An intelligent anti-interference method based on radar-infrared heterogeneous fusion

Abstract

This invention discloses an intelligent anti-jamming method based on radar-infrared heterogeneous fusion, comprising: radar and infrared sensors performing local filtering based on their respective measurements to obtain local state estimates; constructing a set space of track mapping relationships between radar and infrared sensors, and selecting the optimal track matching pair based on a minimum cost function; adaptively configuring the fusion weights of the heterogeneous sensors according to the optimal matching result to complete the radar-infrared anti-jamming fusion; deriving variable-dimensional Mahalanobis distance based on the projection of the reliable dimension measurement space to extract spatial statistical features, and constructing a main lobe active deception interference feature dataset in conjunction with the physical law features of the measurement itself; and building a deep learning model based on a long short-term memory network to achieve accurate identification of radar main lobe active deception interference types. This invention effectively suppresses main lobe active deception interference and achieves stable tracking of real targets through the designed distributed heterogeneous fusion criterion.

Description

Technical Field

[0001] This invention relates to the fields of sensor fusion and electronic countermeasures, and in particular to an intelligent anti-jamming method based on radar-infrared heterogeneous fusion. Background Technology

[0002] In modern electronic warfare environments, radar main lobe jamming can be categorized into suppression jamming and active deception jamming based on its mechanism of action. Among these, active main lobe deception jamming, with its advantages of strong concealment, high controllability, and stable deception effect, has become a mainstream jamming style. Radar-infrared composite anti-jamming technology is considered one of the important approaches to dealing with complex electronic warfare environments. Compared to centralized fusion, distributed fusion technology has significant advantages in terms of communication load, system robustness, and time synchronization accuracy requirements, and has become an important development direction for heterogeneous fusion of radar and infrared systems.

[0003] The Relative Field Synthesis (RFS) theory provides a unified method for describing and modeling uncertainties in multi-source, multi-target systems within the framework of point process theory, by modeling the uncertainties of multi-source measurements and the time-varying characteristics of the unknown number of targets. Based on this, many high-performance distributed fusion algorithms have been proposed. However, most existing distributed fusion algorithms are designed for similar multi-sensor systems. In radar-infrared heterogeneous fusion scenarios, different types of sensors have different observation mechanisms for the same target and significantly different response characteristics to the same interference signals. Existing technologies often directly adopt fusion strategies for homogeneous sensors without fully considering the type differences between heterogeneous sensors, leading to a decline in fusion performance under complex interference environments. Although some studies have attempted to introduce adaptive fault-tolerant or hierarchical track fusion mechanisms, the fundamental problem of fusion model mismatch caused by the differences in heterogeneous sensor characteristics has not been solved.

[0004] To counter active deception jamming on the radar main lobe, accurate jamming type identification is a prerequisite for formulating effective countermeasures. Existing jamming detection and identification methods are mainly limited to determining the "presence" of jamming. For example, detection methods based on the statistical characteristics of measurement residuals or the analytical formula of jamming probability can only distinguish the presence or absence of jamming, but cannot specifically identify the specific pattern of jamming. In real electronic warfare environments, enemy jamming patterns are highly agile, encompassing various complex types such as range deception, velocity deception, combined range-velocity deception, and angle deception. Existing technical solutions lack the ability to identify these specific jamming patterns, making it difficult to support targeted anti-jamming decisions.

[0005] Overall, existing distributed fusion algorithms are mainly designed for homogeneous sensors and do not fully consider the type differences between heterogeneous sensors, thus failing to effectively suppress active spoofing interference on the main lobe. Existing interference type identification algorithms can only identify the presence of specific interferences and lack descriptions of various interference characteristics. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a radar-infrared heterogeneous fusion method under a distributed framework to suppress active interference in the main lobe and identify the type of interference.

