False track identification method and system based on spatio-temporal information
By using the SPAT-ConvLSTM model to identify false tracks on radar data in complex environments, the problem of poor false track recognition in the prior art is solved, and higher track accuracy and robustness are achieved.
Patent Information
- Application Number
- CN202510264581.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-07
- Publication Date
- 2025-06-06
AI Technical Summary
The prior art is difficult to effectively identify false tracks in complex environments, especially in large quantities of clutter environments, resulting in the real track being misjudged as false tracks.
Using a false track recognition method based on spatiotemporal information, a neural network model, especially a SPAT-ConvLSTM model, uses a neural network model, and preprocessing, track correlation, sliding window conversion and annotation of historical radar echo data, and generate latitude and longitude images for training, thereby identifying false tracks.
It improves the accuracy and robustness of target track generation, effectively recognizes and suppresses false tracks, and improves the quality of tracks.
Smart Images
Figure CN120103293A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the fields of radar data processing and artificial intelligence technology, and in particular to a false track identification method and system based on spatiotemporal information. Background Art
[0002] In complex environments, affected by electronic interference, clouds, rain, lightning and other factors, radar measurement points often contain various types of clutter. Therefore, the measurement information received by the radar may come from real flying targets, or from erroneous measurements such as sea clutter and ground clutter. Existing track association and initiation methods, such as the nearest neighbor and global nearest neighbor algorithms, are simple to implement in engineering, but are affected by various clutters and will inevitably produce false tracks; algorithms such as joint probability data association and multi-hypothesis tracking can better cope with the situation of strong clutter in the environment, but they also have the disadvantage of large computational complexity, and these algorithms will still produce false tracks in a large number of clutter environments. Therefore, it is very necessary to design relevant algorithms to distinguish tracks in real time and effectively identify false tracks in the track initiation and association stages.
[0003] As the airspace environment becomes increasingly crowded and complex, by studying the target track data in the airspace, we can grasp the changing laws of airspace characteristics and the temporal and spatial distribution, and deeply understand the airspace situation. As the main means of detecting target flight, radar plays an incomparable role in the study of airspace target tracks. Due to the complexity of the detection environment, radar is often subject to various interferences and produces false tracks. Therefore, the true and false judgment of the detected target track has become the main problem of radar data processing algorithms.
[0004] At present, most of the methods for distinguishing false tracks are based on threshold rules. They set the target's speed, acceleration, direction change range and other parameters through experience, and identify the tracks that do not meet the range as false tracks. This method is difficult to take into account all the rules and restrictions of false tracks and is difficult to adapt to complex and changeable actual scenarios. In addition, some false track identification methods based on adaptive boosting (Adaptive Boosting, Adaboost) and support vector machine (SVM) have been proposed, but the effectiveness of such methods in practical applications still needs to be improved, and there are cases where false tracks are not effectively identified. Summary of the invention
[0005] The embodiment of the present application provides a false track identification method and system based on spatiotemporal information, which is used to use a neural network model to distinguish false tracks, use effective spatiotemporal information in the target track to identify false tracks, and improve the accuracy and robustness of target track generation.
[0006] The present application embodiment provides a false track identification method based on spatiotemporal information, comprising:
[0007] Acquire historical radar echo data and pre-process the acquired echo data;
[0008] Perform track association on the echo data after preprocessing to form the target's track information;
[0009] Select a track segment containing a specified number of consecutive track points from the target's track information and mark it;
[0010] For the annotated track segments, the continuous track points are converted into longitude and latitude images based on sliding window conversion;
[0011] The SPAT-ConvLSTM model is trained using the converted latitude and longitude images;
[0012] The trained SPAT-ConvLSTM model is used to identify false tracks of newly acquired radar target track data.
[0013] An embodiment of the present application also provides a false track identification system based on spatiotemporal information, including a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the false track identification method based on spatiotemporal information as described above are implemented.
[0014] The embodiment of the present application utilizes a neural network model to distinguish false tracks, utilizes effective spatiotemporal information in the target track to identify false tracks, and improves the accuracy and robustness of target track generation.
[0015] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:
[0017] Figure 1 The basic process diagram of the false track identification method based on spatiotemporal information in an embodiment of the present application is shown;
[0018] Figure 2This is a schematic diagram of a target track image of a false track identification method based on spatiotemporal information according to an embodiment of the present application;
[0019] Figure 3 This is a schematic diagram of the SPAT-ConvLSTM model architecture of an embodiment of the present application;
[0020] Figure 4 This is a schematic diagram of the real-time determination process of the method of the embodiment of the present application at the initial stage of the track;
[0021] Figure 5 The present invention is a method for implementing an embodiment of the present invention and illustrating the real-time determination process in the track association stage. DETAILED DESCRIPTION
[0022] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.
