Anti-intelligent modeling method and system based on physical field characteristic information fusion

Abstract

The application discloses a kind of anti-no intelligent modeling method and system based on physical field characteristic information fusion, belong to anti-unmanned aerial vehicle detection technical field, the method includes: obtaining background electromagnetic field reference, unmanned aerial vehicle equipment characteristics, environmental field and electromagnetic disturbance data;Equipment physical AI model and environmental physical AI model are constructed;Quantify the influence of inter-field coupling by multi-physical field coupling analysis model, generate coupling correction parameters;Based on real-time scene parameter dynamic weighted fusion generates multi-physical field fusion physical AI model, input real-time disturbance data deduces reconstruction unmanned aerial vehicle scattering characteristics, position, motion state and type, and high-frequency dynamic update model.Measured results show that the positioning error of the application is ≤10 m, the false detection rate is ≤10%, the hardware cost is reduced by more than 60%, and the silent unmanned aerial vehicle can be effectively detected, which is suitable for key area prevention and control such as airport and nuclear power plant.

Description

Technical Field

[0001] This invention relates to the field of anti-drone detection technology, and more specifically, to an anti-drone intelligent modeling method and system based on the fusion of physical field characteristic information. Background Technology

[0002] Low-altitude unmanned aerial vehicles (UAVs), especially those characterized by low altitude, low speed, and small size, pose significant detection and identification challenges in critical area security due to their small radar cross-section, low flight altitude, flexible flight paths, and low operating costs. Existing counter-UAV detection technologies typically rely on radar, electro-optical, acoustic, radio detection, or multi-sensor combinations. Radio detection schemes usually focus on the UAV's remote control or image transmission links, while radar and electro-optical schemes rely on active detection or visual visibility. However, in scenarios involving electromagnetically silent UAVs, complex urban obstructions, multipath propagation, strong background electromagnetic interference, severe weather conditions, and limited hardware deployment, existing solutions still struggle to balance detection stability, positioning accuracy, and deployment cost.

[0003] At present, utilizing the existing background electromagnetic field within the monitoring area as the basic sensing environment and identifying the disturbances caused by intruding drones to the background electromagnetic field is a promising counter-drone detection approach. The drone's body, rotor, motor, payload, and material structure can scatter, reflect, absorb, and cause phase disturbances to the background electromagnetic waves. These disturbances can be collected by software-defined radio devices and used for target detection, localization, and identification. However, to stably transform these disturbances into usable target parameters, a modeling system capable of simultaneously describing equipment characteristics, background electromagnetic field characteristics, and the influence of complex environments is needed. However, existing modeling techniques suffer from at least the following problems:

[0004] Single-physics-field modeling has obvious limitations. Existing models often only model a single factor among electromagnetic propagation, target scattering, weather effects, or terrain occlusion, without fully considering the influence of weather on electromagnetic propagation, the influence of terrain and buildings on the propagation path of scattered signals, and the influence of dynamic background electromagnetic interference on disturbance feature extraction, resulting in insufficient model accuracy and generalization ability.

[0005] The models are out of touch with real-world application scenarios. Some models rely primarily on idealized simulation data or static parameters, failing to integrate the scattering characteristics of real UAV equipment, such as materials, dimensions, attitude, and rotor status, with the background electromagnetic field reference data, dynamic noise data, and real-time environmental data within the actual monitoring area. This results in poor model adaptability in complex real-world scenarios.

[0006] The model update mechanism is lagging. Traditional model parameters are usually fixed after offline training or updated only at a low frequency, making it difficult to adapt to dynamic factors such as sudden changes in wind speed, humidity, increased electromagnetic interference in communication, and changes in temporary building obstruction in a timely manner, thus affecting the real-time performance of UAV disturbance feature recognition and localization inference.

[0007] Multiphysics fusion lacks a unified framework. In existing technologies, different sub-models often operate independently, lacking a unified mechanism for sub-model access, coupling analysis, fusion optimization, and dynamic updating. This makes it difficult for the outputs of various models to form stable collaboration, ultimately affecting the reliability of anti-drone detection and positioning results.

[0008] Therefore, there is a need to provide an anti-intelligence modeling scheme based on the fusion of physical field characteristic information to solve the problems of insufficient accuracy of single physical field models, disconnect between models and actual scenes, lagging dynamic updates, and insufficient reliability of multi-field fusion in existing technologies. Summary of the Invention

[0009] The purpose of this invention is to provide an anti-drone intelligent modeling method and system based on the fusion of physical field characteristic information, so as to solve the problems of insufficient multi-physical field coupling, insufficient fusion of real equipment characteristics and dynamic environment data, inability of the model to adapt to changes in the actual scene at high frequency, and difficulty in the model output to stably support drone detection and positioning in existing anti-drone modeling technologies.

[0010] According to a first aspect of the present invention, an anti-intelligence modeling method based on the fusion of physical field characteristic information is provided, comprising:

[0011] The system acquires background electromagnetic field reference data, UAV equipment characteristic data, environmental field data, and electromagnetic disturbance data synchronously collected by software radio monitoring points in the monitoring area. The environmental field data includes at least meteorological data, dynamic electromagnetic environment data, and terrain and building data.

[0012] Based on the background electromagnetic field reference data and the UAV equipment characteristic data, an equipment physical AI model is constructed. The equipment physical AI model includes a background electromagnetic field reference sub-model and a UAV electromagnetic scattering disturbance sub-model. The background electromagnetic field reference sub-model is used to characterize the field strength distribution, propagation attenuation and phase distribution of the known background electromagnetic field in the monitoring area. The UAV electromagnetic scattering disturbance sub-model is used to characterize the amplitude distortion, frequency shift and phase disturbance caused by the UAV in the background electromagnetic field.

[0013] An environmental physics AI model is constructed based on the environmental field data. The environmental physics AI model includes a meteorological sub-model, a dynamic electromagnetic environment sub-model, and a terrain and building sub-model, which are used to quantify the effects of meteorological factors, background electromagnetic interference, and terrain and building shading, reflection, or diffraction on the propagation of the background electromagnetic field and the scattering disturbance of the UAV.

[0014] A multiphysics coupling analysis model is established to quantify the coupling effect between the equipment physical AI model and the environmental physical AI model, and to obtain coupling correction parameters that include at least meteorological-electromagnetic propagation coupling terms, electromagnetic noise-disturbance characteristic coupling terms, and terrain and building-scattering propagation coupling terms.

[0015] Based on real-time scene parameters, the equipment physical AI model, the environment physical AI model, and the coupling correction parameters are dynamically weighted and fused to generate a multi-physics field fusion physical AI model.

[0016] The real-time collected electromagnetic disturbance data is input into the multi-physics fusion physical AI model to infer and reconstruct the scattering disturbance characteristics, target position, target motion state and target type of the UAV target, and the model parameters are updated dynamically at high frequency based on the inference and reconstruction results.

[0017] Optionally, obtaining the background electromagnetic field reference data includes:

[0018] During the baseline period when no drone targets are observed, background signal spectrum, field strength, phase, and noise floor data for one or more frequency bands are collected through multiple software radio monitoring points distributed throughout the monitoring area.

