number of targets is ensured. This supports the trained electromagnetic spectrum sensing model in maintaining stable sensing performance under different space target densities, improving the model's generalization ability and scenario adaptability.
[0066] In one optional embodiment of this specification, before sampling the electromagnetic spectrum data emitted by medium- and high-orbit space targets in various space target scenarios using multiple satellite constellations to obtain the sampled electromagnetic spectrum data corresponding to each space target scenario, the method further includes: Obtain multiple sampling times; By utilizing multiple satellite constellations, electromagnetic spectrum data emitted by medium- and high-orbit space targets in various space target scenarios are sampled to obtain sampled electromagnetic spectrum data corresponding to multiple satellite constellations in each space target scenario, including: At any given sampling time, multiple satellite constellations are used to sample the electromagnetic spectrum data emitted by medium- and high-orbit space targets in each space target scenario in parallel, thereby obtaining the sampled electromagnetic spectrum data of multiple satellite constellations in each space target scenario at each sampling time.
[0067] The sampling time is a preset discrete time point, which is used to trigger the satellite constellation to synchronously acquire electromagnetic spectrum data of space targets.
[0068] Optionally, one way to acquire multiple sampling times is to select multiple time nodes at fixed time intervals on the simulation time axis. Another way is to dynamically determine the acquisition time point based on the visibility window between the satellite and the target. This specification does not limit the implementation of this method.
[0069] Parallelism refers to the simultaneous execution of electromagnetic spectrum data acquisition operations by satellite nodes in multiple satellite constellations at the same sampling time.
[0070] For example, multiple sampling times are selected at equal intervals during a continuous simulation period. When each sampling time arrives, all satellite nodes in six low-Earth orbit satellite constellations with different orbital inclinations are simultaneously triggered to perform signal power acquisition on space targets in each space target scenario. Each constellation completes the visibility judgment and received power calculation of all space targets at the same sampling time, and obtains the sampled electromagnetic spectrum data of multiple constellations from their respective observation perspectives at that sampling time.
[0071] In the embodiments of this specification, multiple sampling times are acquired. At any given sampling time, multiple satellite constellations are used to sample the electromagnetic spectrum data emitted by medium- and high-orbit space targets in each space target scenario in parallel, obtaining the sampled electromagnetic spectrum data of multiple satellite constellations in each space target scenario at each sampling time. By triggering multiple groups of satellite constellations to synchronously acquire data in parallel at the same sampling time, the sampled data acquired by each group of constellations under the same spatiotemporal reference has strict temporal consistency, improving the accuracy of subsequent spatiotemporal alignment and fusion operations, and ensuring that the acquired sampled electromagnetic spectrum data forms a continuous sequence in the time dimension, providing a guarantee for the integrity of the fusion reference truth value throughout the entire time period.
[0072] In one optional embodiment of this specification, step 106 includes: The sampled electromagnetic spectrum data of multiple satellite constellations at any sampling time are spatiotemporally aligned and fused to obtain the fusion reference true value of multiple satellite constellations at each sampling time.
[0073] For details on how to perform spatiotemporal alignment and fusion of sampled electromagnetic spectrum data from multiple satellite constellations at any given sampling time, please refer to the implementation method of spatiotemporal alignment and fusion of sampled electromagnetic spectrum data from multiple satellite constellations. These details will not be elaborated upon here.
[0074] For example, for six sets of sampled electromagnetic spectrum data collected by six low-Earth orbit satellite constellations with different orbital inclinations at any sampling time, the six sets of data are spatially aligned according to a spatial distribution grid constructed with a unified longitude and latitude dimension. Then, time alignment is completed using the sampling time as the time reference. For each spatial grid point, all sampled values in the six constellations that have observation conditions for that grid point are extracted and numerically fused according to the observation geometric weights to obtain the fused reference true value covering the entire geosynchronous orbit zone at that sampling time. The above fusion operation is repeated for all sampling times to obtain the fused reference true value sequence at each sampling time.
[0075] In the embodiments of this specification, the sampled electromagnetic spectrum data of multiple satellite constellations at any sampling time are spatiotemporally aligned and fused to obtain the fusion reference true value of multiple satellite constellations at each sampling time. This ensures that the fusion reference true value maintains a one-to-one correspondence with the sampling time in the time dimension, providing a time-aligned input-output pairing reference for the subsequent construction of the supervision sample set, and supporting the model to learn the mapping relationship between sparse observations at any sampling time and the complete true value at the same time.
[0076] In one optional embodiment of this specification, step 108 includes: Using sampled electromagnetic spectrum data from any satellite constellation as input samples and fused baseline ground truth as supervision labels, training sample pairs are constructed to obtain a supervision sample set for training the electromagnetic spectrum sensing model.
[0077] Optionally, one approach to constructing training sample pairs using sampled electromagnetic spectrum data from any satellite constellation as input samples and fusion benchmark ground truth as supervision labels could be: extracting spatial gridded features from the sampled data of any satellite constellation as input tensors, using the fusion benchmark ground truth with the same spatiotemporal range as the output label tensor, and storing the two as a single training sample pair. Another approach could be: traversing the sampled data of all satellite constellations under various space target scenarios and sampling times, matching the corresponding fusion benchmark ground truth for each set of sampled data, forming multiple sets of input-output paired samples. This specification does not limit this approach in its embodiments.
[0078] For example, the sampled electromagnetic spectrum data of the first constellation in one of the six low-Earth orbit satellite constellations at each sampling time are selected. These data are organized according to a spatial grid to form an input feature tensor. The fusion baseline ground truth values belonging to the same sampling time and the same spatial target scene as this group of sampled data are used as label tensors. The input feature tensor and label tensor are paired and stored to form a training sample pair. This process is repeated for all sampling times, all spatial target scenes, and all satellite constellations, constructing corresponding training sample pairs for each group of sampled data. All training sample pairs are then aggregated to form a supervised sample set for training the electromagnetic spectrum sensing model.