[0023] There are many shortcomings in the existing false track identification methods: (1) The traditional false track identification method based on threshold rules is difficult to adapt to complex and changeable actual scenarios; (2) The false track methods based on methods such as Adaboost and SVM fail to effectively utilize the spatiotemporal information in the target track, and the effect in actual use scenarios still needs to be improved. The embodiment of the present application provides a false track identification method based on spatiotemporal information, such as Figure 1 Shown include:
[0024] In step S101, historical radar echo data is acquired, and the acquired echo data is preprocessed. In some embodiments, preprocessing the acquired echo data includes:
[0025] The acquired echo data is subjected to point condensation processing, multiple original points belonging to the same target are merged, and radar measurement points whose positions do not conform to the set range are deleted to form radar measurements.
[0026] In step S102, the echo data after preprocessing is track-associated to form the target track information. In a specific example, based on the radar measurement formed, a new track is generated based on the track start, and the measurement point track and the existing track are associated to obtain the target track.
[0027] In step S103, a track segment containing a specified number of consecutive track points is selected from the target's track information and annotated. For example, nine consecutive points in the target's track are saved as different track segments, and then each track segment is annotated using the target's own GPS data, civil aviation ADS-B data and other true values.
[0028] In step S104, for the annotated track segment, the continuous track points are converted into a latitude and longitude image based on a sliding window conversion. In some embodiments, for the annotated track segment, the continuous track points are converted into a latitude and longitude image based on a sliding window conversion, including:
[0029] The marked track segment containing the first number of points is converted into an image in a sliding window manner according to the latitude and longitude information of the points in the sliding window, and the points are connected in the image, and marks are added to the points and the parts through which the connecting lines pass.
[0030] Specifically, each track segment obtained above can be converted into a longitude and latitude image by sliding window method for every 5 consecutive points, and a total of five longitude and latitude image data can be obtained. The area where there is a track in the image is 1, and the area without a track is 0; then the obtained longitude and latitude image data is sent to the SPAT-ConvLSTM model for training, and the weight parameters after model training are saved.
[0031] In step S105, the SPAT-ConvLSTM model is trained using the converted latitude and longitude images.
[0032] In step S106, the trained SPAT-ConvLSTM model is used to perform false track identification on the newly acquired radar target track data.
[0033] In actual situations, the changes in position, speed, angle, etc. between consecutive track points can reflect the difference between real and false tracks, because the changes of real targets between several consecutive track points in time are relatively stable, while false tracks may have large changes. The method of this application trains the SPAT-ConvLSTM model to allow the model to learn the temporal and spatial change laws between track points, identify and process false tracks, and effectively suppress false tracks, thereby improving the quality of tracks.
[0034] In some embodiments, preprocessing the acquired echo data also includes: in the process of generating the target track, the obtained radar echo is first subjected to point track aggregation processing, and multiple original point tracks belonging to the same target are merged so that each target obtains point track data that uniquely represents the target's physical position.
[0035] After the echo data is condensed to obtain radar measurement, the track is started to generate a new track, and the measurement point track is associated with the existing track to obtain the target track. t The location information is (X t ,Y t ), where X t and Y t They are the longitude and latitude information of the track at time t respectively.
[0036] In some embodiments, for the annotated track segment, converting continuous track points into latitude and longitude images based on sliding window conversion specifically includes:
[0037] After the data is annotated, the track segments of the above-annotated 9 consecutive points are stored every 5 points in a sliding window manner. According to the longitude and latitude information of each point, it is converted into a picture, where the horizontal axis is the longitude and the vertical axis is the latitude, and 5 pictures can be obtained. Each picture is as follows Figure 2 As shown, Figure 2 Points are points in the target track, and the points are connected by straight lines. Figure 2 The part through which the connecting line passes is set to 1, and the rest is set to 0, and the target track image is obtained.
[0038] like Figure 3 As shown, in some embodiments of the SPAT-ConvLSTM model of the present application, training the converted latitude and longitude images in the SPAT-ConvLSTM model includes:
[0039] In the SPAT-ConvLSTM model, the continuous images obtained based on the sliding window are convolved separately;
[0040] For the image after the convolution operation, the spatial attention module of the SPAT-ConvLSTM model is input to obtain the feature information of the spatial position in the image;
[0041] The ConvLSTM model is input to extract the spatiotemporal feature information between the traces.
[0042] like Figure 3 In the figure, the left side shows the data of 5 consecutive pictures obtained above. After the pictures are convolved respectively, they are sent to the spatial attention module to obtain the important feature information of the spatial position in the picture, and then sent to the ConvLSTM model for processing to extract the spatiotemporal feature information between the traces. Finally, the ReLU activation function is used in the output stage to obtain the prediction result of the model.
