AI-based rapid target identification method for interceptor missiles
By proposing a rapid target identification method for interceptor missiles based on GRU networks, the problems of model non-convergence and noise interference in traditional methods are solved, achieving high-precision, low-resource-consumption target identification and adapting to complex cluster combat environments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-03-29
- Publication Date
- 2026-04-03
AI Technical Summary
In complex swarm warfare environments, traditional interceptor missile target identification methods, such as Kalman filtering and neural networks, suffer from problems such as model non-convergence, large noise interference, poor flexibility, and waste of computing resources, making it difficult to achieve high-precision and efficient interceptor target identification.
By employing a neural network based on GRU, training samples are obtained through a simulator to construct an interception probability model. Information fusion is performed by combining time weights to achieve rapid identification of interceptor missiles and aircraft, thereby reducing the performance requirements of airborne computers.
It improves the accuracy and flexibility of target identification, reduces the impact of noise, lowers the computational resource requirements, and is adaptable to different numbers of interceptor missiles and aircraft.
Smart Images
Figure CN116451149B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method for rapid target identification of interceptor missiles based on artificial intelligence, belonging to the field of guidance and control. Background Technology
[0002] In real-world swarm warfare scenarios, there will be situations where a large number of interceptor missiles intercept swarms of aircraft. To improve the survivability of aircraft, it is necessary to identify the interceptor missiles' targets in order to achieve targeted evasion.
[0003] Traditional interceptor missile target identification methods are mostly based on the Kalman Filter (KF) method, which is based on the prediction-correction idea.
[0004] To accurately model enemy attacks and predict the next move of interceptor missiles, key characteristic parameters of the interceptor missiles, such as their velocity and velocity direction, must be accurately acquired. This is difficult to achieve in complex air combat environments. In Kalman filtering, to address this issue, the difference between the true and estimated values is typically treated as significant process noise. However, this approach easily leads to non-convergence of the identification model. Furthermore, Kalman filtering requires high continuity of information acquisition, significantly reducing its engineering application value and limiting the maneuverability of our aircraft.
[0005] With the development of artificial intelligence technology, existing research has proposed using neural networks to identify interceptor targets for interceptor missiles. Compared with the Kalman filter method, neural network methods have significantly improved in terms of online identification speed and generalization ability. However, traditional neural networks are criticized for their low interpretability and susceptibility to noise interference. The main ways for neural networks to combat noise are through noise reduction processing before data is input into the network and through the model's own passive noise resistance. In intense cluster game scenarios, however, data measurement is difficult to guarantee completely accurate and is usually contaminated by noise. The model's output will fluctuate significantly in such cases, which will bring difficulties to subsequent decision-making.
[0006] Furthermore, existing methods for identifying and intercepting targets using neural networks lack flexibility. Under swarm warfare conditions, the number of interceptor missiles and aircraft is uncertain. To ensure training effectiveness, the number of interceptor missiles and aircraft needs to be preset during neural network training and use. When the actual number of aircraft is less than the preset number, the missing aircraft positions need to be filled with zeros. When the number of interceptor missiles is greater than the preset number, batch calculations are required. The former situation will waste a lot of computing resources, while the latter situation will lead to repeated calculations and reduced identification accuracy.
[0007] Therefore, it is necessary to conduct further research on the identification of interceptor missile targets in order to solve the above problems. Summary of the Invention
[0008] To overcome the above problems, the inventors conducted in-depth research and designed a rapid target identification method for interceptor missiles based on artificial intelligence, including the following steps:
[0009] S1. Use a neural network to obtain the interception probability of each interceptor missile against the aircraft at the current moment, as a preliminary identification result;
[0010] S2. Based on the preliminary identification results over a period of time, information is fused to identify the interception target of the interceptor missile.
[0011] Furthermore, S1 includes the following sub-steps:
[0012] S11. Build a simulator and obtain training samples through the simulator;
[0013] S12. Construct a neural network and train it using training samples to obtain a recognition model;
[0014] S13. The identification model is mounted on the aircraft, and the interception probability of each interceptor missile against the aircraft at the current moment is obtained through the identification model.
[0015] Furthermore, in S11, the simulator is equipped with multiple interceptor missiles and multiple aircraft. The process of interceptor missiles intercepting aircraft is simulated by the simulator, and the kinematic information of interceptor missiles and aircraft is extracted from it.