[0079] In the embodiments of this specification, by using sampled electromagnetic spectrum data from any satellite constellation as input samples and fused benchmark ground truth as supervision labels, training sample pairs are constructed to obtain a supervision sample set for training the electromagnetic spectrum sensing model. This allows for the selection of observation data from any single satellite constellation as sparse input samples, which, together with the corresponding fused benchmark ground truth, form paired training samples. This directly supports the supervised learning training of the spectrum reconstruction and prediction algorithm, establishing a supervision mapping relationship between sparse observations from a single constellation and the complete spatial spectrum distribution. This enables the model to optimize parameters based on the differences between input samples and ground truth labels under the same spatiotemporal conditions during training, learning the ability to reconstruct the full-domain spectrum distribution from sparse observation data, and improving the accuracy of the model in spectrum sensing and prediction based on single-constellation measured data.
[0080] In one optional embodiment of this specification, the fused reference true value is the reference electromagnetic spectrum; Before constructing a supervised sample set for training the electromagnetic spectrum sensing model based on fused baseline ground truth and sampled electromagnetic spectrum data from multiple satellite constellations, the following steps are also included: The system acquires spatial location information of multiple satellite constellations and determines the received signal power and equivalent isotropic radiated power based on sampled electromagnetic spectrum data. The spatial location information includes longitude, latitude, and orbital altitude. A spatial location grid is constructed based on longitude and latitude, and an orbital height field is constructed based on orbital height. The received signal power and equivalent isotropic radiation power are resampled to the spatial position grid, the orbital height field is mapped to the spatial position grid, and the position information of the medium-high orbit space target is encoded into a position matrix and mapped to the spatial position grid to obtain the spatially aligned received signal power, equivalent isotropic radiation power, orbital height field and position matrix. Using sampled electromagnetic spectrum data from any satellite constellation as input samples and fused baseline ground truth as supervision labels, training sample pairs are constructed to obtain a supervision sample set for training the electromagnetic spectrum sensing model, including: A multi-feature input tensor is constructed using the spatially aligned received signal power, equivalent isotropic radiated power, orbital height field, and position matrix as input samples. A reference electromagnetic spectrum map is used as a supervision label to construct training sample pairs, resulting in a supervision sample set for training the electromagnetic spectrum sensing model.
[0081] The baseline electromagnetic spectrum is a standard reference image data characterizing the distribution of electromagnetic signal intensity at each grid point.
[0082] Spatial location information refers to the three-dimensional coordinate data of a satellite in space, which may include longitude, latitude, and orbital altitude.
[0083] Optionally, acquiring the spatial position information of multiple satellite constellations and determining the received signal power and equivalent isotropic radiated power based on sampled electromagnetic spectrum data can be achieved by: deriving the longitude, latitude, and orbital altitude of each low-Earth orbit satellite at each sampling time from an orbit simulation tool, and extracting the received signal power value of each satellite's receiver from the sampled electromagnetic spectrum data. Simultaneously, based on the electromagnetic wave free-space propagation loss formula and considering the instantaneous spatial distance between each satellite and the space target, the equivalent isotropic radiated power value of each space target's transmitter is determined from the received signal power value. This specification does not limit the implementation of this method.
[0084] Optionally, the method for constructing a spatial location grid based on longitude and latitude and an orbital altitude field based on orbital altitude can be as follows: the longitude and latitude ranges are divided into uniform grid cells according to a preset spatial resolution, and the orbital altitude values of each satellite are filled into the corresponding grid cells according to their projection positions on the longitude-latitude grid. This specification does not limit this approach in the embodiments.
[0085] The spatial location grid is a regular grid structure formed by discretizing a spatial region according to the longitude and latitude dimensions.
[0086] The orbital height field is the field data formed by filling the spatial location grid with the orbital height values at each spatial location.
[0087] The position information of medium and high orbit space targets is the position coordinate data of the space targets in the airspace.
[0088] The position matrix is an encoded matrix formed by mapping the position information of a spatial target to a spatial position grid through an encoding method.
[0089] Optionally, one way to resample the received signal power and equivalent isotropic radiated power to a spatial location grid is to use cubic spline interpolation to resample the received signal power and equivalent isotropic radiated power to a globally covered regular latitude and longitude grid (e.g., a grid resolution of 512×1024, corresponding to a geographical range of -180° to 180° longitude and -90° to 90° latitude). Another way is to use bilinear interpolation, Kriging interpolation, or radial basis function interpolation to resample the received signal power and equivalent isotropic radiated power to a spatial location grid. This specification does not limit the implementation of this method.
[0090] Optionally, the received signal power, equivalent isotropic radiation power, and orbital altitude can be linearly normalized to scale them to the [0,1] interval, unifying the numerical characteristics of each feature to accelerate model training convergence and improve numerical stability.
[0091] Optionally, one way to map the orbital altitude field to a spatial location grid is to treat the dynamically changing satellite orbital altitude as an independent physical field, and then interpolate and normalize it before mapping it to the spatial location grid to form an altitude information matrix spatially aligned with the signal data. Another approach is to use Kriging interpolation to spatially interpolate and estimate the orbital altitude values of uncovered grid cells. This specification does not limit the implementation of this method.
[0092] Optionally, one way to encode the position information of medium- and high-orbit space targets into a position matrix and map it to a spatial position grid is to encode the position information of the space targets into a sparse matrix. The element values in the matrix represent the presence of a space target at the corresponding grid location and its orbital altitude as h; if no target exists, the value is 0. This can be used as key prior knowledge to participate in normalization and identify the spatial distribution of space targets. Another approach is to use one-hot encoding to map the target position to its corresponding grid cell. This specification does not limit the implementation of this method.
[0093] Spatial alignment refers to mapping multiple data sources from different channels to the same spatial location grid coordinate system, so that the various data have a one-to-one mapping relationship in spatial location.
[0094] A multi-feature input tensor is a multidimensional tensor data structure formed by stacking multiple data channels with different physical meanings along the channel dimension.
[0095] Optionally, one implementation of constructing training sample pairs using spatially aligned received signal power, equivalent isotropic radiated power, orbital height field, and position matrix as input samples, and a reference electromagnetic spectrum as a supervision label, can be as follows: The received signal power, equivalent isotropic radiated power, orbital height field, and position matrix are concatenated along the channel dimension to form a four-channel input tensor, and the reference electromagnetic spectrum is used as the output label. The input tensor and output label are then associated and stored as a single training sample pair. Another implementation can be as follows: The above four types of data are normalized separately and then stacked into a multi-channel tensor, which is then paired with the corresponding reference electromagnetic spectrum to form training samples. This specification does not limit the implementation in this way.
