Methods, devices, and electronic equipment for correlating interrupted tracks by comparing track motion characteristics

Abstract

This disclosure provides embodiments of a method, apparatus, and electronic device for linking interrupted tracks by comparing track motion characteristics. One specific implementation of the method includes: acquiring a new track group; selecting new tracks from the new track group that meet preset screening conditions with historical tracks to obtain a target track group; inputting the target track group and historical tracks into a preset track association neural network to generate a track association probability value sequence; in response to determining that the largest track association probability value in the sequence is less than a preset association probability, recording the cumulative interruption duration of the historical tracks; in response to determining that the cumulative interruption duration is less than a preset interruption duration threshold, updating the new track group and re-executing the track association step. This implementation can improve the accuracy of interrupted track association.

Description

Technical Field

[0001] Embodiments of this disclosure relate to the fields of computer technology and track association, specifically to interrupted track association methods, apparatus, and electronic devices for comparing track motion characteristics. Background Technology

[0002] As a primary situational awareness device, radar's most fundamental task is to stably and continuously track airborne targets. However, due to fluctuations in the radar cross-section of targets, large-angle maneuvers, and Doppler blind spots, radar often experiences tracking interruptions when detecting targets, leading to track gaps and severely impacting the decision-making efficiency of air traffic controllers.

[0003] Current techniques for linking interrupted tracks can be broadly categorized into two types: traditional methods and deep learning methods. Traditional methods include Kalman filter-based linking and interactive multi-model linking, as well as fuzzy track linking based on similarity metrics. These methods require prior knowledge and assumptions about the target motion model, consuming significant resources in parameter tuning and model selection, resulting in low practical value. Deep learning methods, on the other hand, avoid the complex process of model selection and parameter tuning. Existing techniques include Twin Contrast Neural Networks (TSADCNN), deep temporal contrastive methods, generative linking methods, and graph representation linking methods. While these methods have made some progress, they still have significant shortcomings: TSADCNN and graph representation methods only achieve interaction between feature modules through loss functions; generative methods lack consideration for temporal features; and deep temporal contrastive methods only extract features through an encoder structure and perform simple comparisons, failing to fully utilize the spatiotemporal correlation characteristics between new and historical tracks.

[0004] Traditional methods heavily rely on prior assumptions and manual parameter tuning, making them ill-suited for complex and ever-changing real-world environments. Existing deep learning methods fail to adequately consider the interaction between new and historical track features, either employing overly simplistic interaction methods (using only a loss function) or completely ignoring temporal features. Most methods use symmetric network structures to process sequences of new and historical tracks of unequal length, resulting in low feature extraction efficiency, and significant room for improvement in correlation accuracy under long interruption times and multi-target scenarios. These shortcomings directly impact the accuracy and real-time performance of interrupted track correlation, affecting air traffic controllers' understanding and decision-making regarding the situation of aerial targets.

[0005] The information disclosed in this background section is only intended to enhance the understanding of the background of the inventive concept, and therefore may contain information that does not form prior art known to those skilled in the art. Summary of the Invention

[0006] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0007] Some embodiments of this disclosure propose methods, apparatus, and electronic devices for correlating interrupted tracks by comparing track motion characteristics, in order to solve one or more of the technical problems mentioned in the background section above.

[0008] In a first aspect, some embodiments of this disclosure provide a method for linking interrupted tracks by comparing track motion features. The method includes: in response to detecting a historical track interruption, obtaining a new track group; for the historical track and the new track group, performing the following track linking steps: selecting new tracks from the new track group that satisfy preset screening conditions with the historical track to obtain a target track group; inputting the target track group and the historical track into a preset track linking neural network to generate a track linking probability value sequence, wherein the track linking neural network is used to extract motion features of the new track and the historical track for track linking, each track linking probability value corresponds to a target track, and the track linking neural network includes: a preprocessing module, a track feature extraction module, an assisted attention module, and a classifier module; in response to determining that the largest track linking probability value in the track linking probability value sequence is less than a preset linking probability, recording the cumulative interruption duration of the historical track; in response to determining that the cumulative interruption duration is less than a preset interruption duration threshold, updating the new track group, and performing the track linking steps again.