[0016] In a preferred embodiment, the rate of change of the relative distance between the interceptor missile and the aircraft in the simulator is expressed as:
[0017]
[0018] The relative angular acceleration between the interceptor missile and the aircraft is expressed as:
[0019]
[0020] Where i represents different interceptor missiles, j represents different aircraft, and ρ ij V represents the relative distance between the i-th interceptor missile and the j-th aircraft; a,i V represents the velocity of the i-th interceptor missile. t,j γ represents the velocity of the j-th aircraft; a,i γ represents the heading angle of the i-th interceptor missile. t,j λ represents the heading angle of the j-th aircraft. ij This represents the relative angle between the i-th interceptor missile and the j-th aircraft.
[0021] In a preferred embodiment, in the simulator, the interceptor missile employs a proportional guidance law, and its guidance command is as follows:
[0022]
[0023] Among them, u a,i This represents the normal acceleration command for the i-th interceptor missile, where N is the guidance parameter, n represents the number of aircraft, and T... i This represents the target intercepted by the i-th interceptor missile.
[0024] In a preferred embodiment, Gaussian white noise is added to the relative distance and relative angle between the interceptor missile and the aircraft in the simulator to simulate actual measurement conditions.
[0025] In a preferred embodiment, in S12, the neural network is a GRU network.
[0026] In a preferred embodiment, the output layer of the neural network uses softmax as the activation function to output the intercept probability of the interceptor missile against the aircraft, expressed as:
[0027]
[0028] p1(t|t)+p2(t|t)=1
[0029] Where p1(t|t) represents the probability that the current interceptor missile will intercept the aircraft at the current moment, p2(t|t) represents the probability that the current interceptor missile will not intercept the aircraft at the current moment, o1(t) represents the output value of the first neuron in the output layer of the neural network, o2(t) represents the output value of the second neuron in the output layer of the neural network, and o(t) = [o1(t), o2(t)] T .
[0030] In a preferred embodiment, in S13, the kinematic information of each interceptor missile and the aircraft is measured in real time by sensors mounted on the aircraft. The kinematic information of each interceptor missile and the aircraft is input into the identification model. The identification model outputs the interception probability of different interceptor missiles against the aircraft at the current moment. The obtained interception probability is used as the preliminary identification result.
[0031] In a preferred embodiment, in S2, the preliminary identification results from different times are fused by setting time weights.
[0032] The beneficial effects of this invention include:
[0033] (1) The rapid identification method for intercepting targets of interceptor missiles based on artificial intelligence provided by the present invention has high identification accuracy and is less affected by noise;
[0034] (2) The rapid target identification method based on artificial intelligence provided by the present invention is highly flexible, can adapt to different numbers of interceptor missiles and aircraft, and has low requirements for the performance of airborne computers. Attached Figure Description
[0035] Figure 1 A schematic flowchart of a rapid target identification method based on artificial intelligence for interceptor missiles according to a preferred embodiment of the present invention is shown.
[0036] Figure 2 This diagram illustrates the training process of the identification model in Example 1.
[0037] Figure 3 A schematic diagram of the identification results in Example 1 is shown. Detailed Implementation
[0038] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Through these descriptions, the features and advantages of the present invention will become clearer and more apparent.
[0039] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments. Although various aspects of embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless specifically indicated otherwise.
[0040] According to the present invention, a rapid target identification method for interceptor missiles based on artificial intelligence is provided, such as... Figure 1 As shown, it includes the following steps:
[0041] S1. Use a neural network to obtain the interception probability of each interceptor missile against the aircraft at the current moment, as a preliminary identification result;
[0042] S2. Based on the preliminary identification results over a period of time, information is fused to identify the interception target of the interceptor missile.
[0043] Unlike existing technologies that mostly obtain interceptor missile targets directly through neural networks, the inventors discovered that directly using neural networks to display interceptor missile target allocation presents numerous problems. These problems include poor flexibility and accuracy, and high computational demands, requiring a high-performance airborne computer. Because the number of interceptor missiles and aircraft is uncertain in swarm warfare scenarios, the number of interceptor missiles and aircraft needs to be pre-set during neural network training. When the actual number of interceptor missiles or aircraft is less than the preset number, the missing interceptor missile or aircraft positions need to be filled with zeros, which wastes significant computational resources. When the number of interceptor missiles exceeds the preset number, batch calculations are required, leading to redundant calculations and reduced identification accuracy.
