Diffusion model radar track information probability prediction method based on conditional prior

Abstract

The invention discloses a diffusion model radar track information probability prediction method based on conditional prior. The method comprises the following steps: acquiring historical track information of a target; inputting the historical track information into the trained condition prediction network to obtain predicted track information of the target; wherein the condition prediction network is used for extracting non-stationary factors and potential variables when the target moves to obtain non-stationary features and potential variable information, and predicting a future track of the target according to the non-stationary features and the potential variable information; the predicted track information is input into the trained diffusion model to generate a plurality of possible predicted track sequences through random sampling and gradual denoising processes, predicted track distribution of the target is obtained, and the predicted track distribution further comprises uncertainty information of each predicted track sequence. The prediction information provided by the invention is more and comprises the quantitative information of the uncertainty of each prediction track, and reliable probability reference and firm decision support can be provided for subsequent decision tasks.

Description

Technical Field

[0001] This invention belongs to the field of radar signal processing technology, specifically relating to a probabilistic prediction method for radar track information based on a conditional prior diffusion model. Background Technology

[0002] Radar trajectory prediction is a core research direction in the fields of radar data processing and target situational awareness. Essentially, it uses historical target observation data acquired by radar sensors to model target motion patterns and environmental interference characteristics to predict the target's spatial position, motion state, and evolution trend over a future period. It serves as the decision-making basis for critical scenarios such as airspace security, air traffic control, and complex airspace situational awareness. In airspace security scenarios, accurate radar trajectory prediction can predict target flight trajectories several seconds to tens of seconds in advance, providing sufficient reaction time for relevant protection systems. In air traffic control, trajectory prediction needs to avoid flight path conflicts between multiple aircraft in real time to ensure airspace operational safety. In complex airspace situational awareness, trajectory prediction needs to integrate data from multiple radars and sensors to reconstruct the motion relationships and operational logic of multiple targets, supporting relevant scheduling decisions.

[0003] However, in practical applications, targets exhibit maneuverability and complex non-stationary characteristics, which limits the accuracy and consistency of traditional trajectory distribution predictions. Meanwhile, traditional radar trajectory prediction methods often focus on point estimates of the target's future state. Such single predictions cannot fully reflect potential risks and may lead to decision-making biases (such as misjudgments of traffic scheduling timing and conflicts with civil aviation routes). Furthermore, these methods lack precise quantification of the uncertainty of prediction results, making it difficult to provide reliable references for subsequent radar-related decisions and controls. Summary of the Invention

[0004] This invention provides a probabilistic prediction method for radar track information based on a conditional prior model. This method can solve the problem that traditional schemes focus on point estimation and do not quantify the uncertainty of prediction results, resulting in limited prediction information that affects decision-making effectiveness.

[0005] In a first aspect, embodiments of the present invention provide a probability prediction method for radar track information based on a conditional prior diffusion model, the method comprising: Obtain the target's historical flight path information; The historical trajectory information is input into the trained conditional prediction network to obtain the target's predicted trajectory information; The conditional prediction network is used to extract non-stationary factors and latent variables during the target's motion, obtain non-stationary features and latent variable information, and predict the target's future trajectory based on the non-stationary features and latent variable information. The predicted trajectory information is input into a trained diffusion model to generate multiple possible predicted trajectory sequences through random sampling and stepwise denoising processes, thereby obtaining the predicted trajectory distribution of the target. The predicted trajectory distribution also includes uncertainty information for each predicted trajectory sequence.

[0006] Secondly, embodiments of the present invention provide a radar trajectory information probability prediction device based on a conditional prior diffusion model, including an acquisition unit and a processing unit; The acquisition unit is used to acquire the target's historical flight track information; The processing unit is used for: The historical trajectory information is input into the trained conditional prediction network to obtain the target's predicted trajectory information; The conditional prediction network is used to extract non-stationary factors and latent variables during the target's motion, obtain non-stationary features and latent variable information, and predict the target's future trajectory based on the non-stationary features and latent variable information. The predicted trajectory information is input into a trained diffusion model to generate multiple possible predicted trajectory sequences through random sampling and stepwise denoising processes, thereby obtaining the predicted trajectory distribution of the target. The predicted trajectory distribution also includes uncertainty information for each predicted trajectory sequence.