[0019] The background signal spectrum, field strength, phase, and noise floor data are synchronized in time, calibrated in frequency, and registered in space to form a background electromagnetic field reference map corresponding to the coordinates of the monitoring area.

[0020] Optionally, constructing the background electromagnetic field reference sub-model includes: using Maxwell's equations and the electromagnetic wave propagation equation as physical constraints, inputting the monitoring area boundary conditions, frequency band parameters, field strength sampling values ​​and propagation path parameters into a neural network, outputting the spatial field strength distribution, propagation attenuation coefficient, phase distribution and confidence level of the background electromagnetic field, and training the neural network with constraints using measured background electromagnetic field reference data.

[0021] Optionally, constructing the electromagnetic scattering perturbation sub-model of the UAV includes:

[0022] Acquire at least one of the following equipment characteristics of the drone: model, size, material, attitude, rotor speed, and motion speed;

[0023] Electromagnetic scattering simulation data are generated based on physical optics methods and the method of moments.

[0024] The electromagnetic scattering simulation data and the measured scattering perturbation data are used together as training samples to train a hybrid network including a convolutional feature extraction layer and a temporal feature extraction layer, so as to output the scattering cross section, perturbation amplitude, perturbation phase and perturbation frequency shift of the UAV under different attitudes and motion states.

[0025] Optionally, constructing the environmental physical AI model includes:

[0026] Meteorological sub-models are constructed based on wind speed, wind direction, air pressure, temperature, and humidity to output meteorological propagation correction factors;

[0027] A dynamic electromagnetic environment sub-model is constructed based on civilian communication signals, industrial electromagnetic radiation, and transient electromagnetic interference within the monitoring area to output dynamic noise baseline and interference suppression parameters;

[0028] Based on 3D terrain data, building outlines, heights, materials, and spatial locations, a terrain-building sub-model is constructed to output occlusion loss, reflection path, diffraction path, and multipath propagation correction factor.

[0029] Optionally, the environmental impact fusion of the meteorological sub-model, dynamic electromagnetic environment sub-model, and terrain and building sub-model includes:

[0030] The meteorological propagation correction factor, dynamic noise baseline, interference suppression parameter, occlusion loss, reflection path, diffraction path, and multipath propagation correction factor are input into the attention fusion network;

[0031] The attention fusion network generates attention weights for each environmental sub-model based on the real-time environmental change magnitude, the signal-to-noise ratio of monitoring points, and historical inference errors, and outputs the comprehensive environmental correction amount.

[0032] Optionally, the multiphysics coupling analysis model is established in the following manner:

[0033] Construct a joint state vector that includes equipment field variables, meteorological field variables, electromagnetic environment variables, and terrain and building variables;

[0034] Calculate the coupling sensitivity matrix between different physical fields based on the joint state vector;

[0035] The electromagnetic disturbance characteristics are corrected using the coupling sensitivity matrix, so that the corrected electromagnetic disturbance characteristics simultaneously reflect the scattering characteristics of the UAV, changes in background electromagnetic propagation, and the influence of environmental propagation.

[0036] Optionally, the dynamic weighted fusion includes:

[0037] The equipment physical AI model output, the environmental physical AI model output, and the coupling correction parameters are used as fusion inputs;

[0038] Based on the signal-to-noise ratio, weather change rate, background electromagnetic interference intensity, terrain occlusion degree, and historical positioning error in the real-time scene parameters, calculate the equipment weight, environment weight, and coupling weight.

[0039] The fusion output of the multiphysics fusion physical AI model is generated using the equipment weight, environment weight, and coupling weight. When the background electromagnetic interference intensity or meteorological change rate increases, the environment weight and coupling weight are increased.

[0040] Optionally, the high-frequency dynamic update includes:

[0041] Acquire new environmental field data and new electromagnetic disturbance data at a period of no more than 300ms;

[0042] The newly added environmental field data and the newly added electromagnetic disturbance data are subjected to anomaly removal and confidence assessment;

[0043] Based on retaining the historical model parameters, incremental learning is used to update some parameters of the multiphysics fusion physical AI model, and the inference error within the verification window determines whether to roll back the current update or adjust the subsequent update step size.

[0044] In a second aspect, the present invention also provides an anti-unintelligent modeling system based on the fusion of physical field characteristic information, which adopts the above-mentioned anti-unintelligent modeling method based on the fusion of physical field characteristic information, including a perception layer, a transmission layer, a computing layer and a control terminal.

[0045] The sensing layer includes software-defined radio monitoring points, a background electromagnetic field reference calibration module, and environmental parameter sensors, wherein:

[0046] The software-defined radio monitoring point is used to synchronously collect electromagnetic disturbance data in the monitoring area;

[0047] The background electromagnetic field reference calibration module is used to acquire background electromagnetic field reference data of the monitoring area.

[0048] The environmental parameter sensor is used to acquire environmental field data, which includes at least meteorological data, dynamic electromagnetic environment data, and terrain and building data.

[0049] The perception layer is also used to acquire UAV equipment characteristic data;

[0050] The transmission layer connects the perception layer and the computing layer, and is used to transmit the electromagnetic disturbance data, background electromagnetic field reference data, environmental field data, and UAV equipment characteristic data.

[0051] The computing layer includes edge computing nodes and a central server, and the central server is deployed with the following modules:

[0052] The equipment physical AI model is constructed based on the background electromagnetic field reference data and the UAV equipment characteristic data. The equipment physical AI model includes a background electromagnetic field reference sub-model and a UAV electromagnetic scattering disturbance sub-model. The background electromagnetic field reference sub-model is used to characterize the field strength distribution, propagation attenuation and phase distribution of the known background electromagnetic field in the monitoring area. The UAV electromagnetic scattering disturbance sub-model is used to characterize the amplitude distortion, frequency shift and phase disturbance caused by the UAV in the background electromagnetic field.

[0053] An environmental physics AI model is used to construct based on the environmental field data. The environmental physics AI model includes a meteorological sub-model, a dynamic electromagnetic environment sub-model, and a terrain and building sub-model, which are used to quantify the effects of meteorological factors, background electromagnetic interference, and terrain and building shading, reflection, or diffraction on the propagation of the background electromagnetic field and the scattering disturbance of the UAV.

[0054] A multiphysics coupling analysis model is used to establish a coupling analysis model between the equipment physical AI model and the environmental physical AI model, quantify the coupling effect between the two, and obtain coupling correction parameters including at least meteorological-electromagnetic propagation coupling term, electromagnetic noise-disturbance characteristic coupling term, and terrain and building-scattering propagation coupling term.

[0055] The dynamic weighted fusion module is used to dynamically weight and fuse the equipment physical AI model, the environmental physical AI model and the coupling correction parameters according to real-time scene parameters to generate a multi-physics fusion physical AI model.

[0056] The deduction and reconstruction module is used to input the real-time collected electromagnetic disturbance data into the multi-physics fusion physical AI model to deduce and reconstruct the scattering disturbance characteristics, target position, target motion state and target type of the UAV target;

[0057] The incremental update module is used to perform high-frequency dynamic updates on the parameters of the multi-physics fusion physical AI model based on the inference and reconstruction results;

[0058] The control terminal is connected to the computing layer and is used to receive and display the scattering disturbance characteristics, target position, target motion state and target type of the UAV target.