[0009] Secondly, some embodiments of this disclosure provide an interrupted track association device for comparing track motion characteristics. The device includes: an acquisition unit configured to acquire a new track group in response to detecting a historical track interruption; and a track association unit configured to perform the following track association steps for the historical track and the new track group: selecting new tracks from the new track group that satisfy preset screening conditions with the historical track to obtain a target track group; and inputting the target track group and the historical track into a preset track association neural network to generate a track association probability value sequence, wherein... The aforementioned track association neural network is used to extract motion features of new tracks and historical tracks for track association. Each track association probability value corresponds to a target track. The aforementioned track association neural network includes: a preprocessing module, a track feature extraction module, an assisted attention module, and a classifier module. In response to determining that the largest track association probability value in the aforementioned track association probability value sequence is less than a preset association probability, the cumulative interruption duration of the aforementioned historical track is recorded. In response to determining that the aforementioned cumulative interruption duration is less than a preset interruption duration threshold, the aforementioned new track group is updated, and the aforementioned track association step is executed again.

[0010] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first aspect above.

[0011] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0012] This invention proposes a method for linking interrupted tracks based on motion feature comparison. Through an innovative network architecture design (preprocessing module, track feature extraction module, assisted attention module, and classifier module), it solves key problems in existing technologies. The core of this method lies in constructing an asymmetric deep neural network. Through multi-module collaboration, it achieves deep extraction and effective interaction of spatiotemporal features between new and historical tracks, significantly improving the accuracy of linking interrupted tracks. Attached Figure Description

[0013] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0014] Figure 1This is a flowchart of some embodiments of the obstacle information generation method according to the present disclosure;

[0015] Figure 2 This is a schematic diagram of the track preprocessing module;

[0016] Figure 3 This is a schematic diagram of the track feature extraction module;

[0017] Figure 4 This is a schematic diagram of the attention-assisting module;

[0018] Figure 5 This is a schematic diagram of the overall structure of the trajectory association neural network;

[0019] Figure 6 These are schematic diagrams illustrating the structure of some embodiments of the obstacle information generation apparatus according to this disclosure;

[0020] Figure 7 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0021] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0022] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0023] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0024] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0025] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0026] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0027] Figure 1 A flow 100 of some embodiments of the interrupted track association method according to the present disclosure is shown. The interrupted track association method based on comparing track motion features includes the following steps:

[0028] Step 101: In response to the detection of a historical track interruption, acquire a new track group.

[0029] In some embodiments, the entity executing the interrupted track association method by comparing track motion characteristics (e.g., a radar device deployed with a track association neural network) acquires a new track group via the radar device in response to detecting a historical track interruption.

[0030] Specifically, during radar detection, a target (such as an aircraft) is continuously tracked. Due to factors such as Doppler blind spots, the target's path breaks down (i.e., the track is interrupted). After a period of time, multiple new tracks appear. From this, a new track group can be obtained.

[0031] In practice, track interruption correlation is a radar target tracking technique used to address the problem of target track interruptions caused by target maneuvering, obstruction, or other interference. During radar detection, target tracks may be interrupted due to sudden target changes, obstruction by obstacles, or signal interference. After the interruption, the radar system may detect new tracks (i.e., newly appearing target points). The task of track interruption correlation is to match these newly appearing tracks with previously interrupted historical tracks to determine whether they belong to the same target.

[0032] Step 102: For historical tracks and new track groups, perform the following track association steps:

[0033] Step 1021: Select new tracks from the new track group that meet the preset screening conditions with respect to historical tracks to obtain the target track group.

[0034] In some embodiments, the execution entity may select new tracks from the new track group that meet preset screening conditions with the historical tracks to obtain a target track group.

[0035] In some optional implementations of certain embodiments, the execution entity selects new tracks from the new track group that meet preset screening conditions with the historical tracks to obtain a target track group, including:

[0036] The first step is to determine the interruption time point, the coordinates of the interruption end, and the corresponding target interruption velocity value of the aforementioned historical flight paths, as well as the starting point coordinates of each new flight path in the aforementioned new flight path group. The interruption time point can be the moment the historical flight path was broken. The coordinates of the interruption end can be the position of the historical flight path at the time of interruption. The target interruption velocity value can be the velocity value of the aircraft corresponding to the historical flight path at the interruption time point.