[0096] It should be noted that the formula for calculating the equivalent isotropic radiated power (EIRP) is as follows: ,in For transmission power, For antenna gain, This refers to feeder loss.
[0097] Free-space signal propagation follows the Friesian transport formula: ; in For receiving antenna gain, For the signal wavelength, For the transmission and reception distance, This represents the system loss coefficient.
[0098] To illustrate, the input sample consists of four types of data: received signal power ( Figure 6 A schematic diagram of the equivalent isotropic radiated power of a low-Earth orbit satellite constellation is shown; equivalent isotropic radiated power ( Figure 7 This specification shows a schematic diagram of the equivalent isotropic radiated power of a low-Earth orbit satellite constellation according to one embodiment, and the orbital altitude field ( Figure 8 This specification shows a schematic diagram of the orbital altitude field and position matrix of a low-Earth orbit satellite constellation according to one embodiment. Figure 9 This diagram illustrates a position matrix of a medium-to-high orbit space target according to an embodiment of this specification. The dataset can be divided into a training set, a validation set, and a test set in a 7:2:1 ratio. This division is performed at the scene level to ensure that each set contains independent target configurations and observation geometries, preventing data leakage.
[0099] For example, for the sampled electromagnetic spectrum data acquired by a set of low-Earth orbit satellite constellations at a certain sampling time, the longitude, latitude, and orbital altitude of each low-Earth orbit satellite at that time are exported from the orbit simulation tool. Based on the sampled data, the received signal power value of each satellite receiver and the equivalent isotropic radiation power value of each space target transmitter are determined. A spatial grid covering the target observation area is constructed according to a preset longitude and latitude step size. The orbital altitude value, received signal power value, and equivalent isotropic radiation power value are resampled to the same spatial grid by interpolation or direct mapping. At the same time, the instantaneous position of each space target is converted into a position matrix by encoding and mapped to the grid to obtain the spatially aligned received signal power map, equivalent isotropic radiation power map, orbital altitude field map, and position matrix map. The above four types of data are stacked as four input channels to form a multi-feature input tensor, which is paired with the corresponding reference electromagnetic spectrum map and stored as a training sample pair. All sampling times, all space target scenes, and all satellite constellations are traversed, and all sample pairs are summarized to form a supervised sample set.
[0100] In the embodiments of this specification, spatial position information of multiple satellite constellations is acquired, and the received signal power and equivalent isotropic radiation power are determined based on sampled electromagnetic spectrum data. The spatial position information includes longitude, latitude, and orbital altitude. A spatial position grid is constructed based on longitude and latitude, and an orbital altitude field is constructed based on orbital altitude. The received signal power and equivalent isotropic radiation power are resampled to the spatial position grid, the orbital altitude field is mapped to the spatial position grid, and the position information of medium- and high-orbit space targets is encoded into a position matrix and mapped to the spatial position grid, resulting in spatially aligned received signal power, equivalent isotropic radiation power, orbital altitude field, and position matrix. The spatially aligned received signal power, equivalent isotropic radiation power, orbital altitude field, and position matrix are then used to calculate the received signal power, equivalent isotropic radiation power, orbital altitude field, and position matrix. A matrix is used to construct a multi-feature input tensor as input samples, and a reference electromagnetic spectrum map is used as a supervision label to construct training sample pairs, resulting in a supervision sample set for training the electromagnetic spectrum sensing model. This can transform the irregular and sparsely distributed sampling data in the original simulation into regularized grid data, eliminating the problem of incompatible input formats. At the same time, by organizing four types of data—received signal power, equivalent isotropic radiation power, orbital height field, and position matrix—into a multi-feature input tensor, the model can obtain multi-dimensional contextual information such as propagation attenuation, transmitter radiation, observation platform distribution, and target spatial distribution from the same input sample. This improves the model's ability to extract comprehensive features from multimodal inputs, thereby enhancing the model's accuracy in accurately sensing and predicting the electromagnetic spectrum distribution.
[0101] In one optional embodiment of this specification, multiple satellite constellation parameters are obtained, including: The system iterates through the preset satellite constellation parameters within a preset step size to obtain multiple satellite constellation parameters.
[0102] The preset satellite constellation parameter range is a pre-defined range of parameter values used to generate the combination of satellite constellation parameters. For example, the orbital inclination range can be an inclination range, the orbital altitude range can be an altitude range, and the ascending node right ascension range can be a circumferential distribution range.
[0103] The preset step size is the interval between two adjacent parameter values when traversing and sampling within the preset parameter range.
[0104] Traversal refers to extracting the current parameter value at each step node within a preset parameter range according to a preset step size, thereby forming a parameter sequence that covers the entire parameter range.
[0105] Optionally, one way to obtain multiple satellite constellation parameters by traversing within a preset satellite constellation parameter range according to a preset step size is to: for a single parameter dimension, incrementally increase the value from the minimum value of the parameter range to the maximum value according to a preset step size, and collect the parameter values at each step size node to form multiple candidate parameters for that dimension. Another way is to: for multiple parameter dimensions, independently traverse each dimension according to the corresponding step size, and then combine and enumerate the candidate parameters of each dimension to form multiple sets of constellation parameter combinations covering the multi-parameter space. This specification does not limit the embodiments in this way.
[0106] For example, in the orbital inclination traversal dimension, the orbital inclination range is set from 0° to 90°, with an angle step of 10°. Starting from 0°, an inclination value is taken every 10° interval, sequentially obtaining 0°, 10°, 20° up to 90°, resulting in 10 candidate orbital inclination values. In the orbital altitude traversal dimension, the orbital altitude range is set from 400km to 1200km, with an orbital altitude step of 100km. Starting from 400km, an altitude value is taken every 100km interval, sequentially obtaining 40°, 10°, 20° up to 90°, resulting in 10 candidate orbital inclination values. Nine candidate orbital altitude values were obtained from 0km, 500km to 1200km. In the ascending node right ascension traversal dimension, the ascending node right ascension range was set from 0° to 360°, with an angle step of 30°. Starting from 0°, a right ascension value was taken every 30°, resulting in 12 candidate ascending node right ascension values. The candidate parameters in the above three dimensions were fully enumerated to obtain 1080 constellation parameter combinations. Each combination contains an orbital inclination value, an orbital altitude value, and an ascending node right ascension value.