[0059] Compared with the prior art, the present invention has at least the following beneficial effects:

[0060] This invention integrates drone scattering, background electromagnetic propagation, meteorology, electromagnetic interference, and the effects of terrain and buildings into the modeling process by equipping a physical AI model, an environmental physical AI model, and a multiphysics coupling analysis model, thereby improving the model's ability to express and generalize complex scenarios.

[0061] To address the issue of the model being out of sync with the actual scenario, this invention uses background electromagnetic field reference data, real UAV equipment characteristic data, real-time environmental field data, and measured electromagnetic disturbance data together for modeling, so that the model output matches the real physical environment and real equipment disturbance characteristics of the monitoring area.

[0062] This invention employs a high-frequency dynamic update and incremental learning mechanism with a cycle of no more than 300ms, which reduces the overhead of repeated training while enabling the model to quickly adapt to changes in wind speed, humidity, background electromagnetic noise, and propagation path.

[0063] This invention utilizes the coupling sensitivity matrix to correct electromagnetic disturbance characteristics, and dynamically weights and fuses the equipment physical AI model, the environment physical AI model, and the coupling correction parameters according to real-time scene parameters. It adaptively adjusts the contribution ratio of equipment, environment, and coupling terms according to real-time scene parameters, thereby improving the stability and interpretability of the model output.

[0064] This invention can provide a directly callable underlying physical AI model output for anti-drone detection and positioning systems, including scattering disturbance prediction, environmental baseline, disturbance correction features, target position, target motion state and target type, thereby improving the detection, positioning and identification capabilities of low, slow and small drones, and helping to reduce the cost pressure caused by dense hardware deployment.

[0065] Other features and advantages of the invention will become clear from the following detailed description of exemplary embodiments of the invention with reference to the accompanying drawings. Attached Figure Description

[0066] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments of the invention and, together with their description, serve to explain the principles of the invention.

[0067] Figure 1 This is a schematic diagram of the anti-unintelligent modeling method provided in an embodiment of the present invention;

[0068] Figure 2 This is a schematic diagram of the anti-unintelligent modeling system provided in an embodiment of the present invention.

[0069] Explanation of reference numerals in the attached figures:

[0070] 100. Perception Layer; 110. Software Radio Monitoring Point; 120. Background Electromagnetic Field Reference Calibration Module; 130. Environmental Parameter Sensor; 200. Transmission Layer; 300. Computation Layer; 310. Edge Computing Node; 320. Central Server; 321. Equipment Physical AI Model; 322. Environmental Physical AI Model; 323. Multiphysics Coupled Analysis Model; 324. Dynamic Weighted Fusion Module; 325. Inference and Reconstruction Module; 326. Incremental Update Module; 400. Prevention and Control Terminal. Detailed Implementation

[0071] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the invention.

[0072] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0073] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0074] In all the examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0075] Example 1: This invention proposes an example of an anti-intelligence modeling method based on the fusion of physical field characteristic information. Specifically, as follows... Figure 1 As shown, it includes:

[0076] The system acquires background electromagnetic field baseline data, UAV equipment characteristic data, environmental field data, and electromagnetic disturbance data synchronously collected from software-defined radio monitoring points in the monitoring area. The environmental field data includes at least meteorological data, dynamic electromagnetic environment data, and terrain and building data. It should be noted that the UAV equipment characteristic data refers to a set of parameters related to the UAV's own physical properties, specifically including at least one of the following: UAV model, size, material, attitude, rotor speed, and motion speed. These data describe the physical basis for the scattering disturbance generated by the target UAV on the background electromagnetic field. Using computational electromagnetics methods such as Physical Optics (PO) and the Method of Moments (MoM), based on inputs such as the UAV's geometric dimensions and material parameters, corresponding electromagnetic scattering simulation data is generated. This simulation data can cover the scattering characteristics of different models, attitudes, and motion states, solving the problems of high difficulty and cost in acquiring actual measurement data.

[0077] In this embodiment of the invention, acquiring the background electromagnetic field reference data includes: during a reference time period when no UAV target is observed, collecting background signal spectrum, field strength, phase, and noise floor data for one or more frequency bands through multiple software-defined radio monitoring points distributed in the monitoring area; performing time synchronization, frequency calibration, and spatial registration on the background signal spectrum, field strength, phase, and noise floor data to form a background electromagnetic field reference map corresponding to the coordinates of the monitoring area. The comprehensive acquisition of multi-dimensional data provides a complete data foundation for subsequent modeling; the establishment of the background electromagnetic field reference map eliminates equipment errors and time deviations, solves the false detection problem caused by incomplete or inaccurate background data in traditional methods, and ensures the reliability of subsequent disturbance detection.

[0078] Based on the background electromagnetic field reference data and the UAV equipment characteristic data, an equipment physics AI model is constructed. This model includes a background electromagnetic field reference sub-model and a UAV electromagnetic scattering disturbance sub-model. The background electromagnetic field reference sub-model characterizes the field strength distribution, propagation attenuation, and phase distribution of the known background electromagnetic field within the monitoring area. The UAV electromagnetic scattering disturbance sub-model characterizes the amplitude distortion, frequency shift, and phase disturbance induced by the UAV in the background electromagnetic field. By separating the background electromagnetic field and the UAV scattering disturbance, the equipment physics AI model achieves accurate modeling of the interaction mechanism between the UAV and the background environment, avoiding the shortcomings of traditional single models that cannot distinguish between background changes and target disturbances, thus improving the physical interpretability of target identification.

[0079] It should be noted that, in this embodiment of the invention, constructing the background electromagnetic field reference sub-model includes: using Maxwell's equations and the electromagnetic wave propagation equation as physical constraints, inputting the monitoring area boundary conditions, frequency band parameters, field strength sampling values, and propagation path parameters into a neural network, outputting the spatial field strength distribution, propagation attenuation coefficient, phase distribution, and confidence level of the background electromagnetic field, and training the neural network with constraints using measured background electromagnetic field reference data. Introducing physical constraints such as Maxwell's equations and embedding physical laws into the neural network training ensures that the model output conforms to the objective laws of electromagnetic propagation, improving the model's generalization ability and prediction reliability in unknown scenarios.

[0080] In this embodiment of the invention, constructing the UAV electromagnetic scattering perturbation sub-model includes: acquiring at least one equipment characteristic data of the UAV, including its model, size, material, attitude, rotor speed, and motion speed; generating corresponding electromagnetic scattering simulation data based on physical optics methods and the method of moments; and using the electromagnetic scattering simulation data and measured scattering perturbation data as training samples to train a hybrid network including a convolutional feature extraction layer and a temporal feature extraction layer to output the scattering cross section, perturbation amplitude, perturbation phase, and perturbation frequency shift of the UAV under different attitudes and motion states. The convolutional feature extraction layer uses a two-dimensional convolutional kernel sliding on a time-frequency graph to automatically learn and extract local features. The input data for the temporal feature extraction layer is a high-dimensional feature vector sequence output by the convolutional layer, or the original continuous IQ signal sequence, using a recurrent neural network (RNN) or its variants, such as an LSTM long short-term memory network or a GRU gated recurrent unit. The hybrid network architecture represents a deep fusion of deep learning and computational electromagnetics, mining the spatial physical fingerprint of electromagnetic signals through the convolutional layer and capturing the dynamic physical laws of target motion through the temporal layer. The electromagnetic scattering perturbation sub-model of the UAV trained by this hybrid network can output the accurate physical parameters of the UAV under different attitudes and motion states from end to end.