[0044] In this invention, the process of acquiring the interceptor target of the interceptor missile is divided into two stages. In S1, a neural network is used to obtain the interception probability of each interceptor missile against the aircraft at the current moment, thereby achieving the pairing of the interceptor missile and the aircraft. When the interception probability of the interceptor missile against the aircraft is greater than a preset value, the interceptor missile and the aircraft are considered to be paired; otherwise, they are considered not to be paired.
[0045] Obtaining interception probabilities through neural networks, rather than directly outputting interception targets, facilitates distributed computing, maximizing the utilization of computing power on each aircraft, thereby reducing the performance requirements of onboard computers.
[0046] Furthermore, S1 includes the following sub-steps:
[0047] S11. Build a simulator and obtain training samples through the simulator;
[0048] S12. Construct a neural network and train it using training samples to obtain a recognition model;
[0049] S13. The identification model is mounted on the aircraft, and the interception probability of each interceptor missile against the aircraft at the current moment is obtained through the identification model.
[0050] In S11, the simulator is equipped with multiple interceptor missiles and multiple aircraft. Each interceptor missile is paired with an aircraft, and the simulator simulates the process of an interceptor missile intercepting an aircraft, from which the kinematic information of the interceptor missile and the aircraft is extracted.
[0051] Furthermore, for any given aircraft, the kinematic information of each interceptor missile and the aircraft is used as the input of the training sample, and whether the interceptor missile's target is the current aircraft is used as the label of the training sample. Preferably, when the interceptor missile's target is the aircraft, the label of the training sample is 1, otherwise it is 0.
[0052] According to the present invention, the kinematic information of the interceptor missile and the aircraft includes:
[0053] The speed of the aircraft, the speed of the interceptor missile, the relative distance between the interceptor missile and the aircraft, the relative angle between the interceptor missile and the aircraft, and the acceleration of the aircraft.
[0054] In a preferred embodiment, the rate of change of the relative distance between the interceptor missile and the aircraft in the simulator is expressed as:
[0055]
[0056] The relative angular acceleration between the interceptor missile and the aircraft is expressed as:
[0057]
[0058] Where i represents different interceptor missiles, j represents different aircraft, and ρ ij V represents the relative distance between the i-th interceptor missile and the j-th aircraft; a,i V represents the velocity of the i-th interceptor missile. t,j γ represents the velocity of the j-th aircraft; a,i γ represents the heading angle of the i-th interceptor missile. t,j λ represents the heading angle of the j-th aircraft. ij This represents the relative angle between the i-th interceptor missile and the j-th aircraft.
[0059] In a preferred embodiment, in the simulator, the interceptor missile employs a proportional guidance law, and its guidance command is as follows:
[0060]
[0061] Among them, u a,i This represents the normal acceleration command for the i-th interceptor missile, where N is the guidance parameter, n represents the number of aircraft, and T... i This represents the target intercepted by the i-th interceptor missile.
[0062] In a preferred embodiment, Gaussian white noise is added to the relative distance and relative angle between the interceptor missile and the aircraft in the simulator to simulate actual measurement conditions.
[0063] In a preferred embodiment, the simulator performs a pairing of the interceptor missile with the aircraft at the beginning of the simulation, and then no further pairing is performed, meaning that the interceptor missile's target does not change.
[0064] The inventors discovered that the current state of an interceptor missile is affected by changes in the state of its target. The interceptor missile's interception intent is achieved through a series of actions. Therefore, in S12, from the aircraft's perspective, to identify the intercepted target and its reverse trajectory, the identification basis should be the interceptor missile's actions over a period of time.
[0065]
[0066] Among them, O i This is the initial identification result of the i-th interceptor missile against aircraft j; This represents the kinematic information of aircraft j over a period of time. This represents the relative distance and relative angle between the interceptor missile and the aircraft observed by aircraft j over a certain period of time; (t1,t) K () indicates the length of that period of time.