[0043] In computer vision, the attention module plays a very important role in improving the feature extraction ability of the model. Compared with processing an entire image, the attention mechanism can extract the feature information that needs to be paid attention to from the entire image features, thereby improving the model's discrimination ability. In some embodiments, the spatial attention module of the SPAT-ConvLSTM model adopts the following process:
[0044] f=cat(maxpool(I),avgpool(I))
[0045] SA=σ(conv(f))
[0046] Among them, σ is the sigmoid activation function, maxpool(I) means the global maximum pooling of the input feature map I in the channel dimension, avgpool(I) means the global average pooling of the input feature map I in the channel dimension, cat means the concatenation operation along the channel dimension, conv(f) means the convolution operation on the feature map f, and SA means the spatial attention weight matrix.
[0047] In this application, the feature information I of the input image is sent to the maximum pooling and average pooling along the channel dimension respectively, and then the two results are merged along the channel dimension to obtain the feature f, and finally f is sent to the convolution layer and the sigmoid activation function to obtain the final output SA of the spatial attention module.
[0048] After obtaining the spatial attention weight matrix SA, it is multiplied by the input feature map to obtain the output of the spatial attention module:
[0049]
[0050] In some embodiments, the SPAT-ConvLSTM model is trained using the converted latitude and longitude images, wherein the ConvLSTM processing of the SPAT-ConvLSTM model satisfies:
[0051]
[0052] Where σ represents the sigmoid activation function, Indicates the multiplication of corresponding elements of the matrix, * indicates the convolution operation, W xf ,W hf ,W cf represents the weight of the forget gate, W xi ,W hi ,W ci represents the weight of the input gate, W xc ,W hc represents the unit state weight, W xo ,Who ,W co represents the weight of the output gate, b i ,b f ,b c ,b o Indicates the bias of different gates. In the embodiment of the present application, the advantages of ConvLSTM combined with convolutional neural network spatial feature extraction and LSTM temporal feature extraction are utilized, which can make full use of the spatiotemporal information of the target track and realize accurate identification of false tracks.
[0053] At the end of the model, the ReLU activation function is used to get the final prediction result P re , which outputs a value between 0 and 1, indicating the probability that the track is the true track, P re The larger the value, the closer it is to the real track. In some embodiments, training the SPAT-ConvLSTM model using the converted latitude and longitude images further includes:
[0054] During the training process, the L1 loss function is used to satisfy:
[0055]
[0056] Among them, B represents the number of samples in each batch of training, P re Represents the prediction result, L t Indicates the label value of this track. Through this loss function, the model parameters are continuously updated iteratively during the training process, so that the parameters of the real track are constantly close to 1 and the parameters of the false track are constantly close to 0, so that the model can successfully identify the false track.
[0057] In some embodiments, using the trained SPAT-ConvLSTM model to identify false tracks on newly acquired radar echo data includes:
[0058] In the track start phase, the point track that meets the target motion law is used to start a new track and is recorded in the temporary track chain list;
[0059] In the association phase, the obtained track is associated with the radar measurement data and recorded in the confirmed track list.
[0060] In some embodiments, using the trained SPAT-ConvLSTM model to perform false track identification on newly acquired radar echo data further includes:
[0061] For the output of the SPAT-ConvLSTM model, the result predicted to be greater than the set threshold is taken as the true track, and the result predicted to be less than or equal to the set threshold is taken as the false track, and the false track is deleted; and,
[0062] If the real track is not successfully associated within the set time threshold, the corresponding track will be deleted.
[0063] like Figure 4 As shown in the figure, at the beginning of the track, the model weight parameters saved after training are loaded, and the target track image data is sent to the SPAT-ConvLSTM model to obtain the final output O.
[0064]
[0065] According to the above formula, a suitable threshold T is set in the final output stage. When the predicted value is greater than T, the final output result is 1, which is a real track. Otherwise, the output is 0, which means that the track is a false track. In practice, T is set to 0.5. If the track is judged to be a false track by the above formula, the batch of tracks will be deleted directly.
[0066] like Figure 5 As shown in the figure, in the track association stage, the target track image data is sent to the SPAT-ConvLSTM model for false track identification. The setting of the threshold T in the identification process is the same as that in the initial stage. After the false track is identified, the latest false track is deleted, and the optimal point-to-point association is re-searched when the confirmed track is updated next time. If the real point track is not successfully associated within the set time threshold, the track information is deleted.
[0067] The method of the embodiment of the present application constructs a SPAT-ConvLSTM model, first using the spatial attention module to obtain the spatial feature information that needs to be paid attention to in the target track image, and then using the advantages of ConvLSTM spatial and temporal feature extraction to obtain effective spatiotemporal information in the target track for false track discrimination. The method of the embodiment of the present application converts the target track into a continuous image, allowing the neural network model to make full use of the spatiotemporal information contained in the target track image data and improve the accuracy of false track identification.
