Radar target true and false track discrimination method based on confidence classification

Abstract

The invention discloses a method for discriminating true and false tracks of a radar target based on confidence classification, and belongs to the field of radar target recognition. Extracting five-dimensional features of time difference, displacement distance, speed, acceleration and spatial distance in the temporary track set of the given radar target; constructing a deep neural network classifier based on confidence functions to obtain the confidence functions of real, false and uncertain categories to which each temporary track belongs; selecting a plurality of neighbor decision samples according to the feature distance between the temporary tracks, constructing decision evidences based on a corresponding confidence function, and fusing the decision evidences; dividing two subsets according to each temporary track global confidence function obtained by fusion; and outputting a category discrimination result of the temporary track of each radar target. According to the invention, the deep neural network classifier based on a confidence function improves mining learning of an internal relationship between track features and mode categories; the number of decision samples is supplemented through two-step decision, the recognition accuracy of the temporary track of the uncertain category is effectively improved, and rapid track building and stable tracking of a target are achieved.

Description

Technical Field

[0001] This invention relates to radar target recognition technology; specifically, it relates to a method for distinguishing between true and false radar target tracks based on confidence classification. Background Technology

[0002] To address the issue that radar detection information contains a large amount of clutter and uncertainty due to limitations in detection accuracy and complex environmental factors, leading to numerous false tracks during target path establishment and significantly impacting target tracking accuracy, there is an urgent need to research a confidence-based identification framework that can effectively characterize uncertain information. This framework, combined with the strong data feature mining capabilities of deep neural networks, can provide support for improving radar target path differentiation. Traditional target path identification algorithms employ binary classification, lacking the ability to represent and process inaccurate and incomplete data, thus limiting the improvement of identification accuracy. This invention proposes a method for distinguishing real and fake radar target tracks based on confidence classification. First, five-dimensional features—time difference, displacement distance, velocity, acceleration, and spatial distance—are extracted from each temporary track. Second, a deep neural network classifier based on confidence functions is constructed to obtain the confidence functions for each track's classification as real, fake, or uncertain. Then, multiple nearest-neighbor decision samples are selected based on the feature distances between tracks, and decision evidence is constructed and fused. Subsequently, the temporary tracks are divided into two subsets based on the global confidence function for a two-step decision process: temporary tracks with the highest confidence level (real or fake) are directly classified, while those with the highest confidence level (uncertain) are then classified based on the class center distance. Finally, the classification results for each temporary track are output.

[0003] The underlying and relied-upon publicly known technology of this invention is the confidence function theory, also known as evidence theory or DS theory. Confidence function theory has unique advantages in representing and processing imprecise and incomplete information. The basic confidence assignment functions of multiple pieces of evidence can be combined using DS rules to obtain a new global mass function for decision analysis. DS rules are simple and easy to implement; in multi-evidence fusion, the combination order does not need to be considered, allowing for rapid algorithm convergence. However, DS has certain limitations when there is high conflict in the evidence. Therefore, scholars have studied Yager, Smets, Dubois & Prade, PCR5, DSmT, etc. Each combination rule has its own advantages and disadvantages, and should be rationally selected or combined according to the specific application scenario.

[0004] The closest patent to the invention is: A method for identifying false tracks based on deep neural networks. Its specific implementation steps are as follows: (1) The acquired radar measurement data is processed offline, and the target track state is formed through algorithms such as data association and Kalman filtering. (2) The target track state is recorded, the classification label is completed, and training samples are formed. (3) The target track samples are converted into images using a two-dimensional grid model to form a binary image of the target track. The convolutional neural network receives the binary image information and performs convolution and pooling processing to form an image feature input signal, thereby realizing the feature extraction of the target track at the image level. (4) For different state information in the target track samples, second-order fitting processing is carried out to obtain the second-order fitting function of each state. The curve change rate of the second-order fitting function is the data feature input signal, thereby realizing the feature extraction of the target track at the data level. (5) The image and data feature signals are input to the binary classification neural network for training, the deep neural network is constructed, and the offline processing ends. (6) Online processing executes steps (1) to (4) on the target track generated in real time to complete the extraction of image and data features, and inputs the extracted feature signals into the deep neural network to carry out classification calculations to obtain the recognition results, thereby clarifying the true or false attributes of the target track. (7) Based on the recognition results, false tracks are eliminated and true tracks are retained, thereby solving the problem of many false tracks in dense clutter environments.