[0007] To achieve the above objectives, the technical solution provided by this invention is: an intelligent anti-interference method based on radar-infrared heterogeneous fusion, comprising the following steps:

[0008] Step 1: Radar and infrared sensors perform local filtering based on their respective measurements to obtain local state estimates; Step 2: Construct a set space of track mapping relationships between radar and infrared sensors, and select the optimal track matching pair based on the minimum cost function; Step 3: Adaptively configure the fusion weights of heterogeneous sensors according to the results of the optimal track matching pair, complete the anti-interference fusion of radar and infrared sensors, and output the anti-interference fused track; Step 4: Derive the variable-dimensional Mahalanobis distance based on the projection of the reliable dimension measurement space to extract spatial statistical features, and construct a main lobe active deception interference feature dataset in conjunction with the physical law features of the measurement itself; Step 5: Build a deep learning model based on a long short-term memory network, train the network using the constructed dataset, and achieve accurate identification of radar main lobe active deception interference types.

[0009] Optionally, in step 1, the signals received by the radar and infrared sensor are locally filtered using a labeled multi-target Bernoulli filter, and the local posterior probability density distributions obtained by the radar and infrared sensor are both LMB distributions, as shown in equation (1):

[0010] (1);

[0011] in, This represents the local posterior probability density distribution of the radar and infrared sensors. Labels are used to ensure that different target states have differentiated labels; This represents the Kronekerdelta function; Indicates weight; Represents the target probability density; For a set of multi-objective states The corresponding label set; the superscript X represents the multi-objective state set. The product of the probability density functions of all elements within the array;

[0012] ;

[0013] ;

[0014] In practical implementation, Use parameter set To fully characterize. Among them, It is tagged as The probability of the existence of the target; It is tagged as The target status PDF; It is a collection of tags.

[0015] Optionally, step 2 includes: step 21: determining the fusion criteria of the radar and infrared sensor; step 22: solving for the optimal matching between the radar and infrared sensor and the target.

[0016] Optionally, the fusion criterion in step 21 is as shown in equation (2):

[0017] (2);

[0018] in, Let represent the posterior probability density of the s-th sensor, where s = 1, 2; This represents the posterior probability density distribution after fusion; Indicates sensor The fusion weights satisfy the following conditions: ; The differential symbol for a set variable; Indicates the number of sensors.

[0019] Optionally, step 22 includes:

[0020] Step 221: Determine the target association cost function as follows:

[0021] (3);

[0022] in, , This represents the Mahalanobis distance between two distributions; Indicates radar target The probability density; Indicates infrared target The probability density; This represents the distance threshold.

[0023] Step 222: Construct the cost matrix as follows:

[0024] (4);

[0025] in, Here is the cost matrix. Indicates radar target With infrared targets Related costs; Indicates radar target With infrared targets Related costs; Indicates radar target With infrared targets Related costs; Indicates radar target With infrared targets Related costs.

[0026] Step 223: Solve for the optimal matching, including: defining the mapping relationship between radar targets and infrared targets. Mapping and correlation relationships between radar targets and infrared targets All mappings constitute a mapping space, denoted as . ; Represents a set of radar tags; Given a set of infrared tags, any matching mapping can be performed. Represented as Allocation matrix The allocation matrix consists of binary numbers 0 and 1, and the sum of the elements in each row and each column is either 0 or 1; for any , If and only if the radar target and infrared targets When associated ,Right now:

[0027] (5);

[0028] in, For the allocation matrix, Indicates radar target With infrared targets Is it related? Indicates radar target With infrared targets Is it related? Indicates radar target With infrared targets Is it related? Indicates radar target With infrared targets Is it related?

[0029] Therefore, the cost function is represented as a matrix. and The Frobenius inner product form,

[0030] (6);

[0031] in, Indicates the cost associated with the target; Indicate whether the targets are related; The trace of the matrix is ​​represented; the superscript T indicates transpose.

[0032] Optionally, step 3 includes: step 31: adaptively configuring fusion weights for each target based on the matching results, and calculating the posterior probability density of the fused targets using fusion criteria; step 32: feeding back the fusion results from step 31 to the radar and infrared sensors; the radar and infrared sensors perform local prediction and updates based on the feedback results, respectively.