[0081] It leverages the advantage of simulation data covering multiple operating conditions and calibrates the model accuracy through measured data, enabling it to accurately capture the scattering characteristics of UAVs in different attitudes and motion states, thus solving the problem of insufficient description of dynamic target scattering characteristics by traditional models.

[0082] An environmental physics AI model is constructed based on the environmental field data. The environmental physics AI model includes a meteorological sub-model, a dynamic electromagnetic environment sub-model, and a terrain and building sub-model, which are used to quantify the effects of meteorological factors, background electromagnetic interference, and terrain and building shading, reflection, or diffraction on the propagation of the background electromagnetic field and the scattering disturbance of the UAV.

[0083] In this embodiment of the invention, the construction of the environmental physics AI model includes: constructing a meteorological sub-model based on wind speed, wind direction, air pressure, temperature, and humidity to output meteorological propagation correction factors; constructing a dynamic electromagnetic environment sub-model based on civilian communication signals, industrial electromagnetic radiation, and transient electromagnetic interference within the monitoring area to output dynamic noise baselines and interference suppression parameters; and constructing a terrain and building sub-model based on three-dimensional terrain data, building outlines, heights, materials, and spatial locations to output shading loss, reflection paths, diffraction paths, and multipath propagation correction factors. The environmental physics AI model, by subdividing meteorological, electromagnetic, and terrain sub-models, comprehensively quantifies the interference factors of complex environments on electromagnetic propagation, providing multi-dimensional basis for subsequent environmental correction and effectively overcoming the positioning deviation caused by neglecting environmental coupling in traditional methods. Civilian communication signals refer to continuous radiation signals generated by mobile communication infrastructure deployed by operators within the monitoring area, including but not limited to 4G / 5G base station signals, Wi-Fi router signals, and Bluetooth device signals. Industrial electromagnetic radiation refers to broadband or narrowband electromagnetic radiation generated by industrial automation equipment, high-voltage transmission lines, rail transit systems, and medical imaging equipment, excluding civilian communication facilities. Transient electromagnetic interference refers to occasional, pulsed, abnormal electromagnetic radiation, including but not limited to electrostatic discharge, transient motor start-up and shutdown, power supply surges, and illegal radio transmissions. This type of interference manifests as sudden spikes in the time domain and as broad-spectrum energy diffusion in the frequency domain, making it highly susceptible to misinterpretation as UAV target signals.

[0084] Specifically, the construction of the meteorological sub-model includes atmospheric refractive index profile modeling, atmospheric attenuation modeling, and model training and output.

[0085] The meteorological sub-model first calculates the atmospheric refractive index based on measured meteorological data. The refractive index is calculated using the following formula: , Where P is the air pressure (hPa), T is the absolute temperature (K), and e is the water vapor pressure (hPa), a refractive index profile model is established based on this, and the refraction bending and propagation delay of electromagnetic waves are calculated by using the ray tracing method or the mapping function model through the refractive index profile.

[0086] Oxygen and water vapor in the atmosphere absorb and attenuate electromagnetic waves. The meteorological sub-model refers to the oxygen and water vapor attenuation calculation methods given in relevant recommendations of the International Telecommunication Union (ITU-R) (such as ITU-R Recommendation P.676), calculating the atmospheric gas attenuation coefficient based on temperature, humidity, and air pressure parameters. In the rainfall scenario, a rainfall attenuation model is further introduced to calculate the electromagnetic wave amplitude attenuation and phase error caused by rainfall.

[0087] The meteorological sub-model is constructed using a combination of physics-driven and data-driven approaches: electromagnetic wave propagation physics equations (such as ray tracing and parabolic equation methods) serve as physical constraints, and measured meteorological-propagation correlation data are used for parameter calibration or neural network training, ultimately outputting a meteorological propagation correction factor. This correction factor quantifies the impact of meteorological factors on the propagation path, propagation loss, and phase change of the background electromagnetic field. The meteorological propagation correction factor is then input into an attention fusion network, where it, along with the outputs of the environmental and topographic sub-models, generates a comprehensive environmental correction. When meteorological conditions change abruptly, the meteorological sub-model synchronously outputs an updated correction factor and adjusts the model parameters through a high-frequency dynamic update mechanism.

[0088] The input to the dynamic electromagnetic environment sub-model is real-time data from various electromagnetic interference sources within the monitoring area, specifically including civilian communication signals, industrial electromagnetic radiation, and transient electromagnetic interference. The core task of the dynamic electromagnetic environment sub-model is to establish a dynamic baseline of electromagnetic noise and interference in the monitoring area and to provide interference suppression parameters. Firstly, the dynamic electromagnetic environment sub-model establishes a noise floor model for the monitoring area based on long-term, wide-band spectrum monitoring data. This model can refer to the anthropogenic noise prediction model in ITU-R Recommendation P.372, but needs to be localized and calibrated using measured data. The noise baseline can be extracted using methods such as multi-scale morphological filtering. By detecting changes in the noise floor estimate at different filtering scales, the accuracy of the rapidly changing noise floor estimation can be improved. The dynamic noise baseline includes a real-time updated background noise power spectral density, used to distinguish UAV disturbance signals from background noise; interference suppression parameters include interference frequency markers, interference intensity coefficients, and adaptive filter coefficients, used to weaken the impact of non-target interference in subsequent signal processing. The dynamic noise baseline and interference suppression parameters are input into an attention fusion network. When the intensity of background electromagnetic interference increases, the dynamic weighted fusion module will correspondingly increase the environmental weight and coupling weight, thereby weakening the impact of non-target electromagnetic interference on the identification of UAV disturbance features.

[0089] The input to the terrain and architecture sub-model is 3D geographic information data of the monitoring area, specifically including 3D terrain data, building outlines, heights, materials, and spatial locations. This data can be obtained by importing 3D maps, building structural parameters, or laser scanning data. First, based on the input 3D terrain and building data, a 3D electromagnetic propagation scene model of the monitoring area is established. Based on this model, ray-tracing is used to analyze multiple propagation paths of electromagnetic waves from the emission source (or scattering source) to the receiving point. Based on this multi-path analysis, the terrain and architecture sub-model outputs occlusion loss, reflection paths, diffraction paths, and multipath propagation correction factors. In densely built-up areas such as important conference venues, the terrain and architecture sub-model accurately quantifies the occlusion and multipath effects around the venues. Combined with multiphysics fusion simulation, it achieves synchronous output of UAV position, speed, and model.