[0005] The key to rapid radar target establishment and stable tracking lies in quickly and efficiently identifying genuine target tracks from a large number of false tracks. Traditional radar target track identification methods mostly employ rule-based models such as K-nearest neighbors, support vector machines, decision trees, and Bayesian classifiers. These algorithms have limitations such as fixed models and poor generalization. They cannot extract deep features to distinguish between highly similar temporary tracks and are difficult to handle complex scenarios. Artificial intelligence neural network methods can gradually extract the most essential intrinsic correlation features of target data through the complex structure of deep networks, and have been well applied in pattern classification. However, when relying solely on neural networks without expanding the identification framework, all decisions are still based on a probabilistic framework, that is, only binary classification judgments of true or false can be made. However, radar is limited by its own accuracy and is also affected by complex external environments. The detection data inevitably contains a lot of uncertain information. Existing track identification methods, limited by probabilistic frameworks, lack the representation and processing of inaccurate and incomplete information, and can only make hard classifications of true or false, which inevitably leads to misjudgments. In practical applications, sometimes an immediate judgment is not necessary. In such cases, a multi-step decision-making strategy can be adopted. For temporary tracks with clearly defined categories, a judgment result can be given immediately, serving as supplementary evidence to assist in determining the authenticity of remaining tracks. For temporary tracks with ambiguous categories, they can be initially classified into an uncertain category, and then a true / false judgment can be made under the guidance of increasingly rich sample data. Therefore, this invention first overcomes the limitations of traditional rule models based on a probabilistic framework by extending the target track recognition framework through confidence function theory. Secondly, it designs a deep neural network based on confidence functions to improve the mining of the intrinsic relationship between track features and pattern categories. Finally, through a two-step decision-making process, it effectively improves the recognition accuracy of temporary tracks, enabling rapid target navigation and stable tracking. Summary of the Invention

[0006] To address the shortcomings of existing methods for identifying temporary radar target tracks, this paper proposes a confidence-based classification method for distinguishing between true and false radar target tracks, belonging to the field of radar target recognition. The method extracts five-dimensional features—time difference, displacement distance, velocity, acceleration, and spatial distance—from a given set of temporary radar target tracks. A deep neural network classifier based on confidence functions is constructed to obtain the confidence functions for each temporary track's classification as true, false, or uncertain. Based on the feature distances between temporary tracks, multiple nearest-neighbor decision samples are selected and fused after constructing decision evidence using the corresponding confidence functions. The resulting global confidence functions of each fused temporary track are used to divide it into two subsets. The classification results for each temporary radar target track are then output. This invention utilizes a confidence-based deep neural network classifier to enhance the mining and learning of the intrinsic relationship between track features and pattern categories. By supplementing the number of decision samples through a two-step decision process, the method effectively improves the accuracy of identifying uncertain category temporary tracks, enabling rapid target navigation and stable tracking.

[0007] This invention provides a method for distinguishing true and false radar target tracks based on confidence classification; the specific steps are as follows:

[0008] Step 1: Construct a category confidence identification framework for radar target tracks;

[0009] Step 2: Extract radar target track features;

[0010] Step 3: Design a deep neural network classifier based on confidence functions;

[0011] A deep neural network classifier is set up based on radar target trajectory characteristics and category confidence recognition framework;

[0012] Step 4: Construct target trajectory decision evidence using a deep neural network classifier, and fuse the decision evidence to obtain a global confidence function set;

[0013] Step 5: Divide the global confidence function set into two subsets through a two-step classification decision; determine the category of the target track in the two subsets based on the overall confidence distribution to obtain the target track of uncertain category; finally, complete the discrimination between true and false tracks.