[0033] Optionally, step 31 includes: Step 311: Configure different fusion weights for different targets as follows:

[0034] (7);

[0035] in, , Representing radar targets respectively The covariance matrix and the infrared sensor target The covariance matrix; Indicates the fusion weights; The value of the variable that represents the minimum trace of the matrix function;

[0036] Step 312: Calculate the posterior probability density after target fusion;

[0037] The existence probability and state density of the fused radar and infrared sensors are as follows:

[0038] (8);

[0039] (9);

[0040] in, ; Denotes the integration constant; This indicates the probability of the target existing after fusion; This represents the probability density of the fused target. Indicates the fusion weights; Indicates the probability of the presence of a radar target; Represents the probability density of radar targets; Indicates the probability of the presence of an infrared target; This represents the probability density of infrared targets.

[0041] Therefore, the merged LMB distribution can be represented as:

[0042] (10);

[0043] in, This indicates the probability of the target existing after fusion; This represents the probability density of the fused target. This is the merged tag space.

[0044] Optionally, step 4 includes: Step 41: Based on the raw measurement vector received by the radar Constructing a hyperspace measurement set and utilize identity tags Each measurement subset of this hyperspace is identified, where Represents the set of labels corresponding to each physical measurement dimension in a single measurement vector; Step 42: Derive the target and set measurement The generalized objective likelihood function; Step 43: Integrate the confidence dimension projection statistical features and the temporal physical law features to construct the main lobe active deception interference feature dataset.

[0045] Optionally, step 41 includes:

[0046] Step 411: Define the time The radar measurement set is ;in, This represents the set of all measurements at the current moment. Indicates the first A single measurement vector , This indicates the total number of measurements acquired by the radar at the current moment;

[0047] Step 412: Place the first A traditional single-measure vector can be represented as a random set:

[0048] (11);

[0049] in, This represents the full-dimensional measurement space in which a single radar measurement vector resides; Represents the set of real numbers; express 3D real space; Indicates the first The observation components corresponding to each physical measurement dimension Further construct a hyperspace measurement ensemble :

[0050] (12);

[0051] in, Indicates the first Each measurement dimension component This represents the identity label corresponding to the component of this measurement dimension;

[0052] Step 413: Establish identity tags for radar surveillance scenarios. The correspondence between the physical measurement dimensions and the corresponding dimensions, where i=1 corresponds to radial distance. i=2 corresponds to the azimuth angle i=3 corresponds to pitch angle i=4 corresponds to Doppler velocity ;

[0053] Step 414: Utilizing the hyperspace measurement set For the Divide the measurement into segments and define the first... The set of trustworthy dimension labels corresponding to each measurement is: Then a subset of reliable dimension measurements is constructed. With interference dimension measurement subset It is expressed as follows:

[0054] (13);

[0055] (14);

[0056] in, Indicates the first A set of reliable dimension labels that are not affected by active deception in a measurement. Indicates the first A set of dimension labels that are subject to active deception interference in a measurement. and They represent the first A subset of reliable measurement dimension components and a subset of interference measurement dimension components, both satisfying:

[0057] (15).

[0058] Optionally, step 42 includes:

[0059] Step 421: Let Indicates a single-objective state. Represents the set of multi-objective states at the current moment; where, Indicates the motion state of a single target. Representing single-target identity labels; deriving a ensemble measurement method using the predicted LMB distribution sign representation and confidence mass function. Generalized objective likelihood function :

[0060] (16);

[0061] in, Indicates the detection probability; Represents the set of credible dimension label indexes; To measure the probability that the dimension is not disturbed. For the probability of interference, To measure the volume of space, Indicates clutter intensity parameters, Indicates that the target state is Conditions, No. One-dimensional likelihood function of each reliable measurement dimension component; Set goals With reliable measurement subset The correlation mapping between them; this process is called the projection of the credible dimension measurement space.

[0062] Step 422: Generalized objective likelihood function under Gaussian mixture implementation Equivalent to variable-dimensional Mahalanobis distance (VDMD):

[0063] (17);

[0064] in, ; ; ; For radar The m-th measurement at time m; For the target pseudo-measurement; The new information covariance matrix; To select a matrix, This indicates the measurement of the selected dimension. This indicates the dimension measurement of the selected dimension. This represents the covariance matrix of the selected dimension, and the superscript T indicates transpose.