[0107] In the embodiments of this specification, by traversing within a preset satellite constellation parameter range according to a preset step size, multiple satellite constellation parameters are obtained. This enables the formation of uniform and systematic sampling coverage in the parameter space, ensuring that the obtained multiple satellite constellation parameters have sufficient distribution breadth and density in terms of orbital inclination, orbital altitude, and right ascension of the ascending node. As a result, the observation blind spots of each satellite constellation constructed are complementary in space, avoiding the problem that multiple constellation observation geometries cannot form effective complementary coverage. This provides a reliable constellation configuration basis for generating the fusion reference true value with full coverage.
[0108] In one optional embodiment of this specification, the satellite constellation parameters include orbital inclination, and the preset step size includes angular step size; By traversing within a preset satellite constellation parameter range according to a preset step size, multiple satellite constellation parameters are obtained, including: According to the preset angle step size, the system traverses within the preset track inclination range to obtain multiple track inclination angles.
[0109] The orbital inclination is the numerical value of the angle between the satellite's orbital plane and the Earth's equatorial plane.
[0110] The angle step size is the numerical interval between two adjacent angle values when discretely sampling the orbital inclination or the right ascension of the ascending node. For example, iterates through the range of 0° to 90° with a step size of 5°.
[0111] Optionally, to obtain multiple track inclination angles by traversing within a preset track inclination angle range according to a preset angle step size, one approach is to set the starting value, ending value, and angle step size of the track inclination angle in the track simulation tool, and then have the simulation tool automatically generate and export all discrete inclination angle values from the starting value to the ending value according to the step size. Another approach is to calculate the inclination angle value at each step size node using a numerical calculation program in a loop iteration manner based on the preset track inclination angle range and angle step size, and store the calculation results as a candidate inclination angle list. This specification does not limit the implementation of this approach.
[0112] For example, in the orbital inclination traversal dimension, a specific orbital inclination range and a specific angle step size are set. Within the specific orbital inclination range, an inclination value is taken at every specific angle step size, and N candidate orbital inclination values are obtained in sequence. Based on these N orbital inclination values, N groups of low-Earth orbit satellite constellations are constructed. The orbital inclinations of each constellation are different, and the observation coverage areas of each constellation for space targets are staggered in the latitudinal direction. The blind spots of different constellations for space targets are complementary.
[0113] In the embodiments of this specification, by traversing within a preset orbital inclination range according to a preset angle step size, multiple orbital inclination angles are obtained. This enables multiple satellite constellations constructed based on multiple orbital inclination angles to complement each other in spatial latitude for the observation blind spots of space targets. This provides a data foundation with complementary coverage characteristics for the spatiotemporal fusion of subsequent multi-constellation sampling data, and improves the spatial integrity and coverage uniformity of the fusion reference true value.
[0114] In an optional embodiment of this specification, the satellite constellation parameters further include at least one of orbital altitude and right ascension of the ascending node, and the preset step size further includes at least one of orbital altitude step size and right ascension step size of the ascending node; The system iterates through the preset satellite constellation parameters within a preset step size to obtain multiple satellite constellation parameters, including at least one of the following: According to the preset track height step size, traverse within the preset track height range to obtain multiple track heights; According to the preset ascending node right ascension step size, traverse within the preset ascending node right ascension range to obtain multiple ascending node right ascensions.
[0115] The orbital altitude is the vertical height of the satellite above the Earth's surface.
[0116] The right ascension of the ascending node is the numerical value of the right ascension angle of the ascending node of the satellite orbit relative to the vernal equinox.
[0117] The orbital height step size is the numerical interval between two adjacent height values when the orbital height is discretely sampled.
[0118] The ascending node right ascension step size is the numerical interval between two adjacent right ascension values when discretely sampling the ascending node right ascension.
[0119] Optionally, the orbital inclination, orbital height, and right ascension of the ascending node can be combined into parameters, and multiple combined parameters can be obtained by iterating according to a preset step size of the combined parameters.
[0120] For example, in the orbital inclination traversal dimension, the orbital inclination range is set to 0° to 90°, with an angle step of 15°. Starting from 0°, an inclination value is taken every 15° interval, successively obtaining 0°, 15°, 30° up to 90°, resulting in a total of 7 candidate orbital inclination values; in the orbital altitude traversal dimension, the orbital altitude range is set to 400km to 1200km, with an orbital altitude step of 200km. Starting from 400km, an altitude value is taken every 200km interval, successively obtaining 400km, 600km, 150km, 30 ... Five candidate orbital altitude values were obtained at 00km, 800km, 1000km, and 1200km. In the ascending node right ascension traversal dimension, the ascending node right ascension range was set to 0° to 360°, with an angle step of 45°. Starting from 0°, a right ascension value was taken every 45°, resulting in eight candidate ascending node right ascension values. The candidate parameters in the above three dimensions were fully enumerated to obtain 280 constellation parameter combinations. Each combination contains an orbital inclination value, an orbital altitude value, and an ascending node right ascension value.
[0121] In the embodiments of this specification, based on obtaining multiple orbital inclinations by traversing within a preset orbital inclination range according to a preset angle step size, multiple orbital altitudes are obtained by traversing within a preset orbital altitude range according to a preset orbital altitude step size, and multiple ascending node right ascensions are obtained by traversing within a preset ascending node right ascension range according to a preset ascending node right ascension step size. This enables systematic coverage in the three-dimensional constellation parameter space, ensuring that the parameters of each group of satellite constellations have different distributions in the inclination, altitude, and right ascension dimensions. Furthermore, the observation geometry in space is complementary in multiple degrees of freedom, further enhancing the spatial integrity of the fused reference true value.