[0090] In this embodiment of the invention, the environmental impact fusion of the meteorological sub-model, dynamic electromagnetic environment sub-model, and terrain and building sub-model includes: inputting the meteorological propagation correction factor, dynamic noise baseline, interference suppression parameter, occlusion loss, reflection path, diffraction path, and multipath propagation correction factor into an attention fusion network; the attention fusion network generates attention weights for each environmental sub-model based on the real-time environmental change amplitude, monitoring point signal-to-noise ratio, and historical extrapolation error, and outputs a comprehensive environmental correction. The attention fusion network first normalizes the input meteorological propagation correction factor, dynamic noise baseline, interference suppression parameter, occlusion loss, reflection path, diffraction path, and multipath propagation correction factor to unify the dimensions and numerical ranges of each correction factor, eliminating the interference of different physical dimensions on the fusion result; it then maps the correction factors output by each sub-model to the same high-dimensional feature space through a fully connected coding layer, obtaining the feature vectors corresponding to each of the meteorological sub-model, dynamic electromagnetic environment sub-model, and terrain and building sub-model; and uses the real-time environmental change amplitude, monitoring point signal-to-noise ratio, and historical extrapolation error as prior parameters for weight adjustment. The process involves several steps: First, the initial attention score of each sub-model is calculated. Then, the three prior parameters are concatenated into an environment state vector, which interacts with the initial attention score. Finally, the attention weights of each environment sub-model are generated using the Softmax normalization function, with the sum of all sub-model weights being 1. A higher weight indicates a greater contribution of the corresponding sub-model's output to the correction result in the current environment. For example, when the real-time environment changes significantly, the attention weight of the meteorological sub-model automatically increases. When the signal-to-noise ratio of the monitoring point is low, the weight of the dynamic electromagnetic environment sub-model is increased accordingly to enhance the interference suppression effect. The correction factor of each sub-model is weighted and summed with its corresponding attention weight to obtain the comprehensive environmental correction amount. This correction amount comprehensively reflects the combined influence of meteorological, electromagnetic, and terrain factors on the propagation of electromagnetic signals in the current scenario. The attention fusion network is trained using labeled samples from the actual test scenario. The training samples include the true values ​​of the correction factors under different meteorological, electromagnetic, and terrain conditions. During training, the mean square error between the comprehensive environmental correction amount and the true value is used as the main loss function, while physical constraint terms are added. The network parameters are iteratively optimized using the backpropagation algorithm until the model's inference error on the validation set meets the preset requirements.

[0091] The weights of each environmental sub-model are dynamically adjusted, and the environmental correction amount is adaptively optimized according to the real-time scene. This avoids the limitations of fixed weight fusion and further improves the flexibility and accuracy of environmental interference suppression.

[0092] A multiphysics coupling analysis model was established to quantify the coupling effects between the equipment physical AI model and the environmental physical AI model, obtaining coupling correction parameters including at least meteorological-electromagnetic propagation coupling terms, electromagnetic noise-disturbance feature coupling terms, and terrain / building-scattering propagation coupling terms. This approach overcomes the limitations of traditional single-field modeling, enabling the model to more realistically reflect physical processes in complex scenarios and improving the accuracy of the inference results.

[0093] In this embodiment of the invention, the multiphysics coupling analysis model is established as follows: a joint state vector is constructed, comprising equipment field variables, meteorological field variables, electromagnetic environment variables, and terrain and building variables; a coupling sensitivity matrix between different physical fields is calculated based on the joint state vector; and the electromagnetic disturbance characteristics are corrected using the coupling sensitivity matrix, so that the corrected electromagnetic disturbance characteristics simultaneously reflect the scattering characteristics of the UAV, changes in background electromagnetic propagation, and environmental propagation effects. In this embodiment, the terrain and building variables participate in the coupling analysis as static / quasi-static structural features of the environment. In practical engineering applications, terrain and building information can be converted into "numerical variables" consistent with the aforementioned variables in the following way:

[0094] Topographic feature parameterization: Extracting topographic elevation data, surface roughness, mountain slope, etc., into continuous spatial distribution parameters;

[0095] Building geometry and material characteristics: Transforming the height, density, outline distribution of buildings, and the electromagnetic reflection / transmission loss coefficients of building materials (such as concrete, glass, and metal) into specific numerical characteristics;

[0096] Spatial topological features: Using a digital elevation model (DEM) or a 3D urban model, scalar or vector values ​​such as the occlusion rate and multipath propagation prediction coefficient of the target monitoring area relative to the surrounding terrain and buildings are extracted. Through the above processing, the terrain and buildings are transformed into a set of discrete feature variables / parameters with clear physical meaning and numerical range.

[0097] By combining the state vector and the coupling sensitivity matrix, the degree of mutual influence between the various physical fields was clarified, enabling precise correction of the electromagnetic disturbance characteristics and ensuring that the corrected characteristics simultaneously contain comprehensive information on target scattering, background propagation, and environmental influence.

[0098] The equipment physical AI model, the environment physical AI model, and the coupling correction parameters are dynamically weighted and fused based on real-time scene parameters to generate a multi-physics fused physical AI model. This allows the fused model to optimize performance for different scenarios, improving its robustness and scene adaptability.

[0099] In this embodiment of the invention, the dynamic weighted fusion includes: using the equipment physical AI model output, the environment physical AI model output, and coupling correction parameters as fusion inputs; calculating equipment weights, environment weights, and coupling weights based on real-time scene parameters such as signal-to-noise ratio, meteorological change rate, background electromagnetic interference intensity, terrain occlusion degree, and historical positioning error; and generating the fusion output of the multi-physics fusion physical AI model using the equipment weights, environment weights, and coupling weights, wherein the environment weights and coupling weights are increased when the background electromagnetic interference intensity or meteorological change rate increases. This enables the model to adaptively cope with complex environmental changes, avoids the performance degradation of the fixed-weight model in extreme scenarios, and further enhances the stability and reliability of the system.

[0100] Real-time collected electromagnetic disturbance data is input into the multi-physics fusion physical AI model to infer and reconstruct the scattering disturbance characteristics, target position, target motion state, and target type of the UAV target. Based on the inference and reconstruction results, the model parameters are updated dynamically at high frequency. Real-time inference and reconstruction achieves comprehensive perception of the UAV target, providing complete information support for prevention and control decisions. The high-frequency dynamic update mechanism enables the model to adapt to changes in the environment and the target in a timely manner, maintaining long-term operational accuracy.

[0101] In this embodiment of the invention, the high-frequency dynamic update includes: acquiring new environmental field data and new electromagnetic disturbance data at a period of no more than 300ms, with the high-frequency update period of ≤300ms ensuring the model's rapid response to environmental changes; performing anomaly removal and confidence assessment on the new environmental field data and new electromagnetic disturbance data to avoid contaminating the model with erroneous data; updating some parameters of the multiphysics fusion physical AI model using incremental learning while retaining historical model parameters, and determining whether to roll back the current update or adjust the subsequent update step size based on the inference error within the verification window, thereby reducing computational overhead, preventing model drift, and ensuring the stability and security of long-term updates.