[0065] The advantages of this invention over the prior art are:

[0066] 1) This intelligent anti-interference method can adaptively adjust the fusion weights of each target to suppress active deception interference of the main lobe;

[0067] 2) This intelligent anti-jamming method accurately identifies specific jamming patterns such as range deception, velocity deception, combined range-velocity deception, and angle deception through multi-dimensional feature joint analysis. The identification results provide crucial decision-making basis for subsequent development of targeted anti-jamming strategies, greatly enhancing the initiative and effectiveness of electronic countermeasures. Attached Figure Description

[0068] Figure 1 This is a flowchart of an intelligent anti-interference method based on radar-infrared heterogeneous fusion provided by an embodiment of the present invention. Detailed Implementation

[0069] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] Reference Figure 1 This embodiment provides an intelligent anti-jamming method based on radar-infrared heterogeneous fusion, including the following steps:

[0071] Step 1: The radar and infrared sensors perform local filtering based on their respective measurements to obtain local state estimates.

[0072] Specifically, the signals received by the radar and infrared sensors are locally filtered using a labeled multi-target Bernoulli filter. The local posterior probability density distributions obtained by the radar and infrared sensors are both LMB distributions, as shown in equation (1):

[0073] (1);

[0074] in, This represents the local posterior probability density distribution of the radar and infrared sensors. Labels are used to ensure that different target states have differentiated labels; This represents the Kronekerdelta function; Indicates weight; Represents the target probability density; For a set of multi-objective states The corresponding label set; the superscript X represents the multi-objective state set. The product of the probability density functions of all elements within the array;

[0075] ;

[0076] ;

[0077] In this embodiment, Use parameter set To fully characterize. Among them, It is tagged as The probability of the existence of the target; It is tagged as The target status PDF; It is a collection of tags; Representative set The product of the probability density functions of all elements within the array; Labels ensure that different target states have differentiated labels; For multi-objective states The corresponding tag set. Let represent the posterior probability density of the s-th sensor, where s = 1, 2.

[0078] Step 2: Construct a set space of track mapping relationships between radar and infrared sensors, and select the optimal track matching pair from it based on the minimum cost function.

[0079] Specifically, this includes: Step 21: Determining the fusion criteria of the radar and infrared sensor.

[0080] In this embodiment, the fusion criterion is selected as shown in equation (2):

[0081] (2);

[0082] in, Let represent the posterior probability density of the s-th sensor, where s = 1, 2; This represents the posterior probability density distribution after fusion; Indicates sensor The fusion weights satisfy the following conditions: ; The differential symbol for a set variable; Indicates the number of sensors.

[0083] Step 22: Solve for the optimal matching between the radar and infrared sensor and the target.

[0084] In this embodiment, step 22 includes:

[0085] Step 221: Determine the target association cost function as follows:

[0086] (3);

[0087] in, , This represents the Mahalanobis distance between two distributions; Indicates radar target The probability density; Indicates infrared target The probability density; This represents the distance threshold.

[0088] In this embodiment, the radar output label shows a multi-target Bernoulli distribution. Its parameterized characterization is Infrared output labeling for multi-target Bernoulli distribution Its parameterized characterization is .

[0089] Step 222: Construct the cost matrix as follows:

[0090] (4);

[0091] in, Here is the cost matrix. Indicates radar target With infrared targets Related costs; Indicates radar target With infrared targets Related costs; Indicates radar target With infrared targets Related costs; Indicates radar target With infrared targets Related costs.

[0092] Step 223: Solve for the optimal matching, including: defining the mapping relationship between radar targets and infrared targets. Mapping and correlation relationships between radar targets and infrared targets All mappings constitute a mapping space, denoted as . ; Represents a set of radar tags; Given a set of infrared tags, any matching mapping can be performed. Represented as Allocation matrix The allocation matrix consists of binary numbers 0 and 1, and the sum of the elements in each row and each column is either 0 or 1; for any , If and only if the radar target and infrared targets When associated ,Right now:

[0093] (5);

[0094] in, For the allocation matrix, Indicates radar target With infrared targets Is it related? Indicates radar target With infrared targets Is it related? Indicates radar target With infrared targets Is it related? Indicates radar target With infrared targets Is it related?