[0014] Furthermore, in step 1, the framework for constructing the category confidence identification of radar target tracks is as follows:

[0015] Based on the advantages of confidence function theory in representing and processing uncertain information, a category confidence identification framework for radar target tracks is constructed. ,in, Indicates the actual track category. Indicates the category of false flight path. Indicates an uncertain category;

[0016] If the real target track categories and the false target track categories are mutually exclusive, then the category intersection operation results in an empty set, i.e. ;

[0017] The uncertain category is the union of all independent and mutually exclusive categories. Within the category confidence identification framework, for Therefore, the intersection operation of each independent mutually exclusive category and the uncertain category is itself. , .

[0018] Furthermore, in step 2, the process of extracting radar target track features is as follows:

[0019] The radar target trajectory characteristics include time difference. Displacement distance ,speed acceleration and spatial distance Five-dimensional features;

[0020] The radar target track is composed of M radar points arranged in a time sequence, where the radar point feature P at time l is... l :

[0021] (1);

[0022] in, , , These are the target spatial location information after the measurements have been converted into northeast celestial coordinates; t l For time series labeling;

[0023] A set of M radar points for:

[0024] (2);

[0025] Time difference between two radar points in the (k+1)th and kth sequences for:

[0026] (3);

[0027] Displacement distance between two radar points from time k to k+1 for:

[0028] (4);

[0029] The target velocity in time series k for:

[0030] (5);

[0031] target acceleration at time series k for:

[0032] (6);

[0033] The spatial distance from the radar spot to the origin at time k. for:

[0034] (7);

[0035] The five-dimensional feature vector of the k-th time-series radar point for:

[0036] (8);

[0037] Calculate the mean values ​​of each dimension of the features of all time-series radar points contained in the track:

[0038] (9);

[0039] Finally, the five-dimensional feature vector of the radar target trajectory is obtained. for:

[0040] (10).

[0041] Furthermore, in step 3, the deep neural network classifier includes an input layer, a hidden layer, and an output layer;

[0042] The input layer has 5 nodes, which correspond to time difference, displacement distance, target velocity, target acceleration, and spatial distance, respectively. The input to the deep neural network classifier is the five-dimensional feature vector of the radar target trajectory. ;

[0043] The hidden layer has 5 layers and all nodes in the 5th layer are set to 20.

[0044] The output layer has 3 nodes, each corresponding to a different real track category. False track category and uncertain categories confidence level , and And satisfy This means that the confidence sum of all pattern categories given by the deep neural network classifier is 1.

[0045] Furthermore, in step 4, the process of constructing and fusing the target trajectory decision evidence is as follows:

[0046] Step 4.1, Define To distinguish between true and false The target trajectory is obtained according to the radar target trajectory feature extraction method in step 2. The feature set of a target trajectory The similarity d between two different target tracks is represented by the feature distance. ij :

[0047] (11);

[0048] Step 4.2: Obtain the feature matrix D based on the similarity between any two paths. N :

[0049] (12);

[0050] Since the feature distances correspond quite well to each other, that is Therefore, the characteristic matrix DN It is a symmetric matrix, and all elements on the main diagonal are 0;

[0051] Step 4.3, based on the feature matrix D N Arbitrary target trajectory Based on similarity d ij Select The set of target tracks most similar to it and the corresponding set of similar target track features. ,in, ;

[0052] Step 4.4, based on the set of similar target track features By using a deep neural network classifier, obtain of A set of decision-making evidence: ,in, , This is a three-dimensional confidence vector, representing the degree of support given to true, false, and uncertain evidence in decision-making. This represents the Kth piece of decision evidence; based on the DS rule, the two pieces of decision evidence are... and The fusion and combination formula is as follows:

[0053] (13);

[0054] in, and The labels used to represent evidence for decision-making. , ; and If all the confidence levels are true, false, or uncertain, then... , ; The confidence function after fusion. The confidence level of the result obtained from the fusion of decision evidence regarding the authenticity of the target trajectory; The confidence level of the result obtained from the fusion of decision evidence that the target trajectory is false; The confidence level of the result obtained from the fusion of decision evidence for the target trajectory to be of an uncertain category;

[0055] After completing the fusion of all decision-making evidence in sequence, the target trajectory is obtained. Global confidence function ;in, ;

[0056] Step 4.5, for Set of target tracks Perform steps 4.1 to 4.4 to obtain... A set of global confidence functions for each target trajectory .