[0037] The second step involves determining whether each new track in the new track group meets preset screening criteria based on the aforementioned interruption time point, the interruption end coordinates, and the target interruption speed value. The screening criterion is that the distance between the starting coordinates of the new track and the interruption end coordinates is less than or equal to the track distance. The track distance is generated by the aforementioned interruption time point, the target interruption speed value, and a preset speed difference. The following formula can be used to determine whether each new track in the new track group meets the preset screening criteria:

[0038] d es ≤t d ·(v e +Δv)

[0039] Where, d es t represents the distance between the last point of the historical track (the coordinates of the end of the interruption) and the coordinates of the starting point of the new track. d Indicates the time of track interruption, v e Δv represents the target speed when the historical track was interrupted (target interruption speed value), and Δv represents the permissible speed difference (preset speed difference). Here, t d ·(v e +Δv) is the distance threshold used to filter d es New tracks that are less than the distance threshold.

[0040] The third step is to identify new tracks that meet the preset screening criteria as target tracks, thus obtaining a target track group.

[0041] Step 1022: Input the target track group and historical tracks into a preset track association neural network to generate a sequence of track association probability values.

[0042] In some embodiments, the execution entity may input the target track group and the historical track into a preset track association neural network to generate a sequence of track association probability values. The track association neural network can be used to extract motion features from new and historical tracks for track association, with each track association probability value corresponding to a target track. The track association neural network includes a preprocessing module, a track feature extraction module, an assisted attention module, and a classifier module.

[0043] In some optional implementations of certain embodiments, the execution entity inputs the target track group and the historical track into a preset track association neural network to generate a sequence of track association probability values, including:

[0044] For each target track in the above target track group, perform the following steps:

[0045] The first step is to input the target trajectory and the historical trajectory into the preprocessing module of the trajectory association neural network for encoding, so as to obtain the historical route encoding features and the new route encoding features.

[0046] As an example, see Figure 2 The diagram shows a trajectory preprocessing module.

[0047] Here, the preprocessing module includes embedding 3D track vectors into a high-dimensional space using a multilayer perceptron to capture inter-channel relationships. The 3D track vectors include the x-coordinate, y-coordinate, and relative time point of the track. Sine position encoding is applied to add positional information to the sequence. This yields historical and new route encoding features. This allows any position encoding to be represented as a linear function of previous positions, which is beneficial for capturing long-sequence dependencies. Specifically, embedding the 3D track vectors into the high-dimensional space is achieved using a multilayer perceptron, increasing the dimension to 64. Considering the extensive use of the Encoder structure in the Transformer, and the Transformer model's lack of awareness of the input sequence order (i.e., permutation invariance), position encoding is introduced to explicitly add positional information to each element, enabling the model to distinguish elements at different positions.

[0048] The second step involves extracting features from the historical route coding features and the new route coding features using the aforementioned attention-assisted module and track feature extraction module, thereby obtaining historical route feature vectors and new route feature vectors.

[0049] As an example, see Figure 3 The diagram shows a trajectory feature extraction module.

[0050] Here, the trajectory feature extraction module consists of multiple identical stacked layers, each containing a multi-head self-attention sub-layer and a feedforward neural network sub-layer, with residual connections and normalization layers in each layer. Multi-head self-attention captures sequence relationships from different subspaces through multiple parallel attention heads. The feedforward neural network provides non-linear transformation capabilities. The trajectory feature extraction module also includes four identical encoder layers. Deep feature extraction is performed through these four identical encoder layers. Next, a flattening layer flattens each historical and new trajectory feature vector. Finally, a multilayer perceptron is used for dimensionality reduction to obtain the historical and new trajectory feature vectors. In practice, the training of the learnable network parameters in the trajectory feature extraction module and the preprocessing module is primarily dominated by a contrastive loss function.

[0051] As yet another example, see Figure 4 The diagram shown is of an attention-assisting module.