[0095] Therefore, the cost function can be represented as a matrix. and The Frobenius inner product form,

[0096] (6);

[0097] in, Indicates the cost associated with the target; Indicate whether the targets are related; Represents the trace of a matrix.

[0098] It should be noted that this linear allocation problem is solved by minimizing the cost function. To solve for the optimal allocation matrix The resulting optimal allocation matrix can be uniquely mapped to the corresponding optimal label matching relationship. The linear assignment problem for finding the optimal label matching relationship can be solved using the Murty / Gibbs algorithm.

[0099] Step 3: Adaptively configure the fusion weights of heterogeneous sensors based on the optimal matching result, complete the radar and infrared anti-interference fusion, and output the fused track after anti-interference.

[0100] In this embodiment, step 3 includes:

[0101] Step 31: Adaptively configure fusion weights for each target based on the matching results, and calculate the posterior probability density of the fused targets using the fusion criteria.

[0102] Specifically, step 311: Configure different fusion weights for different targets as follows:

[0103] (7);

[0104] in, , Representing radar targets respectively The covariance matrix and the infrared sensor target The covariance matrix; Indicates the fusion weights; The value of the variable that represents the minimum trace of the matrix function.

[0105] Step 312: Calculate the posterior probability density after target fusion.

[0106] The existence probability and state density of the fused radar and infrared heterogeneous sensors are as follows:

[0107] (8);

[0108] (9);

[0109] in, ; Denotes the integration constant; This indicates the probability of the target existing after fusion; This represents the probability density of the fused target. Indicates the fusion weights; Indicates the probability of the presence of a radar target; Represents the probability density of radar targets; Indicates the probability of the presence of an infrared target; This represents the probability density of infrared targets. Therefore, the fused LMB distribution can be expressed as:

[0110] (10);

[0111] in, This indicates the probability of the target existing after fusion; This represents the probability density of the fused target. This is the merged tag space.

[0112] Step 32: The fusion result from Step 31 is fed back to the radar and infrared sensors; the radar and infrared sensors perform local prediction and updates respectively based on the feedback result. The distribution of the radar target state prediction is represented as follows: The target state after update is characterized as .

[0113] Step 4: Based on the spatial projection of the reliable dimension measurement, derive the variable dimension Mahalanobis distance to extract spatial statistical features, and combine it with the physical law features of the measurement itself to construct the main lobe active deception interference feature dataset.

[0114] In this embodiment, step 4 includes:

[0115] Step 41: Based on the raw measurement vector received by the radar Constructing a hyperspace measurement set and utilize identity tags Each measurement subset of this hyperspace is identified, where This represents the set of labels corresponding to each physical measurement dimension in a single measurement vector.

[0116] Specifically, step 41 includes:

[0117] Step 411: Define the time The radar measurement set is ;in, This represents the set of all measurements at the current moment. Indicates the first A single measurement vector , This represents the total number of measurements obtained by the radar at the current moment.

[0118] Step 412: Place the first A traditional single-measure vector can be represented as a random set:

[0119] (11);

[0120] in, This represents the full-dimensional measurement space in which a single radar measurement vector resides; Represents the set of real numbers; express 3D real space; Indicates the first The observation components corresponding to each physical measurement dimension Further construct a hyperspace measurement ensemble :

[0121] (12);

[0122] in, Indicates the first Each measurement dimension component This represents the identity label corresponding to the component of this measurement dimension;

[0123] Step 413: Establish identity tags for radar surveillance scenarios. The correspondence between the physical measurement dimensions and the corresponding dimensions, where i=1 corresponds to radial distance. i=2 corresponds to the azimuth angle i=3 corresponds to pitch angle i=4 corresponds to Doppler velocity ;

[0124] Step 414: Utilizing the hyperspace measurement set For the Divide the measurement into segments and define the first... The set of trustworthy dimension labels corresponding to each measurement is: Then a subset of reliable dimension measurements is constructed. With interference dimension measurement subset It is expressed as follows:

[0125] (13);

[0126] (14);

[0127] in, Indicates the first A set of reliable dimension labels that are not affected by active deception in a measurement. Indicates the first A set of dimension labels that are subject to active deception interference in a measurement. and They represent the first A subset of reliable measurement dimension components and a subset of interference measurement dimension components, both satisfying:

[0128] (15).