[0122] Corresponding to the above-described data processing method embodiments for electromagnetic spectrum sensing, this application also provides a model training method for electromagnetic spectrum sensing. Figure 10 A flowchart illustrating a model training method for electromagnetic spectrum sensing according to one embodiment of this specification is shown. Figure 10 As shown, the method includes: Step 1002: Obtain the supervised sample set, which is constructed based on the above-described data processing method for electromagnetic spectrum sensing.
[0123] Step 1004: Using the sampled electromagnetic spectrum data as training input and the fused baseline true value as supervision label, perform autoregressive training on the initial neural network model to obtain the electromagnetic spectrum sensing model.
[0124] For example, after obtaining the supervised sample set constructed based on the above data processing method, any training sample pair is extracted from the supervised sample set. The sampled electromagnetic spectrum data in the training sample pair is used as the input of the initial neural network model, and the fusion benchmark true value in the training sample pair is used as the supervision label of the model output. After forward propagation calculation, the model outputs the predicted spectrum distribution. The mean square error loss function is used to calculate the difference between the prediction result and the supervision label. Based on the difference, the model parameters are updated through backpropagation. After multiple trainings, the prediction result converges to the supervision label, and the electromagnetic spectrum sensing model is obtained.
[0125] In the embodiments of this specification, multiple satellite constellation parameters are acquired, and satellite constellations corresponding to these parameters are constructed. These multiple satellite constellations have different observation blind zones for medium- and high-orbit space targets. Electromagnetic spectrum data emitted by medium- and high-orbit space targets are sampled using these multiple satellite constellations to obtain sampled electromagnetic spectrum data for multiple satellite constellations. The sampled electromagnetic spectrum data from multiple satellite constellations are then spatiotemporally aligned and fused to obtain a fusion reference ground truth for multiple satellite constellations. Based on the fusion reference ground truth and the sampled electromagnetic spectrum data from multiple satellite constellations, a supervisory sample set is constructed for training the electromagnetic spectrum sensing model. It is possible to construct satellite constellations with different spatial geometric distributions using different constellation parameters, enabling each satellite constellation to complement each other in the observation blind zone. By using multi-constellation collaborative sampling, the observation blind zone that exists when observing a single constellation is compensated for. This allows the generated fusion reference ground truth to have complete spatial coverage characteristics across the entire airspace, eliminating spatial gaps in the reference ground truth. This improves the integrity and reliability of the ground truth data used in training the electromagnetic spectrum sensing model, enabling the deep learning model to learn and infer the spectrum distribution based on complete spatial context information. This enhances the accuracy and robustness of the trained model in sensing and predicting electromagnetic spectrum data in the airspace.
[0126] In an optional embodiment of this specification, the fused reference true value is the reference electromagnetic spectrum; Before step 1004, the method further includes: The system acquires spatial location information of multiple satellite constellations and determines the received signal power and equivalent isotropic radiated power based on sampled electromagnetic spectrum data. The spatial location information includes longitude, latitude, and orbital altitude. A spatial location grid is constructed based on longitude and latitude, and an orbital height field is constructed based on orbital height. The received signal power and equivalent isotropic radiation power are resampled to the spatial position grid, the orbital height field is mapped to the spatial position grid, and the position information of the medium-high orbit space target is encoded into a position matrix and mapped to the spatial position grid to obtain the spatially aligned received signal power, equivalent isotropic radiation power, orbital height field and position matrix. A multi-feature input tensor is constructed based on the spatially aligned received signal power, equivalent isotropic radiated power, orbital height field, and position matrix. Step 1004 includes: The multi-feature input tensor is input into the initial neural network model so that the initial neural network model can perform multimodal reasoning on the input tensor and output a predicted electromagnetic spectrum. Based on the loss value between the baseline electromagnetic spectrum and the predicted electromagnetic spectrum, the initial neural network model is trained autoregressively to obtain the electromagnetic spectrum sensing model.
[0127] For example, for any training sample pair in the supervised sample set, the longitude, latitude, and orbital altitude of each low-Earth orbit satellite at the sampling time are obtained, and the received signal power and equivalent isotropic radiated power are determined from the sampled electromagnetic spectrum data. A spatial grid is constructed according to a preset longitude and latitude step size. The orbital altitude, received signal power, and equivalent isotropic radiated power are resampled to grid cells. The instantaneous positions of each spatial target are converted into position matrices using Gaussian kernel encoding and mapped to the same grid, resulting in four spatially aligned two-dimensional data channels. The four data channels are stacked into a multi-feature input tensor and input into the initial neural network model. After feature extraction from each channel, the model performs cross-modal feature interaction through a fusion layer and outputs a predicted electromagnetic spectrum map through a decoding layer. A structural similarity loss combined with a mean squared error loss is used to calculate the comprehensive loss value between the predicted map and the reference electromagnetic spectrum map. Based on this loss value, the model parameters are updated through gradient descent. The above process is iteratively executed on all training sample pairs until the model converges, resulting in the trained electromagnetic spectrum sensing model.
[0128] In the embodiments of this specification, a multi-feature input tensor containing received signal power, equivalent isotropic radiation power, orbital altitude field, and position matrix is constructed. This multi-feature input tensor is then input into an initial neural network model, enabling the initial neural network model to perform multimodal inference on the input tensor and output a predicted electromagnetic spectrum map. Based on the loss value between the reference electromagnetic spectrum map and the predicted electromagnetic spectrum map, the initial neural network model is subjected to autoregressive training to obtain an electromagnetic spectrum sensing model. This model enables joint inference learning using four types of data during training: propagation loss, emitted radiation, spatial geometric distribution of satellite constellations, and spatial distribution of targets. This process uncovers the intrinsic correlations among multimodal data and improves the model's reconstruction accuracy and generalization ability of the electromagnetic spectrum map under complex spatial situations.
[0129] The above is an illustrative scheme of a model training method for electromagnetic spectrum sensing according to this embodiment. It should be noted that the technical solution of this model training method for electromagnetic spectrum sensing belongs to the same concept as the technical solution of the data processing method for electromagnetic spectrum sensing described above. Details not described in detail in the technical solution of the model training method for electromagnetic spectrum sensing can be found in the description of the technical solution of the data processing method for electromagnetic spectrum sensing described above.