[0102] Example 2: This invention also provides an example of an anti-intelligence modeling system based on the fusion of physical field characteristic information, as detailed below. Figure 2 As shown, the above-mentioned anti-intelligence modeling method based on the fusion of physical field characteristic information includes a perception layer 100, a transmission layer 200, a computing layer 300, and a prevention and control terminal 400.

[0103] The sensing layer 100 includes a software-defined radio monitoring point 110, a background electromagnetic field reference calibration module 120, and an environmental parameter sensor 130. The software-defined radio monitoring point 110 is used to synchronously collect electromagnetic disturbance data of the monitoring area; the background electromagnetic field reference calibration module 120 is used to acquire background electromagnetic field reference data of the monitoring area; the environmental parameter sensor 130 may include a meteorological sensor, an electromagnetic environment monitoring device, and a terrain and building data acquisition device. The meteorological sensor can collect wind speed, wind direction, air pressure, temperature, and humidity; the electromagnetic environment monitoring device can collect noise floor, spectral occupancy rate, and interference intensity; the terrain and building data acquisition device can import 3D maps, building structure parameters, or laser scan data.

[0104] The transmission layer 200 connects the sensing layer 100 and the computing layer 300, and is used to transmit the electromagnetic disturbance data, background electromagnetic field reference data, environmental field data, and UAV equipment characteristic data. The transmission layer 200 can employ a local area network, gigabit Ethernet, 5G industrial communication module, or a combination thereof. The transmission layer 200 performs timestamp calibration, data packet integrity verification, and transmission priority management on data from different monitoring points and different sensors. Preferably, electromagnetic disturbance data has a higher transmission priority, and environmental field data is uploaded synchronously according to the model update cycle.

[0105] The computing layer 300 includes edge computing nodes 310 and a central server 320. The edge computing nodes 310 are deployed close to the perception layer 100 and are used for preprocessing and preliminary disturbance detection of the raw electromagnetic disturbance data. Preprocessing includes filtering, denoising, detrending, time-frequency conversion, feature enhancement, and outlier removal. Preliminary disturbance detection is used to determine whether the current data contains disturbance characteristics distinct from background noise and ordinary electromagnetic interference, thereby reducing the unnecessary computational load on the central server 320. The central server 320 is deployed with the following modules:

[0106] Equipment physical AI model 321 is used to construct based on the background electromagnetic field reference data and the UAV equipment characteristic data. The equipment physical AI model 321 includes a background electromagnetic field reference sub-model and a UAV electromagnetic scattering disturbance sub-model. The background electromagnetic field reference sub-model is used to characterize the field strength distribution, propagation attenuation and phase distribution of the known background electromagnetic field in the monitoring area. The UAV electromagnetic scattering disturbance sub-model is used to characterize the amplitude distortion, frequency shift and phase disturbance caused by the UAV in the background electromagnetic field.

[0107] An environmental physics AI model 322 is used to construct based on the environmental field data. The environmental physics AI model 322 includes a meteorological sub-model, a dynamic electromagnetic environment sub-model, and a terrain and building sub-model, which are used to quantify the effects of meteorological factors, background electromagnetic interference, and terrain and building shading, reflection, or diffraction on the propagation of the background electromagnetic field and the scattering disturbance of the UAV.

[0108] Multiphysics coupling analysis model 323 is used to establish a coupling analysis model between the equipment physical AI model 321 and the environmental physical AI model 322, quantify the coupling effect between the two, and obtain coupling correction parameters including at least meteorological-electromagnetic propagation coupling term, electromagnetic noise-disturbance characteristic coupling term, and terrain and building-scattering propagation coupling term;

[0109] The dynamic weighted fusion module 324 is used to dynamically weight and fuse the equipment physical AI model 321, the environmental physical AI model 322 and the coupling correction parameters according to real-time scene parameters to generate a multi-physics fusion physical AI model.

[0110] The deduction and reconstruction module 325 is used to input the real-time collected electromagnetic disturbance data into the multi-physics field fusion physical AI model to deduce and reconstruct the scattering disturbance characteristics, target position, target motion state and target type of the UAV target;

[0111] The incremental update module 326 is used to perform high-frequency dynamic updates on the parameters of the multiphysics fusion physical AI model based on the inference and reconstruction results.

[0112] The control terminal 400 is connected to the computing layer 300 and is used to display UAV target detection results, location coordinates, trajectory, UAV type, confidence level, and alarm information. The control terminal 400 can also feed back manual confirmation results or handling results to the central server 320 for further optimization of the model sample library and update strategies.

[0113] Example 3: Sparse Grid Deployment and Model Application. In practical applications, to reduce hardware deployment costs, the system can adopt a sparse grid deployment strategy. Before deployment, the system uses computational electromagnetics simulation to determine the locations of candidate monitoring points based on the monitoring area, background electromagnetic field intensity distribution, terrain and building obstruction, and target control level. Subsequently, the minimum set of monitoring points is selected based on coverage, signal-to-noise ratio, geometric configuration between monitoring points, and data redundancy. By combining multi-dimensional constraints with computational electromagnetics simulation to screen monitoring points, the number of hardware deployments is compressed to the minimum set while ensuring monitoring coverage and signal-to-noise ratio. Hardware costs are reduced by more than 60% compared to dense deployment, while avoiding the transmission and computational burden caused by redundant data.

[0114] Under sparse deployment conditions, direct sampling data may be insufficient in some areas. This invention utilizes a background electromagnetic field reference sub-model and a multi-physics fusion physical AI model to perform spatial interpolation and perturbation reconstruction on sparse monitoring data. Specifically, based on the real-time perturbation characteristics of deployed monitoring points, the spatial coordinates of the monitoring points, the background electromagnetic field reference spectrum, and the comprehensive environmental correction, the equivalent perturbation distribution of undeployed locations is reconstructed. Thus, while reducing the number of monitoring points, the ability to predict target positions and trajectories can be maintained, enabling the system to maintain a positioning accuracy of ≤10m even under sparse grid conditions. This solves the problem of target misses or trajectory breaks that are easily caused by traditional sparse deployment.

[0115] In a specific implementation scenario, the monitoring area is the area surrounding an important conference venue. The system first collects background electromagnetic field baseline data in the absence of drones and establishes a background electromagnetic field baseline map. Subsequently, the system imports a 3D model of the buildings surrounding the venue and deploys meteorological sensors and electromagnetic environment monitoring equipment. When a drone enters the monitoring area, its fuselage and rotor structure generate scattering disturbances to the background electromagnetic field. Multiple software-defined radio monitoring points simultaneously collect disturbance signals, and edge computing nodes extract disturbance amplitude, frequency shift, and phase jitter characteristics. The central server combines equipment physics AI models, environmental physics AI models, and coupling correction parameters to extrapolate and reconstruct the drone's position, speed, and model, and outputs the results to the control terminal. This accurately quantifies the obstruction and multipath effects around the venue. Combined with multi-physics fusion extrapolation, it achieves synchronous output of drone position, speed, and model, with a extrapolation latency ≤500ms and a model recognition accuracy ≥80%, providing real-time and accurate control data for important event security.