[0067] In S12, preferably, the neural network is a GRU network. The GRU network is a simplified and merged neural network based on the Long Short-Term Memory (LSTM) network. It has been widely used in the field of time series prediction. It can accelerate the training speed while ensuring accuracy and solve the problems of "gradient explosion" or "gradient vanishing" that exist in the original RNN network during model training.
[0068] The GRU network comprises multiple GRU neurons, which include:
[0069] Two inputs: the neuron output from the previous time step and the neuron output x from the previous layer. t It is used to inherit the output of the neuron in the previous moment and to receive the output of the neuron in the previous layer;
[0070] Two gates: the reset gate, used to calculate how much of the previous memory needs to be retained; and the update gate, used to calculate the update content at the current moment.
[0071] One state: Candidate hidden layer state Used to calculate the internal state of a neuron at the current moment;
[0072] One output: the final output h t The output is calculated based on the current input and internal state.
[0073] Tanh is set as the activation function of the input layer to compress the input range to the range [-1, 1], i.e.
[0074] y t =tanh(wx t +b)
[0075] In this invention, the above-described GRU neuron structure is as known to those skilled in the art and is not particularly limited thereto.
[0076] In this invention, the output layer of the neural network uses softmax as the activation function to output the interception probability of the interceptor missile against the aircraft, expressed as:
[0077]
[0078] p1(t|t)+p2(t|t)=1
[0079] Where p1(t|t) represents the probability that the current interceptor missile will intercept the aircraft at the current moment, p2(t|t) represents the probability that the current interceptor missile will not intercept the aircraft at the current moment, o1(t) represents the output value of the first neuron in the output layer of the neural network, o2(t) represents the output value of the second neuron in the output layer of the neural network, and o(t) = [o1(t), o2(t)] T .
[0080] Where p1(t|t) is the preliminary identification result of the neural network output. Based on this, through step S2, the neural network output identification over a period of time is considered, and the final identification result is obtained by weighting according to the time weight.
[0081] Preferably, in a neural network, the input data needs to be converted into dimensionless values with the same range before input, that is, it needs to be standardized and normalized to eliminate the influence of different indicators due to different attributes. This can be done using currently widely used processing methods, such as maximum-minimum linear transformation, Logistic / Softmax transformation, arctangent function transformation, Z-score normalization, and logarithmic transformation. More preferably, maximum-minimum linear transformation is used to scale the data to the [0,1] range, which is represented as:
[0082]
[0083] Where, minA x maxA is the minimum value of the x-th attribute. x This represents the maximum value of the attribute; a xi The raw data before processing, a' xi This is the processed data used for training.
[0084] In a preferred embodiment, the neural network is trained using supervised learning.
[0085] In a preferred embodiment, the neural network is trained based on the Adam algorithm, which combines the Momentum algorithm with the RMSprop algorithm, incorporates momentum, and enables adaptive parameter training. Compared with the BP algorithm, the Adam algorithm can effectively avoid the network getting trapped in local optima, accelerate network convergence, and improve the robustness of the algorithm. The Adam algorithm has been widely used in neural network training. The specific Adam algorithm process will not be described in detail in this invention.
[0086] In this invention, network training only utilizes 1v1 paired kinematic data (i.e., aircraft being intercepted by interceptor missiles), and does not involve swarm adversarial scenarios. Therefore, it can flexibly handle different swarm sizes, and when constructing the training set, only 1v2 adversarial models (aircraft intercepted or not intercepted) need to be built. This is because the established neural network only applies 1v1 (aircraft intercepted) kinematic data, while 1v2 already encompasses both intercepted and uninterrupted scenarios. On the other hand, when constructing the database, it should include as many different scenarios as possible. In swarm adversarial scenarios, as the swarm size increases, the amount of data generated increases dramatically, and the required computational resources and time also increase. Therefore, to ensure sample reliability while maximizing the generalization of the sample database and minimizing computational load and time, only the 1v2 mode is used when constructing the sample database.
[0087] In S13, sensors onboard the aircraft measure and obtain the kinematic information of each interceptor missile and the aircraft in real time. This kinematic information is then input into an identification model, which outputs the interception probability of different interceptor missiles against the aircraft at the current moment, denoted as p. ij The interception probability obtained is used as the initial identification result.
[0088] In S2, preferably, the preliminary identification results from different times are fused by setting time weights.