[0057] Furthermore, in step 5, the process of classifying the target tracks of the two subsets based on the overall confidence distribution is as follows:

[0058] Step 5.1: Based on whether the maximum confidence level belongs to the uncertain category, divide the global confidence function set of the target trajectory into two subsets. and :

[0059] (14);

[0060] (15);

[0061] in, for The category with the highest confidence. ;

[0062] like Then Classified to middle;

[0063] like Then classify it into middle;

[0064] Step 5.2, will All target tracks are directly assigned to the class with the highest confidence level as true. Or false category ,right Perform category labeling, Category tags for:

[0065] (16);

[0066] Step 5.2, for After all target tracks have been classified into patterns, calculate the following: Category Center and Category Center :

[0067] (17);

[0068] in, The maximum confidence level is the number of true target tracks. The number of target tracks with the highest confidence level as false;

[0069] Step 5.3, for Calculate the global confidence function for the target trajectory in the target path. respectively with and The confidence distance to the category center is used to assign the pattern to the nearest center. for:

[0070] (18);

[0071] )for The category center;

[0072] For target tracks of uncertain category, the category is determined based on the overall confidence distribution. At this point, all target tracks have been determined to be real or fake.

[0073] The technical effects of this invention are as follows:

[0074] This invention overcomes the limitations of traditional algorithms based on probabilistic frameworks and designs a target trajectory recognition framework based on confidence functions. This invention designs a deep neural network classifier based on confidence functions to improve the mining and learning of the intrinsic relationship between trajectory features and pattern categories. This invention supplements the number of decision samples through two-step decision-making, which can effectively improve the recognition accuracy of target trajectories of uncertain categories and realize rapid target navigation and stable tracking. Attached Figure Description

[0075] Figure 1 The flowchart shows a method for distinguishing true and false radar target tracks based on confidence classification.

[0076] Figure 2 This is a target real / false track classifier based on deep neural networks. Detailed Implementation

[0077] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0078] A method for distinguishing true and false radar target tracks based on confidence classification; such as Figure 1 As shown, the specific steps are as follows:

[0079] Step 1: Construct a category confidence identification framework for radar target tracks;

[0080] Based on the advantages of confidence function theory in representing and processing uncertain information, a category confidence identification framework for radar target tracks is constructed. ,in, Indicates the actual track category. Indicates the category of false flight path. Indicates an uncertain category;

[0081] If the real target track categories and the false target track categories are mutually exclusive, then the category intersection operation results in an empty set, i.e. ;

[0082] The uncertain category is the union of all independent and mutually exclusive categories. Within the category confidence identification framework, for Therefore, the intersection operation of each independent mutually exclusive category and the uncertain category is itself. , .

[0083] Step 2: Extract radar target track features;

[0084] The radar target trajectory characteristics include time difference. Displacement distance ,speed acceleration and spatial distance Five-dimensional features;

[0085] The radar target track is composed of M radar points arranged in a time sequence, where the radar point feature P at time l is... l :

[0086] (1);

[0087] in, , , These are the target spatial location information after the measurements have been converted into northeast celestial coordinates; t l For time series labeling;

[0088] A set of M radar points for:

[0089] (2);

[0090] Time difference between two radar points in the (k+1)th and kth sequences for:

[0091] (3);

[0092] Displacement distance between two radar points from time k to k+1 for:

[0093] (4);

[0094] The target velocity in time series k for:

[0095] (5);

[0096] target acceleration at time series k for:

[0097] (6);

[0098] The spatial distance from the radar spot to the origin at time k. for:

[0099] (7);

[0100] The five-dimensional feature vector of the k-th time-series radar point for:

[0101] (8);

[0102] Calculate the mean values ​​of each dimension of the features of all time-series radar points contained in the track:

[0103] (9);

[0104] Finally, the five-dimensional feature vector of the radar target trajectory is obtained. for:

[0105] (10).