[0007] Thirdly, embodiments of the present invention provide an electronic device, including a processor and a memory, wherein the memory is used to store a computer program; the processor can be used to execute a calculator program (instructions) stored in the memory to implement the method of the first aspect described above.

[0008] The beneficial effects of this invention compared to existing technologies are as follows: This invention, through a multiple sampling mechanism of a diffusion model, outputs a complete distribution of future trajectories including multiple predicted trajectory sequences and their uncertainty information. This quantifies the uncertainty of each predicted trajectory sequence, providing reliable probabilistic references and solid decision support for downstream tasks such as determining the timing of UAV interception and avoiding aircraft flight path conflicts. Furthermore, by extracting non-stationary factors and latent variable information of target motion through a conditional prediction network for prediction, the accuracy and adaptability of predictions can be improved. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the structure of a conditional prediction network provided in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating a scenario where a conditional prediction network extracts predicted trajectory information, as provided in an embodiment of the present invention. Figure 3 A schematic diagram illustrating the operating principle of a diffusion model provided in an embodiment of the present invention; Figure 4This is a schematic diagram illustrating the training process of a conditional prediction network and a diffusion model provided in an embodiment of the present invention; Figure 5 A flowchart illustrating the implementation of a probability prediction method for radar track information based on a conditional prior diffusion model, provided in an embodiment of the present invention. Figure 6 A schematic diagram of the structure of a radar track information probability prediction device based on a conditional prior diffusion model provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0010] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0011] It should be understood that, when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0012] It should also be understood that the term “and / or” as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0013] As used in this specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [described condition or event] is detected" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once [described condition or event] is detected," or "in response to detection of [described condition or event]."

[0014] Furthermore, in the description of this invention and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0015] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of the invention include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0016] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0017] Example 1 Figure 1 The diagram shown illustrates the structure of a conditional prediction network according to an embodiment of the present invention. As an example and not a limitation, the conditional prediction network may include a non-stationary factor learning module, an embedding layer, an encoder, a latent variable processing module, and a decoder.

[0018] In one possible implementation, the non-stationary factor learning module can extract local features from historical track information and learn the non-stationary factors therein to obtain non-stationary factor features; the embedding layer can perform numerical, temporal, and positional embedding processing on historical track information to obtain a first embedding vector; the encoder can extract high-dimensional latent information representation of historical track information based on the first embedding vector and non-stationary factor features; then, the latent variable processing module can extract latent variable information based on the high-dimensional latent information representation; the embedding layer performs embedding processing on the track sequence vector to obtain a second embedding vector; finally, the decoder determines the predicted track information of the target based on the latent variable information, non-stationary factor features, and the second embedding vector.

[0019] For example, non-stationary factors refer to the phenomenon that the statistical characteristics of a track or radar signal change over time due to changes in the target's motion state or the influence of the external environment. For example, sudden maneuvers of the target can cause changes in the characteristics of the radar signal over time.

[0020] For example, latent variables are implicit space variables used to characterize the uncertainty of future radar tracks, which can implicitly characterize the uncertainties that may exist in the evolution of the target's future track, such as maneuvering changes, random disturbances and measurement noise.

[0021] For example, a track sequence vector is composed of placeholders for historical track information and predicted track information.

[0022] For example, a conditional prediction network can be based on the target's historical trajectory over the previous N+1 time steps: Predict the trajectory at time M+1 after the target is reached. Then the track sequence vector can be , where 0 is used as a placeholder for the predicted trajectory information.