[0116] In another specific implementation scenario, the monitoring area is the perimeter of an airport. Due to the presence of communication signals, navigation signals, and electromagnetic radiation from vehicles and equipment around the airport, the background electromagnetic environment is highly dynamic. The system's dynamic electromagnetic environment sub-model continuously outputs a dynamic noise baseline, and the dynamic weighted fusion module increases the environmental weight and coupling weight based on the intensity of background electromagnetic interference, thereby weakening the impact of non-target electromagnetic interference on UAV disturbance feature recognition. If meteorological conditions change abruptly, such as strong winds or rainfall causing changes in the propagation environment, the system's meteorological sub-model synchronously outputs a meteorological propagation correction factor and adjusts the model parameters through a high-frequency dynamic update mechanism to keep the inference and reconstruction results stable. This effectively suppresses the impact of background interference such as communication and navigation on target recognition, maintaining a false detection rate of ≤10% even under strong interference and abrupt weather changes.

[0117] Example 4: During the model training phase, the system can use experimental data, simulation data, and real-time monitoring data to jointly build a training sample library. Experimental data may include no fewer than 100,000 experimental data points on UAV scattering and background electromagnetic field characteristics, covering UAVs of different materials, sizes, attitudes, rotor states, and model types, as well as background electromagnetic fields of different frequency bands, field strengths, and noise levels. Simulation data may include no fewer than 500,000 multiphysics coupling simulation data points, covering different meteorological conditions, different electromagnetic interference intensities, and different terrain and building obstruction conditions.

[0118] During training, the system trains a background electromagnetic field baseline sub-model, a UAV electromagnetic scattering disturbance sub-model, a meteorological sub-model, a dynamic electromagnetic environment sub-model, and a terrain and building sub-model, respectively. After the sub-models are trained, the attention fusion network, the coupling sensitivity matrix, and the dynamic weighted fusion module are trained. The training loss may include disturbance feature prediction error, environmental correction error, positioning error, aircraft type recognition error, physical constraint loss, and update stability loss.

[0119] During the verification phase, the system can be tested and verified for model adaptability in complex outdoor scenarios. Verification metrics include UAV scattering disturbance prediction accuracy, target detection false detection rate, positioning error, inference delay, aircraft type recognition accuracy, and model update overhead. Preferably, under typical application conditions, the system can achieve a UAV scattering disturbance prediction accuracy of no less than 85%, a planar positioning error of no more than 10m, an inference delay of no more than 500ms, a false detection rate of no more than 10%, and an aircraft type recognition accuracy of no less than 80%. The above performance metrics are not intended to limit the scope of protection, but rather to illustrate the technical effects achievable by the present invention in a preferred embodiment.

[0120] Example 5: To avoid model drift caused by high-frequency dynamic updates, this example further provides a model update safety mechanism. The incremental update module 326 establishes an update sample set and a validation sample set before each update. The update sample set consists of environmental field data and electromagnetic disturbance data from the latest period, while the validation sample set consists of data with high confidence from the most recent several periods.

[0121] After updating the model parameters, the incremental update module 326 calculates the change in inference error using the validation sample set. If the updated inference error is lower than the original inference error, or the error increment does not exceed a preset tolerance threshold, the update is accepted. If the updated inference error exceeds the preset threshold, or the target trajectory exhibits abrupt changes that do not conform to the motion continuity constraint, the parameters are rolled back to their original state, and the next update step size is reduced or the data confidence threshold is increased. This safety mechanism enables the model to maintain its high-frequency adaptability while reducing the risk of model drift caused by abnormal sensor data, transient strong interference, or mislabeled feedback.

[0122] The above description of the structure, features, and effects of the present invention is based on the embodiments shown in the figures. However, the above are only preferred embodiments of the present invention. It should be noted that the technical features involved in the above embodiments and their preferred methods can be reasonably combined and matched by those skilled in the art to form a variety of equivalent solutions without departing from or changing the design concept and technical effects of the present invention. Therefore, the present invention is not limited to the scope of implementation shown in the figures. Any changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments, that do not exceed the spirit covered by the specification and figures, should be within the protection scope of the present invention.

Claims

1. A method for anti-intelligence modeling based on the fusion of physical field characteristic information, characterized in that, include: The system acquires background electromagnetic field reference data, UAV equipment characteristic data, environmental field data, and electromagnetic disturbance data synchronously collected by software radio monitoring points in the monitoring area. The environmental field data includes at least meteorological data, dynamic electromagnetic environment data, and terrain and building data. Based on the background electromagnetic field reference data and the UAV equipment characteristic data, an equipment physical AI model is constructed. The equipment physical AI model includes a background electromagnetic field reference sub-model and a UAV electromagnetic scattering disturbance sub-model. The background electromagnetic field reference sub-model is used to characterize the field strength distribution, propagation attenuation, and phase distribution of the known background electromagnetic field in the monitoring area. The UAV electromagnetic scattering disturbance sub-model is used to characterize the amplitude distortion, frequency shift, and phase disturbance induced by the UAV in the background electromagnetic field. An environmental physics AI model is constructed based on the environmental field data. The environmental physics AI model includes a meteorological sub-model, a dynamic electromagnetic environment sub-model, and a terrain and building sub-model, which are used to quantify the effects of meteorological factors, background electromagnetic interference, and terrain and building shading, reflection, or diffraction on the propagation of the background electromagnetic field and the scattering disturbance of the UAV. A multiphysics coupling analysis model is established to quantify the coupling effect between the equipment physical AI model and the environmental physical AI model, and to obtain coupling correction parameters that include at least meteorological-electromagnetic propagation coupling terms, electromagnetic noise-disturbance characteristic coupling terms, and terrain and building-scattering propagation coupling terms. Based on real-time scene parameters, the equipment physical AI model, the environment physical AI model, and the coupling correction parameters are dynamically weighted and fused to generate a multi-physics field fusion physical AI model. The real-time collected electromagnetic disturbance data is input into the multi-physics fusion physical AI model to infer and reconstruct the scattering disturbance characteristics, target position, target motion state and target type of the UAV target, and the model parameters are updated dynamically at high frequency based on the inference and reconstruction results.

2. The anti-intelligence modeling method based on the fusion of physical field characteristic information according to claim 1, characterized in that, Obtaining the background electromagnetic field reference data includes: During the baseline period when no drone targets are observed, background signal spectrum, field strength, phase, and noise floor data for one or more frequency bands are collected through multiple software radio monitoring points distributed throughout the monitoring area. The background signal spectrum, field strength, phase, and noise floor data are synchronized in time, calibrated in frequency, and registered in space to form a background electromagnetic field reference map corresponding to the coordinates of the monitoring area.

3. The anti-intelligence modeling method based on the fusion of physical field characteristic information according to claim 1, characterized in that, The construction of the background electromagnetic field reference sub-model includes: using Maxwell's equations and the electromagnetic wave propagation equation as physical constraints, inputting the monitoring area boundary conditions, frequency band parameters, field strength sampling values ​​and propagation path parameters into a neural network, outputting the spatial field strength distribution, propagation attenuation coefficient, phase distribution and confidence level of the background electromagnetic field, and training the neural network with constraints using measured background electromagnetic field reference data.