[0106] Step 3: Design a deep neural network classifier based on confidence functions;

[0107] A deep neural network classifier is set up based on radar target trajectory characteristics and category confidence recognition framework;

[0108] The deep neural network classifier includes an input layer, hidden layers, and an output layer; for example... Figure 2 As shown;

[0109] Fully connected neural network structures are intuitive, allowing for flexible adjustments to the number of network layers and neurons based on task complexity. They effectively correlate global features during data classification. Therefore, the deep neural network classifier described is designed based on the task complexity of radar target trajectory recognition and the characteristics of the confidence recognition framework. Figure 2 As shown;

[0110] The input layer has 5 nodes, which correspond to time difference, displacement distance, target velocity, target acceleration, and spatial distance, respectively. The input to the deep neural network classifier is the five-dimensional feature vector of the radar target trajectory. ;

[0111] The hidden layer has 5 layers and all nodes in the 5th layer are set to 20.

[0112] The setting of hidden layers is related to the complexity of the task. If the deep neural network structure is too simple, it will not be able to effectively explore the correlation between the features of the sample data and the output category; if the deep neural network structure is too complex, it will lead to limitations such as overfitting and gradient vanishing. Considering that the target trajectory has already undergone effective feature extraction before being input into the classifier, the number of hidden layers is set to 5 and the number of nodes in each layer is set to 20.

[0113] The output layer has 3 nodes, each corresponding to a different real track category. False track category and uncertain categories confidence level , and And satisfy This means that the confidence sum of all pattern categories given by the deep neural network classifier is 1.

[0114] The output of the deep neural network classifier is the confidence level of the target trajectory belonging to each pattern category, namely, real, false, and uncertain categories in the confidence recognition framework constructed in step 1; therefore, the number of nodes in the output layer of the deep neural network classifier is set to 3.

[0115] Step 4: Construct target trajectory decision evidence using a deep neural network classifier, and fuse the decision evidence to obtain a global confidence function set;

[0116] Step 4.1, Define To distinguish between true and false The target trajectory is obtained according to the radar target trajectory feature extraction method in step 2. The feature set of a target trajectory The similarity d between two different target tracks is represented by the feature distance. ij :

[0117] (11);

[0118] Step 4.2: Obtain the feature matrix D based on the similarity between any two paths. N :

[0119] (12);

[0120] Since the feature distances correspond quite well to each other, that is Therefore, the characteristic matrix D N It is a symmetric matrix, and all elements on the main diagonal are 0; this is because each track is completely similar to itself, and the feature distance is 0.

[0121] Step 4.3, based on the feature matrix D NArbitrary target trajectory Based on similarity d ij Select The set of target tracks most similar to it and the corresponding set of similar target track features. ,in, ; The choice of value is related to the distribution characteristics of the task data. A larger value better represents the overall correlation, but it will greatly increase the computational complexity. In specific engineering applications, the following should be selected: and ;

[0122] Step 4.4, based on the set of similar target track features By using a deep neural network classifier, obtain of A set of decision-making evidence: ,in, , This is a three-dimensional confidence vector, representing the degree of support given to true, false, and uncertain evidence in decision-making. This represents the Kth piece of decision evidence; based on the DS rule, the two pieces of decision evidence are... and The fusion and combination formula is as follows:

[0123] (13);

[0124] in, and The labels used to represent evidence for decision-making. , ; and If all the confidence levels are true, false, or uncertain, then... , ; The confidence function after fusion. The confidence level of the result obtained from the fusion of decision evidence regarding the authenticity of the target trajectory; The confidence level of the result obtained from the fusion of decision evidence that the target trajectory is false; The confidence level of the result obtained from the fusion of decision evidence for the target trajectory to be of an uncertain category;

[0125] After completing the fusion of all decision-making evidence in sequence, the target trajectory is obtained. Global confidence function ;in, ;

[0126] Step 4.5, for Set of target tracks Perform steps 4.1 to 4.4 to obtain... A set of global confidence functions for each target trajectory .