[0023] In one example, see Figure 2 The system can normalize historical trajectory information to obtain the mean, variance, and corresponding normalized historical trajectory information. Then, the mean and variance of the historical trajectory information are input into the non-stationary factor learning module, and the normalized historical trajectory information is input into the embedding layer for encoding.

[0024] In one example, the non-stationary factor learning module can be composed of two stacked one-dimensional convolutional layers and a multilayer perceptron.

[0025] In one example, the encoder can consist of several stacked encoder units, with an inner layer containing an attention layer based on non-stationary self-attention, used to extract global feature correlations from historical track information to obtain a high-dimensional latent information representation.

[0026] For example, the encoder and decoder can be a Transformer encoder and a Transformer decoder, respectively.

[0027] In one example, see Figure 2 The latent variable processing module can first generate the mean and log-variance of the posterior distribution of the latent variables, and then reparameterize them to obtain the latent variable information.

[0028] In one example, the decoder can perform cross-attention and masked self-attention operations on latent variable information, non-stationary factor features, and a second embedding vector to obtain predicted track information.

[0029] For example, the decoder can perform masked self-attention operation on the second embedding vector, and under the modulation of non-stationary factor features, perform cross-attention operation on the second embedding vector as a query vector, latent variable information as a key vector and value vector to generate predicted trajectory information.

[0030] Optionally, the encoder can have 2 layers, and the decoder can have 1 layer. Each layer of the encoder and decoder can employ an 8-head attention mechanism, with a latent space dimension of 512.

[0031] With this structural setup, the encoder and decoder can effectively process non-stationary sequence data, improving the accuracy and adaptability of predictions.

[0032] Figure 3The diagram shown illustrates the operating principle of a diffusion model provided in an embodiment of the present invention.

[0033] In some embodiments, see Figure 3 The diffusion model can represent the diffusion time step using linear embedding to obtain a time step embedding vector. Then, the noisy historical track information and prior prediction results are concatenated to form an input vector, which is then passed to three fully connected layers. Hadamard multiplication is performed between the output of each layer and the time step embedding vector, and a Softplus nonlinear activation function is applied before it is input to the next layer. Finally, the result is mapped to a one-dimensional output (the noise estimation result with the same dimension as the predicted track information) through a fourth fully connected layer. This allows the prediction of the noise component in the forward diffusion process, achieving effective noise removal.

[0034] For example, the predicted trajectory distribution may include multiple possible predicted trajectory sequences for the target. The predicted trajectory sequences collectively characterize the probability distribution of the target's future trajectory, including information such as the mean and confidence interval, providing uncertainty information for trajectory prediction in addition to point prediction.

[0035] Traditional prediction models rely on broad attention weights to passively adapt to trajectory changes, lacking a dedicated non-stationary modeling module and lagging in response to sudden target maneuvers. This invention, through a combined architecture of a "non-stationary factor learning module + Transformer self-attention," can actively extract non-stationary features such as speed abrupt changes and trajectory changes from the trajectory. Simultaneously, by injecting conditional priors throughout the entire diffusion model's forward noise addition and reverse denoising process, it continuously maintains consistency between historical information and future predictions, significantly improving prediction stability in non-stationary scenarios.

[0036] Furthermore, traditional prediction models can only output a single predicted trajectory and cannot characterize complex trajectory distributions such as multi-peaked or heavy-tailed paths. The diffusion generation system of this invention can accurately capture such complex distribution characteristics, covering more potential risk scenarios, and is especially suitable for fields with extremely high risk coverage requirements.

[0037] Figure 4 The diagram shown is a schematic representation of the training process of a conditional prediction network and diffusion model provided in an embodiment of the present invention.