[0052] Here, the assisted attention module consists of a Bidirectional Long Short-Term Memory (BiLSTM) network, a self-attention layer, and a feedforward layer (feedforward neural network). The preprocessed features of the historical track are first captured by the BiLSTM to capture bidirectional temporal dependencies, then highlighted by the self-attention layer to highlight key time step information, and finally processed by residual connections and the feedforward neural network to obtain the historical track assisted feature vector. This historical track assisted feature vector is used as the key and value vectors input to the attention calculation. The query vector required for this attention calculation comes from the preprocessed feature vector of the new track. The key and value vectors generated from the historical track interact with the query vector of the new track through the attention mechanism. The feature vector of the new track can determine the importance of different positions in its sequence based on the temporal features of the historical track, allowing the track feature extraction to focus more on important parts and ignore unimportant parts.

[0053] The third step involves classifying the historical route feature vectors and new route feature vectors using the aforementioned classifier module to generate track association probability values.

[0054] Specifically, the classifier module consists of a feature concatenation layer, a fully connected layer, and a softmax function. The track vectors (historical route feature vector and new route feature vector) extracted from the new and historical tracks are directly concatenated along the channel dimension by the feature concatenation layer. This is then input into a fully connected layer with an output dimension of 2. The two output values ​​represent the degree to which the new and historical tracks are related (corresponding) and the degree to which they are unrelated (not related). Finally, the softmax function normalizes the output vector into a probability distribution vector, so that the network's final two outputs represent the probability of track association and the probability of track non-association, respectively. The training of the learnable parameters of the classifier module is mainly dominated by cross-entropy loss.

[0055] As an example, see Figure 5 The diagram shows the overall structure of the trajectory association neural network. In the entire network structure, the lower branch (the branch for the new route input) is only used to process the feature vector of the new trajectory. Assisted attention does not fuse the features of the historical trajectory and the new trajectory, but rather uses the features of the historical trajectory to help express the features of the new trajectory.

[0056] In practice, the loss function design in this application combines contrastive loss and cross-entropy loss. Contrastive loss is applied after the new and old track feature extraction module, calculating the feature similarity between new and old tracks (i.e., new tracks and historical tracks). This ensures that the track features of the same target are spatially similar while the features of different targets are far apart. Therefore, the loss function should be designed to reduce the difference between the extracted features of new and old tracks for the same target and increase the difference between the extracted features of new and old tracks for different targets. The contrastive loss is expressed as:

[0057]

[0058] Where D represents the Euclidean distance between the feature vectors output by the old and new track feature extraction modules (historical route feature vector and new route feature vector). Margin > 0 represents the edge distance, which is the shortest distance between two track segments belonging to different targets in high-dimensional space. Y + Indicates the network prediction label. Loss C This indicates the comparison loss value.

[0059] Here, cross-entropy loss is mainly used to guide the training of the classifier module, assisting it in transforming the distance information between the feature vectors of new and old tracks into probability information of whether they are related, thus optimizing the classification decision boundary. Its formula can be:

[0060] Loss BCE =-[Y + log(p(Y+ =1))+(1-Y + )log(1-p(Y + =1))]

[0061] Among them, Loss BCE This represents the cross-entropy loss value. The label value is 1 or 0; a value of 1 indicates a correlation between new and old tracks, while 0 indicates no correlation. p(Y + =1) represents the probability that the network predicts the label to be 1.

[0062] Then, the contrastive loss and cross-entropy loss values ​​mentioned above can be weighted and summed using the following formula:

[0063]

[0064] Loss represents the total loss value.

[0065] In addition, the neural network training in this application uses the following parameter configurations: the multilayer perceptron has an embedding dimension of 64, a hidden layer dimension of 256, and the position encoding dimension is the same as the embedding dimension; the number of self-attention heads is 8, and the hidden layer dimension of the feedforward neural network is 256. The number of hidden units in the BiLSTM is 64. The multilayer perceptron has a dimensionality reduction of 128 for the hidden layer and 32 for the output dimension. The edge distance in the contrastive loss is set to 15.0; training uses the Adam optimizer with an initial learning rate of 0.0001 and a batch size of 128. The training dataset needs to use real detection tracks from the radar to be deployed and simulated tracks from the radar. Positive sample data is obtained by processing tracks at different time intervals and with different new track lengths. Negative sample data is obtained by randomly combining new and old tracks of different targets. The dataset needs to maintain a balance between positive and negative samples.