[0127] Step 5: Divide the global confidence function set into two subsets through a two-step classification decision; determine the category of the target track in the two subsets based on the overall confidence distribution to obtain the target track of uncertain category; finally, complete the discrimination between true and false tracks;

[0128] Step 5.1: Based on whether the maximum confidence level belongs to the uncertain category, divide the global confidence function set of the target trajectory into two subsets. and :

[0129] (14);

[0130] (15);

[0131] in, for The category with the highest confidence. ;

[0132] like Then Classified to middle;

[0133] like Then classify it into middle;

[0134] Step 5.2, will All target tracks are directly assigned to the class with the highest confidence level as true. Or false category ,right Perform category labeling, Category tags for:

[0135] (16);

[0136] Step 5.2, for After all target tracks have been classified into patterns, calculate the following: Category Center and Category Center :

[0137] (17);

[0138] in, The maximum confidence level is the number of true target tracks. The number of target tracks with the highest confidence level as false;

[0139] Step 5.3, for Calculate the global confidence function for the target trajectory in the target path. respectively with and The confidence distance to the category center is used to assign the pattern to the nearest center. for:

[0140] (18);

[0141] )for The category center;

[0142] For target tracks of uncertain category, the category is determined based on the overall confidence distribution. At this point, all target tracks have been determined to be real or fake.

[0143] All target tracks that have completed category determination are used as supplementary samples to optimize the designed confidence function-based deep network classifier, thereby continuously improving the performance of the deep network classifier.

Claims

1. A method for distinguishing true and false radar target tracks based on confidence classification, characterized in that, The specific steps are as follows: Step 1: Construct a category confidence identification framework for radar target tracks; Step 2: Extract radar target trajectory features; Step 3: Design a deep neural network classifier based on confidence functions; A deep neural network classifier is set up based on radar target trajectory characteristics and category confidence recognition framework; Step 4: Construct target trajectory decision evidence using a deep neural network classifier, and fuse the decision evidence to obtain a global confidence function set; Step 5: Divide the global confidence function set into two subsets through a two-step classification decision; determine the category of the target track in the two subsets based on the overall confidence distribution to obtain the target track of uncertain category; finally, complete the discrimination between true and false tracks.

2. The radar target real / false track discrimination method based on confidence classification according to claim 1, characterized in that, In step 1, the framework for constructing the category confidence identification of radar target tracks is as follows: Based on the advantages of confidence function theory in representing and processing uncertain information, a category confidence identification framework for radar target tracks is constructed. ,in, Indicates the actual track category. Indicates the category of false flight path. Indicates an uncertain category; If the real target track categories and the false target track categories are mutually exclusive, then the category intersection operation results in an empty set, i.e. ; The uncertain category is the union of all independent and mutually exclusive categories. Within the category confidence identification framework, for Therefore, the intersection operation of each independent mutually exclusive category and the uncertain category is itself. , .