[0038] In some embodiments, see Figure 4The process involves inputting historical flight path information as samples into a conditional prediction network to obtain predicted flight path information. Simultaneously, the decoder outputs the KL divergence between the latent variables and the standard normal distribution. Then, the predicted and actual flight path information are input into a diffusion model. The diffusion model uses the predicted flight path information as the conditional mean, transforming the diffusion chain into a conditional diffusion chain, and continuously adding noise to the actual flight path information to construct the forward diffusion process. Simultaneously, the diffusion model also learns a reverse denoising process, given the noisy flight path, predicted flight path information, and the number of diffusion steps, to predict noise and gradually recover the actual flight path distribution. Finally, the loss of the conditional prediction network for this training iteration is determined based on the actual flight path information, predicted flight path information, and the KL divergence of the latent variables. The loss of the diffusion model for this training iteration is determined based on the actual flight path information, predicted flight path information, predicted noise, and actual noise. The total loss is determined based on the loss values ​​of the two networks, and the network parameters are updated under the constraint of the total loss. This process can be repeated until convergence is achieved, resulting in two trained networks.

[0039] For example, the total loss can be a weighted sum of the losses of the conditional prediction network and the diffusion model.

[0040] In one possible implementation, the loss function of the conditional prediction network can be a weighted sum of the expected loss term and the KL divergence loss term. The former can be used to improve the prediction accuracy of the network, while the latter is used to constrain the distribution of latent variables, preventing the latent variables from deviating too much from the standard normal prior, thereby avoiding overfitting and maintaining the continuity of the latent space structure.

[0041] Specifically, when calculating the loss of a conditional network, a weighting coefficient can be assigned to the KL divergence loss term. To balance reconstruction accuracy with the constraints of potential variable distribution.

[0042] In one example, the expected loss term can be approximated using Monte Carlo sampling.

[0043] For example, the expected loss term can satisfy the following formula:

[0044] in, For the expected loss term, Representing latent variables, Indicates historical flight track information. The time step length for each historical track in the historical track information. This represents a mapping function for feature extraction and encoding of historical flight track information. This represents the posterior distribution of latent variables obtained by encoding historical flight track information; This indicates predicted flight path information. The time step length for each predicted track, Indicates that given latent variables Predicting the conditional probability distribution of flight path sequences under given conditions. To determine the number of samples to approximate the desired number, To obtain the distribution of latent variables The first sample obtained from the middle One sample, Indicates that in a given number Sample of latent variables Predicting track sequences under the given conditions The conditional probability distribution.

[0045] In one example, the KL divergence loss term can satisfy the following formula:

[0046] in, For KL divergence loss term, Representing latent variables, Indicates historical flight track information. The time step length for each historical track in the historical track information. This represents a mapping function for feature extraction and encoding of historical flight track information. This represents the posterior distribution of latent variables obtained by encoding historical flight track information. Let be the prior distribution of the latent variables; , Let represent the mean and variance corresponding to the j-th latent variable dimension, respectively.

[0047] In one possible implementation, during the forward noise addition process, the diffusion model can obtain the current noise-added true trajectory by adding Gaussian noise to the linear interpolation of the previous noise-added true trajectory and the conditional mean.

[0048] In one example, the conditional diffusion process performed by the diffusion model during forward diffusion can be represented as:

[0049] in, For the diffusion process in the 1st Noise intensity coefficient at each time step As a unit array, For diffusion model to real trajectory The track information obtained after the t-th noise addition. This indicates predicted flight path information. This represents the track state at the previous diffusion time step. and predicted flight path information Under the conditions, the current diffusion time step flight path The conditional probability distribution.

[0050] Therefore, arbitrary diffusion stages can be directly sampled from real flight paths using the following derived formula. :

[0051] in, .

[0052] In one example, the corresponding reverse denoising process performed by the diffusion model can be represented as:

[0053] in, , , , .

[0054] Compared to traditional diffusion models, the diffusion model provided in this invention introduces... , This ensures that the entire reverse denoising process is guided by conditional information at every step, thereby ensuring that the generated distribution remains consistent with the conditions in terms of both timing and meaning.

[0055] For example, due to the true posterior distribution Since it cannot be calculated directly, we can use variational inference to parameterize the inversely generated distribution as follows: .