[0089] More preferably, the time weighting is greater for later times.
[0090] More preferably, the information fusion is represented as:
[0091]
[0092] Where, p ij (t p |[t1,t p ]) represents the final interception probability of the i-th interceptor missile intercepting the j-th aircraft, k p k is the time weighting coefficient. p The larger the value of p, the greater the weight of the subsequent identification results. ij (t) represents the probability that the i-th interceptor missile intercepts the j-th aircraft at time t, which is the preliminary identification result obtained at time t in S1.
[0093] Example
[0094] Example 1
[0095] The model experiment includes the following steps:
[0096] S1. Use a neural network to obtain the interception probability of each interceptor missile against the aircraft at the current moment, as a preliminary identification result;
[0097] S2. Integrate the information from the preliminary identification results over a period of time to identify the interception target of the interceptor missile;
[0098] S1 includes the following sub-steps:
[0099] S11. Build a simulator and obtain training samples through the simulator;
[0100] S12. Construct a neural network and train it using training samples to obtain a recognition model;
[0101] S13. Obtain the interception probability of each interceptor missile against the aircraft at the current moment through the identification model.
[0102] In S11, the simulator is equipped with multiple interceptor missiles and multiple aircraft. The process of interceptor missiles intercepting aircraft is simulated by the simulator, and the kinematic information of interceptor missiles and aircraft is extracted from it.
[0103] In the simulator, the rate of change of the relative distance between the interceptor missile and the aircraft is expressed as:
[0104]
[0105] The relative acceleration between the interceptor missile and the aircraft is expressed as:
[0106]
[0107] Where i represents different interceptor missiles, j represents different aircraft, and ρ ij V represents the relative distance between the i-th interceptor missile and the j-th aircraft; a,i V represents the velocity of the i-th interceptor missile. t,j γ represents the velocity of the j-th aircraft; a,i γ represents the heading angle of the i-th interceptor missile. t,j λ represents the heading angle of the j-th aircraft. ij Let represent the relative velocity between the i-th interceptor missile and the j-th aircraft.
[0108] In the simulator, the interceptor missile employs a proportional guidance law, and its guidance command is as follows:
[0109]
[0110] In the simulator, Gaussian white noise is added to the relative distance and relative angle between the interceptor missile and the aircraft to simulate the actual measurement situation.
[0111] In S12, the neural network is a GRU network.
[0112] The output layer of the neural network uses softmax as the activation function to output the intercept probability of the interceptor missile against the aircraft, expressed as:
[0113]
[0114] p1(t|t)+p2(t|t)=1
[0115] In S13, the kinematic information of each interceptor missile and the aircraft is measured in real time by the sensors on the aircraft. The kinematic information of each interceptor missile and the aircraft is input into the identification model. The identification model outputs the interception probability of different interceptor missiles against the aircraft at the current moment. The obtained interception probability is used as the preliminary identification result.
[0116] In S2, the preliminary identification results from different times are fused by setting time weights. The information fusion is represented as follows:
[0117]
[0118] Wherein, the time weighting coefficient k p The value is 4.
[0119] Furthermore, five interceptor missiles and three aircraft (numbered U1 to U3) were set in the simulator. The initial distance between the interceptor missiles and the aircraft was [12, 15] km, and the guidance law parameter N of the interceptor missiles was [3, 5]. The interception targets of the five interceptor missiles were U1, U2, U3, U1, and U2, respectively. The simulation was performed to obtain 100,000 trajectory simulation data. 80,000 of these data were selected as the training set for neural network training, and the rest were used as the test set.
[0120] The GRU neural network has two hidden layers with 96 neurons each. During training, the dropout rate is 50%, and the learning rate is 0.00125. The training process is as follows: Figure 2 As shown in the figure, it can be seen from the training process that the neural network can converge quickly, has a low loss function, and has high recognition accuracy after training.
[0121] The trajectory data in the test set is used as the measurement value of the aircraft, and then the interception probability of each interceptor missile against the aircraft at the current moment is obtained through the identification model.
[0122] After fusing information from the initial identification results over a period of time, the final identification results of the interceptor missile's interception target are as follows: Figure 3 As shown, from Figure 3It can be seen that a relatively accurate identification result can be obtained after 3-6 seconds, and the identification accuracy can be greatly improved after 10 seconds. When the identification result is compared with the label in the test set, the identification result is completely accurate.