3. The radar target real / false track discrimination method based on confidence classification according to claim 1, characterized in that, In step 2, the process of extracting radar target track features is as follows: The radar target trajectory characteristics include time difference. Displacement distance ,speed acceleration and spatial distance Five-dimensional features; The radar target track is composed of M radar points arranged in a time sequence, where the radar point feature P at time l is... l : （1）； in, , , These are the target spatial location information after the measurements have been converted into northeast celestial coordinates; t l For time series labeling; A set of M radar points for: （2）； Time difference between two radar points in the (k+1)th and kth sequences for: （3）； Displacement distance between two radar points from time k to k+1 for: （4）； The target velocity in time series k for: （5）； target acceleration at time series k for: （6）； The spatial distance from the radar spot to the origin at time k. for: （7）； The five-dimensional feature vector of the k-th time-series radar point for: （8）； Calculate the mean values ​​of each dimension of the features of all time-series radar points contained in the track: （9）； Finally, the five-dimensional feature vector of the radar target trajectory is obtained. for: （10）。 4. The radar target real / false track discrimination method based on confidence classification according to claim 1, characterized in that, In step 3, the deep neural network classifier includes an input layer, a hidden layer, and an output layer; The input layer has 5 nodes, which correspond to time difference, displacement distance, target velocity, target acceleration, and spatial distance, respectively. The input to the deep neural network classifier is the five-dimensional feature vector of the radar target trajectory. ; The hidden layer has 5 layers and all nodes in the 5th layer are set to 20. The output layer has 3 nodes, each corresponding to a different real track category. False track category and uncertain categories confidence level , and And satisfy That is, the confidence sum of all pattern categories given by the deep neural network classifier is 1.

5. The radar target real / false track discrimination method based on confidence classification according to claim 1, characterized in that, In step 4, the process of constructing and fusing the target trajectory decision evidence is as follows: Step 4.1, Define To distinguish between true and false The target trajectory is obtained according to the radar target trajectory feature extraction method in step 2. The feature set of a target trajectory The similarity d between two different target tracks is represented by the feature distance. ij : （11）； Step 4.2: Obtain the feature matrix D based on the similarity between any two paths. N : （12）； Since the feature distances correspond quite well to each other, that is Therefore, the characteristic matrix D N It is a symmetric matrix, and all elements on the main diagonal are 0; Step 4.3, based on the feature matrix D N Arbitrary target trajectory Based on similarity d ij Select The set of target tracks most similar to it and the corresponding set of similar target track features. ,in, ; Step 4.4, based on the set of similar target track features By using a deep neural network classifier, obtain of A set of decision-making evidence: ,in, , This is a three-dimensional confidence vector, representing the degree of support given to true, false, and uncertain evidence in decision-making. This represents the Kth piece of decision evidence; based on the DS rule, the two pieces of decision evidence are... and The fusion and combination formula is as follows: （13）； in, and The labels used to represent evidence for decision-making. , ; and If all the confidence levels are true, false, or uncertain, then... , ; The confidence function after fusion. The confidence level of the result obtained from the fusion of decision evidence regarding the authenticity of the target trajectory; The confidence level of the result obtained from the fusion of decision evidence that the target trajectory is false; The confidence level of the result obtained from the fusion of decision evidence for the target trajectory to be of an uncertain category; After completing the fusion of all decision-making evidence in sequence, the target trajectory is obtained. Global confidence function ;in, ; Step 4.5, for Set of target tracks Perform steps 4.1 to 4.4 to obtain... A set of global confidence functions for each target trajectory .

6. The radar target real / false track discrimination method based on confidence classification according to claim 1, characterized in that, In step 5, the process of classifying the target tracks of the two subsets based on the overall confidence distribution is as follows: Step 5.1: Based on whether the maximum confidence level belongs to the uncertain category, divide the global confidence function set of the target trajectory into two subsets. and : （14）； （15）； in, for The category with the highest confidence. ; like Then Classified to middle; like Then classify it into middle; Step 5.2, will All target tracks are directly assigned to the class with the highest confidence level as true. Or false category ,right Perform category labeling, Category tags for: （16）； Step 5.2, for After all target tracks have been classified into patterns, calculate the following: Category Center and Category Center : （17）； in, The maximum confidence level is the number of true target tracks. The number of target tracks that are considered false with the highest confidence level; Step 5.3, for Calculate the global confidence function for the target trajectory in the target path. respectively with and The confidence distance to the category center is used to assign the pattern to the nearest center. for: （18）； )for The category center; For target tracks of uncertain category, the category is determined based on the overall confidence distribution. At this point, all target tracks have been determined to be real or fake.

7. An electronic device, characterized in that, include: One or more processors; Memory; One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs being configured to perform the method as described in any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program code that can be invoked by a processor to execute the method as described in any one of claims 1 to 6.

Images