[0056] In one example, the inverse denoising process of the diffusion model can be approximated by minimizing the KL divergence (i.e., the loss function of the diffusion model).

[0057] For example, the loss function of the diffusion model can satisfy the following formula:

[0058] in, Let be the loss function of the diffusion model. The actual injected noise, The noise predicted by the diffusion model, For diffusion model to real trajectory The track information obtained after the t-th noise addition. This indicates predicted flight path information. The time step length for each predicted track, The total number of diffusion steps, Represents the L2 norm. This represents the true posterior distribution of the forward diffusion process, i.e., given the true trajectory and the conditionally predicted trajectory, from the ... Step back to the first The true distribution of steps Indicates historical flight track information. The time step length for each historical track in the historical track information. This represents the inverse denoising distribution learned by the diffusion model, given historical track information and conditionally predicted tracks, starting from the... Step to restore to the first step.

[0059] Specifically, the diffusion step number T can be set to 1000, using linear noise scheduling, in , The conditional prediction network and diffusion model were trained using the PyTorch framework and an NVIDIA RTX 3090 24GB GPU, with a latent variable dimension of 512, an Adam optimizer, a learning rate of 0.0001, and a batch size of 32.

[0060] Current mainstream training methods only use the error between predicted and true values ​​as the sole optimization objective, which easily leads to the bias problem of "disconnect between historical features and future predictions". This invention constructs a joint optimization system of "prior prediction loss + diffusion generation loss (noise prediction error)" and reduces model instability and distribution bias through coordinated constraints of weight parameters. It ensures that the feature learning in the training phase is consistent with the distribution generation behavior in the inference phase, and significantly improves the prediction accuracy of long windows.

[0061] Example 2 The probability prediction method for radar track information based on conditional priors provided in this invention can be applied to electronic devices such as mobile terminals, personal laptops, and supercomputers. This invention does not impose any restrictions on the specific type of electronic device.

[0062] Figure 5 The diagram shown illustrates a probabilistic prediction method for radar track information based on a conditional prior diffusion model, provided by an embodiment of the present invention. As an example and not a limitation, the method may include steps S501-S503, which are described below.

[0063] S501, Obtain the target's historical flight path information.

[0064] S502 inputs historical trajectory information into a trained conditional prediction network to obtain the target's predicted trajectory information.

[0065] For example, a conditional prediction network can extract non-stationary factors and latent variables during target motion, obtain information on non-stationary features and latent variables, and predict the target's future trajectory based on the information on non-stationary features and latent variables.

[0066] S503 inputs the predicted trajectory information into the diffusion model to generate multiple possible predicted trajectory sequences through random sampling and stepwise denoising processes, thereby obtaining the predicted trajectory distribution of the target.

[0067] For example, the predicted trajectory sequence collectively represents the probability distribution of the target's future trajectory, including information such as the mean and confidence interval, providing uncertainty information for trajectory prediction in addition to point prediction.

[0068] This invention, through the multiple sampling mechanism of the diffusion model, outputs the complete distribution of future trajectories, including multiple predicted trajectory sequences and their uncertainty information. It can quantify the uncertainty of each predicted trajectory sequence, providing reliable probabilistic references and solid decision support for subsequent downstream tasks such as judging the timing of UAV interception and avoiding aircraft flight path conflicts.

[0069] Furthermore, compared to traditional prediction methods that require pre-setting attention breadth parameters for different prediction window scenarios, resulting in limited adaptability and reliance on manual adjustments, the end-to-end architecture of this invention requires no external feature engineering or scenario-based parameter pre-setting. It can directly output future distribution results by inputting historical radar track data, and can seamlessly adapt to multiple scenarios such as drone interception (short window) and air traffic control (medium to long window), significantly reducing the adaptation cost of industrial-grade deployment.

[0070] Example 3 Figure 6 The diagram shown illustrates the structure of a probabilistic prediction device for radar track information based on a conditional prior diffusion model, provided in an embodiment of the present invention. As an example and not a limitation, the device may include an acquisition unit and a processing unit.