[0129] Step 42: Deriving the target and set measurement The generalized objective likelihood function.

[0130] Specifically, step 42 includes:

[0131] Step 421: Let Indicates a single-objective state. Represents the set of multi-objective states at the current moment; where, Indicates the motion state of a single target. Representing single-target identity labels; deriving a ensemble measurement method using the predicted LMB distribution sign representation and confidence mass function. Generalized objective likelihood function :

[0132] (16);

[0133] in, Indicates the detection probability; Represents the set of credible dimension label indexes; To measure the probability that the dimension is not disturbed. For the probability of interference, To measure the volume of space, Indicates clutter intensity parameters, Indicates that the target state is Conditions, No. One-dimensional likelihood function of each reliable measurement dimension component; Set goals With reliable measurement subset The mapping between them;

[0134] Step 422: Generalized objective likelihood function under Gaussian mixture implementation Equivalent to variable-dimensional Mahalanobis distance (VDMD):

[0135] (17);

[0136] in, ; ; ; For radar The m-th measurement at time m; For the target pseudo-measurement; The new information covariance matrix; To select a matrix, This indicates the measurement of the selected dimension. This indicates the dimension measurement of the selected dimension. This represents the covariance matrix of the selected dimension, and the superscript T indicates transpose.

[0137] Step 43: Integrate the credible dimension projection statistical features and the temporal physical law features to construct the main lobe active deception interference feature dataset.

[0138] Specifically, step 43 includes:

[0139] Step 431: Based on the credible dimension measurement space projection, extract the variable-dimensional Mahalanobis distance statistical features of the main lobe active spoofing interference. Specifically, define the VDMD values ​​corresponding to the following four typical dimension combinations:

[0140] ;

[0141] in, Represents the Mahalanobis distance across all dimensions; Indicates the distance to the Vimarcante distance; Indicates the velocity-wise distance; This represents the distance-velocity Vimarcante distance. The four VDMD values ​​mentioned above form a statistical feature vector. :

[0142] ;

[0143] Step 432: Based on the characteristics of the measurement itself, extract the physical characteristics of the main lobe active deception interference and define the physical characteristics. for:

[0144] ;

[0145] in, Indicates distance; Indicates Doppler velocity; Indicates the signal amplitude.

[0146] Step 433: Combine the statistical and physical characteristics of the main lobe active deception interference to obtain the joint feature vector. :

[0147] ;

[0148] A feature dataset for main lobe active deception interference is constructed based on feature vectors.

[0149] Step 5: Build a deep learning model based on a long short-term memory network, train the network using the constructed dataset, and achieve accurate identification of active deception interference types on the radar main lobe.

[0150] Specifically, a deep cascaded architecture based on convolutional neural networks (CNN) and long short-term memory networks (LSTM) is constructed, and an attention mechanism is embedded in this architecture. The local spatial features of the interference signal are extracted by the CNN layer, the temporal evolution of the features are captured by the LSTM layer, and the key spatiotemporal feature components are dynamically weighted by the attention mechanism, so as to achieve accurate classification and identification of active deception interference types of the main lobe.

[0151] Then, using the dataset constructed in step 4, the constructed neural network is trained to identify the active deception jamming type of the radar main lobe.

[0152] Finally, based on the output posterior LMB distribution, the target state is extracted; then steps 1 to 11 are repeated.

[0153] In this embodiment, the target state estimation result is extracted as follows:

[0154] ;

[0155] in, Extract a threshold for the target state; Indicates the status of the target being extracted; This represents the input value at which the objective function reaches its maximum value.

[0156] The intelligent anti-interference method provided in this embodiment effectively suppresses the main lobe active deception interference through the designed distributed heterogeneous fusion criterion, achieving stable tracking of the real target. At the same time, based on the constructed two-layer feature dataset, the LSTM deep learning grid can output accurate interference types, significantly improving the system's anti-interference robustness and the accuracy of interference type identification.