4. The anti-intelligence modeling method based on the fusion of physical field characteristic information according to claim 1, characterized in that, The construction of the UAV electromagnetic scattering perturbation sub-model includes: Acquire at least one of the following equipment characteristics of the drone: model, size, material, attitude, rotor speed, and motion speed; Electromagnetic scattering simulation data are generated based on physical optics methods and the method of moments. The electromagnetic scattering simulation data and the measured scattering perturbation data are used together as training samples to train a hybrid network including a convolutional feature extraction layer and a temporal feature extraction layer, so as to output the scattering cross section, perturbation amplitude, perturbation phase and perturbation frequency shift of the UAV under different attitudes and motion states.

5. The anti-intelligence modeling method based on the fusion of physical field characteristic information according to claim 1, characterized in that, Constructing the environmental physics AI model includes: Meteorological sub-models are constructed based on wind speed, wind direction, air pressure, temperature, and humidity to output meteorological propagation correction factors; A dynamic electromagnetic environment sub-model is constructed based on civilian communication signals, industrial electromagnetic radiation, and transient electromagnetic interference within the monitoring area to output dynamic noise baseline and interference suppression parameters; Based on 3D terrain data, building outlines, heights, materials, and spatial locations, a terrain-building sub-model is constructed to output occlusion loss, reflection path, diffraction path, and multipath propagation correction factor.

6. The anti-intelligence modeling method based on the fusion of physical field characteristic information according to claim 5, characterized in that, The environmental impact fusion of the meteorological sub-model, dynamic electromagnetic environment sub-model, and terrain and building sub-model includes: The meteorological propagation correction factor, dynamic noise baseline, interference suppression parameter, occlusion loss, reflection path, diffraction path, and multipath propagation correction factor are input into the attention fusion network; The attention fusion network generates attention weights for each environmental sub-model based on the real-time environmental change magnitude, the signal-to-noise ratio of monitoring points, and historical inference errors, and outputs the comprehensive environmental correction amount.

7. The anti-intelligence modeling method based on the fusion of physical field characteristic information according to claim 1, characterized in that, The multiphysics coupling analysis model is established in the following way: Construct a joint state vector that includes equipment field variables, meteorological field variables, electromagnetic environment variables, and terrain and building variables; Calculate the coupling sensitivity matrix between different physical fields based on the joint state vector; The electromagnetic disturbance characteristics are corrected using the coupling sensitivity matrix, so that the corrected electromagnetic disturbance characteristics simultaneously reflect the scattering characteristics of the UAV, changes in background electromagnetic propagation, and the influence of environmental propagation.

8. The anti-intelligence modeling method based on the fusion of physical field characteristic information according to claim 1, characterized in that, The dynamic weighted fusion includes: The equipment physical AI model output, the environmental physical AI model output, and the coupling correction parameters are used as fusion inputs; Based on the signal-to-noise ratio, weather change rate, background electromagnetic interference intensity, terrain occlusion degree, and historical positioning error in the real-time scene parameters, calculate the equipment weight, environment weight, and coupling weight. The fusion output of the multiphysics fusion physical AI model is generated using the equipment weight, environment weight, and coupling weight. When the background electromagnetic interference intensity or meteorological change rate increases, the environment weight and coupling weight are increased.

9. The anti-intelligence modeling method based on the fusion of physical field characteristic information according to claim 1, characterized in that, The high-frequency dynamic update includes: Acquire new environmental field data and new electromagnetic disturbance data at a period of no more than 300ms; The newly added environmental field data and the newly added electromagnetic disturbance data are subjected to anomaly removal and confidence assessment; Based on retaining the historical model parameters, incremental learning is used to update some parameters of the multiphysics fusion physical AI model, and the inference error within the verification window determines whether to roll back the current update or adjust the subsequent update step size.

10. An anti-intelligence modeling system based on fusion of physical field characteristic information, employing the anti-intelligence modeling method based on fusion of physical field characteristic information as described in any one of claims 1 to 9, characterized in that, It includes a perception layer (100), a transmission layer (200), a computing layer (300), and a prevention and control terminal (400). The sensing layer (100) includes a software-defined radio monitoring point (110), a background electromagnetic field reference calibration module (120), and an environmental parameter sensor (130), wherein: The software radio monitoring point (110) is used to synchronously collect electromagnetic disturbance data in the monitoring area; The background electromagnetic field reference calibration module (120) is used to acquire background electromagnetic field reference data of the monitoring area; The environmental parameter sensor (130) is used to acquire environmental field data, which includes at least meteorological data, dynamic electromagnetic environment data, and terrain and building data. The perception layer (100) is also used to acquire UAV equipment characteristic data; The transmission layer (200) connects the perception layer (100) and the computing layer (300) and is used to transmit the electromagnetic disturbance data, background electromagnetic field reference data, environmental field data and UAV equipment characteristic data. The computing layer (300) includes edge computing nodes (310) and a central server (320), the central server (320) being deployed with the following modules: The equipment physical AI model (321) is used to construct based on the background electromagnetic field reference data and the UAV equipment characteristic data. The equipment physical AI model (321) includes a background electromagnetic field reference sub-model and a UAV electromagnetic scattering disturbance sub-model. The background electromagnetic field reference sub-model is used to characterize the field strength distribution, propagation attenuation and phase distribution of the known background electromagnetic field in the monitoring area. The UAV electromagnetic scattering disturbance sub-model is used to characterize the amplitude distortion, frequency shift and phase disturbance caused by the UAV in the background electromagnetic field. An environmental physics AI model (322) is used to construct based on the environmental field data. The environmental physics AI model (322) includes a meteorological sub-model, a dynamic electromagnetic environment sub-model, and a terrain and building sub-model, which are used to quantify the effects of meteorological factors, background electromagnetic interference, and terrain and building shading, reflection, or diffraction on the propagation of the background electromagnetic field and the scattering disturbance of the UAV. A multiphysics coupling analysis model (323) is used to establish a coupling analysis model between the equipment physical AI model (321) and the environmental physical AI model (322), quantify the coupling effect between the two, and obtain coupling correction parameters including at least meteorological-electromagnetic propagation coupling term, electromagnetic noise-disturbance characteristic coupling term and terrain-building-scattering propagation coupling term; The dynamic weighted fusion module (324) is used to dynamically weight and fuse the equipment physical AI model (321), the environmental physical AI model (322) and the coupling correction parameters according to real-time scene parameters to generate a multi-physics fusion physical AI model. The deduction and reconstruction module (325) is used to input the real-time collected electromagnetic disturbance data into the multi-physics field fusion physical AI model to deduce and reconstruct the scattering disturbance characteristics, target position, target motion state and target type of the UAV target; The incremental update module (326) is used to perform high-frequency dynamic updates on the parameters of the multi-physics fusion physical AI model based on the inference and reconstruction results; The control terminal (400) is connected to the computing layer (300) and is used to receive and display the scattering disturbance characteristics, target position, target motion state and target type of the UAV target.

Images