[0068] An embodiment of the present application also proposes a false track identification system based on spatiotemporal information, including a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the false track identification method based on spatiotemporal information as described above are implemented.
[0069] It should be noted that in the various embodiments of the present application, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.
[0070] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.
[0071] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.
[0072] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.
Claims
1. A false track identification method based on spatiotemporal information, characterized in that: include: Acquire historical radar echo data and pre-process the acquired echo data; Perform track association on the echo data after preprocessing to form the target's track information; Select a track segment containing a specified number of consecutive track points from the target's track information and mark it; For the annotated track segments, the continuous track points are converted into longitude and latitude images based on sliding window conversion; The SPAT-ConvLSTM model is trained using the converted latitude and longitude images; The trained SPAT-ConvLSTM model is used to identify false tracks of newly acquired radar target track data.
2. The false track identification method based on spatiotemporal information as claimed in claim 1, characterized in that: The preprocessing of the acquired echo data includes: Perform point condensation processing on the acquired echo data, and merge multiple original points belonging to the same target to form radar measurements; Based on the radar measurement formed, a new track is generated based on the track start, and the measurement point track is associated with the existing track to obtain the target track.
3. The false track identification method based on spatiotemporal information as claimed in claim 1, characterized in that: For the annotated track segments, the continuous track points are converted into latitude and longitude images based on the sliding window conversion, including: The marked track segment containing the first number of points is converted into an image in a sliding window manner according to the latitude and longitude information of the points in the sliding window, and the points are connected in the image, and marks are added to the points and the parts through which the connecting lines pass.
4. The false track identification method based on spatiotemporal information as claimed in claim 1, characterized in that: Training the converted latitude and longitude images in the SPAT-ConvLSTM model includes: In the SPAT-ConvLSTM model, the continuous images obtained based on the sliding window are convolved separately; For the image after the convolution operation, the spatial attention module of the SPAT-ConvLSTM model is input to obtain the feature information of the spatial position in the image; The ConvLSTM model is input to extract the spatiotemporal feature information between the traces.
5. The false track identification method based on spatiotemporal information as claimed in claim 4, characterized in that: The spatial attention module of the SPAT-ConvLSTM model adopts the following process: f=cat(maxpool(I),avgpool(I)) SA=σ(conv(f)) Wherein, σ is the sigmoid activation function, maxpool(I) represents the global maximum pooling of the input feature map I in the channel dimension, avgpool(I) represents the global average pooling of the input feature map I in the channel dimension, cat represents the concatenation operation along the channel dimension, conv(f) represents the convolution operation on the feature map f, and SA represents the spatial attention weight matrix; After obtaining the spatial attention weight matrix SA, it is multiplied by the input feature map to obtain the output of the spatial attention module:
6. The false track identification method based on spatiotemporal information as claimed in claim 5, characterized in that: The SPAT-ConvLSTM model is trained using the converted latitude and longitude images, where the ConvLSTM processing of the SPAT-ConvLSTM model satisfies: In the formula, σ represents the sigmoid activation function, Indicates the multiplication of corresponding elements of the matrix, * indicates the convolution operation, W xf ,W hf ,W cf represents the weight of the forget gate, W xi ,W hi ,W ci represents the weight of the input gate, W xc ,W hc represents the unit state weight, W xo ,W ho ,W co represents the weight of the output gate, b i ,b f ,b c ,b o Indicates the bias of different gates.
7. The false track identification method based on spatiotemporal information as claimed in claim 6, characterized in that: Training the SPAT-ConvLSTM model using the converted latitude and longitude images also includes: During the training process, the L1 loss function is used to satisfy: Among them, B represents the number of samples in each batch of training, P re Represents the prediction result, L t Indicates the label value of this track segment.
8. The false track identification method based on spatiotemporal information as claimed in claim 1, characterized in that: Using the trained SPAT-ConvLSTM model, the false track identification of the newly acquired radar echo data includes: In the track start phase, the point track that meets the target motion law is used to start a new track and is recorded in the temporary track chain list; In the association phase, the obtained track is associated with the radar measurement data and recorded in the confirmed track list.
9. The false track identification method based on spatiotemporal information as claimed in claim 8, characterized in that: Using the trained SPAT-ConvLSTM model, false track identification of newly acquired radar echo data also includes: For the output of the SPAT-ConvLSTM model, the result predicted to be greater than the set threshold is taken as the true track, and the result predicted to be less than or equal to the set threshold is taken as the false track, and the false track is deleted; and, If the real track is not successfully associated within the set time threshold, the corresponding track will be deleted.
10. A false track identification system based on spatiotemporal information, characterized in that: The method comprises a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the false track identification method based on spatiotemporal information as described in any one of claims 1 to 9 are implemented.