[0157] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A smart anti-interference method based on radar-infrared heterogeneous fusion, characterized in that, Includes the following steps: Step 1: The radar and infrared sensors perform local filtering based on their respective measurements to obtain local state estimates; Step 2: Construct a set space of track mapping relationships between radar and infrared sensors, and select the optimal track matching pair from the set space of track mapping relationships based on the minimum cost function; Step 3: Adaptively configure the fusion weights of heterogeneous sensors based on the results of the optimal track matching pair, complete the fusion of radar and infrared anti-sensor interference, and output the anti-interference fused track; Step 4: Based on the spatial projection of the reliable dimension measurement, derive the variable dimension Mahalanobis distance to extract spatial statistical features, and combine it with the physical law features of the measurement itself to construct the main lobe active deception interference feature dataset; Step 5: Build a deep learning model based on a long short-term memory network, train the network using the constructed dataset, and achieve accurate identification of active deception interference types on the radar main lobe.

2. The intelligent anti-interference method according to claim 1, characterized in that, In step 1, the signals received by the radar and infrared sensor are locally filtered using a labeled multi-target Bernoulli filter. The local posterior probability density distributions obtained by the radar and infrared sensor are both LMB distributions, as shown in equation (1): （1）； in, This represents the local posterior probability density of the radar and infrared sensors. Labels are used to ensure that different target states have differentiated labels; This represents the Kroneker delta function; Indicates weight; Represents the target probability density; For a set of multi-objective states The corresponding label set; the superscript X represents the multi-objective state set. The product of the probability density functions of all elements within the array.

3. The intelligent anti-interference method according to claim 1, characterized in that, Step 2 includes: Step 21: Determine the fusion criteria for the radar and infrared sensor; Step 22: Solve for the optimal matching between the radar and infrared sensor and the target.

4. The intelligent anti-interference method according to claim 3, characterized in that, The fusion criterion in step 21 is shown in equation (2): （2）； in, Let represent the posterior probability density of the s-th sensor, where s = 1, 2; This represents the posterior probability density distribution after fusion; Indicates sensor The fusion weights satisfy the following conditions: ; The differential symbol for a set variable; Indicates the number of sensors.

5. The intelligent anti-interference method according to claim 3, characterized in that, Step 22 includes: Step 221: Determine the target association cost function as follows: （3）； in, , This represents the Mahalanobis distance between two distributions; Indicates radar target The probability density; Indicates infrared target The probability density; Indicates the distance threshold; Step 222: Construct the cost matrix as follows: （4）； in, Let be the cost matrix. Indicates radar target With infrared targets Related costs; Indicates radar target With infrared targets Related costs; Indicates radar target With infrared targets Related costs; Indicates radar target With infrared targets Related costs; Step 223: Solve for the optimal matching, including: defining the mapping relationship between radar targets and infrared targets. Mapping and correlation relationships between radar targets and infrared targets All mappings constitute a mapping space, denoted as . ; Represents a set of radar tags; Represents a set of infrared tags; then any matching mapping... Represented as Allocation matrix The allocation matrix consists of binary numbers 0 and 1, and the sum of the elements in each row and each column is either 0 or 1; for any , If and only if the radar target and infrared targets When associated ,Right now: （5）； in, For the allocation matrix, Indicates radar target With infrared targets Is it related? Indicates radar target With infrared targets Is it related? Indicates radar target With infrared targets Is it related? Indicates radar target With infrared targets Is it related? Therefore, the cost function is represented as a matrix. and Frobenius inner product form: （6）； in, Indicates the cost associated with the target; Indicate whether the targets are related; The trace of the matrix is ​​represented; the superscript T indicates transpose.

6. The intelligent anti-interference method according to claim 1, characterized in that, Step 3 includes: Step 31: Adaptively configure fusion weights for each target based on the results of the optimal track matching pair, and calculate the posterior probability density of the fused targets using the fusion criterion; Step 32: Feed back the fusion result from step 31 to the radar and infrared sensors; the radar and infrared sensors perform local prediction and update based on the feedback result.