[0066] In practice, the key innovations of this application include the following aspects: (1) Asymmetric network architecture design, abandoning the traditional symmetric dual network structure, adopting differentiated feature extraction paths according to the characteristics of different lengths of new and old tracks, and adding an assisting attention module to the old track branch, which is more in line with the needs of actual application scenarios. (2) Assisting attention mechanism, after preprocessing the old track vector, extracting the temporal features of the old track through BiLSTM and self-attention, and influencing the temporal expression of the new track through attention calculation in the form of key-value pairs. Since the vectors output by the new and old tracks after passing through the feature extraction module are to be used as a measure of similarity comparison, the interaction between the new and old tracks placed after feature extraction will affect the training of the feature extraction module. Moreover, the old track only needs to affect the temporal features of the new track, and there is no need to over-extract all the features of the old track. Using a single layer of BiLSTM and self-attention is sufficient, avoiding the increase in network complexity caused by stacking the encoder structure multiple times, and realizing the effective interaction of the features of the new and old tracks, solving the problem of insufficient feature interaction in the existing methods. (3) The multi-layer feature processing flow, including the systematic combination of techniques such as embedding dimension expansion, position encoding, multi-head self-attention, residual connection, and layer normalization, realizes the full extraction of the spatiotemporal features of the track. (4) The dual loss function design, which combines contrastive loss and cross-entropy loss, not only ensures the discriminativeness of the feature space, but also optimizes the classification performance.

[0067] Specifically, the technical effects brought about by this application are mainly reflected in the following aspects: (1) It significantly improves the accuracy of interrupted track association, achieving a correct association rate of 96.4%-99.2% under different interruption times (4-20 seconds), and maintaining an accuracy rate of 95.2% in multi-target scenarios (50 targets). (2) It enhances the adaptability to long interruption times and complex multi-target scenarios, effectively utilizing the rich temporal information of old tracks through the assisted attention mechanism. (3) It reduces the algorithm's dependence on manual parameter tuning and empirical knowledge, realizing end-to-end automated track association. (4) It improves the real-time performance of the algorithm, optimizing computational efficiency through asymmetric network design and feature dimensionality reduction strategy. (5) It has good scalability, and the modular network design facilitates adjustment and optimization for different application scenarios.

[0068] In addition, to comprehensively evaluate the technical effectiveness of this invention, a systematic experimental scheme was designed, including dataset construction, comparison algorithm selection, evaluation metric setting, and result analysis. Regarding the dataset, 3000 target tracks with four typical motion modes (uniform speed, constant acceleration, cooperative turning, and accelerated turning) were simulated and generated within a 10km radius radar measurement area. A balanced dataset containing 48,000 samples was constructed using different interruption times and interception strategies to ensure sufficient training and testing. For the comparison algorithm, the currently best-performing TSADCNN method was selected as the main benchmark. An ablation experiment (removing the assisting attention module) was also set to verify the contribution of the assisting attention module. The evaluation metric used was the track association correctness rate, i.e., the proportion of correctly associated targets to the total number of targets, which is the most direct and effective performance measure in the track association problem. Experimental results show that this invention performs excellently in all test scenarios. In the interruption time sensitivity test, 25 target tracks were randomly generated. As the interruption time increased from 4 seconds to 20 seconds, this invention maintained a significant advantage, especially maintaining an accuracy of 96.4% even with a 20-second long interruption, nearly 20 percentage points higher than TSADCNN. This result verifies the effectiveness of the temporal feature extraction and interaction mechanism designed in this invention for long-interruption scenarios. In multi-target scenario tests, as the number of targets increases from 10 to 50, the performance degradation of this invention is minimal, maintaining an accuracy of 95.2% even with 50 targets, while TSADCNN drops to 76.6%. This demonstrates that the feature representation designed in this invention has better discriminative power and can effectively handle track confusion in dense target scenarios. Ablation experiments show that the assisted attention module brings an average performance improvement of 1.42% in different scenarios, with a more significant improvement in the most challenging scenarios (long interruptions, multiple targets), indicating that this module is particularly important for handling complex situations. In new track length variation tests, the performance of this application and the ablation experiments can maintain a high level, while the accuracy of the TSADCNN method is on average 11.6% lower than that of this invention.