[0071] For example, the acquisition unit can acquire the target's historical trajectory information; the processing unit can input the historical trajectory information into a trained conditional prediction network to obtain the target's predicted trajectory information; wherein, the conditional prediction network is used to extract non-stationary factors and latent variables during the target's motion, obtain non-stationary features and latent variable information, and predict the target's future trajectory based on the non-stationary features and latent variable information; the predicted trajectory information is input into a trained diffusion model to generate multiple possible predicted trajectory sequences through random sampling and stepwise denoising processes to obtain the target's predicted trajectory distribution, and the predicted trajectory distribution also includes the uncertainty information of each predicted trajectory sequence.

[0072] This invention, through the multiple sampling mechanism of the diffusion model, outputs the complete distribution of future trajectories, including multiple predicted trajectory sequences and their distribution information. It can quantify the uncertainty of each predicted trajectory sequence, providing a reliable probabilistic reference and solid decision support for subsequent downstream tasks such as judging the timing of UAV interception and avoiding aircraft flight path conflicts.

[0073] Example 3 Figure 7 The diagram shown is a structural schematic of an electronic device provided in an embodiment of the present invention. Figure 7 The illustrated electronic device 700 may include: at least one processor 710 ( Figure 7 The diagram shows only one processor, a memory 720, and a computer program 730 stored in the memory 720 and executable on the at least one processor 710, which, when executing the computer program 730, implements the steps in any of the above method embodiments.

[0074] The electronic device 700 can be a robot or other processing device capable of implementing the above methods. This embodiment of the invention does not impose any restrictions on the specific type of electronic device.

[0075] Those skilled in the art will understand that Figure 7 This is merely an example of electronic device 700 and does not constitute a limitation on the electronic device. It may include more or fewer components than shown, or combine certain components, or use different components. For example, the electronic device 700 may also include input / output interfaces.

[0076] The processor 710 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASTCs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0077] In some embodiments, the memory 720 may be an internal storage unit, such as a hard disk or RAM. In other embodiments, the memory 720 may be an external storage device, such as a plug-in hard disk, a smart memory card (SMC), a secure digital card (SD), or a flash card. Furthermore, the memory 720 may include both internal and external storage units. The memory 720 is used to store the operating system, applications, a boot loader, data, and other programs, such as the program code of the computer program. The memory 720 can also be used to temporarily store data that has been output or will be output.

[0078] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0079] Those skilled in the art will understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the functions described above can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this invention. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0080] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0081] This invention provides a computer program product that, when run on an electronic device, enables the electronic device to perform the steps described in the various method embodiments above.

[0082] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0083] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0084] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

Claims

1. A probabilistic prediction method for radar track information based on a conditional prior diffusion model, characterized in that, include: Obtain the target's historical flight path information; The historical trajectory information is input into the trained conditional prediction network to obtain the target's predicted trajectory information; The conditional prediction network is used to extract non-stationary factors and latent variables during the target's motion, obtain non-stationary features and latent variable information, and predict the target's future trajectory based on the non-stationary features and latent variable information. The predicted trajectory information is input into a trained diffusion model to generate multiple possible predicted trajectory sequences through random sampling and stepwise denoising processes, thereby obtaining the predicted trajectory distribution of the target. The predicted trajectory distribution also includes uncertainty information for each predicted trajectory sequence.

2. The method according to claim 1, characterized in that, The conditional prediction network includes a non-stationary factor learning module, an embedding layer, an encoder, a latent variable processing module, and a decoder. The non-stationary factor learning module is used to extract local features of historical flight track information and learn non-stationary factors to obtain the non-stationary factor features. The embedding layer is used to embed the historical track information to obtain a first embedding vector; The encoder is used to extract high-dimensional latent layer information representation of the historical trajectory information based on the first embedding vector and the non-stationary factor features. The latent variable processing module is used to extract the latent variable information based on the high-dimensional latent layer information representation. The embedding layer is also used to embed the track sequence vector to obtain a second embedding vector, wherein the track sequence vector is composed of placeholders for historical track information and predicted track information; The decoder is used to determine the predicted trajectory information based on the latent variable information, the non-stationary factor features, and the second embedding vector.