7. The intelligent anti-interference method according to claim 6, characterized in that, Step 31 includes: Step 311: Configure different fusion weights for different targets as follows: （7）； in, , Representing radar targets respectively The covariance matrix and the infrared sensor target The covariance matrix; Indicates the fusion weights; The value of the variable that represents the minimum trace of the matrix function; Step 312: Calculate the posterior probability density after target fusion; The existence probability and state density of the fused radar and infrared sensors are as follows: （8）； （9）； in, ; Denotes the integral constant; Indicates the probability of the existence of the target after fusion; This represents the probability density of the fused target. Indicates the probability of the presence of a radar target; Represents the probability density of radar targets; Indicates the probability of the presence of an infrared target; Represents the probability density of infrared targets; The resulting merged LMB distribution is represented as follows: （10）； in, Indicates the probability of the existence of the target after fusion; This represents the probability density of the fused target. This is the merged tag space.

8. The intelligent anti-interference method according to claim 1, characterized in that, Step 4 includes: Step 41: Based on the raw measurement vector received by the radar Constructing a hyperspace measurement set and utilize identity tags Each measurement subset of this hyperspace is identified, where This represents the set of labels corresponding to each physical measurement dimension in a single measurement vector; Step 42: Deriving the target and set measurement The generalized objective likelihood function; Step 43: Integrate the credible dimension projection statistical features and the temporal physical law features to construct the main lobe active deception interference feature dataset.

9. The intelligent anti-interference method according to claim 8, characterized in that, Step 41 includes: Step 411: Define the time. The radar measurement set is ;in, This represents the set of all measurements at the current moment. Indicates the first A single measurement vector , This indicates the total number of measurements acquired by the radar at the current moment; Step 412: Place the first A traditional single-measure vector can be represented as a random set: （11）； in, This represents the full-dimensional measurement space in which a single radar measurement vector resides; Represents the set of real numbers; express 3D real space; Indicates the first The observation components corresponding to each physical measurement dimension Further construct a hyperspace measurement ensemble : （12）； in, Indicates the first Each measurement dimension component This represents the identity label corresponding to the component of this measurement dimension; Step 413: Establish identity tags for radar surveillance scenarios. The correspondence between the physical measurement dimensions and the corresponding dimensions, where i=1 corresponds to radial distance. i=2 corresponds to the azimuth angle i=3 corresponds to pitch angle i=4 corresponds to Doppler velocity ; Step 414: Utilizing the hyperspace measurement set For the Divide the measurement into segments and define the first... The set of trustworthy dimension labels corresponding to each measurement is: Then a subset of reliable dimension measurements is constructed. With interference dimension measurement subset It is expressed as follows: （13）； （14）； in, Indicates the first A set of reliable dimension labels that are not affected by active deception in a measurement. Indicates the first A set of dimension labels that are subject to active deception interference in a measurement. and They represent the first A subset of reliable measurement dimension components and a subset of interference measurement dimension components, both satisfying: （15）。 10. The intelligent anti-interference method according to claim 8, characterized in that, Step 42 includes: Step 421: Let Indicates a single-objective state. Represents the set of multi-objective states at the current moment; where, Indicates the motion state of a single target. Representing single-target identity labels; deriving a ensemble measurement method using the predicted LMB distribution sign representation and confidence mass function. Generalized objective likelihood function : （16）； in, Indicates the detection probability; Represents the set of credible dimension label indexes; To measure the probability that the dimension is not disturbed. For the probability of interference, To measure the volume of space, Indicates clutter intensity parameters, Indicates that the target state is Conditions, No. One-dimensional likelihood function of each reliable measurement dimension component; Set goals With reliable measurement subset The mapping between them; Step 422: Generalized objective likelihood function under Gaussian mixture implementation Equivalent to variable-dimensional Mahalanobis distance: （17）； in, ; ; ; For radar The m-th measurement at time m; For the target pseudo-measurement; The new information covariance matrix; To select the matrix; Indicates the measurement of the selected dimension; Indicates the dimension measurement of the selected dimension; This represents the covariance matrix of the selected dimension; the superscript T indicates transpose.

Images