[0069] Based on the combined experimental results, this application achieves significant performance improvements in addressing the critical issue of track association during interruptions through its innovative network architecture and feature interaction mechanism, particularly in the most challenging scenarios of long-interruption, multi-target engagement. This enables it to better meet the stable tracking requirements of highly maneuverable targets with large variations in radar cross-section leading to discontinuous tracking, thus possessing significant military application value. Therefore, this invention can be widely applied to various air defense radar systems, and is especially suitable for handling the tracking interruption problem of highly maneuverable, long-duration discontinuous targets. In actual deployment, the trained model can be integrated into the radar signal processing chain to process newly detected track segments in real time and perform association matching with historical track databases. The radar system can adjust the association threshold according to actual needs to balance the ratio of missed associations and false associations.

[0070] Step 1023: In response to the determination that the largest trajectory association probability value in the trajectory association probability value sequence is less than the preset association probability, record the cumulative interruption duration of the historical trajectory.

[0071] In some embodiments, the executing entity may, in response to determining that the largest track association probability value in the track association probability value sequence is less than a preset association probability, record the cumulative interruption duration of the historical track. The cumulative interruption duration is the duration from the interruption time to the current time. The fact that the largest track association probability value in the track association probability value sequence is less than the preset association probability indicates that no new track is associated with the historical track.

[0072] Step 1024: In response to determining that the cumulative interruption duration is less than the preset interruption duration threshold, update the new track group and execute the track association step again.

[0073] In some embodiments, the execution entity may, in response to determining that the cumulative interruption duration is less than a preset interruption duration threshold, update the new track group and re-execute the track association step. The condition that the cumulative interruption duration is less than the preset interruption duration threshold (batch break time threshold) can be interpreted as meaning that within the batch break time threshold, currently acquired new tracks are not associated, thereby allowing new tracks to be acquired and the association step to continue.

[0074] Optionally, in response to determining that the largest track association probability value in the above track association probability value sequence is greater than or equal to the above preset association probability, the track association between the new track corresponding to the largest track association probability value and the above historical track is determined to be complete.

[0075] Optionally, in response to determining that the cumulative interruption duration is greater than or equal to the preset interruption duration threshold, the above-mentioned track association step is terminated, and the track association of the above-mentioned historical track is determined to have failed. Here, a cumulative interruption duration less than the preset interruption duration threshold can indicate that no new track associated with the historical track was found within the batch interruption time threshold.

[0076] The technical advantages of this invention are mainly reflected in the following aspects: First, it achieves a breakthrough in performance indicators, increasing the accuracy of interrupted track association to over 95%, significantly outperforming existing best methods. This advancement has direct significance for improving air traffic control decision-making efficiency. Second, it innovatively proposes a assisted attention mechanism, guiding the feature learning of new tracks through old tracks (i.e., historical tracks), solving the problem of feature interaction in unequal-length sequences and providing a new solution for similar temporal correlation problems. Third, the network architecture design fully considers practical application needs; the asymmetric structure adapts to the differences in length between old and new tracks, and the modular design facilitates adjustment and optimization for different scenarios. Fourth, the algorithm is robust, exhibiting stable performance under varying factors such as interruption time, number of targets, and length of new tracks, making it suitable for complex combat environments. Fifth, it realizes a shift from experience-driven to data-driven approaches, significantly reducing the need for manual parameter tuning and improving the system's automation level and deployment efficiency. Compared with existing technologies, this invention has three significant features: First, the feature interaction is more comprehensive, achieving deep interaction between new and old track features through an assisted attention mechanism, rather than a simple loss function connection; second, the spatiotemporal feature extraction is more comprehensive, taking into account both spatial motion patterns and temporal evolution laws; and third, it has stronger practical applicability, specifically designed for unequal track segments commonly encountered in air defense operations, overcoming the limitations of symmetric networks in this scenario.

[0077] Further reference Figure 6 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of an interrupted track association device for comparing track motion characteristics. These device embodiments are similar to... Figure 1 Corresponding to the method embodiments shown, this interrupted track association device for comparing track motion characteristics can be specifically applied to various electronic devices.