[0130] Corresponding to the above data processing method embodiments, this application also provides an electromagnetic spectrum sensing method. Figure 11 A flowchart illustrating an electromagnetic spectrum sensing method according to one embodiment of this specification is shown. Figure 11 As shown, the method includes: Step 1102: Use the satellite constellation to sample the electromagnetic spectrum data emitted by medium- and high-orbit space targets in the airspace to be measured, and obtain the electromagnetic spectrum data to be measured.
[0131] Electromagnetic spectrum sampling data of space target transmitted signals collected by satellite constellations in the airspace to be tested.
[0132] For example, a constellation of low-Earth orbit (LEO) satellites is deployed in the target space. During operation, each LEO satellite continuously receives signals from space targets in the geosynchronous orbit zone and the medium Earth orbit region according to a preset sampling period. Each satellite collects electromagnetic wave signals emitted by the space targets through its receiving antenna. After processing by the receiver, the received signal power values at each frequency point are recorded. The received signal power values collected by all satellites at the same sampling time are summarized to obtain electromagnetic spectrum data reflecting the current electromagnetic signal distribution in the space to be measured.
[0133] Step 1104: Input the electromagnetic spectrum data to be measured into the electromagnetic spectrum sensing model and output the electromagnetic spectrum map corresponding to the spatial domain to be measured. The electromagnetic spectrum sensing model is trained based on the above-mentioned model training method for electromagnetic spectrum sensing.
[0134] The airspace to be measured is the target spatial region where electromagnetic spectrum sensing and prediction are required.
[0135] For example, after the electromagnetic spectrum data to be measured is converted into a regularized input format according to the same gridded preprocessing process as in the training phase, it is input into the electromagnetic spectrum sensing model pre-trained based on the above model training method. After forward inference calculation, the model outputs a complete electromagnetic spectrum map corresponding to the spatial domain to be measured. The spectrum map is presented in the form of a spatial grid, and the value of each grid point represents the predicted value of the electromagnetic signal intensity at that location, thereby achieving the goal of recovering the complete spatial electromagnetic spectrum distribution using limited measured sampling data.
[0136] In the embodiments of this specification, electromagnetic spectrum data emitted by medium- and high-orbit space targets in the airspace to be measured is sampled using a satellite constellation to obtain electromagnetic spectrum data to be measured. The electromagnetic spectrum data to be measured is input into an electromagnetic spectrum sensing model, which outputs an electromagnetic spectrum map corresponding to the airspace to be measured. The electromagnetic spectrum sensing model is trained based on the above-mentioned model training method. It can use the limited measured data collected by a single satellite constellation to obtain the electromagnetic spectrum distribution results of the complete airspace through the inference of the trained model. This avoids the difficulty of not being able to directly obtain the full-domain spectrum due to insufficient constellation coverage in actual measurements, and realizes the perception and prediction of the electromagnetic spectrum of the entire airspace based on sparse measured data.
[0137] In an optional embodiment of this specification, prior to step 1104, the following step is further included: The spatial location information of the satellite constellation is obtained, and the received signal power and equivalent isotropic radiated power are determined based on the electromagnetic spectrum data to be measured. The spatial location information includes longitude, latitude and orbital altitude. A spatial location grid is constructed based on longitude and latitude, and an orbital height field is constructed based on orbital height. The received signal power and equivalent isotropic radiation power are resampled to the spatial position grid, the orbital height field is mapped to the spatial position grid, and the position information of the medium-high orbit space target is encoded into a position matrix and mapped to the spatial position grid to obtain the spatially aligned received signal power, equivalent isotropic radiation power, orbital height field and position matrix. A multi-feature input tensor is constructed based on the spatially aligned received signal power, equivalent isotropic radiated power, orbital height field, and position matrix. Step 1104 includes: A multi-feature input tensor is input into the electromagnetic spectrum sensing model so that the initial neural network model can perform multimodal reasoning on the multi-feature input tensor and output the electromagnetic spectrum map corresponding to the spatial domain to be measured.
[0138] For example, the longitude, latitude, and orbital altitude of each satellite in the satellite constellation at the sampling time are obtained, and the received signal power value and the equivalent isotropic radiation power value of each space target are extracted from the electromagnetic spectrum data to be measured. A spatial grid is constructed according to a preset longitude and latitude step size. The orbital altitude value, received signal power value, and equivalent isotropic radiation power value are resampled to the spatial grid. The instantaneous position of each space target is encoded into a position matrix and mapped to the same spatial grid. The obtained received signal power map, equivalent isotropic radiation power map, orbital altitude field map, and position matrix map are stacked into a multi-feature input tensor. This tensor is fed into a pre-trained electromagnetic spectrum sensing model. The model extracts features from each channel through the encoding layer, performs cross-modal feature interaction through the fusion layer, and outputs the electromagnetic spectrum prediction map corresponding to the space domain to be measured by the decoding layer.
[0139] In the embodiments of this specification, the spatial position information of the satellite constellation is obtained, and the received signal power and equivalent isotropic radiation power are determined based on the electromagnetic spectrum data to be measured. The spatial position information includes longitude, latitude, and orbital altitude. A spatial position grid is constructed based on longitude and latitude, and an orbital altitude field is constructed based on orbital altitude. The received signal power and equivalent isotropic radiation power are resampled to the spatial position grid, the orbital altitude field is mapped to the spatial position grid, and the position information of the mid-to-high orbit space target is encoded into a position matrix and mapped to the spatial position grid to obtain the spatially aligned received signal power and equivalent isotropic radiation power. The system constructs a multi-feature input tensor based on the spatially aligned received signal power, equivalent isotropic radiated power, orbital height field, and position matrix. This multi-feature input tensor is then input into the electromagnetic spectrum sensing model, enabling the initial neural network model to perform multimodal inference on the multi-feature input tensor and output the electromagnetic spectrum map corresponding to the measured spatial domain. This multimodal input provides the model with multi-dimensional contextual information such as observation geometry, transmission characteristics, propagation loss, and target distribution, allowing the model to comprehensively utilize the complementarity of various modal data for joint inference and improving the output accuracy of the electromagnetic spectrum map.