3. The method according to claim 2, characterized in that, The decoder is specifically used to perform masked self-attention operation on the second embedding vector, and under the modulation of the non-stationary factor features, to perform cross-attention operation on the second embedding vector as a query vector, the latent variable information as a key vector and a value vector, so as to generate the predicted trajectory information.

4. The method according to claim 1, characterized in that, The loss function used during training of the conditional prediction network includes an expectation loss term and a KL divergence loss term. The expectation loss term is used to improve the prediction accuracy of the conditional prediction network, and the KL divergence loss term is used to constrain the distribution of latent variables.

5. The method according to claim 4, characterized in that, The expected loss term satisfies the following formula: in, For the expected loss term, Indicates the latent variable, Indicates historical flight track information. The time step length of each historical track in the historical track information. This represents a mapping function for feature extraction and encoding of historical flight track information. This represents the posterior distribution of latent variables obtained by encoding historical flight track information; This refers to the predicted flight path information. The time step length for each predicted track, Indicates that given latent variables Conditional probability distribution for predicting flight path sequences under given conditions. To approximate the desired number of samples, To the distribution of the latent variables The first sample obtained from the middle One sample, Indicates that in a given number Sample of latent variables Predicting track sequences under the given conditions The conditional probability distribution.

6. The method according to claim 4, characterized in that, The KL divergence loss term satisfies the following formula: in, For the KL divergence loss term, Indicates the latent variable, Indicates historical flight track information. The time step length of each historical track in the historical track information. This represents a mapping function for feature extraction and encoding of historical flight track information. This represents the posterior distribution of latent variables obtained by encoding historical flight track information. Let be the prior distribution of the latent variables; , Let represent the mean and variance corresponding to the j-th latent variable dimension, respectively.

7. The method according to claim 1, characterized in that, During the training process, the diffusion model uses real track sequences as target samples and gradually injects random noise into the real track sequences under the conditional constraints of predicted track information to construct a positive diffusion process. It also learns the inverse denoising process that predicts noise and gradually restores the true track distribution given a noisy track, predicted track information, and number of diffusion steps.

8. The method according to claim 7, characterized in that, The loss function used during training of the diffusion model satisfies the following formula: in, Let be the loss function of the diffusion model. The actual injected noise, The noise predicted by the diffusion model, For the diffusion model on the real trajectory The track information obtained after the t-th noise addition. This refers to the predicted flight path information. The time step length for each predicted track, The total number of diffusion steps, Describing the L2 norm, This represents the true posterior distribution of the forward diffusion process, i.e., given the true trajectory and the conditionally predicted trajectory, from the ... Step back to the first The true distribution of steps Indicates historical flight track information. The time step length of each historical track in the historical track information. This represents the inverse denoising distribution learned by the diffusion model, given historical track information and conditionally predicted tracks, starting from the... Step to restore to the first step.

9. A device for probabilistic prediction of radar track information based on a conditional prior model, characterized in that, Includes an acquisition unit and a processing unit; The acquisition unit is used to acquire the target's historical flight track information; The processing unit is used for: The historical trajectory information is input into the trained conditional prediction network to obtain the target's predicted trajectory information; The conditional prediction network is used to extract non-stationary factors and latent variables during the target's motion, obtain non-stationary features and latent variable information, and predict the target's future trajectory based on the non-stationary features and latent variable information. The predicted trajectory information is input into a trained diffusion model to generate multiple possible predicted trajectory sequences through random sampling and stepwise denoising processes, thereby obtaining the predicted trajectory distribution of the target. The predicted trajectory distribution also includes uncertainty information for each predicted trajectory sequence.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1-8.

Images