[0078] like Figure 6As shown, the interrupted track association device 600 for comparing track motion characteristics in some embodiments includes: an acquisition unit 601 and a track association unit 602. The acquisition unit 601 is configured to acquire a new track group in response to detecting a historical track interruption. The track association unit 602 is configured to perform the following track association steps for the historical track and the new track group: selecting new tracks from the new track group that meet preset screening conditions with the historical tracks to obtain a target track group; inputting the target track group and the historical tracks into a preset track association neural network to generate a track association probability value sequence, wherein the track association neural network is used to extract motion features of the new tracks and historical tracks for track association, and each track association probability value corresponds to a target track. The track association neural network includes a preprocessing module, a track feature extraction module, an assisted attention module, and a classifier module; in response to determining that the largest track association probability value in the track association probability value sequence is less than a preset association probability, recording the cumulative interruption duration of the historical track; in response to determining that the cumulative interruption duration is less than a preset interruption duration threshold, updating the new track group and performing the track association steps again.

[0079] It is understandable that the units described in the interrupted track correlation device 600 for comparing track motion characteristics are related to the reference. Figure 1 The steps in the described method correspond accordingly. Therefore, the operations, features, and beneficial effects described above for the method also apply to the interrupted track association device 600 and the units contained therein for comparing track motion characteristics, and will not be repeated here.

[0080] The following is for reference. Figure 7 It shows a schematic diagram of the structure of an electronic device (such as a radar device) suitable for implementing some embodiments of the present disclosure. Figure 7 The electronic device shown is merely an example and should not be construed as limiting the functionality or scope of the embodiments of this disclosure. Figure 7 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The memory may include a non-volatile storage medium and internal memory. The non-volatile storage medium may store an operating system and a computer program. The computer program includes program instructions that, when executed, cause the processor to perform any of the methods described above. The processor provides computational and control capabilities to support the operation of the entire computer device. The internal memory provides an environment for the execution of the computer program in the non-volatile storage medium; when executed by the processor, the computer program causes the processor to perform any of the methods described above. The network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 7The structure shown is merely a block diagram of a portion of the structure related to the present disclosure and does not constitute a limitation on the computer device to which the present disclosure is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0081] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0082] In one embodiment, the processor is configured to run a computer program stored in a memory to perform the following steps: in response to detecting a historical track interruption, a new track group is acquired; for the historical track and the new track group, the following track association steps are performed: selecting new tracks from the new track group that meet preset screening conditions with the historical tracks to obtain a target track group; inputting the target track group and the historical tracks into a preset track association neural network to generate a track association probability value sequence, wherein the track association neural network is used to extract motion features of the new tracks and historical tracks for track association, each track association probability value corresponds to a target track, and the track association neural network includes: a preprocessing module, a track feature extraction module, an assistive attention module, and a classifier module; in response to determining that the largest track association probability value in the track association probability value sequence is less than a preset association probability, the cumulative interruption duration of the historical track is recorded; in response to determining that the cumulative interruption duration is less than a preset interruption duration threshold, the new track group is updated, and the track association steps are performed again.

[0083] This disclosure also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, and the method implemented when the program instructions are executed can be referred to the various embodiments of the methods described above.

[0084] The aforementioned computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. Alternatively, the aforementioned computer-readable storage medium may be an external storage device of the computer device, such as a plug-in hard disk, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.