[0140] The above is an illustrative scheme for electromagnetic spectrum sensing according to this embodiment. It should be noted that the technical solution for electromagnetic spectrum sensing and the technical solution for the data processing method for electromagnetic spectrum sensing described above belong to the same concept. For details not described in detail in the technical solution for electromagnetic spectrum sensing, please refer to the description of the technical solution for the data processing method for electromagnetic spectrum sensing described above.
[0141] Figure 12 A structural block diagram of a computing device according to one embodiment of this specification is shown. The components of the computing device 1200 include, but are not limited to, a memory 1210 and a processor 1220. The processor 1220 is connected to the memory 1210 via a bus 1230, and a content database 1250 is used to store content data.
[0142] The computing device 1200 also includes an access device 1240, which enables the computing device 1200 to communicate via one or more networks 1260. Examples of these networks include a Public Switched Telephone Network (PSTN), a Local Area Network (LAN), a Wide Area Network (WAN), a Personal Area Network (PAN), or a combination of communication networks such as the Internet. The access device 1240 may include one or more of any type of wired or wireless network interface (e.g., a Network Interface Controller (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, or a Near Field Communication (NFC) interface.
[0143] In one embodiment of this specification, the aforementioned components of the computing device 1200 and Figure 12 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 12 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.
[0144] The computing device 1200 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, content-based computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers (PCs). The computing device 1200 can also be a mobile or stationary server.
[0145] The processor 1220 is used to execute the following computer program / instructions, which, when executed by the processor, implement the above-mentioned data processing method for electromagnetic spectrum sensing, the model training method for electromagnetic spectrum sensing, and the above-mentioned steps for electromagnetic spectrum sensing.
[0146] The above is a schematic representation of a computing device according to this embodiment. It should be noted that the technical solution of this computing device belongs to the same concept as the technical solution of the data processing method for electromagnetic spectrum sensing described above. Details not described in detail in the technical solution of the computing device can be found in the description of the technical solution of the data processing method for electromagnetic spectrum sensing described above.
[0147] An embodiment of this specification also provides a computer-readable storage medium storing a computer program / instructions that, when executed by a processor, implement the aforementioned data processing method for electromagnetic spectrum sensing, the model training method for electromagnetic spectrum sensing, and the aforementioned steps for electromagnetic spectrum sensing.
[0148] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium belongs to the same concept as the technical solution of the data processing method for electromagnetic spectrum sensing described above. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the data processing method for electromagnetic spectrum sensing described above.
[0149] An embodiment of this specification also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the above-described data processing method for electromagnetic spectrum sensing, the model training method for electromagnetic spectrum sensing, and the steps of electromagnetic spectrum sensing.
[0150] The above is an illustrative scheme of a computer program product according to this embodiment. It should be noted that the technical solution of this computer program product and the technical solution of the above-described data processing method for electromagnetic spectrum sensing belong to the same concept. For details not described in detail in the technical solution of the computer program product, please refer to the description of the technical solution of the above-described data processing method for electromagnetic spectrum sensing.
[0151] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.
[0152] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.
[0153] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.
[0154] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.
[0155] The preferred embodiments disclosed above are merely illustrative of this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described herein. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.
Claims
1. A data processing method for electromagnetic spectrum sensing, characterized in that, include: Obtain parameters for multiple satellite constellations and construct satellite constellations corresponding to these parameters, wherein the multiple satellite constellations have different blind spots for observing medium and high orbit space targets; Using the multiple satellite constellations, the electromagnetic spectrum data emitted by the medium- and high-orbit space target is sampled to obtain the sampled electromagnetic spectrum data of the multiple satellite constellations; The sampled electromagnetic spectrum data of the multiple satellite constellations are spatiotemporally aligned and fused to obtain the fusion reference true value of the multiple satellite constellations; Based on the fusion baseline truth values of the multiple satellite constellations and the sampled electromagnetic spectrum data, a supervised sample set is constructed for training the electromagnetic spectrum sensing model.
2. The method according to claim 1, characterized in that, Before obtaining the parameters of multiple satellite constellations, the process also includes: Construct multiple space target scenarios, in which medium and high orbit space targets with different densities are constructed in each space target scenario; The step of sampling the electromagnetic spectrum data emitted by the medium- and high-orbit space target using the multiple satellite constellations to obtain sampled electromagnetic spectrum data of the multiple satellite constellations includes: By utilizing the aforementioned multiple satellite constellations, electromagnetic spectrum data emitted by medium- and high-orbit space targets in various space target scenarios are sampled to obtain sampled electromagnetic spectrum data corresponding to the multiple satellite constellations in each space target scenario.
3. The method according to claim 2, characterized in that, Before sampling the electromagnetic spectrum data emitted by medium- and high-orbit space targets in each space target scenario using the multiple satellite constellations to obtain the sampled electromagnetic spectrum data of the multiple satellite constellations corresponding to each space target scenario, the method further includes: Obtain multiple sampling times; The method of sampling electromagnetic spectrum data emitted by medium- and high-orbit space targets in various space target scenarios using the multiple satellite constellations to obtain sampled electromagnetic spectrum data corresponding to the multiple satellite constellations in each space target scenario includes: At any given sampling time, the electromagnetic spectrum data emitted by medium- and high-orbit space targets in each space target scenario are sampled in parallel using the multiple satellite constellations to obtain the sampled electromagnetic spectrum data of the multiple satellite constellations in each space target scenario at each sampling time.
4. The method according to claim 3, characterized in that, The process of performing spatiotemporal alignment and fusion of the sampled electromagnetic spectrum data of the multiple satellite constellations to obtain the fusion reference truth value of the multiple satellite constellations includes: The sampled electromagnetic spectrum data of the multiple satellite constellations at any sampling time are spatiotemporally aligned and fused to obtain the fusion reference true value of the multiple satellite constellations at each sampling time.
5. The method according to any one of claims 1-4, characterized in that, The supervised sample set for training the electromagnetic spectrum sensing model is constructed based on the fused baseline ground truth of the multiple satellite constellations and the sampled electromagnetic spectrum data, including: Using the sampled electromagnetic spectrum data of any satellite constellation as input samples and the fusion benchmark true value as supervision labels, training sample pairs are constructed to obtain a supervision sample set for training the electromagnetic spectrum sensing model.