[0123] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," "outer," "front," and "rear," etc., indicate the orientation or positional relationship based on the orientation or positional relationship in the working state of this invention, and are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention. Furthermore, the terms "first," "second," "third," and "fourth" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.
[0124] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.
[0125] The present invention has been described above with reference to preferred embodiments; however, these embodiments are merely exemplary and illustrative. Various substitutions and modifications can be made to the present invention based on these embodiments, all of which fall within the scope of protection of the present invention.
Claims
1. A method for rapid target identification of interceptor missiles based on artificial intelligence, characterized in that, Includes the following steps: S1. Use a neural network to obtain the interception probability of each interceptor missile against the aircraft at the current moment, as a preliminary identification result; S2. Integrate the information from the preliminary identification results over a period of time to identify the interception target of the interceptor missile; S1 includes the following sub-steps: S11. Build a simulator and obtain training samples through the simulator; S12. Construct a neural network and train it using training samples to obtain a recognition model; S13. Mount the identification model onto the aircraft and obtain the interception probability of each interceptor missile against the aircraft at the current moment through the identification model; In S11, the simulator is equipped with multiple interceptor missiles and multiple aircraft. The simulator simulates the process of interceptor missiles intercepting aircraft, and extracts the kinematic information of the interceptor missiles and aircraft. The kinematic information of the interceptor missiles and aircraft includes: The speed of the aircraft, the speed of the interceptor missile, the relative distance between the interceptor missile and the aircraft, the relative angle between the interceptor missile and the aircraft, and the acceleration of the aircraft; In the simulator, the rate of change of the relative distance between the interceptor missile and the aircraft is expressed as: The relative angular acceleration between the interceptor missile and the aircraft is expressed as: Where i represents different interceptor missiles, j represents different aircraft, and ρ ij V represents the relative distance between the i-th interceptor missile and the j-th aircraft; a,i V represents the velocity of the i-th interceptor missile. t,j γ represents the velocity of the j-th aircraft; a,i γ represents the heading angle of the i-th interceptor missile. t,j λ represents the heading angle of the j-th aircraft. ij This represents the relative angle between the i-th interceptor missile and the j-th aircraft. In the simulator, the interceptor missile employs a proportional guidance law, and its guidance command is as follows: Among them, u a,i This represents the normal acceleration command for the i-th interceptor missile, where N is the guidance parameter, n represents the number of aircraft, and T... i Indicates the target intercepted by the i-th interceptor missile; In S12, the neural network is a GRU network; In S13, the kinematic information of each interceptor missile and the aircraft is measured in real time by the sensors on the aircraft. The kinematic information of each interceptor missile and the aircraft is input into the identification model. The identification model outputs the interception probability of different interceptor missiles against the aircraft at the current moment. The obtained interception probability is used as the preliminary identification result. In S2, the preliminary identification results from different times are fused by setting time weights. The information fusion is represented as follows: Where, p ij (t p |[t1,t p ]) represents the final interception probability of the i-th interceptor missile intercepting the j-th aircraft, k p k is the time weighting coefficient. p The larger the value of p, the greater the weight of the subsequent identification results. ij (t) represents the probability that the i-th interceptor missile intercepts the j-th aircraft at time t, which is the preliminary identification result obtained at time t in S1.
2. The method for rapid target identification of interceptor missiles based on artificial intelligence according to claim 1, characterized in that, In the simulator, Gaussian white noise is added to the relative distance and relative angle between the interceptor missile and the aircraft to simulate the actual measurement situation.
3. The method for rapid target identification of interceptor missiles based on artificial intelligence according to claim 1, characterized in that, The output layer of the neural network uses softmax as the activation function to output the intercept probability of the interceptor missile against the aircraft, expressed as: p1(t|t)+p2(t|t)=1 Where p1(t|t) represents the probability that the current interceptor missile will intercept the aircraft at the current moment, p2(t|t) represents the probability that the current interceptor missile will not intercept the aircraft at the current moment, o1(t) represents the output value of the first neuron in the output layer of the neural network, o2(t) represents the output value of the second neuron in the output layer of the neural network, and o(t) = [o1(t), o2(t)] T .