[0085] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0086] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for correlating interrupted tracks by comparing their motion characteristics, characterized in that, include: In response to the detection of a historical track interruption, a new track group is acquired; For the historical track and the new track group, perform the following track association steps: Select new tracks from the new track group that meet preset screening conditions with the historical tracks to obtain the target track group; The target track group and the historical track are input into a preset track association neural network to generate a sequence of track association probability values. The track association neural network is used to extract the motion features of the new track and the historical track for track association. Each track association probability value corresponds to a target track. The track association neural network includes: a preprocessing module, a track feature extraction module, an assisted attention module, and a classifier module. In response to determining that the largest trajectory association probability value in the trajectory association probability value sequence is less than a preset association probability, the cumulative interruption duration of the historical trajectory is recorded; In response to determining that the cumulative interruption duration is less than a preset interruption duration threshold, the new track group is updated, and the track association step is executed again; The step of inputting the target track group and the historical track into a preset track association neural network to generate a sequence of track association probability values ​​includes: For each target track in the target track group, perform the following steps: The target track and the historical track are input into the preprocessing module of the track association neural network for encoding processing to obtain historical route encoding features and new route encoding features. The attention-assisted module and the trajectory feature extraction module are used to extract features from the historical route coding features and the new route coding features to obtain historical route feature vectors and new route feature vectors. The classifier module classifies the historical route feature vector and the new route feature vector to generate a trajectory association probability value.

2. The method according to claim 1, characterized in that, The method further includes: In response to determining that the largest track association probability value in the sequence of track association probability values ​​is greater than or equal to the preset association probability, it is determined that the track association between the new track corresponding to the largest track association probability value and the historical track is completed.

3. The method according to claim 1, characterized in that, The method further includes: In response to determining that the cumulative interruption duration is greater than or equal to the preset interruption duration threshold, the track association step is terminated, and the track association of the historical track is determined to have failed.

4. The method according to claim 1, characterized in that, The step of selecting new tracks from the new track group that meet preset screening conditions with the historical tracks to obtain the target track group includes: Determine the interruption time point, the coordinates of the end of the interruption, and the corresponding target interruption speed value of the historical track; and determine the starting coordinates of each new track in the new track group. Based on the interruption time point, the interruption end coordinates, and the target interruption speed value, determine whether each new track in the new track group meets the preset screening conditions. The screening condition is that the distance between the starting coordinates of the new track and the interruption end coordinates is less than or equal to the track distance. The track distance is generated by the interruption time point, the target interruption speed value, and the preset speed difference. New tracks that meet the preset screening conditions are identified as target tracks, thus obtaining a target track group.

5. The method according to claim 1, characterized in that, The trajectory feature extraction module consists of multiple identical stacked layers. Each layer contains a multi-head self-attention sub-layer and a feedforward neural network sub-layer. Each layer is equipped with residual connections and normalization layers. The multi-head self-attention captures sequence relationships from different subspaces through multiple attention heads in parallel. The feedforward neural network provides nonlinear transformation capabilities. The trajectory feature extraction module flattens each historical route feature vector and the new route feature vector, and performs dimensionality reduction processing through a multilayer perceptron to obtain the historical route feature vector and the new route feature vector.

6. A device for correlating interrupted tracks by comparing track motion characteristics, comprising: The acquisition unit is configured to acquire a new track group in response to the detection of an interruption in the historical track; The track association unit is configured to perform the following track association steps for the historical track and the new track group: Select new tracks from the new track group that meet preset screening conditions with the historical tracks to obtain the target track group; The target track group and the historical track are input into a preset track association neural network to generate a sequence of track association probability values. The track association neural network is used to extract motion features from new and historical tracks for track association. Each track association probability value corresponds to a target track. The track association neural network includes a preprocessing module, a track feature extraction module, an assisted attention module, and a classifier module. The step of inputting the target track group and the historical track into the preset track association neural network to generate the sequence of track association probability values ​​includes: For each target track in the target track group, perform the following steps: The target track and the historical track are input into the preprocessing module of the track association neural network for encoding processing to obtain historical route encoding features and new route encoding features. The attention-assisted module and the trajectory feature extraction module are used to extract features from the historical route coding features and the new route coding features to obtain historical route feature vectors and new route feature vectors. The classifier module classifies the historical route feature vector and the new route feature vector to generate a track association probability value. In response to determining that the largest trajectory association probability value in the trajectory association probability value sequence is less than a preset association probability, the cumulative interruption duration of the historical trajectory is recorded; In response to determining that the cumulative interruption duration is less than a preset interruption duration threshold, the new track group is updated, and the track association step is executed again.

7. An electronic device, comprising: One or more processors; Storage device, on which one or more programs are stored, When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-5.

8. A computer-readable medium having a computer program stored thereon, wherein, When the program is executed by the processor, it implements the method as described in any one of claims 1-5.

Images