6. The method according to claim 5, characterized in that, The fusion reference true value is the reference electromagnetic spectrum diagram; Before constructing the supervised sample set for training the electromagnetic spectrum sensing model based on the fused baseline truth values of the multiple satellite constellations and the sampled electromagnetic spectrum data, the method further includes: The spatial position information of the multiple satellite constellations is obtained, and the received signal power and equivalent isotropic radiated power are determined based on the sampled electromagnetic spectrum data, wherein the spatial position information includes longitude, latitude and orbital altitude; A spatial location grid is constructed based on the longitude and latitude, and an orbital height field is constructed based on the orbital height. The received signal power and equivalent isotropic radiation power are resampled to the spatial position grid, the orbital height field is mapped to the spatial position grid, and the position information of the medium-high orbit space target is encoded into a position matrix and mapped to the spatial position grid to obtain the spatially aligned received signal power, equivalent isotropic radiation power, orbital height field and position matrix. The process involves using sampled electromagnetic spectrum data from any satellite constellation as input samples and the fused baseline ground truth as supervision labels to construct training sample pairs, resulting in a supervision sample set for training the electromagnetic spectrum sensing model. This set includes: Using the spatially aligned received signal power, equivalent isotropic radiation power, orbital height field, and position matrix to construct a multi-feature input tensor as input samples, and using the reference electromagnetic spectrum map as supervision labels, training sample pairs are constructed to obtain a supervision sample set for training the electromagnetic spectrum sensing model.
7. The method according to claim 1, characterized in that, The acquisition of multiple satellite constellation parameters includes: The system iterates through the preset satellite constellation parameters within a preset step size to obtain multiple satellite constellation parameters.
8. The method according to claim 7, characterized in that, The satellite constellation parameters include orbital inclination, and the preset step size includes angular step size; The process involves traversing within a preset satellite constellation parameter range according to a preset step size to obtain multiple satellite constellation parameters, including: According to the preset angle step size, the system traverses within the preset track inclination range to obtain multiple track inclination angles.
9. A model training method for electromagnetic spectrum sensing, characterized in that, include: Obtain a supervised sample set, wherein the supervised sample set is constructed based on the method described in any one of claims 1 to 8; Using the sampled electromagnetic spectrum data as training input and the fused baseline ground truth as supervision label, the initial neural network model is trained autoregressively to obtain the electromagnetic spectrum sensing model.
10. The method according to claim 9, characterized in that, The fusion reference true value is the reference electromagnetic spectrum diagram; Before using the sampled electromagnetic spectrum data as training input and the fused benchmark ground truth as a supervision label to perform autoregressive training on the initial neural network model, the method further includes: The spatial position information of the multiple satellite constellations is obtained, and the received signal power and equivalent isotropic radiated power are determined based on the sampled electromagnetic spectrum data, wherein the spatial position information includes longitude, latitude and orbital altitude; A spatial location grid is constructed based on the longitude and latitude, and an orbital height field is constructed based on the orbital height. The received signal power and equivalent isotropic radiation power are resampled to the spatial position grid, the orbital height field is mapped to the spatial position grid, and the position information of the medium-high orbit space target is encoded into a position matrix and mapped to the spatial position grid to obtain the spatially aligned received signal power, equivalent isotropic radiation power, orbital height field and position matrix. A multi-feature input tensor is constructed based on the spatially aligned received signal power, equivalent isotropic radiation power, orbital height field, and position matrix. The process of using the sampled electromagnetic spectrum data as training input and the fused baseline ground truth as a supervision label to perform autoregressive training on the initial neural network model to obtain the electromagnetic spectrum sensing model includes: The multi-feature input tensor is input into an initial neural network model, so that the initial neural network model performs multimodal reasoning on the input tensor and outputs a predicted electromagnetic spectrum. Based on the loss value between the baseline electromagnetic spectrum and the predicted electromagnetic spectrum, the initial neural network model is trained autoregressively to obtain the electromagnetic spectrum sensing model.
11. An electromagnetic spectrum sensing method, characterized in that, include: The electromagnetic spectrum data emitted by medium- and high-orbit space targets in the airspace to be measured is sampled using a satellite constellation to obtain the electromagnetic spectrum data to be measured. The electromagnetic spectrum data to be measured is input into the electromagnetic spectrum sensing model, and the electromagnetic spectrum map corresponding to the spatial domain to be measured is output. The electromagnetic spectrum sensing model is trained based on the method described in claim 9 or 10.
12. The method according to claim 11, characterized in that, Before inputting the electromagnetic spectrum data to be measured into the electromagnetic spectrum sensing model and outputting the electromagnetic spectrum map corresponding to the spatial domain to be measured, the method further includes: The spatial location information of the satellite constellation is obtained, and the received signal power and equivalent isotropic radiated power are determined based on the electromagnetic spectrum data to be measured. The spatial location information includes longitude, latitude, and orbital altitude. A spatial location grid is constructed based on the longitude and latitude, and an orbital height field is constructed based on the orbital height. The received signal power and equivalent isotropic radiation power are resampled to the spatial position grid, the orbital height field is mapped to the spatial position grid, and the position information of the medium-high orbit space target is encoded into a position matrix and mapped to the spatial position grid to obtain the spatially aligned received signal power, equivalent isotropic radiation power, orbital height field and position matrix. A multi-feature input tensor is constructed based on the spatially aligned received signal power, equivalent isotropic radiation power, orbital height field, and position matrix. The step of inputting the electromagnetic spectrum data to be measured into the electromagnetic spectrum sensing model and outputting the electromagnetic spectrum map corresponding to the spatial domain to be measured includes: The multi-feature input tensor is input into the electromagnetic spectrum sensing model so that the initial neural network model performs multimodal reasoning on the multi-feature input tensor and outputs the electromagnetic spectrum map corresponding to the spatial domain to be measured.
13. A computing device, characterized in that, include: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 12.
14. A computer-readable storage medium, characterized in that, It stores a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 12.
15. A computer program product, characterized in that, Includes a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 12.
